Abstract: Dissolved oxygen is an essential parameter for monitoring water quality in rice-prawn farming, as it plays a significant role in crayfish feeding and metabolism. Accurately predicting dissolved oxygen content is critical for maintaining optimal farming conditions and preventing environmental damage. However, dissolved oxygen levels can be challenging to predict due to the complexity of the factors affecting them. A BiLSTM-GRU fusion neural network prediction model that can overcome these challenges was proposed. The model combined the benefits of BiLSTM, which extracted more feature factors, and GRU, which achieved fast and accurate prediction. The sensors and corrected historical data were cleaned and calibrated based on the offset to ensure accuracy. A comprehensive analysis of the resource consumption and prediction performance of the model under different sampling periods was conducted and it was determined that 30 minutes was the optimal sampling period. The proposed model was compared with traditional LSTM, GRU, BiLSTM, and BiGRU models, which was found that the model was demonstrated better prediction performance, with mean absolute error, root mean square error, and determination coefficient of 0.275 9 mg/L, 0.616 0 mg/L, and 0.954 7, respectively. These values were 25.14%, 13.25%, and 2.22% higher than those of the traditional LSTM neural network model. Overall, the proposed BiLSTM-GRU fusion neural network prediction model had significant potential for improving the accuracy of dissolved oxygen content prediction in rice-prawn farming.