A Water Quality Prediction Model Based on Long Short-Term Memory Networks and Optimization Algorithms

Abstract

Currently, the water environment is complicated, and the water quality factors are non-linear and non-stationary, so the accuracy of traditional water quality prediction model is restricted. In order to improve the model performance and the prediction accuracy of the water quality factors (PH, dissolved oxygen, ammonia nitrogen and total phosphorus), a prediction model based on AWPSO-LSTMAT is proposed in this paper. The Zhou River in Haihe River Basin will be the major research objective of this study, therefore, the monitoring data of four types of water quality factors collected from the Xitunqiao water quality monitoring section is considered as the data set. In comparison with the previous prediction models such as SVR, LSTM, CNN-LSTM and CNN-GRU, obviously, the prediction effect of AWPSO-LSTMAT is significantly improved by means of modifying and optimizing the original algorithm. The mean absolute error (MAE) in predicting PH, dissolved oxygen (DO), ammonia nitrogen (AN) and total phosphorus (TP) are 0.039, 0.34, 0.119 and 0.019, respectively. Moreover, the designed prediction model can own higher prediction accuracy and stronger generalization than other existed models. According to above-mentioned illustration, this study can provide reliable technical support for the early warning of water quality, and has a good application value for the water environment treatment in the Zhou River.