Can WSNs be Emotional? 1

Can wireless sensor networks be emotional? A survey of computational models of and their applications for wireless sensor networks

Tahir Emre Kalayci* Department of Computer Engineering, Celal Bayar University, Manisa, Turkey E-mail: [email protected] *Corresponding author Majid Bahrepour, Nirvana Meratnia and Paul J. M. Havinga Pervasive Systems Group, University of Twente, Enschede, The Netherlands E-mail: m [email protected] E-mail: [email protected] E-mail: [email protected]

Abstract Advances in psychology have revealed that emotions and rationality are inter- linked and emotions are essential for rational behaviour and decision making. Therefore, integration of emotions with intelligent systems has become an important topic in en- gineering. The integration of emotions into intelligent systems requires computational models to generate emotions from external and internal sources. This paper first provides a survey of current computational models of and their applications in engineer- ing. Finally, it assesses potential of integrating emotions in wireless sensor networks by listing some use scenarios and by giving one model application. In this model applica- tion performance of a neural network for event detection has been improved using Brain Emotional Learning Based Intelligent Controller (BELBIC).

Keywords: artificial intelligence; BELBIC; emotions; emotional learning; wireless sensor networks.

Reference to this paper should be made as follows: Kalayci, T.E. and Bahrepour, M. and Meratnia, N. and Havinga, P. J. M, (2017) ‘Can Wireless Sensor Networks be Emotional?A survey of computational models of emotions and their applications for Wireless Sensor Networks’, International Journal of Ad Hoc and Ubiquitous Computing, Vol. 25, No. 3, pp.133–146. DOI: 10.1504/ijahuc.2017.10001736

Biographical notes: T. E. Kalayci is assistant professor in Celal Bayar University Com- puter Engineering department. He finished his PhD in 2011 at Ege University Computer Engineering department on optimisation problems in wireless sensor networks and his MSc in 2006 in Ege University Computer Engineering department.

M. Bahrepour is currently working as Data Scientist at Jibes B.V. in The Netherlands. He holds his PhD from the Pervasive Systems Group at the University of Twente where he conducted data analysis and event detection in wireless sensor networks. During his mas- ter study he applied artificial emotions in financial domain and in his PhD he introduced emotions to wireless sensor networks.

N. Meratnia is associate professor in the Pervasive Systems Group at the University of Twente. Her research interests are in the area of smart sensor systems, Internet of Things (IoT), cyber physical systems, (underwater) wireless sensor networks, and wearable com- puting.

P. J. M. Havinga is full professor and chair of the Pervasive Systems Group at the Uni- versity of Twente. He received his PhD at the University of Twente on the thesis entitled “Mobile Multimedia Systems” in 2000, and was awarded with the “DOW Dissertation Energy Award” for this work. His research focuses on wireless sensor networks, large-scale distributed systems, and energy-efficient wireless communication. Int. J. of Ad Hoc and Ubiquitous Computing, Vol. x, No. x, xxxx 2

1 Introduction (2004), emotion is “a mental state marked by prominent patterns that; are not controllable to any reasonable ex- With the proliferation in Micro-Electro-Mechanical Sys- tent by the virtual multi-verse modelling subsystem, or tems technology which has facilitated the development have the property that their state at each time is far more of smart sensor nodes, wireless sensor networks (WSNs) easily interpretable by integration of past and future in- have gained worldwide attention in recent years. These formation. Such patterns will often, though not always, smart sensing devices are small, with limited process- involve complex and broad physiological changes”. ing and computing resources, and are inexpensive com- There is a widespread agreement between psycholo- pared to traditional sensing devices (Yick et al., 2008). gists in using “emotion” term to represent “a complex High density and large distribution of sensor nodes en- state of diffuse physical changes marked by strong feel- able monitoring and controlling physical environments ings and accompanied by a behavioural impulse towards from remote locations fast and with high accuracy. An- achieving a specific goal” (Statt, 1998). The identifica- other unique feature of sensor networks is the coopera- tion and labelling of particular emotions involve a large tion between wireless sensor nodes (Akyildiz et al., 2002). element of social learning and vary widely across time These properties allow WSNs to have great potential for and space (Statt, 1998). In psychology, emotions are of- many applications in a number of scenarios such as mil- ten viewed as heuristics that give an individual the ten- itary target tracking and surveillance, natural disaster dency to perform particular actions (Steunebrink et al., relief, biomedical health monitoring, hazardous environ- 2009). According to Steunebrink et al. (2009), emotions ment exploration and seismic sensing (Yick et al., 2008). can be used to determine a set of actions that an agent Wireless sensor networks are integrated with intel- tends to perform in a certain situation, which is typically ligent techniques to solve complex problems, to under- a subset of all possible actions it can perform. Moreover, stand environment better and to work more efficiently. action tendency can provide a measure to order this sub- There are many finished and ongoing studies to increase set, thereby indicating the most likely action(s) an agent intelligence of wireless sensor networks for sensing the tends to perform (Steunebrink et al., 2009). environment and reacting to the events accurately and in By using the term emotion we especially mean a real time (Bahrepour et al., 2010a,b; Horst, 2010; Marin- heuristics which effects a decision making process for a Perianu et al., 2010). To this end, many artificial intelli- more rational outcome. Our expectation is that ratio- gence techniques have been adapted. nal outcome will help developed intelligent system to Looking at decision making process in human beings, improve its operation in different aspects. We will inves- one realise that emotions influence the way we perceive, tigate this situation for the wireless sensor networks. In learn, remember, and think. Studies show that emotions this study we use emotional learning term to define a are important part of human decision process (Damasio, learning mechanism in which emotions are produced by 1994; El-Nasr et al., 2000). To make systems more intel- the output and are used as a reinforcement mechanism ligent and more rational, many computer scientists have in the process. tried to adopt emotions in computation. In this paper, we review emotional-based intelligent systems, current 2.2 Emotions and Human Intelligence computational models of emotion, their applications in engineering, and investigate applicability of emotions in Emotions are important aspects of human intelligence WSNs. and have been shown to play a significant role in the human decision-making process (El-Nasr et al., 2000). Studies that investigate importance of emotions for hu- 2 Emotion man decisions produced significant results (Damasio, 1994; Ekman and Davidson, 1994; Frijda, 1987; LeDoux, 2.1 What is Emotion? 1996; Oatley et al., 2006). Researchers in different areas are trying to better understand the role of emotions in As Cabanac (2002) states “there is no consensus in the humans and to incorporate results into computation. Es- literature on a definition of emotion”, and “although an pecially in the field of artificial intelligence, outcomes of enormous literature exists on the psychobiology of affect, these investigations are used to make artificial decision there is no singular or even preferred definition of emo- process better and more effective (Bates, 1994; Bosse tion”. et al., 2010; Martinez-Miranda and Aldea, 2005; Pfis- According to Encyclopaedia Britannica, emotion is ter and B¨ohm,2008; Picard, 1997). Psychologists explo- “a complex experience of consciousness, body sensation, ration of the role of emotions as a positive component in and behaviour that reflects the personal significance of a human cognition and intelligence resulted in a wide vari- thing, an event, or a state of affairs” (Solomon, 2014). ety of evidences which show that emotions have a major According to the APA dictionary of psychology, emotion impact on memory, thinking, and judgement (El-Nasr is a “complex reaction pattern, involving experiential, be- et al., 2000). Neurological studies by Damasio (1994) havioural, and physiological elements, by which the indi- and others have demonstrated that people who lack the vidual attempts to deal with a personally significant mat- capability of emotional response often make poor deci- ter of event” (Vandenbos, 2006). According to Goertzel sions that can seriously limit their functioning in society.

Copyright ⃝c 2015 Inderscience Enterprises Ltd. Can WSNs be Emotional? 3

For example, one patient of the Damasio is a good ex- • Theories of emotion can be used to underpin com- ample of the importance of emotions for correct reason- putational autonomy, to direct and inform percep- ing of the brain (Martinez-Miranda and Aldea, 2005). tion and behaviour selection and to form a better This patient lost the ability to prioritise the tasks and model of computational reasoning (Dias and Paiva, could not feel anything respect to events that are hap- 2005). pening to him. Although his intelligence was not altered • and his brain worked correctly, he was unable to make As research shows, emotion is an important part decisions because of the absence of feelings (Martinez- of human rational thinking. This is a strong mo- Miranda and Aldea, 2005). With the help of Damasio’s tivation to include an emotional component in studies, emotions are regarded as an important part in an intelligent system to make it more human-like the human rational thinking process (Martinez-Miranda (Martinez-Miranda and Aldea, 2005). and Aldea, 2005). Damasio’s work also can be regarded Apart from these motivations, a valid question is: “do as an important evidence indicating that emotions are an we really need intelligent software systems incorporate essential part of human intelligence, and play a crucial an emotional component in their response” (Martinez- role in perception, rational decision-making and learning Miranda and Aldea, 2005)? Martinez-Miranda and Aldea (Davis and Lewis, 2003). Most of the current theories on (2005) answer this question by stating that “it is not emotion agree that emotions constitute a very powerful always required, it depends on the problem that the sys- motivational system that influences perception and cog- tem deals with” and they also claim that if emotions nition in many important ways (Davis and Lewis, 2003). such as anxiety, fear, and stress are incorporated into In addition to Damasio’s hypothesis, Pfister and intelligent systems dealing with complex critical tasks B¨ohm(2008) proposed four-fold classification of emo- (e.g., air traffic control, diagnosis of failures in an elec- tions with respect to their functions in decision making. tricity power plant), the result could be disastrous. On They argue that emotions are not homogeneous con- the contrary, if those emotions are included in systems cerning their role in decision making and they explain that aim to simulate the human behaviour in certain cir- four distinct functionality of the emotions as (Pfister and cumstances (e.g., human computer interfaces, education, B¨ohm,2008): entertainment, etc.), the system will be more friendly to 1. providing information about pleasure and pain for the user and its responses will be more similar to human preference construction; behaviour (Martinez-Miranda and Aldea, 2005). Despite this claim, as seen in example applications, some critical 2. enabling rapid choices under time pressure; tasks are controlled by emotional learning applications. 3. focusing attention on relevant aspects of a decision problem; 3 Related Work 4. generating commitment concerning morally and socially significant decisions. This section provides information about computational models of emotions and their applications. 2.3 Emotions and Artificial Intelligence 3.1 Models There are important reasons for increasing interest in bringing emotions to artificial intelligence. These reasons Emotion has been modelled in various fields and from dif- include: ferent perspectives. Here, we mention some of these mod- els. We focus on the emotional models that are not agent • An obvious application of emotions is to make ar- based architectures, nor models for emotional characters. tificial agents and robots more believable (persua- Interested readers can look into (Bosse et al., 2010; Dias sive, “one that provides the illusion of life” (Bates, and Paiva, 2005; Gratch et al., 2009; Imbert and de Anto- 1994) human users (both in terms of actions they nio, 2005; Kazemifard et al., 2011; Marsella and Gratch, perform and affective expressions they show based 2009) for the agent based architectures or emotional on emotions) (Steunebrink et al., 2009). character models like EMA (Marsella and Gratch, 2009), • From a more theoretical perspective, it is inves- COGNITIVA (Imbert and de Antonio, 2005), CoMERG tigated what the role of emotions is in models of (Bosse et al., 2010), I-PEFICADM (Bosse et al., 2010), human decision-making and how they may be em- FearNot! (Dias and Paiva, 2005), GeMA (Kazemifard ployed to make them more accurate and effective et al., 2011). (Steunebrink et al., 2009). 3.1.1 Picard’s Computational Model • There exists psychological (Ekman and Davidson, 1994; Frijda, 1987; LeDoux, 1996; Oatley et al., Picard (1997) suggested the term affective computing for 2006) and neurological (Damasio, 1994) evidences an approach in the field of artificial intelligence to build that emotions are not only relevant but even nec- computers that recognise and show human affections. essary for rational behaviour. Preliminary results in affective computing can be found 4 T. E. Kalayci et al. in facial expression, and voice recognition and synthesis Figure 1 Schematic of the framework for Emotion-Based (Martinez-Miranda and Aldea, 2005). Control proposed by Velasquez (1998) In her investigations on affective computing, Picard (1997) considered Damasio (1994) research on the im- portance of emotions in human decision-making process. She claimed that the most suitable computing tool for decision-making is the one that includes emotional mech- anisms as well as a knowledge base. To simulate and recognise the human affections, Picard proposed to use Hidden Markov Models (HMM) and rule-based models (Martinez-Miranda and Aldea, 2005). Picard et al. (2001) proposed that machine intelli- gence needs to include emotional intelligence and demon- strated results toward this goal by developing a ma- chine’s ability to recognise human affective state given four physiological signals. They described difficulties in obtaining reliable affective data. After collecting a large data set from a subject trying to elicit and experience each of the eight emotional states daily over multiple self interested emotional systems representing different weeks, they presented and compared multiple families of related affective states. The Behaviour Sys- for feature-based recognition of emotional state. They tem is responsible for deciding next action when an agent concluded that the features of different emotions on the faces a situation. The Behaviour System is a distributed same day tend to cluster more closely than the features of network of self interested behaviour, such as “approach the same emotion on different days. They found out that human”, “play”, “request attention”, and “avoid obsta- the technique of seeding a Fisher Projection with the re- cle”. The Emotional Systems assess the emotional rel- sults of Sequential Floating Forward Search improves the evance of stimuli and bias behavioural responses and performance of the Fisher Projection and provided the future perception accordingly. Relevant Behaviour Sys- highest recognition rates reported to date for the classi- tems generate and execute appropriate Motor Actions fication of affect from physiology: 81 percent recognition (Velasquez, 1998). accuracy on eight classes of emotion including no emo- Main characteristics of this model are (Velasquez, tion state. 1998): 3.1.2 Velasquez’s Computational Model • Support for non-linear behaviour by two different thresholds. The first one, α, is used to determine Velasquez (1998) describes a computational framework when an emotion occurs. The second threshold, ω, for Emotion-Based Control, which is inspired by work specifies the level of saturation for that emotion. In in psychology, neuroscience, and ethology. This frame- addition to these parameters, each Emotional Sys- work captures important aspects of emotional systems tem has a function, Ψ(), which controls the tem- and integrates these with other models of perception, poral decay of its intensity. motivation, behaviour, and motor control. Modular na- ture of the framework allows an incremental approach in • Distinction between emotions, as the system con- which new features are added as soon as more is known tains explicit models for six different primary about emotional process in the brain. Schematic of the emotions, i.e., anger, fear, distress/sadness, en- proposed framework can be seen in Figure 1. joyment/happiness, disgust, and surprise. Primary The Sensory Systems obtain information from the emotions are modelled as the activation of one par- world and provide the Emotion Generation and Be- ticular Emotional System such as sadness or dis- haviour Systems with stimuli. These systems also receive gust. Emotion blends, such as jealousy, emerge as error signals from the Drive Systems. Drives are motiva- the co-activation of two or more of these Emotional tional systems representing urges that impel the agent Systems. into action. For instance, a hunger drive will aid in con- • Containing a distributed network of self interested trolling behaviours that directly affect the level of food emotional systems representing different families intake by the agent. The Emotion Generation System of related affective states, such as fear and panic. bears resemblance to some of the aspects in which the Each member of an emotion family shares certain interactions between neural systems involving the amyg- mechanisms and characteristics, including similar- dala, the hippo-campus, and the pre-frontal cortices have ities in antecedent events, expressions, likely be- been assumed to mediate emotions, such as assigning an havioural responses and physiological patterns. emotional valence to different stimuli, activation of emo- tional behaviours, and emotional learning. The Emotion • Temperaments are modelled through the different Generation System consists of a distributed network of values that parameters (e.g., thresholds, gains, and Can WSNs be Emotional? 5

Figure 2 FLAME’s abstract agent architecture and • Use of fuzzy logic as representation mechanism to emotional process component (El-Nasr et al., map events and observations to emotional states. 2000) • Incorporating machine learning methods, such as associations among objects, sequences of events, and expectations about the user to learn a variety of things about the environment. This allows an agent to adapt its responses dynamically, which will in turn increase its believability.

• Use of a number of planning or rational decision- making algorithms.

• Before linking the model to a different application, aside from specifying the new goals of the agent, some parameters used in the model might need to be adjusted.

• FLAME must be extended to incorporate some of the social agent concepts as the model lacks per- sonality concept and ability to simulate a group of agents that interact with each other.

3.1.4 Moren and Balkenius’s Computational Model

Moren and Balkenius (2000) present a neurologically in- spired computational model of the amygdala and the or- bitofrontal cortex that aims to partially reproduce the same characteristics as the biological system. This model is divided into two main parts: one corresponding to the decay rates) within each Emotional System can amygdala and one corresponding to the orbitofrontal cor- have. The result is a flexible, distributed model tex. The internal structures of these two parts are not that can synthesise a variety of affective phenom- exactly modelled as their biological counterparts, but ena simultaneously. share their organisation. The combined system works at • Implementation as part of an object-oriented two levels: the amygdala learns to predict and react to framework for building autonomous agents. a given reinforcer. This subsystem can never unlearn a connection; once learned, it is permanent, giving the sys- 3.1.3 FLAME tem the ability to retain emotional connections for as long as necessary. The orbitofrontal system tracks mis- El-Nasr et al. (2000) propose a computational model of matches between the base system predictions and the emotions, illustrated in Figure 2, that can be incorpo- actual received reinforcer and learns to inhibit the sys- rated into intelligent agents and other complex interac- tem output in proportion to the mismatch (Moren and tive programs. The model, called Fuzzy Logic Adaptive Balkenius, 2000). Model of Emotions (FLAME), is based on an event- The authors consider three classical states in their appraisal physiological model. It also includes several experiments, i.e., acquisition-extinction-reacquisition, inductive learning algorithms for learning patterns of blocking and conditioned inhibition. Acquisition and ex- events, associations among objects, and expectations. tinction is a basic learning experiment, in which the The proposed model consists of three major components, model is expected to associate a stimulus with a re- i.e., an emotional component, a learning component, and ward/reinforcer, disassociate the stimulus once the rein- a decision-making component. The decision in this model forcer is absent and then re-associate them again (Moren is made according to the situation, the agent’s mood, and Balkenius, 2000). Experimental results indicate that the emotional states and the emotional behaviour and an the model at least has the basic features needed for as- action is then triggered accordingly. FLAME also uses sociative learning. past experience to set the value of emotions. It always By having a close look at the model in Figure 3, one saves the agent response to specific events to use it again can see that there is one A node for every stimulus S in similar situations that occur in the future (Martinez- (including one for the thalamic stimulus). There is also Miranda and Aldea, 2005). one O node for each of the stimuli (except for the tha- Most important characteristics of the model are (El- lamic node). There is one output node in common for Nasr et al., 2000): all outputs of the model, called E. The E node simply 6 T. E. Kalayci et al.

Figure 3 Moren and Balkenius (2000) computational Figure 4 BELBIC structural view in SIMULINK (Jamali model et al., 2009)

learning” for the emotional control process of the BEL- BIC model. In emotional learning, emotions are pro- duced by the performance of the output and are used as a reinforcement mechanism to the learning process. The sums the outputs from the A nodes, then subtracts the emotional learning has the following remark- inhibitory outputs from the O nodes. The result is the able properties: output from the model. Main characteristics of this model as described by • One can use very complicated definitions for emo- Moren and Balkenius (2000) are: tional signal without increasing the computational complexity of algorithm or worrying about differ- • It can handle most common associative learning entiability or ability to render into recursive for- experiments and it is easy to implement and be mulation problems (Babaie et al., 2006). used as part of a larger system. • The parameters can be adjusted in a simple intu- • The model is not a complete learning system. Since itive way to obtain the best performance (Babaie it is an emotional evaluator of stimuli, it needs sev- et al., 2006). eral components to handle “real” learning tasks. • The training is very fast and efficient (Babaie et al., • The two most important missing parts are a con- 2006). text model and some form of motor learning system that can use the output of this model. • Learning rules are very simple and therefore the computational speed is high (Khorramabadi et al., 3.1.5 BELBIC 2008). • Lucas et al. (2005) have developed a computational As a free controller with learning ability BELBIC model based on the in the mammalian may cause low performance in the early stages brain. The model, called BELBIC, uses the network of learning process, and this preliminary phase model developed by Moren and Balkenius (2000) and of learning can result in instability in some cases shares same structure shown in Figure 3. The non-linear (Roshtkhari et al., 2008). controller BELBIC’s structural view in SIMULINK is il- • Emotional learning has been improved by model lustrated in Figure 4. As seen in the figure, sensory input driven development approach as a bio-inspired al- signals are received via thalamus. Processed input signal gorithm for embedded purposes (Jamali et al., will be sent to amygdala and sensory cortex after pre- 2009) processing in sensory input. amygdala and outputs are computed based on emotional signal received from environment. Final output is calculated by 3.2 Applications of Emotions in Engineering subtracting amygdala’s output from orbitofrontal cor- tex’s output (Jamali et al., 2009). General applications of emotion in intelligent systems BELBIC has been employed as a controller can be categorised in applications based on affective in control design problems (Mehrabian et al., 2008) and agents and applications based on emotional learning (see during the past few years it was utilized for several in- Figure 5). Affective agents contains emotional software dustrial applications and control purposes (Jamali et al., agents including emotional multi-agent systems, emo- 2009). Lucas and his colleagues use the term “emotional tional robots, and human-like agents. Emotional learning Can WSNs be Emotional? 7

Figure 5 Classification of applications of emotions in gies (specially intelligent control methods) are sometimes intelligent systems problematic (Lee et al., 1995). Daglarli et al. (2009) present an artificial emotional cognitive system for autonomous mobile robots. Accord- ing to emotional and behavioural state transition prob- abilities, artificial emotions determine sequences of be- haviours for long-term action planning of robots. Dehkordi et al. (2011b) use BELBIC to control the switched reluctance motor (SRM) speed. Motor pa- rameter changes, operating point changes, measurement noise, open circuit fault in one phase and asymmetric phases in SRM are also simulated to show the robust- ness and superior performance of BELBIC. They com- pare BELBIC performance with a Fuzzy Logic Con- troller (FLC) they developed. Dehkordi et al. (2011a) also present a detailed com- parison of various intelligent based controllers for flux weakening speed control of an Interior Permanent Mag- is an emotion-based decision-making, which covers emo- net Synchronous Motor drive. In this paper, BELBIC, tion supported prediction, emotion supported learning, Genetic-Fuzzy Logic Based Controller (GFLBC) and and emotion supported control. genetic-PI based controller are considered and compared. There are many application examples which use emo- They verify the effectiveness of the proposed BELBIC tions. These applications mainly use models discussed in controller-based IPMSM drive by simulation results at Section 3.1 and can be in general grouped into control different operating conditions. and prediction applications. As it can be seen from Ta- Dorrah et al. (2011) propose a BELBIC to replace ble 1, majority of control applications are based on BEL- conventional proportional-integral-derivative (PID) con- BIC model and utilise emotional learning techniques. trollers, which have a very difficult tuning when the pro- cess is subject to external unknown factors. They also 3.2.1 Control Applications optimize the the values of BELBIC and PID gains us- ing a particle swarm optimisation (PSO) technique with Studies such as (Velasquez, 1998) have successfully used minimization of Integral Square Error (ISE) for all loops. the Velasquez’s computational model to develop and According to the performance comparison with con- control several different autonomous agents, including ventional PSO-PID controllers, proposed PSO-BELBIC both synthetic agents and physical robots. prove their usefulness in improving time domain be- Lucas et al. (2005) have adapted BELBIC, a compu- haviour with keeping robustness for all loops. tational model based on the limbic system in the mam- Farhangi et al. (2012) propose an approach based on malian brain, for control engineering applications. Their the emotional learning for improving the loadfrequency results demonstrate excellent control action, disturbance control (LFC) system of a two-area interconnected power handling, and system parameter robustness. BELBIC system with the consideration of generation rate con- has been tested for many different applications (Dehko- straint (GRC). The controller includes a neuro-fuzzy sys- rdi et al., 2011b,a; Dorrah et al., 2011; Farhangi et al., tem with power error and its derivative as inputs. A 2012; Jafari et al., 2013b,a; Jafarzadeh et al., 2012; Ja- fuzzy critic evaluates the present situation, and provides mali et al., 2008; Khadem et al., 2014; Khalghani and the emotional signal (stress). The controller modifies its Khooban, 2014; Khalilian et al., 2012; Khorramabadi characteristics so that the critics stress is reduced. The et al., 2008; Mehrabian et al., 2008; Parsapoor et al., convergence and performance of the proposed controller, 2014; Sadeghieh et al., 2012; Sharbafi et al., 2006; Sheik- both in presence and absence of GRC, are compared with holeslami et al., 2006; Valikhani and Sourkounis, 2014). those of proportional integral (PI), fuzzy logic (FL), and Fatourechi et al. (2003) present an approach for the hybrid neuro-fuzzy (HNF) controllers. control of dynamical systems based on the agent concept. Jafarzadeh et al. (2012) investigate the application The presented control system consists of a set of neuro- of BELBIC to induction motor (IM) drives. They de- fuzzy controllers whose weights are adapted according signed A BELBIC-based direct torque controlled (DTC) to emotional signals provided by blocks called emotional IM drive in which the torque controller and the stator critics. Simulation results are provided for the control of flux magnitude controller are implemented as BELBICs dynamical systems with various complexities in order to in stator flux reference frame. The controllers include an show the effectiveness of the proposed method. The main exponential reward function which determines the con- contribution of the proposed generalisation is to provide troller dynamics. They present the controller design and an easy to implement emotional learning technique for investigate the drive performance and extensive experi- dealing with dynamic (especially multivariable) control mental results which show that the BELBIC controller systems, in which the use of other control methodolo- performs well. 8 T. E. Kalayci et al. al 1 Table Other Prediction Control nlss S:Pril wr piiain te:onmdladtcnqeo oifrainaalbeaotmdlo technique) or model about available information no or technique and model own Other: Optimisation, Swarm Particle PSO: Analysis, vriwo oeapiain sn mtos(A ffcieAet,E:EoinlLann,N:Nr-uz,N:Nua ewr,PA rnil Component Principle PCA: Network, Neural NN: Nero-fuzzy, NF: Learning, Emotional EL: Agents, Affective (AA: emotions using applications some of Overview eeepeso irarycasfiain(ofiadKsaaz 2014) Keshavarz, and (Lotfi 2014b) classification Akbarzadeh-T., microarray and expression 2013) (Lotfi Gene (Lotfi, networks 2012) classification neural al., image emotional et for Practical 2011) (Yang learning al., emotion emotional et artificial inspired (Kalayci with Brain BELBIC algorithm using optimisation WSN chaos in Hybrid detection event NN 2008) Improving 2000) (Petruseva, al., problem et path (El-Nasr Shortest pet a of Simulation 2012) 2009) al., (Khashman, et recognition 2014) Facial (Abdi al., traffic-flow et short-term 2006) (Parsapoor of al., Forecasting storms et geomagnetic (Babaie 2014a) of conditions Akbarzadeh-T., prediction Solar and the (Lotfi for indices Model activity Prediction geomagnetic 2014) of al., prediction et (Yong Online guidance 2004) reentry al., predictor–corrector 2003) et Adaptive al., (Gholipour 2003) et phenomena (Fatourechi al., weather concept et Space agent (Gholipour the index on activity based Geomagnetic systems 2009) dynamical 2014) al., of Khooban, et Control and (Daglarli (Khalghani control compensator robot DVR Autonomous the 1998) for 2006) (Velasquez, controller al., agents Self-tuning et autonomous (Sharbafi for robots 2012) systems wheeled al., Control three et 2011) (Farhangi of al., system control et power Motion (Dorrah interconnected 2014) process of al., column control et distillation Loadfrequency (Khadem two-coupled instrument for 2014) laparoscopic scheme Sourkounis, robotic PSO-BELBIC and a (Valikhani of system control 2013a) turbine Force al., wind 2013b) et DFIG-based al., (Jafari for et Quadrotor Controller (Jafari a system of 2012) servo control al., digital Attitude et a 2012) (Khalilian of al., motor 2012) control et stepper al., Speed (Sadeghieh hybrid et actuator of (Jafarzadeh rotary control motor servo-hydraulic Position induction a 2011a) for of al., design control et controller Position (Dehkordi flux drives and IPMSM torque 2008) of Direct al., control 2011b) et speed al., (Khorramabadi for et controller Controller (Dehkordi core motor reactor 2008) reluctance nuclear al., switched PWR et of (Jamali control crane speed of Sensorless 2008) control al., positioning et and (Mehrabian Anti-swing autopilot vehicle launch 2006) Aerospace al., et (Sheikholeslami systems HVAC Applications te GdnoadHla,2001) Hallam, and (Gadanho Other oe n Balkenius and Moren Velasquez BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC BELBIC MODEL FLAME Other Other Other Other TECHNIQUE L PCA EL, L PSO EL, A NN AA, A NF AA, L NN EL, L NF EL, NF EL, EL NF, Other NN NN AA AA AA AA EL EL EL EL EL EL EL EL EL EL EL EL EL EL EL EL EL EL EL EL Can WSNs be Emotional? 9

Khalilian et al. (2012) present an implementation of This index mainly used in warning and alert systems. BELBIC for position control of hybrid stepper motor The proposed model with brain emotional learning al- drive. They verify the the effectiveness of the proposed gorithm is introduced to make purposeful prediction of BELBIC controller-based hybrid stepper motor drive by Kp index. They claim both the prediction accuracy dur- simulation results. ing geomagnetic disturbances and sub-storms, and the Sadeghieh et al. (2012) utilize a modified version rate of associated correct warning messages show the effi- of BELBIC for position controlling a real laboratorial ciency of this algorithm. Gholipour et al. (2004) also pro- rotary electro-hydraulic servo (EHS) system which are pose a approach towards purposeful prediction problems, known to be non-linear and non-smooth due to many derived from a recently developed model of emotional factors such as leakage, friction, hysteresis, null shift, learning in human brain. The proposed algorithm inher- saturation, dead zone, and especially fluid flow expres- ently emphasizes learning to predict future peaks, and sion through the servo valve. They have compared the produces remarkably accurate predictions among the im- results those obtained from an optimal PID controller, portant regions, features, or objectives. They claim that an auto-tuned fuzzy PI controller (ATFPIC), and a neu- space weather forecasting is an excellent example of us- ral network predictive controller (NNPC) under similar ing their methodology, and their motivation was to intro- circumstances to prove the effectiveness of the modified duce purposeful prediction via multi-objective learning BELBICs online learning ability in reducing the over- algorithm in their research. Three examples of predict- all tracking error. They argue that results demonstrate ing solar activity, geomagnetic activity, and geomagnetic excellent improvement in control action and less energy storms show the characteristics of the suggested algo- consumption. rithm and its usefulness to space weather warning and Jafari et al. (2013b) use BELBIC for speed control alert systems. of a Digital Servo System. Their proposed controller Babaie et al. (2006) propose a physiologically moti- is applied experimentally to a laboratory Digital Servo vated prediction methodology for trading off general pre- System via MATLAB external mode. Their compari- cision with improved predictive performance in regions son of the proposed controllers’ results with conventional of high interest. Results are compared to very accurate PID controller shows satisfactory performance including forecasts of ground based geomagnetic activity index (K) faster response and lower overshoot. and sunspot number obtained by a neuro-fuzzy tech- Jafari et al. (2013a) propose the usage of BELBIC for nique. In this study it has been shown that powerful alert the attitude control of a Quadrotor. They present the system can be designed using computationally intelligent simulation result of controlling the Quadrotor with pro- forecasting techniques. posed controller and they compare the results to a well Abdi et al. (2012) proposed a new method based on tuned PID. Their results show satisfactory performance. neuro-fuzzy and limbic system structure such as Locally Valikhani and Sourkounis (2014) analyse the dynamic Linear Neuro-Fuzzy network (LLNF) and Brain Emo- behaviour of a pulse width modulation (PWM) based tional Learning Based Intelligent Controller (BELBIC) vector controlled doubly fed induction generator (DFIG) models. They use that method for forecasting short-term and improve this behaviour using BELBIC. of traffic-flow. Khadem et al. (2014) propose a new approach based Lotfi and Akbarzadeh-T. (2014a) propose an adaptive on emotional learning to control the force interactions of brain-inspired emotional decayed learning to predict in- a robotic surgical instrument with delicate soft tissues. dices that characterize the chaotic activity of the earth’s The proposed controller is devised using a neuro-fuzzy magnetosphere. Their learning algorithm is a neural ba- regulator that receives the tracking error and its deriva- sis computational model of AmygdalaOFC in a super- tive as inputs, and a proportional-derivative (PD) critic vised manner and a decay rate is considered in Amygdala that evaluates the actual pinch force and produces an learning rule. They made various comparisons between emotional signal. their proposed method, Multilayer Perceptron (MLP) Khalghani and Khooban (2014) propose an emotional (Balasubramanian, 2010), Adaptive Neuro-Fuzzy Infer- controller which is based on emotional learning of hu- ence System (ANFIS) (Jang, 1993) and Locally Linear man brain for controlling the dynamic voltage restorer Neuro-Fuzzy (LLNF) (Mehrabian et al., 2006) in their (DVR) compensator. The proposed emotional controller experiments. is based on Moren and Balkenius (2000) computational Parsapoor et al. (2014) propose a suitable model for model. They report that the performance of the emo- predicting geomagnetic storms based on the brain emo- tional controller depends on the selection of the values tional learning. They employ the proposed model to pre- of its coefficients. They tune these coefficients by an op- dict geomagnetic storms using the Disturbance Storm timisation algorithm in order to improve the proposed Time (Dst) index. They evaluate the performance of the controller. proposed predictor by comparing the results with the re- sults of the adaptive neuro-fuzzy inference system (AN- 3.2.2 Prediction Applications FIS). Yong et al. (2014) propose an adaptive predictorcor- Gholipour et al. (2003) propose usage of BELBIC to rector re-entry guidance algorithm with self-defined way- predict Kp, most popular index of geomagnetic activity. points. They implement their predictor-corrector algo- 10 T. E. Kalayci et al. rithm based on BELBIC instead of the general numerical applying the activation function hardlim in the model. predictorcorrector algorithm that is based on Newton Various comparisons are made between BELPIC and iteration method, which is very sensitive to the initial multilayer perceptron (MLP) to classify the images of a value and is difficult to implement in real time on the typical dataset. According to the experimental results, computer of re-entry vehicles at present. BELPIC shows higher accuracy and lower time complex- ity in image classification. 3.2.3 Other Applications Lotfi and Akbarzadeh-T. (2014b) propose a limbic- based artificial emotional neural network (LiAENN) for El-Nasr et al. (2000) implemented a simulation of a pet a pattern recognition problem. LiAENN is a novel com- named PETEEI - a PET with Evolving Emotional In- putational neural model of the emotional brain that telligence - to demonstrate and evaluate the capabilities models emotional situations such as anxiety and confi- of FLAME. They performed an ablation experiment in dence in the learning process, the short paths, the forget- which they asked users to perform tasks with several ting processes, and inhibitory mechanisms of the emo- variations of the simulation and then they reviewed their tional brain. In the model, the learning weights are ad- assessments of various aspects of PETEEI’s behaviour. justed by the proposed anxious confident decayed brain They found that the adaptive component of the model emotional learning rules (ACDBEL). was critical to the believability of the agent in the simu- Lotfi and Keshavarz (2014) propose a novel hybrid lation. method based on Principal Component Analysis (PCA) Khashman (2009) presents a modified back- and Brain Emotional Learning (BEL) network for the propagation (BP) learning algorithm, namely, the classification tasks of gene-expression micro-array data. emotional back- propagation (EmBP). The presented In the experimental studies, the proposed model is uti- algorithm has additional emotional weights that are lized for the classification problems of the small round updated using two additional emotional parameters: blue cell tumours (SRBCTs), high grade gliomas (HGG), anxiety and confidence. The proposed emotional neu- lung, colon and breast cancer datasets. ral network was applied to a facial recognition problem and the results were compared to a similar application using a conventional neural network. Experimental re- 4 Emotions in Wireless Sensor Networks sults show that the addition of the two novel emotional parameters improves the performance of the neural In this section we investigate applicability of emotions network yielding higher recognition rates and faster for WSNs and present some possible articulations to cope recognition time. In another work, Kashman presents with dynamicity and optimised performance of WSNs. an emotional neural network which is applied to a facial recognition problem using images of faces with differ- 4.1 Applications of Emotions in WSN ent orientations and contrast levels, and compares its To the best of our knowledge, there are only a few ex- performance to that of a conventional neural network amples of use of emotions in wireless sensor networks. (Khashman, 2009). Experimental results suggest that artificial emotions can be successfully modelled and ef- 4.1.1 Emotional ants for intrusion detection ficaciously applied to improve neural networks learning and generalisation. Banerjee et al. (2005) proposed an approach for the Kalayci et al. (2011) investigated applicability of intrusion detection based on emotional ants. The ba- Brain Emotional Based Intelligent Controller (BELBIC) sic idea is to identify the affected path of intrusion in to improve neural network performance for the event de- the sensor network by investigating the pheromone con- tection. Their empirical results show that incorporating centration. The work also emphasises the emotional as- the BELBIC with neural networks improves the accu- pects of agents, in which they can communicate the racy of event detection in many circumstances. characteristics of particular path among them through Yang et al. (2012) present a new unconstrained global pheromone update. To do so, emotional ants are placed optimisation method, hybrid chaos optimisation algo- in a sensor network. They can keep track of the changes rithm with artificial emotion (HCOAAE). Their pro- in the network path, following certain rules depicting posed algorithm avoids trapping to local minima, and the probable possibilities of attack. Once the particu- improves convergence in large space and high-dimension lar path among nodes is detected by the spy emotional optimisation problems. Their main purpose of artificial ant, it can communicate the characteristics of the path emotion is to mimic decision making behaviour pro- through pheromone balancing to the other ants; and cess of humans, to choose most suitable parameters of thereafter the network administrator could be alerted. HCOAAE and decide whether to change current search Authors define the emotional ants as adaptive ants with strategy or not in the next iteration. variable conflict tendencies that can adjust their schema Lotfi (2013) proposes fast image classifier based with phenomena communication. Work of Banerjee et al. on emotional learning. Proposed model is named as (2005) uses emotions as reinforcement system to detect Brain Emotional Learning (BEL) based Picture Clas- intruder since a unique attribute of emotions is transfer- sifier (BELPIC). The distinctive feature of BELPIC is ability between ants. Emotions are used to help decision Can WSNs be Emotional? 11 making in case there is any intruders in the sensor net- 2. Efficiency: in terms of determining when and who work. This application can be considered as a modifica- should perform a specific task based on state of tion of standard ant colony algorithm to be applied in individual nodes and global state of the network. WSNs.

4.1.2 Intelligent monitoring system based on 3. Reliability: in terms of dealing better with data uncertainty and outliers as well communication in- emotional model terferences. Kubota et al. (2009) introduced an emotional model composed of emotion, feeling, and mood components as 4. Robustness: in terms of coping better with physical well as a monitoring system performing human-like mon- and network failure or malicious attacks. itoring. The monitoring system uses image processing to detect and track humans. Emotional model is used to change emotional state of the system based on human de- 4.3 Application: Improving neural network tection and tracking. The emotional state change results in behavioural change of the system. This system collects performance using emotional learning visual, audio and emotional information from environ- ment and converts this information to emotional input In our previous work (Kalayci et al., 2011), we have al- value according to the predefined rules. Each emotion ready investigated the applicability of BELBIC to im- occurs dependent on specific perceptual information and prove performance of the neural network for event detec- each feeling is updated as the integrations of emotion. tion. Inherent errors in sensor data causes training phase Feeling is the short-time state updated by the change of of a neural network to always suffer from some degrees emotions. Yet the mood is the long-term state updated of error, so as a result the event detection and classi- by the change of feelings and governs the change of feel- fication accuracy could be lower (Kalayci et al., 2011). ings. Kubota et al. (2009) work is not actually an emo- So, to compensate this error BELBIC could be used as tional supported decision system. It only collects infor- a feedback mechanism. mation to change emotional state of monitoring system Before conducting experiments, some steps had to and the emotional state helps system act emotionally, be performed to be able to find the correct configura- which only changes the behaviour of the system. This tion for the BELBIC. Many experiments were performed study is a good initiation of emotional WSNs in the way with different parameters (sensory input, stress signal, that events bring the network into emotional states and and learning rate constant values) and different BELBIC operation is varied in various emotional states. The work and neural network combinations. According to these can be considered as affective agents according to the configuration experiments, error has been decided to be definitions in 2.3. used as sensory input and error × error as stress signal. As learning rate constant values define the performance 4.2 Potential Use Scenarios speed of BELBIC, it is also an important value to con- sider. As examples in previous section show, incorporating After finding the correct configuration, BELBIC emotions and emotional learning in wireless sensor net- should be integrated with neural network. In this appli- works have a promising future. As we explained in previ- cation BELBIC has been integrated with neural network ous sections, early studies show that human rationality is in three different ways (before the input layer, after the obtained by the help of emotions (Damasio, 1994). Based output layer and both) and its output is summed accord- on these studies we can say that to have a fully rational ing to these positions. According to these integrations, and intelligent WSN, we may need emotions. However different experiments have been performed using three the question is what types of emotions are applicable to different data sets (forest fire, residential fire and activity WSNs? data sets). Experimental results suggest that improve- An example emotion could be anxiety generated by ment strongly relies on dataset properties. The most im- a low battery situation, which makes WSN behave more portant drawback of using BELBIC with ANN is the erroneous. Other examples of emotions and effect of us- necessity of finding accurate BELBIC parameters and ing these emotions in WSN can be found in Table 2. this depends on empirical research and these parameters The main reason of introducing these emotions is to en- generally are problem/application specific and dataset hance performance of WSN’s decision process through dependent (Kalayci et al., 2011). adaptation of its behaviour upon experiencing specific sit- This application is a good example of using emotional uations. learning system as a feedback and reinforcement mecha- Generally speaking, use of proper emotions in WSNs nism. As BELBIC and neural networks are both compu- could offer the following advantages: tationally light, they are good candidates for implement- 1. Adaptability: correct implementation of emotions ing in wireless sensor node platforms which enable situ- help better adapt to changing network character- ational awareness at the point of action (Kalayci et al., istics and dynamic environments. 2011). 12 T. E. Kalayci et al.

Table 2 Examples of using emotions in WSN Emotion Effect of using emotion Interest Having more interest in some objects can improve accuracy of localisation. Being applicable to asset monitoring and track and trace applications in particular, more accurate localisation can also be awarded, which in turn can increase interest. Boredom Introducing this emotion in time critical event detection applications of WSN can improve scheduling and duty cycle. Not detecting any events or changes for a long time increases the boredom of the node and forces the node to sleep. Anxiety Hardware or link failure can cause nodes experience anxiety. Less importance should be given to decision or data coming from nodes experiencing anxiety. This can improve network reliability and robustness. Curiosity This emotion can enhance service discovery as well as cooperation and collaboration between nodes. Happiness Having high energy, good reputation or being rewarded for good performance can cause node enter happy emotion. This can enhance reliability and decision making capability of the network. Impatience Experiencing impatience can lead nodes to better optimise their routes (e.g. shortest or fastest path) or perform tasks with less latency. Weakness Not capable of performing tasks optimally, being more on sleep mode

5 Challenges After trying to understand emotions and their con- tribution to intelligent systems as well as investigation Technical and open challenges towards application of of important computational models and applications of emotions in WSNs include: emotion in WSNs, we can conclude that: • Complexity of existing emotional based models is • Emotional learning concept is very similar to re- an important challenge towards wide application inforcement learning. We can argue that emotions of these models in wireless sensor networks. act just like another reinforcement and refer to emotional learning as internal reinforced learning. • Uncertainty and error associated with sensor data Despite of similarity, however, there is a differ- pose another challenge towards accurate applica- ence between emotional learning and reinforcement tion of emotional models for wireless sensor net- learning, that is the evaluation of R (reinforcement works. signal in reinforcement learning that is evaluated to • Real-timeness is another requirement of wireless failure/success, stress signal in emotional learning sensor network applications being challenged by that is continuous values) (Khorramabadi et al., application of emotional based computation mod- 2008). els. • Emotions are different heuristic parameters that help solve problems easily. They help to limit so- 6 Conclusions lution space that is searchable in more reasonable time. But of course, emotions are not as simple as Emotions are essential for the human beings to make heuristic parameters. They, with the help of other rational decisions. Significant results of studies about components, possess properties of a more complete emotions and human intelligence have encouraged com- learning and decision support tool. puter scientists to incorporate emotions into intelligent • Systems that use emotions can become more or less systems. This motivation brought some important com- aware of their environment and context. Also use putational models and applications to the field of arti- of emotions can help to make system more concen- ficial intelligence. One should note that different appli- trated to the task at hand and ignore the useless cations may require different emotion models based on distractions from the environment. Therefore, use their requirements and type of the problems they are try- of emotion can increase performance. ing to solve. But in general, applications of emotion can be categorised into two main types, i.e., (i) applications Research on use of emotional learning in wireless sen- based on affective agents, especially human-like agents sor networks has just began and many open questions or robots, and (ii) applications based on emotional learn- and challenges are ahead of research community. ing. The direction of the former category seems to be in integrating emotions to agents, multi-agent systems, and human-like agents. Emotions in the latter type of 7 Acknowledgements applications are used as an internal biasing mechanism to make control (and decision) more successful and more Tahir Emre Kalayci is supported by the DTP OYP efficient. Project 05-DPT-003/05 and TUBITAK 2211 Yurt Ici Can WSNs be Emotional? 13

Doktora scholarship. This paper also describes work in E. Daglarli, H. Temeltas, and M. Yesiloglu. Behav- part undertaken in the context of the SENSEI project, ioral task processing for cognitive robots using artifi- “Integrating the Physical with the Digital World of the cial emotions. Neurocomputing, 72(13-15):2835–2844, Network of the Future” (www.senseiproject.eu). SENSEI 2009. is a Large Scale Collaborative Project supported by the European 7th Framework Programme, contract number: A. R. Damasio. Descartes’ Error: Emotion, Reason, and 215923. the Human Brain. G. P. Putnam, 1994. D. N. Davis and S. C. Lewis. Computational models of emotion for autonomy and reasoning. Informatica References (Slovenia), 27(2):157–164, 2003.

J. Abdi, B. Moshiri, B. Abdulhai, and A. K. Sedigh. B. M. Dehkordi, A. Kiyoumarsi, P. Hamedani, and Forecasting of short-term traffic-flow based on im- C. Lucas. A comparative study of various intelligent proved neurofuzzy models via emotional temporal dif- based controllers for speed control of {IPMSM} drives ference learning algorithm. Eng. Appl. Artif. In- in the field-weakening region. Expert Syst. Appl., 38 tel., 25(5):1022 – 1042, 2012. ISSN 0952-1976. doi: (10):12643 – 12653, 2011a. ISSN 0957-4174. doi: 10.1016/j.engappai.2011.09.011. 10.1016/j.eswa.2011.04.052.

I.F. Akyildiz, W. Su, Y. Sankarasubramaniam, and B. M. Dehkordi, A. Parsapoor, M. Moallem, and C. Lu- E. Cayirci. Wireless sensor networks: a survey. Com- cas. Sensorless speed control of switched reluc- puter Networks, 38(4):393 – 422, 2002. ISSN 1389- tance motor using brain emotional learning based 1286. doi: 10.1016/S1389-1286(01)00302-4. intelligent controller. Energy Convers. and Man- T. Babaie, R. Karimizandi, and C. Lucas. Prediction age., 52(1):85 – 96, 2011b. ISSN 0196-8904. doi: of solar conditions by emotional learning. Intelligent 10.1016/j.enconman.2010.06.046. Data Analysis, 10(6):583–597, 2006. J. Dias and A. Paiva. Feeling and reasoning: A compu- M. Bahrepour, N. Meratnia, and P. J.M. Havinga. Fast tational model for emotional characters. In C. Bento, and accurate residential fire detection using wireless A. Cardoso, and G. Dias, editors, Progress in Artifi- sensor networks. Environ. Eng. Manag. J., 9(2):215– cial Intelligence, volume 3808 of LNCS, pages 127–140. 221, 2010a. Springer Berlin Heidelberg, 2005. ISBN 978-3-540- 30737-2. doi: 10.1007/11595014 13. M. Bahrepour, B. J. van der Zwaag, N. Meratnia, and P. J. M. Havinga. Fire data analysis and feature H. T. Dorrah, A. M. El-Garhy, and M. E. El-Shimy. reduction using computational intelligence methods. Pso-belbic scheme for two-coupled distillation column In G. Phillips-Wren, L.C. Jain, and K. Nakamatsu, process. J. of Advanced Research, 2(1):73 – 83, 2011. editors, 2nd KES Int. Symp. on Intelligent Decision ISSN 2090-1232. doi: 10.1016/j.jare.2010.08.004. Technologies, volume 4 of Smart Innovation, Systems P. Ekman and R. J. Davidson. The Nature of Emotion: and Technologies, pages 289–298, Berlin Heidelberg, Fundamental Questions. Oxford University Press, 2010b. Springer-Verlag. 1994. R. Balasubramanian. Forecasting geomagnetic activity indices using the Boyle index through artificial neu- M. S. El-Nasr, J. Yen, and T. R. Ioerger. Flame - fuzzy ral networks. PhD thesis, Rice University, Houston, logic adaptive model of emotions. Autonomous Agents Texas, April 2010. and Multi-Agent Systems, 3(3):219–257, 2000. S. Banerjee, C. Grosan, A. Abraham, and P. Mahanti. R. Farhangi, M. Boroushaki, and S. H. Hosseini. Intrusion detection on sensor networks using emo- Load-frequency control of interconnected power sys- tional ants. Int. J. of Applied Science and Computa- tem using emotional learning-based intelligent con- tions, 12(3):152–173, 2005. troller. Int. J. of Electrical Power & Energy Sys- tems, 36(1):76–83, 2012. ISSN 0142-0615. doi: J. Bates. The role of emotion in believable agents. Com- 10.1016/j.ijepes.2011.10.026. mun. ACM, 37(7):122–125, July 1994. ISSN 0001- 0782. doi: 10.1145/176789.176803. M. Fatourechi, C. Lucas, and A. K. Sedigh. Emotional learning as a new tool for development of agent-based T. Bosse, J. Gratch, J. F. Hoorn, M. Portier, and G. F. systems. Informatica, 27(2):137–144, 2003. Siddiqui. Comparing three computational models of affect. Advances in Soft Computing, 70/2010:175–184, N. H. Frijda. The Emotions. Cambridge University 2010. Press, 1987. M. Cabanac. What is emotion? Behavioural Pro- S. C. Gadanho and J. Hallam. Robot learning driven by cesses, 60(2):69 – 83, 2002. ISSN 0376-6357. doi: emotions. Adaptive Behavior, 9(1):42–64, March 2001. 10.1016/S0376-6357(02)00078-5. ISSN 1059-7123. doi: 10.1177/105971230200900102. 14 T. E. Kalayci et al.

A. Gholipour, C. Lucas, and D. Shahmirzadi. Predicting T. E. Kalayci, M. Bahrepour, N. Meratnia, and P. J. M. geomagnetic activity index by brain emotional learn- Havinga. How wireless sensor networks can ben- ing algorithm. In Digest of the Proc. of the WSEAS, efit from brain emotional learning based intelli- 2003. gent controller (belbic). Procedia Computer Sci- ence, 5:216–223, 2011. ISSN 1877-0509. doi: A. Gholipour, C. Lucas, and D. Shahmirzadi. Purposeful 10.1016/j.procs.2011.07.029. prediction of space weather phenomena of simulated emotional learning. Int. J. of Modelling and Simula- M. Kazemifard, N. Ghasem-Aghaee, and T. I. ren. De- tion, 24(2):65–72, 2004. sign and implementation of gema: A generic emotional agent. Expert Syst. Appl., 38(3):2640 – 2652, 2011. Ben Goertzel. A general theory of emotion in hu- ISSN 0957-4174. doi: 10.1016/j.eswa.2010.08.054. mans and other intelligences, 2004. Available: http://www.goertzel.org/dynapsyc/2004/Emotions.htm. S. M. Khadem, S. Behzadipour, M. Boroushaki, F. Farahmand, and M. Tavakoli. Design and imple- J. Gratch, S. Marsella, N. Wang, and B. Stankovic. As- mentation of an emotional learning controller for force sessing the validity of appraisal-based models of emo- control of a robotic laparoscopic instrument. Frontiers tion. In 3rd Int. Conf. on Affective Computing and in Biomedical Technologies, 1(3), 2014. ISSN 2345- Intelligent Interaction and Workshops, 2009. 5837. Arie Horst. Sport coach : online activity using wireless M. R. Khalghani and M. H. Khooban. A novel sensor network. Master’s thesis, University of Twente, self-tuning control method based on regulated bi- Electrical Engineering, Mathematics and Computer objective emotional learning controller’s structure Science, July 2010. http://essay.utwente.nl/59619/. with {TLBO} algorithm to control {DVR} compen- sator. Appl. Soft Comput., 24(0):912 – 922, 2014. ISSN R. Imbert and A. de Antonio. When emotion does not 1568-4946. doi: 10.1016/j.asoc.2014.08.051. mean loss of control. In Intelligent Virtual Agents, volume 3661 of LNCS, pages 152–165. Springer Verlag, M. Khalilian, A. Abedi, and A. D. Zadeh. Position con- 2005. doi: 10.1007/11550617 14. trol of hybrid stepper motor using brain emotional controller. Energy Procedia, 14:1998 – 2004, 2012. M. Jafari, A. M. Shahri, and S. B. Shouraki. Attitude ISSN 1876-6102. doi: 10.1016/j.egypro.2011.12.1200. control of a quadrotor using brain emotional learning based intelligent controller. In 13th Iranian Conf. on A. Khashman. Application of an emotional neural net- Fuzzy Systems (IFSC), pages 1–5, Aug 2013a. doi: work to facial recognition. Neural Computing and Ap- 10.1109/IFSC.2013.6675672. plications, 18:309–320, 2009. doi: 10.1007/s00521-008- 0212-4. M. Jafari, A. M. Shahri, and S. B. Shuraki. Speed control of a digital servo system using brain emotional learn- S. S. Khorramabadi, M. Boroushaki, and C. Lucas. Emo- ing based intelligent controller. In 4th Power Elec- tional learning based intelligent controller for a pwr tronics, Drive Systems and Technologies Conf. (PED- nuclear reactor core during load following operation. STC), pages 311–314, Feb 2013b. doi: 10.1109/PED- Annals of Nuclear Energy, 35(11):2051 – 2058, 2008. STC.2013.6506724. N. Kubota, T. Obo, and T. Fukuda. An intelligent moni- S. Jafarzadeh, M. S. Fadali, and C. Lascu. An emotional toring system based on emotional model in sensor net- learning intelligent direct torque and flux controller works. In The 18th IEEE Int. Symp. on Robot and Hu- design for induction motor. In IEEE Energy Conver- man Interactive Communication, pages 346–351, 2009. sion Congress and Exposition, pages 3988–3992, sept. J. E. LeDoux. The emotional brain: mysterious under- 2012. doi: 10.1109/ECCE.2012.6342158. pinnings of emotional life. Simon & Schuster, 1996. M. Jamali, A. Arami, B. Hosseini, B. Moshiri, and C. Lu- P. G. Lee, K. K. Lee, and G. J. Jeon. An index of ap- cas. Real time emotional control for anti-swing and po- plicability for the decomposition method of multivari- sitioning control of simo overhead travelling crane. Int. able fuzzy systems. IEEE Trans. Fuzzy Syst., 3(3):364 J. of Innovative Computing, Information and Control, –369, 1995. 4(9):2334–2344, 2008. E. Lotfi. Brain-inspired emotional learning for image M.R. Jamali, A. Arami, M. Dehyadegari, C. Lucas, and classification. Majlesi Journal of Multimedia Process- Z. Navabi. Emotion on fpga: Model driven approach. ing, 2(3), 2013. ISSN 2251-6255. Expert Syst. Appl., 36(4):7369 – 7378, 2009. E. Lotfi and M. R. Akbarzadeh-T. Adaptive brain J.-S.R. Jang. Anfis: adaptive-network-based fuzzy in- emotional decayed learning for online prediction of ference system. IEEE Trans. Syst., Man, Cybern., geomagnetic activity indices. Neurocomputing, 126 23(3):665–685, May 1993. ISSN 0018-9472. doi: (0):188 – 196, 2014a. ISSN 0925-2312. doi: 10.1109/21.256541. 10.1016/j.neucom.2013.02.040. Can WSNs be Emotional? 15

E. Lotfi and M.-R. Akbarzadeh-T. Practical emo- R. W. Picard, E. Vyzas, and J. Healey. Toward machine tional neural networks. Neural Networks, 59 emotional intelligence: analysis of affective physiolog- (0):61 – 72, 2014b. ISSN 0893-6080. doi: ical state. IEEE Trans. Pattern Anal. Mach. Intell., 10.1016/j.neunet.2014.06.012. 23(10):1175 –1191, oct 2001. ISSN 0162-8828.

E. Lotfi and A. Keshavarz. Gene expression microarray M. J. Roshtkhari, A. Arami, and C. Lucas. Imitative classification using pcabel. Computers in Biology and learning based emotional controller for unknown sys- Medicine, 54:180 – 187, 2014. ISSN 0010-4825. doi: tems with unstable equilibrium. Int. J. of Intelligent 10.1016/j.compbiomed.2014.09.008. Computing and Cybernetics, 3(2):334–359, 2008. A. Sadeghieh, H. Sazgar, K. Goodarzi, and C. Lu- C. Lucas, D. Shahmirzadi, and N. Sheikholeslami. In- cas. Identification and real-time position control of troducing belbic: Brain emotional learning based in- a servo-hydraulic rotary actuator by means of a neu- telligent controller. Intelligent Automation and Soft robiologically motivated algorithm. {ISA} Transac- Computing, 10:11–22, 2005. tions, 51(1):208 – 219, 2012. ISSN 0019-0578. doi: M. Marin-Perianu, S. Bosch, R. S. Marin-Perianu, 10.1016/j.isatra.2011.09.006. J. Scholten, and P. J. M. Havinga. Autonomous ve- M. Sharbafi, C. Lucas, and A. Mohammadinejad. De- hicle coordination with wireless sensor and actuator signing a football team of robots from beginning to networks. ACM Trans. Auton. Adap., 5(4), 2010. end. Int. J. of Information Technology, 3(2), 2006. S. C. Marsella and J. Gratch. Ema: A process model N. Sheikholeslami, D. Shahmirzadi, E. Semsar, C. Lucas, of appraisal dynamics. Cogn. Syst. Res., 10(1):70–90, and M. J. Yazdanpanah. Applying brain emotional 2009. learning algorithm for multivariable control of hvac systems. J. Intell. Fuzzy Syst., 17:35–46, 2006. J. Martinez-Miranda and A. Aldea. Emotions in human and artificial intelligence. Comput. Hum. Behav., 21 R. C. Solomon. emotion. Encyclopedia Britannica. En- (2):323 – 341, 2005. cyclopedia Britannica Online Academic Edition, 2014. http://global.britannica.com/EBchecked/topic/ A. R. Mehrabian, C. Lucas, and J. Roshanian. Aerospace 185972/emotion. launch vehicle control: an intelligent adaptive ap- D. Statt. The Concise Dictionary of Psychology. Rout- proach. Aerosp. Sci. and Technol., 10(2):149 – 155, ledge, London, 1998. 2006. ISSN 1270-9638. doi: 10.1016/j.ast.2005.11.002. B. Steunebrink, M. Dastani, and J. J. Meyer. A formal A. R. Mehrabian, C. Lucas, and J. Roshanian. Design of model of emotion-based action tendency for intelligent an aerospace launch vehicle autopilot based on opti- agents. In Progress in Artificial Intelligence, volume mized emotional learning algorithm. Cybernet. Syst., 5816 of LNCS, pages 174–186. Springer Verlag, 2009. 39:284–303, 2008. doi: 10.1080/01969720801944364. M. Valikhani and C. Sourkounis. A novel intelligent con- J. Moren and C. Balkenius. A computational model of troller for dfig-based wind turbine system. In IEEE emotional learning in the amygdala. In In Proc. of Int. Energy Conf. (ENERGYCON), pages 44–50, May the 6th Int. Conf. on the Simulation of Adaptive Be- 2014. doi: 10.1109/ENERGYCON.2014.6850404. haviour, 2000. G. R. Vandenbos, editor. The APA Dictionary of Psy- K. Oatley, D. Keltner, and J. M. Jenkins. Understanding chology. American Psychological Association, 2006. Emotions, 2nd ed. Wiley-Blackwell, 2006. J. D. Velasquez. Modeling emotion-based decision- making. In In AAAI Fall Symp. Emotional and Intel- M. Parsapoor, U. Bilstrup, and B. Svensson. A brain ligent: The Tangled Knot of Emotion, 1998. emotional learning-based prediction model for the pre- diction of geomagnetic storms. In Federated Conf. Y. Yang, Y. Wang, X. Yuan, and F. Yin. Hybrid chaos on Computer Science and Information Systems, pages optimization algorithm with artificial emotion. Appl. 35–42, Sept 2014. doi: 10.15439/2014F231. Math. and Comput., 218(11):6585 – 6611, 2012. ISSN 0096-3003. doi: 10.1016/j.amc.2011.09.028. Silvana Petruseva. Emotion learning: Solving a shortest path problem in an arbitrary deterministic environ- J. Yick, B. Mukherjee, and D. Ghosal. Wireless ment in linear time with an emotional agent. Int. J. sensor network survey. Comput. Netw., 52(12): Appl. Math. Comput. Sci., 18:409–421, 2008. 2292–2330, August 2008. ISSN 1389-1286. doi: 10.1016/j.comnet.2008.04.002. H. R. Pfister and G. B¨ohm. The multiplicity of emo- E. Yong, W. Qian, and K. He. An adaptive predictorcor- tions: A framework of emotional functions in decision rector reentry guidance based on self-definition way- making. Judgm. Decis. Mak., 3:5–17, 2008. points. Aerosp. Sci. and Technol., 39(0):211 – 221, R. W. Picard. Affective computing. MIT Press, 1997. 2014. ISSN 1270-9638. doi: 10.1016/j.ast.2014.08.004.