Research on water quality prediction method based on AE-LSTM

Abstract

Aiming at the traditional prediction methods of related parameters that affect water quality, they usually only consider the temporal characteristics of the related parameters of water quality and ignore the problem that water quality changes are multivariate related, A prediction method of spatiotemporal correlation water quality parameters based on automatic encoder (AE) dimensionality reduction and long and short time memory (LSTM) neural network is proposed. Firstly, considering that water quality parameter changes have obvious time characteristics, a time series prediction model of water quality parameters is established based on LSTM. Secondly, considering that the water quality changes have multiple correlations, the upstream water quality will also affect the downstream water quality. If all the water quality parameters of the upstream station are added to the prediction model, redundant features will reduce the accuracy of parameter prediction. Therefore, the automatic encoder is used to reduce the dimensionality of the relevant parameters. Finally, the data set of Lang fang Water Quality Automatic Monitoring Station is applied to monitor the effectiveness of the method. By predicting the concentration of total phosphorus (TP) and total nitrogen (TN), the method is found to have better prediction accuracy and robustness.