Simulation optimization of police patrol districting plans using response surfaces

Abstract

In this study, a police patrol discrete-event simulation model is implemented to evaluate the patrol districting plans. Sensitivity analysis is conducted to study how simulation inputs affect performance measures. Evaluation of districting plans is compared between discrete-event and agent-based simulation. A response surface approach is developed to find optimal or sub-optimal districting plans to minimize average response time and workload variation. An iterative searching procedure based on experimental design is proposed to study the relationship between parameters in a redistricting algorithm and performance measures of districting plans. Latin hypercube design and central composite design are used for the discrete-event model. Experimental results show that districting plans generated by improved districting parameters have significantly better performances than randomly generated plans. Compared with an adjusted simulated annealing approach, the response surface method has better searching efficiency. Thus, good districting plans can be generated more efficiently with designed experiments and provide better alternatives for police departments. Similar procedure also works for agent-based simulations.