融合多头自注意力与双向LSTM的工厂化循环水溶氧预测模型研究OA

Dissolved oxygen is a critical parameter in factory-based recirculating aquaculture systems,closely related to the effectiveness of aquaculture operations.To address the monitoring and early warning of dissolved oxygen,this study developed a multidimensional data fusion prediction model by integrating Bidirectional Long Short-Term Memory(BiLSTM)with a multi-head self-attention mechanism.Using Pearson correlation analysis,four predictive factors-water temperature,air temperature,turbidity,and pH were selected for training the dissolved oxygen model.The BiLSTM was employed to capture the temporal characteristics of the parameters,while the multi-head self-attention mechanism was used to establish the nonlinear correlation between dissolved oxygen and the other parameters.Experimental data collected from a factory-based recirculating aquaculture system were used to train the model,resulting in a dissolved oxygen prediction model with a Root Mean Square Error(RMSE)of 0.165,a Mean Absolute Error(MAE)of 0.132,and a coefficient of determination(R2)of 0.965.Compared to the standard LSTM model,the RMSE and MAE were reduced by 42.1％and 41.9％,respectively,and the R2 value increased by 6.8％.The research demonstrates that the proposed multi-head self-attention BiLSTM model outperforms Multilayer Perceptron(MLP),Support Vector Regression(SVR),and LSTM in terms of prediction performance,providing an effective reference for monitoring and early warning of dissolved oxygen in factory-based recirculating aquaculture systems.