In aquaculture, the dissolved oxygen (DO) content of a water body is important for the growth of aquatic products and thus needs to be precisely controlled. For this purpose, through an open-loop experiment of DO aeration under different airflow rates, empirical transfer function models that can fully describe the dynamic response relationship between aeration flow rate and DO content were established. Based on these models, this study proposed a differential evolution (DE) algorithm-optimised radial basis function (RBF) neural network proportional integral derivative (PID) controller (DE-RBF-PID). The proposed controller has two optimisation parts. The first part is devoted to finding the optimal initial parameters of PID using an improved DE algorithm. The second part utilises the powerful learning ability of the RBF neural network to adjust the PID parameters online, which can not only eliminate overshoots, but also improve the controller adaptability. The simulation results for a typical DO nonlinear control system demonstrated the superiority of the proposed DE-RBF-PID controller over conventional PID and RBF-PID controllers. Therefore, this controller can be applied for precise tracking control of DO in complex circulating aquaculture systems.

在水产养殖中，水体的溶解氧（DO）含量对水产品的生长很重要，因此需要精确控制。为此，通过不同风量下溶解氧曝气的开环实验，建立了能够充分描述曝气流量与溶解氧含量动态响应关系的经验传递函数模型。基于这些模型，本研究提出了一种基于微分进化 (DE) 算法优化的径向基函数 (RBF) 神经网络比例积分微分 (PID) 控制器 (DE-RBF-PID)。所提出的控制器有两个优化部分。第一部分致力于使用改进的 DE 算法寻找 PID 的最佳初始参数。第二部分利用RBF神经网络强大的学习能力在线调整PID参数，既可以消除超调，又可以提高控制器的适应性。典型 DO 非线性控制系统的仿真结果证明了所提出的 DE-RBF-PID 控制器优于传统 PID 和 RBF-PID 控制器。因此，该控制器可用于复杂循环水产养殖系统中溶解氧的精确跟踪控制。