Abstract:

In view of the problem that the wastewater treatment process has complex nonlinear characteristics and the effluent BOD is difficult to accurately measure, a soft measurement method based on VW-IGRBF neural network is proposed in this study. The activation function of the neural network is a linear combination of the inverse square root function and the Gaussian function, which makes up for the saturation of a single activation function in certain intervals, and improves the expression and self-adaptability of the hidden activation function. Since the width of the activation function has a greater impact on the generalization performance of the model, a variable width strategy based on kernel density is introduced to improve the generalization ability of the network. In this study, the improved LM algorithm is used to realize the online learning of neural network parameters. Simulation experiments based on actual operating data of the wastewater treatment process show that the proposed VW-IGRBF method has higher prediction accuracy and better adaptive ability for effluent BOD.

Key words: variable width, combinational functionn, RBF, soft-sensing, effluent BOD, activation function, kernel density, LM algorithm