Research Of Water Quality Prediction Methods Based On LSTM

The water quality prediction is the application of the models to predict the trend of water quality in the future based on historical water quality monitoring data.The changes of water quality are closely related to water pollution.Water quality prediction is a basic work in the prevention and control of water environmental pollution.It has important practical significance for promoting the sustainable utilization of water resources and timely pollution prevention.It plays a positive role in anti-pollution early warning system,water supply system and beach water quality prediction system.Time series refers to a set of observations arranged in time sequence.The water quality data obtained by the water quality monitoring station is a time series,so the water quality data is sequential.The water environment system is a gray system affected by various factors,each water quality index has complex multivariate correlation under the interaction of various factors.On the basis of previous studies,the Long-Short Term Memory neural network(LSTM NN),which is commonly used to deal with the application of time series,is introduced into water quality prediction based on the time series of water quality data,and a water quality prediction model based on LSTM is established to predict the trend of water quality indicators.Based on the multivariate correlation of water quality data,an improved grey relational analysis(IGRA)is used to analyze the correlation of the predicted water quality indicators and the correlation indicators are used as input features to predict the predicted water quality indicators.The maily works of this thesis as follows:(1)In this thesis,the common methods of water quality prediction are studied.The methods of water quality prediction based on different theories are summarized and analyzed.Besides,the development of LSTM neural network is studied,and the variants of LSTM in different historical periods are summarized and analyzed.(2)A water quality prediction method based on LSTM neural network is proposed in this thesis.The water quality prediction model with three layers of input,implicit and output was established,and monitoring data set of Taihu Lake is used for training.The experimental results show that the method can effectively mine the time sequence of water quality indicators and improve the accuracy of prediction.(3)This thesis proposes an improved grey relational analysis algorithm IGRA based on the traditional grey relational analysis algorithm.The associated water quality indicators selected by IGRA and the historical monitoring data of the water quality indicators to be predicted are used as input to the model.The monitoring data of Taihu Lake and Hong Kong Victoria Harbour are used for training.The experimental results show that the method can make full use of multivariate correlation of water quality indicators and improve the accuracy of prediction compared with the single-featured prediction model on the two datasets.