Objective: To improve the energy efficiency of mango drying in the air source heat pump system so as to save energy. Methods: The process of drying mangoes was subdivided, and a variable structure control was used to adjust the temperature and humidity of drying room intelligently and dynamically to improve energy efficiency. Each drying process stage was divided into three parts, namely far away from the conversion point, near the conversion point, and closing to the conversion point. For the first two parts, a constrained nonlinear autoregressive neural network (NARX) with external inputs was used to intelligently adjust the temperature and humidity settings so as to save electricity, while for the third part, a PI controller was used to accurately control the dehumidification amount at the conversion point of the drying process so as to ensure the quality of mango drying. Results: Compared with conventional segmented constant temperature and humidity drying methods, the proposed control method could save 8.63% of electricity with a guaranteed quality of mango drying. Conclusion: The proposed subdivided variable structure control method can significantly improve the energy efficiency of heat pump drying systems, and achieve drying quality similar to conventional segmented constant temperature and humidity methods.