Determining optimal police patrol deployments: a simulation-based optimisation approach combining agent-based modelling and genetic algorithms

One of the most important tasks faced by police agencies concerns the strategic deployment of patrols in order to respond to calls whilst also deterring crime. Current deployment strategies typically lack robustness as they are often based on tradition. As police agencies are encouraged to improve the effectiveness and efficiency of their services, it is essential to devise advanced patrol deployments that are based on recent scientific evidence.
Most existing models of patrol deployments are too simplistic, and are thus unable to provide a realistic representation of the complexity of patrol activities. Furthermore, past studies have tended to focus on individual aspects of patrol deployment such as efficiency, reactive effectiveness or proactive effectiveness, rather than consider them all together as part of the same problem.
This thesis proposes to develop a decision-support tool for informing better patrol deployment designs. This tool consists of a simulation-based optimisation approach combining two key components: (1) an agent-based model (ABM) of patrol activities used to evaluate the performance of the system under a given deployment configuration and (2) a genetic algorithm (GA) which seeks to speed up the search for optimal deployments. While the developed framework is designed to be applicable to any police force, a case study is provided for the city of Detroit in order to demonstrate its potential.
The developed decision-support tool shows considerable potential in informing more cost-effective patrol deployments. First, the ABM of patrol activities allows for exploration of the impact of various deployment decisions that police agencies are unable to experiment with in the real world. Second, the GA makes it possible to optimise patrol deployments by identifying 'good' solutions, which provide faster responses to incidents and deter crime in key areas, in reasonable time.