When the neural network is applied to a dynamic system controller, the stability of the neural network controller must be guaranteed. Stability is related to the optimum sigmoid function shape. An autotuning method for the optimum sigmoid function is proposed. The automating method is applied to a learning type direct controller in order to confirm its characteristics. The neural network of this controller has three layers with no inner feedback loop and no direct connection from the input layer to the output layer. Both the input and the hidden layers have four neurons and the output layer has one neuron. Both the hidden and output layers have a sigmoid function to provide the nonlinear mapping capability. The autotuning method uses the steepest descent method in order to apply it to servo control systems. Simulation results using the learning type direct controller confirm that the autotuning method is useful in combination with weight tuning for dynamic systems.
展开▼
