When the control function of a variable universe fuzzy controller is transmitted to the offspring, there are usually some 'distortions' which lead to the error of the algorithms. To solve this problem, this paper proposes a novel optimal method of variable universe fuzzy control based on Q learning algorithm. This algorithm gives an idea of adjusting universes by contraction-expansion factors and geometric proportional factor, and then optimizing the parameters through Q learning algorithm to minimize the performance indexes of the controller for the purpose of reducing the 'distortion rate' in the control process, and improving control performance. Finally, this paper applies the algorithm to non-minimum phase system. Results indicate that this algorithm not only has good robustness and dynamic performance but also has better control performance than the variable universe fuzzy controller.