An Improved Attention-based Bidirectional LSTM Model for Cyanobacterial Bloom Prediction

Cyanobacterial blooms are one of the most serious water pollution problems for freshwater lakes. The treatment of blooms requires a lot of material and financial resources, so an early accurate prediction of cyanobacterial blooms is a very important way to deal with the outbreak of them. But it is challenging to predict the cyanobacterial blooms due to the uncertainty and complexity of their growth process. To deal with this problem, an improved attention-based bidirectional long short-term memory (LSTM) model is proposed in this paper, to make multistep predictions of chlorophyll-a concentration, which is a recognized characterization of algae activity. Firstly, the convolutional neural network (CNN) is used to extract data features and spatiotemporal correlation. Secondly, the bidirectional LSTM network (BiLSTM) is used to predict the concentration of chlorophyll-a based on the extracted features. Finally, the attention mechanism is used to calculate the weights for the characteristic factors that affect the chlorophyll-a concentration. At last, some experiments are carried out based on the real monitoring data of a platform in the Taihu Lake area. Compared with the prediction results of the other four state-of-the-art deep learning methods, the results show that the proposed method in this paper has the highest prediction accuracy.

Automated summary

This paper proposes an improved deep learning model for accurate prediction of cyanobacterial blooms. The model utilizes a convolutional neural network to extract data features and spatiotemporal correlation, and a bidirectional LSTM network for predicting chlorophyll-a concentration. Attention mechanism is employed to calculate the weights for factors influencing the concentration. Experimental results demonstrate that the proposed method achieves the highest prediction accuracy among state-of-the-art deep learning approaches.