BP Neural Network Structure Optimization Algorithm Based on Polynomial Regression

To design a streamlined network structure is a commonly used method for BP neural network to guarantee the neural network´s generalization. Self- configuration algorithm deletes the redundant nodes of the hidden layer to achieve the optimized structure. But it isn´t effective in solving the non-linear problem due to the linear regression theory basis. Thus, a self-configuring algorithm based on polynomial regression is presented. The simulations of the modified algorithm in MATLAB indicate that an improved BP network is achieved, with the optimum number of neurons of the hidden layers.