Variable Sampling Compensation of Networked Control Systems With Delays Using Neural Networks.

摘要

In networked control systems (NCS) information or packets usually flow from a sensor or a set of sensors to a remotely located controller. Then the controller processes the received information and sends a series of control commands to the actuators through a communication network which could be either wireless or wired. For any type of communication network, time delays are an inherent problem and depending on the conditions of the network they can be constant, variable or even of random nature. Time-delays occurring from sensor to controller and from controller to actuators may cause important system performance degradation or even instability. This work proposes a novel strategy of using the predictive capabilities of artificial neural networks (NN), particularly the application of an adaptive NN, to minimize the effects of time delays in the feedback control loop of NCS. We adopt an adaptive time delay neural network (TDNN) to predict future time-delays based on a given history of delays that are particularly present on the network where the corresponding system belongs to. The adaptive nature of a TDNN allows the prediction of unexpected variations of time-delays which might not be present in the training set of a known history of delays. This is an important characteristic for real time applications. Using predicted time delays, different methodologies can be used to alleviate effects of such delays on NCS. Our focus here is on the development of an observer-based variable sampling period model, and this dissertation describes how this method can be used as an effective solution for this problem. Generally speaking, the predicted time-delay values are used for the discretization of a continuous-time linear time invariant system model transforming it into a discrete-time linear time variant system model. In this dissertation, the practical phenomenon of packet dropout is also addressed.