Multi-objective evolutionary algorithms and multiagent models for optimizing police dispatch

In this article we investigate Multi-agent simulation and Multi-objective Evolutionary Algorithms for optimizing resource allocation in Public Safety. We describe a tool that helps Law Enforcement authorities to evaluate, in a controlled environment, different strategies for allocating and dispatching resources, aiming at reducing conflicting goals such as response time, the number of unattended calls and cost of displacement of police cars. This tool is a multi-agent model to represent police cars that lives in a grid in which emergency occurrences appear. A comparison of the strategies for resource dispatch in this environment shows that serving first those calls with low estimated attendance times delivers the best overall performance in terms of waiting time. However this is practically impossible since prioritization of certain crime types is necessary leading to the increase of the waiting time in the queue. Instead of manually trying to identify the best allocation strategy to apply, we have coupled a multi-objective evolutionary algorithm to the simulation model in order to uncover automatically a function to rank the calls in the best order for attendance satisfying multiple and sometimes conflicting goals.