River Water Quality Parameters Prediction Method Based on LSTM-RNN Model

Abstract

This paper investigates the characteristics of dynamic nonlinearity and correlation of water quality parameter information, as well as the gradient disappearance and gradient explosion caused by the training data of traditional RNN network model, etc. The long short-term memory network structure (LSTM) is introduced to optimize the structure of RNN network and the connection weight and threshold of hidden layer. A new water quality parameter prediction model of LSTM-RNN network based on improved RNN network structure is proposed by setting the number of storage units in the hidden layer of the network, the number of structural layers of the network model, and adjusting the time window size of the data training set. Combined with the water quality monitoring data of the River in Shanghai, the model is used to predict and verify the main pollutant index COD (potassium permanganate index) in the River. The simulation results show that compared with the traditional GM (grey model) and RNN network water quality prediction model, the sample approximation accuracy and generalization ability of the training prediction based on LSTM-RNN network model is higher and better than that of the traditional GM (grey model) and RNN network model. Good comprehensive prediction performance of river water quality is presented.