Explorations in swarm algorithms: Hybrid particle swarm optimization and adaptive culture model algorithms.

摘要

Swarm Intelligence refers to the approach of performing a complex task with a large number of simple agents following simple rules. This dissertation proposes two new swarm optimization algorithms namely the gradient based particle swarm optimization (GPSO) algorithm and the continuous adaptive culture model (CACM) algorithm. The GPSO algorithm and the CACM algorithm are then shown to be computationally efficient and converge faster than existing swarm optimization techniques.; Stochastic optimization algorithms like genetic algorithms (GA) and simulated annealing (SA) algorithm perform global optimization but waste computational effort by doing a random search. On the other hand, deterministic algorithms like gradient descent converge rapidly but may get stuck in local minima of multimodal functions. Thus, an approach that combines the strengths of stochastic and deterministic optimization schemes but avoids their weaknesses is of interest. This dissertation presents a new hybrid gradient based PSO (GPSO) algorithm that converges to a significantly more accurate solution than existing PSO based techniques for a variety of test functions.; The utility of the GPSO algorithm is demonstrated by applying it to a target location estimation problem in wireless sensor networks (WSNs). It was observed that the use of the GPSO algorithm provides significantly higher position estimation accuracy throughout the sensor field than classical deterministic schemes.; Optimization of large scale problems cannot be done in reasonable time by serial machines, thus schemes for parallel implementation are of interest. The classical PSO algorithm uses the global best solution to update each particle which leads to unacceptably high communication levels in parallel implementations. This dissertation explores new PSO schemes which do not use the global best solution to update each particle, thereby avoiding communication bottlenecks inherent in the serial PSO algorithm.; GAs are typically used for optimization of discontinuous and nondifferentiable functions that cannot be optimized with derivative based techniques. However, GAs are computation intensive and suffer from slow convergence rates. A new computationally inexpensive alternative to GAs, the continuous adaptive culture model (CACM), is proposed in this dissertation and shown to perform competitively on a variety of benchmark test functions from the literature.