In this paper, we propose an intelligent control method that integrates fuzzy mathematical theory and generative adversarial network (GAN) to address the problems of data scarcity, model complexity and environmental uncertainty in modern control systems. The system fuzzy parameters are quantified by the affiliation function of the fuzzy set, combined with the adversarial training framework of the GAN, and the generatordiscriminator’s minimal-extremely large game is used to dynamically generate high-fidelity data and optimize the control strategy. In the experiments in the field of electrical engineering, the simulated temperature rise of 69.41K at the C-phase temperature measurement point of the temperature rise control of the high-voltage switchgear cabinet has an error of only 1.5K (the allowed value is 72K) with the actual value of 67.91K, which verifies the model accuracy. The response time of fuzzy GAN controller for intelligent speed control of fan is more than 50% shorter than that of traditional GAN, and the amount of overshooting is significantly reduced. In permanent magnet synchronous motor control, fuzzy GAN reduces the steady state error by 67%-82% (from 2.07% to 0.55% under sudden load change condition), speeds up the regulation time by 45%-50% (from 80.37ms to 40.25ms for rated startup), compresses overshooting by 55%-58%, and improves the efficiency by 2.66-3.75 percentage points. The average loss of fuzzy GAN in coal mine control system energy consumption is only 51.16 kW/h in 8 wiring lines, which is 81.2%-85.0% lower than that of 319.12 kW/h in traditional GAN, 341.8 kW/h in integrated AI technology and 272.3 kW/h in PID control, and the energy consumption in high load scenario (Y6) is only 10.7% of the comparison method. It is shown that the proposed method effectively breaks through the bottleneck of traditional control in terms of accuracy, response speed and energy consumption through the adaptivity of fuzzy set and the dynamic optimization ability of GAN.
