I. Introduction
Neuro PID controllers, a deep learning controller, and a lot of artificial intelligence based have been used as controllers in the recent times in control system. In traditional AI applications of neural network methods in image processing, video processing, text recognition etc. use many input data in the form of pixels and texts focusing on accuracy of recognition rather than processing time. However, Neuro PID controller uses fewer inputs such as errors, input signals and outputs of PID controller to obtain high following capability and less processing time. Thus, the need for faster convergence is important in the Neuro PID controller. Weight initialization methods of neural network such as He initialization and Xavier initialization have been proven to be effective in neural networks for faster convergence, however, it has not been implemented in the Neuro PID Control. This paper investigates weight initializations of neural network in Neuro PID Controller to show the importance of weight initialization by case studying with zero initialization, constant initialization, gaussian distributed initialization, uniform distributed initialization, He initialization and Xavier initialization in typical first-order lag elements, Integrator elements and dead time elements to obtain suitable weight initialization, which reduces settling time.