The BP neural networks with data clustering enhancement-an emerging optimization tool

The present paper examines enhancements to a backpropagation (BP) neural networks to use them efficiently in the optimization problems. A BP algorithm was extended with the aim of improving both the network training and its generalization capability. A clustering algorithm was implemented by using the Euclidean distances technique; clustering the input patterns in n-dimensional space provides increased efficiency in terms of computational time required to train the network, and better network performance in generalizing new input patterns. This improved function approximation capability of BP networks is proposed to use in optimization problems to avoid expensive exact analysis of the system for objective and constraint evaluations during each cycle of optimization process.