Automated neural network structure determination via discrete particle swarm optimization (for non-linear time series models)
A. Kalos · 2005
Abstract:- Due to their universal function approximation capability, artificial neural networks have enjoyed widespread use from research and engineering, to finance and banking applications. However, there are two barriers to their acceptance as mainstream data mining tools: 1) the trial-and-error nature of designing the optimal structure, suitable for a particular application and 2) difficulty of interpretation of results. In this paper, we attempt to address the former aspect by describing the use of a discrete version of Particle Swarm Optimization for automating the design of neural networks. The methodology is applied to a case study for selecting the optimal structure for multivariate nonlinear time series models for the day-ahead forecasting of electricity prices. Key-Words:- Neural network structure optimization, Discrete Particle Swam Optimization