Research on out-of-sample prediction method of water quality parameters based on dual-attention mechanism

Abstract

Predicting water quality data is an important measure for ecological environment protection in watersheds. Aiming at the problem that existing prediction algorithms rarely analyze the characteristics of future changes in water quality indicators, this paper proposes an out-of-sample prediction model for water quality parameters based on the dual-attention mechanism. The model adopts the Encoder-Decoder architecture to realize the prediction of data series, and combines the dual attention of dimension and time step to improve the prediction performance of out-of-sample data. The model is used to predict the water quality parameters of a multi-parameter river, analyze the trend of the out-of-sample data, and compare the prediction results with the traditional LSTM network and Encoder-Decoder LSTM network, the prediction accuracies of the water quality indicators are improved, and the prediction accuracy of the out-of-sample data of the water quality indicators reaches 80%. This will be of great significance to the comprehensive management of river waters and the high-quality development of ecological environment.