An Integrated Deep Neural Network Approach For Large-scale Water Quality Time Series Prediction

Water quality prediction is an important aspect of water environment pollution prevention and control.The water quality monitoring data collected over a long period of time can be used to predict water pollution trends,which is of great significance to the management and planning of water environment.Aiming at the problem of how to predict water quality in a timely and effective manner,this study proposes an integrated water quality prediction model.This model can predict the water quality at multi-step in the future based on historical water quality monitoring data for providing advance data guidance of effective regulation and management of water resources.The contributions of this study are mainly reflected in two aspects,including the preprocessing of water quality data and the proposal of sequence-to-sequence deep learning model based on attention mechanism for water quality prediction,as follows:The first is the preprocessing of water quality time series.In this study,the water quality indicator data set released by the national surface water quality automatic monitoring real-time data publishing system is used as the research object.After obtaining the original data,data preprocessing is required.This process includes missing value imputation,standardization and noise reduction.Missing values are inserted to ensure equal intervals of sequence data,that is,water quality indicator distribution at the same time granularity;In the standardized time series,the numerical range of each water quality index is consistent,which can make the subsequent prediction model have better accuracy.The abnormal events in the natural environment will make the water quality abrupt,and the noise data generated will affect the regularity of the water quality sequence in the temporal sequence distribution,which will greatly affect the accuracy of the training generation model.Therefore,this study needs to choose an appropriate noise reduction method to filter the anomalies and errors in the data and reduce the interference to the prediction model.The second is the training of the proposed prediction models.In this study,three deep learning algorithms and three traditional time series algorithms for comparison were selected to predict water quality time series,and the performance of the model was measured based on multiple evaluation indicators and the optimal algorithm was selected.Among them,the traditional sequential algorithms are autoregressive differential moving average algorithm,support vector machine,artificial neural network.Deep learning algorithm includes three kinds of models: long-short-term memory neural network,sequence-to-sequence model,and sequence-to-sequence model based on attention mechanism.This study designed a series of experiments and found that the prediction accuracy of deep learning model is relatively better.At the same time,considering the influence of various factors on the prediction indicator,feature selection method and deep learning model are integrated to further improve the prediction accuracy of the model.In summary,Through the analysis and pretreatment of historical water quality monitoring data,combined with deep learning,this study established an integrated water quality prediction model.Experiments show that compared with other prediction methods,the method proposed in this study can more accurately predict the water quality status in a period of time in the future,and provide reliable data guidance for the effective protection and comprehensive management of water resources.