The main contribution of this paper is the combination of a novel online training neural network-based algorithm for wind speed estimation with block back-stepping scheme to regulate the optimum equilibrium point of wind turbine system. The wind speed is calculated with the optimal mechanical torque value, that is approximated with a neural network identifier. A real-time dynamic nonlinear learning law (as opposed to off-line training pro-cedure) of the weight vector is proposed for the neural network , and uniform asymptotic stability of the error origin is proved using Lyapunov arguments. The nonlinear learning law makes that, the neural network can approximate very fast changing data. In this form, off-line training with extensive input-data is not necessary. Also, good accuracy in any operation condition is guaranteed and continue learning is achieved. A Block-Backstepping controller is derived in order to regulate the optimum equilibrium point. Uniform asymptotic stability of the tracking error origin for the overall system is proved using Lyapunov arguments, and the performance of this controller is compared with an SPBC that was proposed for the same system.