Neural Network Weight Initialization

Proper initialization of neural network weights is critical problem. Many methods have been proposed for initialization of neural network weights. In this paper a direct neural control strategy is used to control the process. The study of effect of initialization of weights in neural network control structure is carried out. The paper describes performance comparison of the controller with reference to continuous stirred tank reactor (CSTR) which is a nonlinear process and simple linear DC Motor system for a unity and random weight initialization. The performance comparison of these two neural controller configurations has been given in terms of ISE and IAE.