Most industrial processes have inherent uncertainties, and are non-linear. The use of the conventional controller is particularly difficult if the system model is slightly known or unknown. In the control of complex nonlinear systems, the adaptive T-S fuzzy system shows high performance. However, it is difficult to train the fuzzy system in practice. This paper proposes a new two-phased training method for a Neuro-Fuzzy controller using reinforcement learning. The methodology is comprised of two distinct phases. First, the premise parameters and rules of the Neuro-Fuzzy controller are explored and initialized. Second, the controller undergoes training via reinforcement learning. The test results demonstrate that our proposed technique leads to a significant decrease in training time and a notable enhancement in the controller’s performance.