A stochastic configuration network modeling method based on improved hidden layer output matrix and supervision mechanism

Abstract

To improve the generalization performance and prediction accuracy of stochastic configuration network (SCN) modeling, a novel SCN modeling method is proposed. First, the first- and second-order directional derivatives of the hidden layer output matrix are solved, the key factors in the directional derivatives are linearly added to the original hidden layer output matrix, and a new formula for obtaining the hidden layer output matrix is constructed. Second, a spatial angle adaptive supervision mechanism is established to improve the quality of the parameter configuration of the hidden layer nodes. The experimental results show that the proposed method improves the generalization performance and prediction accuracy. This work is a beneficial exploration of the standard SCN algorithm.