An Algorith for Time Series Prediction Using Particle Swarm Optimization (pso)

Although Particle Swarm Optimization (PSO) is used in variety of applications; it has limitations in the training phase. In this work, a new enhancement for PSO is proposed to overcome such limitations. The proposed PSO optimization consists of two stages. In the first stage, a Gaussian Maximum Likelihood (GML) is added to PSO to update the last 25% of swarm particles, while in the second stage, a Genetic Algorithm is applied whenever there is lethargy or no change in the fitness evaluation for two consecutive iterations. Finally, the proposed PSO is applied in time series predictions using Local Linear Wavelet Neural Network (LLWNN). The work is evaluated with three different data sets. Implementation of the proposed PSO shows better results than conventional PSO and many other hybrid PSOs proposed by others.