Prediction Based on Online Extreme Learning Machine in WWTP Application

摘要

Predicting the plant process performance is essential for controlling in wastewater treatment plant (WWTP), which is a complex non-linear time-variant system. Extreme learning machine (ELM) is a single-hidden layer feed-forward neural network (SLFN), which randomly generates the feed-forward parameters without tuning the parameters from the input to the output layer. The output weights are calculated via the theory of Moore-Penrose generalized inverse and the minimum norm least-squares. In this paper, online extreme learning machine (Online ELM) is proposed as a predictor in WWTP, which trains the output weights and predicts the next outputs according to the real-time data collected from the process in an online manner. Furthermore, extensive comparison studies have been conducted by using other four neural network structures, including extreme learning machine, ELM with kernel, online sequential ELM (OSELM) and back propagation (BP) neural network.