Research on a Pearson-LSTM-AM-based water quality prediction model for freshwater aquaculture

Abstract

In the field of freshwater aquaculture, water quality significantly impacts the aquaculture products. Fluctuations in water quality can hinder the growth of cultured organisms, lead to frequent diseases, and even cause mass mortality. Therefore, accurately predicting water quality is crucial. To reduce the error rate that may occur when using traditional long short-term memory networks (LSTM) models for water quality prediction, this article proposes a Pearson-LSTM-AM water quality prediction model. Initially, the Pearson correlation test algorithm is used for the input feature selection, and then an attention mechanism is integrated to enhance the LSTM neural network's ability to learn the key features, specifically for predicting the dissolved oxygen (DO) indicator of water quality. Experimental results demonstrate that the proposed method significantly improves upon the Pearson-LSTM model and the LSTM model in terms of root mean square error (RMSE), mean absolute percentage error (MAPE), mean absolute error (MAE), and the coefficient of determination (R²) metrics.