Filtering And Data Fusion For Systems With Incomplete Information

With the advanced development of the computer and communication technology, com-puter network technology has been increasingly used to realize the remote installation,control and maintenance of the plant in modern control system. However, due to the phys-ical limitations of communication technologies, the computer network technology has beenconstrained by a lot of issues, like limited communication bandwidth and node energy,dynamic topology, etc. These constraints lead to many problems during the signal acquisi-tion, transmission and processing, for example, packet dropouts, delay, asynchronous dataand fading and so on, which leads to the deterioration of the system performance due tothe fact that much incomplete information has been received. We need to reduce the in-fuence of the incomplete information as well as share the technological innovation broughtby the computer network in modern control systems. A series of fltering, data fusion stateestimation algorithm aimed at these problems are presented in this thesis research, andthe stability or performance of the proposed algorithm is analyzed. The main results arelisted as follows:Chapter2. Some primary knowledge and common algorithms about fltering anddata fusion state estimate are introduced, which has been used in the following studies,including Projection Theorem, Kalman fltering, information fltering, modifed Kalmanfltering, federated Kalman fltering and information fusion fltering.Chapter3. An optimal flter, predictor and smoother based on a pseudo measurementmodel are presented, and a sufcient condition for the convergence of the proposed flteris given. Two independent Bernoulli variables are used to model the delay and packetdropouts in the networked system, based on which a pseudo measurement model is de-rived. The optimal flter, predictor and smoother are obtained by using the ProjectionTheorem on the pseudo measurement model, and they are only relevant to the statisticalproperties of the delay and dropouts, i.e., the distributions of the two Bernoulli variables,and independent of the exact values of them, so the estimator can be designed of-line.Finally, the simulation results show the efectiveness of the proposed algorithms. Chapter4. This chapter is devoted to the time delay in the multi-sensor multi-channelnetworked system. An optimal centralized fusion algorithm and a suboptimal distributedalgorithm are proposed in a centralized and distributed frame respectively. The buferwith a certain length is used to re-order the received data, and an independent identicallydistributed (i.i.d) random variable is applied to model the time delay process. In thecentralized architecture, an optimal fusion method is obtained through the measurementaugmentation. In the distributed architecture, a suboptimal fusion method is obtained byfusing the local state estimation through the federated Kalman flter. The stability of theproposed algorithms are analyzed at last. Two numerical examples are given to show thevalidity of the proposed techniques.Chapter5. Based on the work of Chapter4, a new centralized fusion algorithmis proposed with the bufer with a certain length. Meanwhile, a probabilistic metric toevaluate the performance of the system is presented, and the relationship between theperformance and the length of the bufer is given under this metric. Compared withthe centralized fusion method in Chapter4, the method in Chapter5can avoid largematrix operations, especially the large matrix inverse operations because of no state ormeasurement augmentation, so it can reduce the computation load efectively and improvethe system operating efciency. Compared with the assumption that no time delay ordropout from the local flter to the fusion center in Chapter4, the method in Chapter5ismore practical. Simulation results show the efectiveness of the proposed method.Chapter6. To fuse information observed by asynchronous multirate sensors, a hybriddata fusion framework is presented. By use of the presented framework, information fromdiferent sensors may be fused efectively. To generate the optimal state estimate, themethod is implemented by prediction and two times update in sequence. The informationobserved by the sensor with the highest sampling rate in the fnest scale is used to updatethe state prediction, and the re-innovation is taken by use of the sensors with lower samplingrates at coarser scales. The process is carried out successively, and the fused state estimateat the fnest scale is generated. The stability of the algorithm is analyzed in the last. Theefectiveness of the algorithm is illustrated through the theoretical proof and simulation results.