IMPROVISED PSO ALGORITHMS AND THEIR APPLICATIONS IN SEISMOLOGY

Abstract

Particle Swarm Optimization, Differential Evolution (DE) and Gravitational Search Algorithm (GSA) are some effective natural computing schemes with reduced memory requirements, computatiorially effective and easier to irnplenierit over coniputational platforms. This Thesis aims to design improvised, more reliable, capable and efficient natural computing methods for various type of optimization problems. The main contribution of this Thesis is the proposal of three new PSO algorithms for unconstrained and constrained optimization problems. A new Shrinking Hypersphere based PSO (SHPSO) algorithm is designed for the unconstrained optimization problems. The proposed SHPSO is tested over thirty benchmark problems taken from IEEE CEC sessions 2008 and 2010. To justify the performance of the SHPSO, it is compared with the five state-of-the-art PSO algorithms namely Basic P50, Trelea I, Trelea II, Clerc PSO and SPSO 2011 on various measures including statistical validation by implementing statistical test. The efficiency and capability of the proposed SHPSO is examined theoretically as well as numerically. The applicability of SHPSO over constrained problem is investigated. Two new hybrid algorithms called CSHPSO and SHPSOGSA for small scale and large scale optimization are designed by hybridizing SHPSO with DE and GSA respectively. The performance of these two proposed algorithms is tested over twenty four constrained optimization problems coined in IEEE CEC 2006 and the results are compared with the original versions of DE and GSA. For comparison, various measures are taken into account to justify the efficieiicy of the proposed hybrids. The very basic aim of the development of these algorithms is to apply them over real life optimization problems, hence the problem of determination of hypocentral parameters of an earthquake is modeled as an optimization problem. It is solved with the help of the proposed SHPSO. Based on single and multilayered Earth crust structure model the problem is modeled in two phases. A real earthquake data observed at North West Himalayan region is employed to perform the experiments, eight and twelve earthquake events are considered for the analysis of single layer model and multilayered model, respectively. These models corresponds to the shallow earthquakes and deep earthquakes. The results of these events are compared with the same set of algorithms namely Basic PSO, Trelea I, Trelea II, Clerc PSO and SPSO 2011. Finally the Thesis is concluded with recommendation of three optimization algorithms namely SHPSO for unconstrained optimization problems, SHPSO-DE (CSHPSO) for small scale constrained optimization problems and SHPSOGSA for large scale constrained optimization problems. The application of these algorithms in seismology is demonstrated and it is shown that SHPSO outperforms to its contemporary algorithms namely Basic PSO, Trelea I, Trelea II, Clerc PSO and SPSO 2011.