The model-free adaptive algorithm is a data-driven method, which does not require an accurate model of the system, has small calculation cost and is easy to implement. The existing model-free adaptive control algorithm often adopts the cut-and-trial method or uses a fixed constant when selecting the penalty factor, which is difficult to achieve a satisfactory control performance. To solve this problem, this paper proposes a method to online optimize the penalty factor of the control law and the pseudo partial derivative. By using the idea of iterative optimization in the steepest descent method, the penalty factor is optimized, the convergence speed is increased significantly, and better system performance parameters are obtained. Hence, the optimal performance of the system can be achieved. On this basis, the stability of the closed-loop system is strictly proved. Finally, Matlab simulation results show that the proposed method has better control quality and stronger anti-disturbance abilities than the existing model-free adaptive control method.