Wastewater Quality Prediction Model Based on DBN-LSTM via Improved Particle Swarm Optimization Algorithm

For the key water quality of wastewater treatment process, Chemical Oxygen Demand (COD) is difficult to be monitored online. In this paper, a wastewater quality prediction model based on improved Particle Swarm Optimization algorithm Deep Belief Network–Long Short-Term Memory (DBN-PSO-LSTM) is proposed, in which DBN can be used as an unsupervised learning framework to retain original features of input data as much as possible, while reducing the input feature dimension. However, the learning accuracy of DBN for time series with strong volatility is not high. Therefore, the advantages of LSTM in sequence modeling can be used to model and predict output COD. Inappropriate selection of DBN parameters will lead to convergence to local optimal solution. Therefore, the improved PSO algorithm is used to optimize DBN structure to achieve the optimal structure. The optimized DBN is used as the feature extraction part of input variable, and DBN extracted the result is used as the input to LSTM network for prediction. The experimental results show that the proposed method is used to establish the COD soft sensing model of wastewater treatment process, which has better prediction accuracy than other models.