On the Implementation of a Predictive Outage Compensation

Abstract: Nowadays wireless networks are used widely in industry due to the advantagesthat they provide. However, these networks are not as reliable as wired networks.This unreliability is the result of problems like delays, packet lossesand limited bandwidth. One should take care of these problems in advance,to get the best performance of the systems.A packet loss, which is the main focus of this thesis, occurs when a packet ofdata gets lost, either from the sensors to the controller or from the controllerto the actuators in the process. There are many methods to deal with packetlosses. In some methods it is tried to improve the connection, to get lesspacket losses. In other methods, a predictive outage compensator (POC) canbe used before the actuator or the controller to compensate for the eects of thepacket losses on the network control system's performance. The POC sends areplacement signal if it does not receive the sent data in the appropriate time.In this thesis, we consider a system with an unreliable wireless network connectionbetween the controller and the actuator. We study the eect of threePOC methods on the performance of this system: Zero (where a signal witha value equal to zero is sent to the system during the time the connection islost), Hold (where the last received signal is used instead of the lost signal)and optimal method. In our proposed method, knowing some parameters ofthe system together with the monitoring of the control history, we estimatethe lost signal. To nd the systems parameters (or to initialize the system), weuse two dierent system identication methods, oine (or batch) and online(or recursive).To model the connection, we consider two packet loss models, outages and theGilbert Elliot model. An outage is dened as a period when several consecutivepackets are lost.  We show that the optimal method improves the performance of the networkedcontrol system remarkably, compared to the other two methods. The conclusionsare veried by simulation results.