CNN-BiLSTM for water quality prediction method based on attention mechanism

Accurate prediction of river water quality is of great help for intelligent management and control of the water environment. Water quality data have complex correlations and time series characteristics, but the existing method in the introduction of LSTM model for water quality prediction, only the water quality series data forward processing, did not consider the reverse processing data, and the ability to focus on local feature extraction and different moment features is not enough. Therefore, this paper proposes a river water quality prediction method based on AM-CNN-BiLSTM. First, the CNN layer is used to feature extract and data dimensionality reduction of the input water quality sequence data and input it to the BiLSTM layer. Secondly, the BiLSTM layer is used to learn the relevant information in the data provided by the CNN layer, and then an attention mechanism is introduced to give more attention to the important features at different times to finally achieve the prediction of river water quality. Comparing the proposed method with other river water quality prediction methods in the national river cross-sectional water quality dataset, the results show that the river water quality prediction method based on AM-CNN-BiLSTM can be more accurate for river water environment prediction.