A Game Theoretic Approach to Emergency Units Location Problem Vito FRAGNELLI 1 - Stefano GAGLIARDO 2 Abstract The location of the emergency units may have a deep influence on the intervention time, which in its turn may be relevant for a positive solution of an emergency situation. In this paper we analyze a particular location problem with a discontin- uous intervention time function. The problem is tackled from different points of view each one resulting in a different game. The paper refers to a toy model and to a real world situation of locating some additional ambulance in the province of Milan. Key Words Emergency Management, Cooperative Game, Location Problem 1 Introduction Emergency management represents a hard task in several situations. It has to be viewed under many different lights, so that a large number of parameters is involved. Consequently, many different experiences and expertizes are needed in order to have an efficient result. First of all, it is necessary to distinguish among different types of emergency: medical, environmental, urban, and so on. Each of them has particular features that have to be tackled in different ways, with different means and tools. Before going deeper in our analysis we want to devote 1 Corresponding author: Universit`a del Piemonte Orientale - Dipartimento di Scienze e Tecnologie Avanzate - Via V.Bellini 25/G - 15100 Alessandria - Italy. e-mail: [email protected] 2 Universit`a di Genova - Dipartimento di Matematica e-mail: [email protected] 1 some words to make clear the characteristics of the kind of situations that are more suitable for our approach. The idea of emergency is strongly correlated to the idea of urgency, and then with the idea of fast intervention. On the other hand the measure of the time re- quired is different for the various types of emergency. Medical intervention should be carried out in minutes, while environmental emergency may allow longer time and the rescue missions after an earthquake can be organized also some days after the event. In this work we are interested in those particular situations for which the utility of the intervention has a discontinuity, i.e. after a given period there is a sudden and large reduction of the necessity of the intervention. A classical example is represented by the ambulances that are allowed for a maximal time for the most difficult situations, the so-called red codes; for example in Italy the time allowed is 8 minutes, including the time for answering the call. It is clear that if an ambulance arrives after 9 minutes the intervention is anyhow completed (for example an injured person is transported to the hospital), but when an ambulance is located on the territory only the area that can be reached in 8 minutes is considered as covered, when a statistical analysis is performed. Another interesting situation is represented by fire planes or helicopters, for which it is necessary to take into account the fuel tank capacity and the distance from a place for charging water. Summarizing, in the standard location problem the utility of a service located in a fixed point decrease when the distance from the point increases, while in this work we refer to situations in which the utility can be consider maximal until a given distance and then it can be considered equal to zero. We want to remark that the distance may be measured in different ways, e.g. ground distance, time distance, linear distance. The main idea that has driven us in this problem was to consider the different candidate locations as agents interacting among them. In fact the choice of deploying an emergency unit in a candidate location cannot take into account only its own characteristics, i.e. the extension of the area that can be covered in the maximal allowed time, the probability of a call in that area, but should consider also the location of the other emergency units. This starting point enables us to have a game theoretic approach to the problem, in order to emphasize the 2 interactions among the ambulances deployed on the area under analysis. Location problems where already studied with the instruments of Game Theory, see [2], where cost sharing problems where studied, but our situation is the first case in which the game theoretic approach is used to support the localization of the facilities. Finally, we want to stress that the complexity of the problem of emergency management allows and requires the synergic use of different methods and ap- proaches. Referring to various aspects of the problem, we can mention, besides Game Theory, statistical analysis, simulation, topography and GIS (Geographical Information System) , graph theory, mathematical programming, and stochastic optimization. The present paper is organized as follows: in the next section we recall some definitions and solutions of cooperative game theory; in Section 3 we introduce two TU games, each of them focusing on particular features of the problem, and analyze a simple way for aggregating the results of them; Section 4 is devoted to a simple theoretical model for testing our approach; a more complex situation arising from the real world is presented in Section 5; Section 6 concludes. 2 Preliminaries on Game Theory In this section we introduce some basic game theoretical concepts that we use throughout the paper. A cooperative game with transferable utility or TU-game, is a pair (N, v), where N = {1, 2, ..., n} denotes the finite set of players and v :2N → IR is the characteristic function,withv(∅) = 0. Often, a TU-game (N, v) is identified with the corresponding characteristic function v. A group of players S ⊆ N is called a coalition and v(S)istheworth of the coalition. n Given a TU-game (N, v), an allocation is a vector (xi)i∈N ∈ IR assigning to player i ∈ N the amount xi. An allocation (xi)i∈N is efficient if i∈N xi = v(N). A solution for a TU-games is a function ψ that assigns an allocation ψ(v)to every TU-game in the class of games with player set N. Game theoretic literature proposes several solutions for TU-games; in this paper we consider the Shapley value and the Banzhaf value, introduced by Shapley [3] and by Banzhaf [1], respectively. 3 The Shapley value assigns to each player his average marginal contribution over all the possible permutations of the players. Formally, given a game (N, v), the Shapley value assigns to player i ∈ N: 1 φi(v)= (v(P (π; i) ∪{i}) − v(P (π; i)) n! π where π is a permutation of the players and P (π; i) is the set of players that precede player i in the permutation π. The Banzhaf value assigns to each player his average marginal contribution over all the possible coalitions he does not belong to. Formally, given a game (N, v), the Banzhaf value assigns to player i ∈ N: β v 1 v S ∪{i} − v S i( )= n−1 ( ( ) ( )) 2 S∈N\{i} The Shapley value is an efficient allocation, while the Banzhaf value is not. In general, it is necessary to normalize the Banzhaf value in order to reach efficiency, when it is used to allocate a profit, a cost or simply to estimate the value of a player in a voting situation. Here, we look for ranking the candidate locations for the emergency units, so that no normalization is necessary. In the following section we will introduce two TU games, each focusing on different features of the problems, and will use the Shapley value and the Banzhaf value to estimate the relevance of each candidate location. 3TheGames The first idea for tackling the problem is to take into account the number of emergency calls that each subset of candidate locations may satisfy, In order to simplify we model we suppose that all the calls from the area covered by an ambulance may be satisfied by the ambulance itself, i.e. the time between two calls is larger than the time required for completing the first intervention. We may define a TU game (N, v1), where the player set N corresponds to the set of candidate locations and the characteristic function v1 associates to each coalition S ⊆ N the worth v1(S)thatisgivenbythenumberofcallscovered locating the ambulances only in the locations represented by S. Consider the following example. 4 Example 1 Consider an area with 9 calls, represented by the bullets; there exist 3 candidate locations, 1, 2 and 3, represented by a circle, and each of them may cover the area represented by the rectangle centered in it. ' $ ' $ y 3 y ' $ y y y y 1 & % 2 y y & %y & % In this case it is possible to associate the game v1(1) = 4; v1(2) = 6; v1(3) = 4; v1(12) = 8; v1(13) = 7; v1(23) = 7; v1(123) = 9 In order to determine the relevance of each candidate location we decided to use the Shapley value and the Banzhaf value. The motivation is that these solutions are rooted in the concept of marginal contribution and this is a very good way for representing the interaction of the ”players” because, as we said in the Introduction the relevance of a location depends on the number of satisfied call that may be added, on the average. For the situation in Example 1 we obtain: φ =(2.833, 3.833, 2.333) β =(2.750, 3.750, 2.250) and in both cases the ordering is 2, 1, 3. We had the doubt that the results of this game could be affected when two different locations overlap for a large number of calls, so that both may have a great relevance even if only one of them is actually important.