Effluent ammonia nitrogen prediction of wastewater treatment process via Tikhonov regularized echo state network

Abstract

To predict the effluent ammonia nitrogen $(NH_{4}-N)$, a Tikhonov regularized echo state network (TRESN) is proposed. TRESN uses the Tikhonov regularization method instead of linear regression to train the model, and transforms the selection of Tikhonov regularization parameters into a statistical inference of hyperparameters. The simulation results show that TRESN can well solve effluent $NH_{4}-N$ prediction compared with other ESNs, also has higher prediction accuracy and generalization ability.