Multi-layered security investment optimization using a simulation embedded within a genetic algorithm

Abstract

The performance of a multi-layered security system, such as those protecting high-value facilities or critical infrastructures, is characterized using several different attributes including detection and interruption probabilities, costs, and false/nuisance alarm rates. The multitude of technology options, alternative locations and configurations for those technologies, threats to the system, and resource considerations that must be weighed make exhaustive evaluation of all possible architectures extremely difficult. This paper presents an optimization model and a computationally efficient solution procedure to identify an estimated frontier of system configuration options which represent the best design choices for the user when there is uncertainty in the response time of the security force, once an intrusion has been detected. A representative example is described.

A Stochastic Programming Approach to the Design Optimization of Layered Physical Protection Systems

“…The model developed in this paper extends the model given in [4] in two significant dimensions. First, scenarios are used to describe the connection between weather, visibility conditions and intruder capabilities with detection probability.…”

“…Many investments are only effective if they are deployed as a cycle or a contiguous boundary instead of possibly disconnected, single-link investments as described in [4]. Consequently, we make use of "layered" investments where a single investment layer consists of a collection of link investments of a single type (e.g., sensor "sX" investments) forming a contiguous boundary around the target at a fixed radius.…”

“…Since this probability becomes vanishingly small when the initial detection is close to the target, we set a threshold to ignore low probability paths (PI = 0.1) which mitigates the multipath issue described in [4] when trying to find the MVP (the path with the lowest PI).…”

“…The solution procedure for generating the candidate security architectures uses an investment planning optimization, similar to the one presented in [4]. A genetic algorithm (GA) determines the best mix of investments to apply to the network based on an objective composed of investment cost, NAR/FAR, and PI.…”

“…Since the investments are now collected in synergistic layers, there is no need to use the region crossover method for genetic crossover as done previously in [4], hence a traditional random cut procedure is used. Each child produced is postprocessed to guarantee feasibility by ensuring that at least one layer of each investment type (delay, detection, and ASO) is included, since anything less creates an infeasible solution.…”