Artificial-life under the waves : an evolutionary approach to the target motion analysis problem

Abstract

This thesis details the development of a system to solve the Target Motion Analysis prob¬lem, which is the problem of tracking a moving target using passive sonar. The information available for the solution of this problem is usually limited to a series of target bearings, calculated by a sonar array. Consequently the problem's solution requires the determination the target's bearing and range just from its bearing, which, as should be immediately ap¬parent, is not an easy task. Traditional methods of solution require the target motion to be constant, which is not a realistic situation. In this thesis, a novel system incorporating Computational Intelligence was developed to solve the Target Motion Analysis problem. This involved formulating the Target Motion Analysis problem as an optimisation problem, and then using a Genetic Algorithm to find the optimal modelling of the target motion. A Survivalist-Based Genetic Algorithm was used due to its resistance to premature convergence and its more powerful search capabilities. Through careful tuning of the Survivalist-Based Genetic Algorithm, excellent results were produced. The developed system is not only able to model non-constant target motion, but is also able, unlike tradition methods of solution, to ignore the effects of the commonly occurring phenomenon of multipath propagation, which introduces a non-measurable bias to our measured target bearings. Finally, in an attempt to produce more consistent results, a novel parallel implementation of the Survivalist-Based Genetic Algorithm tracker was developed. This approach takes advantage of the PVM or Parallel Virtual Machine software package, which allows parallel computing to be performed over a collection of single processor networked machines. The final result is a unique, reliable and robust system, utilising the Survivalist-Based Ge-netic Algorithm and the Parallel Virtual Machine, that consistently produces results that accurately model actual target motion for maneuvering targets in the presence of multipath propagation.