Soft-Sensor of Water Quality Based on Integrated ELM with Meta-Learning

A soft-sensor of water quality for wastewater treatment plants, which is based on an integrated model, is presented. The proposed soft-sensor aims to address the difficulty in using a single model to represent the characteristics of wastewater treatment processes with varying operating regimes.The soft-sensor is composed of three layers, in which a predictive sub-model based on FCM-ELMs are the bottom layer, adaptive weighted fusion method fusing predictive values of the sub-model are the middle layer, and a meta-learning mechanism based on information entropy updating fusion weights is the top layer. The meta-learning mechanism can track the dynamic trend of operating conditions of wasterwater treatment plants. The quick learning advantage of ELM results in the soft-sensor showing excellent real-time performance. The adaptive weighted fusion method and meta-learning mechanism improve the model generalization. Simulation results show that the integrated model for COD is more accurate than other models.