A Pruning Stochastic Configuration Network Algorithm Based on Node Similarity and Contribution and Its Application in Wastewater Treatment Processes

Stochastic configuration networks (SCNs) have been extensively employed in the modeling of nonlinear system regression tasks. The SCN model stops training by setting the maximum number of nodes in the hidden layer or the expected error tolerance. If the parameters are not set properly, it is easy to cause the model to overfit. To address this problem, a pruning SCN algorithm based on node similarity and contribution (SCPSCN) is proposed. First of all, we introduce the definitions of node similarity. Then, basing on Garson algorithm, a method that can effectively judge the contribution of hidden layer nodes is introduced. Finally, the redundant and low-quality nodes are deleted based on the similarity and contribution so that the optimal network structure of SCN is obtained. The proposed method has been successfully applied to prediction of BOD in a wastewater treatment plant. The experimental results show that SCPSCN has better prediction performance than the original SCN method.