Command-based Autopilot System for Ships using Neural Network - PID Controller

PID (Proportional-Integral-Derivative) control is one of the most popular approaches that widely applied in the control aspect, due to its simple concepts and efficiency. However, unless users are really experienced or already knew all about the system, it is very difficult to define the best gains {K-p, K-i, K-d} theoretically that could satisfy their demand the most, since the feedback data always includes unexpected noises. One of the most typical examples is the autopilot system for ships, where most of noises are random in amplitudes and not periodical. From that point, in this paper, we manage to build an algorithm that helps the conventional PID controller self-tune the gains to minimize the tracking directional error by combining Neural Network and PID into a controller. First of all, we would make a mathematical model for a ship by using dynamic equations and Nomoto's model. From that model, the next thing we would do is to design a neural network that acts like a PID controller, and then to analyze how the gains in the network can learn themselves, with proper optimizers. Finally, we would implement the network with those optimizers to simulate, using MATLAB, and evaluate the response as well as compare with each other.