Research On Performance Optimization For Networked Control Systems With Transmission Constraints

With the increasing development of wireless sensor and communication technol-ogy, wireless sensor and actuator networks (WSANs) have enabled a wealth of appli-cations. Once the measurement and/or the control channel of a control system is re-placed by WSANs, the resulting system is known as networked control system (NCS).Compared with the conventional wired control, NCS is more attractive due to its manyadvantages, such as reduced cabling and installation cost, easy relocation and rapidreconstruction. Despite these advantages, inserting WSANs into a control system al-so induces new problems: from the perspective of estimation and control, packet losscausedbyunreliabletransmissiondramaticallydegradesthesystemperformance; fromthe perspective of WSANs, limited sensor duty cycle and network bandwidth shouldbe properly scheduled to optimize the system performance.With the considerations mentioned above, this dissertation is aiming at propos-ing appropriate strategies to substantially improve the system performance. The w-hole thesis investigates the estimation error covariance minimization problem over apacket-dropping network; the sensor periodic scheduling problem under both networkbandwidth and sensor duty cycle constraint, as well as the linear quadratic Gaussianstate regulation problem over a packet-dropping network. More details are as follows:1. To deal with the packet loss existing in the network connecting the local sensorand the remote estimator, a novel mechanism called linear temporal coding is pro-posed, in which the linear combination of the current measurement and the one of thelast step is used. For the case that the packet arrival sequence is unavailable, a nov-el two-stage estimation algorithm is proposed using innovation sequence approach. When measurement noise plays the dominating role, simulations show the beneft ofusing measurement combination; By contrary, when communication noise plays thedominating role, we prove there is no beneft to use this coding strategy for frst-ordersystems. For the case that the packet arrival sequence is available, an on-line strategyis proposed. A novel state observer is proposed to better reveal the recursive relation-ships of the error covariances. The equivalency of the two expressions is also verifedand the optimal coding weight is further obtained for frst-order systems.2. The optimal periodic sensor schedules are proposed for a one-plant-two-sensorsystem for minimizing the remote estimation error covariance. When duty cycles oftwo sensors are surplus, the optimal duty cycle pair is derived, based on which is ouroptimal periodic schedule; When duty cycles are not enough, we prove the “as uni-formly as possible” rule still holds true, but the optimal steps of open-loop predictionrely on system parameters. In both situations, our optimal periodic schedules dependon system dynamics, which distinguishes this study from that of the previous one.3. A linear predictive compensation mechanism is proposed to deal with the s-tochastic packet loss existing in the network connecting the local sensor and the remotecontroller. The control command at the actuator side is frstly expressed as the randomcombinations of diferent steps of predictions. On that basis, the information set ofthe system is divided into several subsets and the accurate expression of prediction er-ror covariance of each subset is provided. By this means, the quantitative relationshipbetweeninfnite-horizonquadraticperformanceindexand thepredictionstepsisestab-lished. Simulation verifes that there exists a limited length of the prediction steps tominimize the performance index.