As the air compressor consumes a large amount of electric energy while providing driven-energy, its energy saving and efficiency enhancement have attracted much attention. The demand of compressed air is with the property of periodicity and intermittency due to the production rhythm, planned scheduling and other factors, which might cause the operational mode change of air compressor units in a wide range. The manual scheduling has the property of slow speed and energy consumption. Thus, aiming at the air compressor system of an industrial park, a coordinated predictive control method of air compressor units based on fuzzy reinforcement learning is proposed. First, a multiple operational mode model of compressed air group based on fuzzy identification model is structured by using the historical data of industrial site and equipment operation mechanism. On this basis, a variable load predictive control method of the air compressor group based on fuzzy reinforcement learning is established to ensure the safe and stable operation of the system under complex operational modes, which aims to minimize the energy consumption in the production process and combines the production process conditions, equipment safety and other constraints. Finally, the proposed method is verified by the air compression simulation system in an industrial park and a good control result is achieved.