Research on Surface Water Quality Prediction based on a CNN-BiLSTM-Attention Combined Deep Learning Approach

Abstract

The ability to predict the environmental conditions of surface water is crucial for prompting the refined management of surface water pollution in China. This paper carried out research on the prediction method of surface water quality based on deep learning algorithms and combined with the real-time data of national automatic monitoring of surface water quality. Under the encoder-decoder framework, the research proposed a CNN-BiLSTM-Attention water quality prediction model which contains CNN, bidirectional LSTM, and attention mechanism. To evaluate the performance of the proposed hybrid model, the research also compared the model with LSTM and CNN-LSTM models, carrying out a comparative analysis of the prediction results of each model through three performance metrics. The research results showed that compared with other models, the CNN-BiLSTM-Attention water quality prediction model can effectively take advantages of each neural network layer and has better prediction ability and higher stability for forecasting future water quality, which can provide strong technical support for water environment management and early warning.