Cascading Neural Network with Particle Swarm Optimization

This paper presents a system that combines the neural network learning and meta-heuristic techniques sequentially one after another to have better results than individual ones. The main novelty of the approach is to get benefit from heterogeneous learning environment. Initial good single individual is prepared from NN learning and this solution is injected to PSO as a good global position since at starting PSO does not have good knowledge of global space. It is named as NN-PSO. The technique is applied to expressional chaotic time series data and power consumption data. NN-PSO exhibits remarkable results in comparison with other available methods in terms of mean squared error and convergence speed.