An LSTM-Based Online Prediction for Effluent BOD in Wastewater Treatment Process

To achieve the real-time dynamic prediction of effluent biochemical oxygen demand (BOD) in the wastewater treatment process (WWTP), an online prediction method based on LSTM neural network (OPLSTM) is proposed in this paper. The online prediction is realized by sliding window strategy using the LSTM model, with the sliding window size adaptively adjusted by an adaptive method based on model performance and data distribution, and BPTT learning algorithm is used to update the parameters of the LSTM. The OPLSTM is first tested on two benchmark datasets and then applied to BOD prediction in WWTP. The experimental results show that the method can realize online prediction and obtain satisfactory prediction accuracy, and the adaptive method for sliding window size would improve the generalization performance of OPLSTM.