The relative eddy current approximation error characteristics of the fluid in centrifugal pumps result in external disturbances such as water flow fluctuations and changes in pipeline resistance. A single flow control method is difficult to adapt to the uncertainty of the centrifugal pump"s operating state, leading to overshoot problems and even oscillation, resulting in poor control effectiveness. Therefore, a fuzzy self disturbance rejection control method based on multivariable neural network PID is proposed. Analyze the flow characteristics of centrifugal pumps in hydraulic engineering and calculate the total flow rate of centrifugal pumps. Construct a multivariable neural network PID fuzzy self disturbance rejection control structure, design a first-order inertia link, calculate the error signal through a negative feedback mechanism, adjust the equivalent gain and inertia time based on the error signal, and achieve preliminary control of the variable centrifugal pump flow rate. By finely adjusting the control gains of kp, ki, and kd, self-tuning second-order parameters, and facing the complex and variable operating states of centrifugal pumps, pole placement and cyclic adjustment strategies are used to avoid negative effects such as overshoot and oscillation caused by external disturbance parameters on the control process, achieving centrifugal pump flow control. In the experiment, three different pipe gallery positions were set up to simulate interference states such as different water flow pressures. The application of this method to control the results showed that the maximum flow rate did not exceed 105m3/h, 102m3/h, and 103m3/h at all three positions, and the centrifugal pump flow rate could be effectively and stably controlled.