A Multi-agent Stochastic Control Model for Adversarial Planning in Naval Operations

Abstract

Potential adversary navies have longer range anti-ship missiles than those currently aboard United States Navy warships. Given the United States' policies towards diplomacy and warfare, the United States has focused on the development of advanced defensive capabilities for Navy ships. This thesis uses Princeton Professor Warren Powell's five-component unified mathematical framework for stochastic optimization problems (state variables, decision variables, exogenous information, transition function, and objective function) to shed light on a current USN challenge, introduced by the Naval Postgraduate School and OPNAV N96 staff. Results in this thesis show the value of using offensive weapons as a defensive capability.

This thesis analyzes a two-ship missile engagement coded in Python. A visual of a single iteration of the missile engagement was coded in Java. The simulation iterates through each state of the mathematical model to imitate a basic version of the decisions and uncertainty that is present in a single missile engagement. The model allows for changing inputs and policies for decision making throughout the simulation by manipulating tunable parameters including, but not limited to: the number and range of offensive missiles in a salvo, the use and effectiveness of Intelligence, Surveillance, and Reconnaissance (ISR) systems, and effectiveness of both defensive and offensive missiles. The cost analysis incorporates figures for all missiles fired averaged over thousands of simulations and for the cost of the ship only when it is a casualty of the engagement. Through the model, this thesis shows that using an offensive strike missile as a defense tactic, effectively shooting first, can be cost effective and crucial to success in an engagement.