A method for regulating dissolved oxygen in industrial aquaculture based on predictive PID

Water quality is very important to aquaculture, among which DO, as a key water quality parameter, directly affects the survival and growth of organisms, and is crucial to improve the efficiency of aquaculture. Especially in high-density intensive culture modes, to ensure sufficient dissolved oxygen content, liquid oxygen is frequently employed as an oxygen source during the control process, leading to higher costs. PID controller is widely used in DO control because of its simple structure, strong robustness and fast response. However, in the face of nonlinear systems, fixed parameters are difficult to adapt, or the performance deteriorates and the system is unstable. Researchers aim to integrate intelligence into PID for online parameter tuning, enhancing adaptability and robustness. In addition, some studies have integrated predictive technology into PID control to predict the future dynamics of the system, accurately calculate the control input combined with the current state, effectively respond to system changes, and reduce delay or interference effects. In the process of DO control, the DO data input is dynamic, and the environment changes easily cause the input-output relationship to change with time. The online prediction model can predict the future dynamic changes of the system more accurately, and provide strong support for the real-time adjustment of PID controller. Firstly, the transfer function model of the relationship between aeration flow and DO concentration was established by experimental modeling, and the parameters were determined by the experimental data of liquid oxygen aeration to obtain the DO control system model. Then, on the basis of PID controller, a quantum neural network (QNN) PID controller (FSSCINET-QNN-PID) based on fast and slow learning sample convolutional interactive network prediction (FSSCINET) is proposed to improve its fast response ability and anti-interference ability. In FSSCINET-QNN-PID, FSSCINET dissolved oxygen online prediction model is designed, and adaptors and memory modules in fast and slow learning networks (FSNET) are introduced into sample Convolutional interactive network (SCINET) to improve the model's adaptability and prediction accuracy in dynamic environment. DO is a kind of time series data with time series features. SCINET uses convolution and interaction to extract features and exchange information for data with different time resolutions, thus improving the prediction accuracy. The adaptor can dynamically train model parameters according to changes in the factory farming environment, reducing prediction errors caused by external changes. The memory module can remember periodic changes during breeding, helping the model adapt quickly and accurately predict future DO concentrations. QNN has strong learning ability and has advantages in controlling nonlinear dynamic systems. Therefore, QNN is used to update the PID control parameters online to improve the adaptive ability of the controller. Finally, a DO monitoring system is set up in the factory farming environment, which can collect DO data in real time and automatically regulate the DO aeration flow, verifying the prediction performance of FSSCINET and the effect of FSSCINET-QNN-PID controller. The results show that compared with SCINET and FSNET, FSSCINET has a better prediction effect, and its mean square error, mean absolute error and root mean square error are 0.037 5 mg l-1, 0.155 4 mg l-1 and 0.193 7 mg l-1, respectively. In the DO setpoint tracking simulation experiment, compared with PID and QNN-PID, the proposed FSSCINET-QNN-PID adjustment time is reduced to 1 642s, and the overshoot is smaller. In the DO control experiment, the FSSCINET-QNN-PID controller reaches stability faster than the conventional PID controller and can maintain small fluctuations. The results show that FSSCINET-QNN-PID has better prediction accuracy and adjustment performance in DO control. This study can provide a new idea for automatic regulation of DO in factory farming.