ABSTRACT
This chapter investigates a three-input feed-forward power-activation neurone. It deals with a new type of three-input feed-forward power-activation neuronet and for function data approximation. The chapter presents a new type of feed-forward three-input power-activation neurone in order to the abilities of multi-input neuronets. It analyzes the model of the three-input power-activation neuronet, and proposes the weights-and-structure-determination (WASD) algorithm to achieve the superior performance of the three-input power-activation neuronet. With the weights-direct-determination subalgorithm exploited, the WASD algorithm can obtain the optimal weights of the three-input power-activation neuronet between hidden layer and output layer directly. The WASD algorithm determines the optimal structure of the three-input power-activation neuronet adaptively by growing and pruning hidden-layer neurons during the training process. Numerical results substantiate that the proposed three-input power-activation neuronet equipped with the WASD algorithm has superior performances in terms of function data approximation, testing and prediction.
