Central air-conditioning systems predominantly operate under partial load conditions. The optimization of a differential pressure setpoint in the chilled water system of a central air-conditioning system leads to a more energy-efficient operation. Determining the differential pressure adjustment value based on the terminal user’s real-time demand is one of the critical issues to be addressed during the optimal control process. Furthermore, the online application of the differential pressure setpoint optimization method needs to be considered, along with the stability of the system. This paper proposes a variable differential pressure reset method with an adaptive adjustment algorithm based on the Mamdani fuzzy model. The proposed method was compared with differential pressure reset methods with reference to the chilled water differential temperature, outdoor temperature, and linear model based on the adjustment algorithm. The energy-saving potential, temperature control effect, and avoidance of the most unfavorable thermodynamic loop effects of the four methods were investigated experimentally. The results indicated that, while satisfying the terminal user’s energy supply demand and ensuring the avoidance of the most unfavorable thermodynamic loop, the proposed adaptive adjustment algorithm also decreased the differential pressure setpoint value by 25.1%-59.1% and achieved energy savings of 10.6%-45.0%. By monitoring the valve position and supply air temperature of each terminal user, the proposed method exhibited suitable online adaptability and could be flexibly applied to buildings with random load changes.