Research On Water Quality Prediction Models Based On Deep Learning Algorithm

Water is an indispensable source for human being and other living species,and it has significant socio-economic and eco-environmental values to establish water quality prediction models to predict water situations.In this paper,two prediction models,W-LSTM and W-Seq2 Seq,are proposed based on the research concerning the prediction of water quality series data using Recurrent Neural Network(RNN).The two models proposed have their own advantages,which are suitable for different application scenarios.The main research contents of this paper include the following two aspects:1.In order to accurately predict the changeable situations of water quality data over times,a water quality prediction model(W-LSTM)based on LSTM and wavelet decomposition is proposed.Firstly,the W-LSTM model uses wavelet decomposition technology to decompose the water quality series data into one low-frequency coefficient and three high-frequency coefficient.Then,the wavelet reconstruction technique is used to reconstruct the four dimensionless frequency coefficients into four dimensional frequency data series.Finally,a group of LSTM models is built,and the four reconstructed data series are used as the input of the model group to generate the final prediction results.The comparative experiment is undertaken to compare the results of W-LSTM with the results of LSTM model where the data is directly input into the model without the process of wavelet transform.Based on four water quality data series of the Wangjiaba Reservoir in Anhui Province,the experimental results show that the prediction results of W-LSTM are obviously better than these of the traditional LSTM,which proves the effectiveness of the models.2.For the application scenarios of trend prediction with large amount of data,a Seq2 Seq water quality prediction model based on wavelet decomposition(W-Seq2Seq)is proposed.First of all,the W-Seq2 Seq model uses wavelet transform as a means of noise reduction,and regards the fourth-order low-frequency signals decomposed from the original data by wavelet as the trend of the original data series.Then,a two-layers bidirectional Seq2 Seq model is built to predict the trend of water quality data.Finally,Comparative experiment is undertaken using the Seq2 Seq models with three different architectures based on four water quality data series of the Menlou Reservoir in Yantai city,Shandong Province.The experimental results show that the two-layer bidirectional Seq2 Seq model is robust and has higher prediction accuracy.