Unknown Input Estimation For Multi-agent Systems Based On Distributed Filters

As a special kind of control system,the multi-agent system technology could solve many complex tasks which could not be solved by the single system based on the same unit calculation capacity.The most important chasteristic of the multi-agent system is the information-sharing between the agents.In industry,there are many kinds of unknown inputs such as faults or disturbances,which could affect the stability of the multi-agent system.The unknown input estimation is a fundamental issue for the multi-agent systems.In this study,some distributed-filter-based methods are proposed to estimate the unknown inputs for the multi-agent systems,the specific studies and results are listed below.In Chapter 2,a distributed cooperative minimum-variance unbiased(MVU)filter is proposed for a linear discrete-time multi-agent system with homologous unknown inputs.Compared with the conventional dencentralized states and unknown inputs filter,the distributed cooperative filter has a much looser existence condition and better estimation performance.In Chapter 3,the distributed cooperative MVU filter proposed in chapter 2 is modified to a distributed semi-cooperative MVU filter.Compared with the distributed cooperative filter,the improved semi-cooperative filter needs less calculation quantity,less communication and better estimation performance.The asymptotic stability of the improved filter is also proved.The simulation results verified the asymptotical stability of the distributed semi-cooperative filter.Chapter 4 focuses on a linear discrete-time multi-agent system with both homologous unknown inputs and heterogeneous unknown inputs.To get the MVU estimation of the homologous unknown inputs as well as the heterogeneous unknown inputs,the filter needs to decouple the homologous unknown inputs and the heterogeneous unknown inputs,but the decoupling condition is very strict.To relax the existence condition of the distributed semi-cooperative filter,the following two improvements are proposed:first,the communication method of the distributed filter is improved to make every filter utilize the global information;second,the decoupling method of the heterogeneous and homologous unknown inputs is improved.Chapter 5 deals with the nonlinear multi-agent system with unknown inputs.First,based on the extended Kalman filter(EKF)method,a nonlinear distributed semi-cooperative MVU filter is proposed for the multi-agnet system with homologous unknown inputs.Second,the unscented Kalman filter(UKF)-based distributed semi-coopeative MVU filter is proposed for the multi-agent system with homologous unknown inputs.Finally,the EKF-based and UKF-based distributed semi-cooperative MVU filters are proposed for the multi-agent system with heterogeneous unknown inputs.The usability and the performance of these two kinds filters are compared.In Chapter 6,a practical experiment is conducted to verify the effectiveness of the filter.This chapter models the LiFePO4 battery pack and designs the nonlinear distributed semi-cooperative MVU filter to estimate the ambient temperature of the battery pack.The results validate the effectiveness of the filter.As the filter only uses the terminal voltage of the battery instead of using the thermocouples to estimate the ambient temperature,the filter has great application value in industry.