Study on Prediction Model of Effluent Total Nitrogen Based on Bidirectional Recurrent Neural Network

Total nitrogen (TN) in effluent of wastewater treatment plants is an important index to evaluate the effect of water treatment. Accurate prediction of effluent TN in the short term can provide data support for the optimized operation of wastewater treatment plants. In order to improve the prediction accuracy of effluent TN in sewage treatment process, this paper studied an effluent TN prediction model based on bidirectional recurrent neural network, which can capture more comprehensive context information by considering both past and future sequence information. The experimental results showed that compared with the coefficient of determination (R2) of RNN neural network on the test set was 0.423, the simulation and prediction accuracy of bidirectional recurrent neural network was higher, and the R2 of BiGRU was the highest (0.657). At the same time, adding attention mechanism to BiGRU could further improve the prediction accuracy, and the R2 was 0.671. In other recurrent neural networks, R2 from high to low is 2-GRU > BiLSTM > 2-LSTM > BiRNN > 2-RNN.