Robust State Estimation And Fault Diagnosis Research Based On Particle Filter For Measurement And Parameter Uncertain Systems

The scale and complexity of modern control system are increasing,and its reliability and safety requirements are getting higher.Accurate state estimation is the key link and necessary foundation for realizing intelligent control and fault prognosis.Particle filter has been widely used in the field of nonlinear and non-Gaussian system state estimation because of the advantage of avoiding system linearity and Gaussian distribution noise constraints.However,the uncertainty of system measurement and parameters,especially the existence of gross error such as bias?drift and outlier may lead to larger deviation,and affect the robustness of state estimation based on common particle filter,when used for fault prognosis,maybe lead to false alarm.In this dissertation,we focus on some key issues in state estimation and fault diagnosis of measurement and parameter uncertain systems.The main research contents include:Firstly,this dissertation discusses the basic idea,significance and application of state estimation and particle filter algorithm,analyzes and compares the advantages and disadvantages of various state estimation methods,summarizes the research status and development trend of particle filter algorithm for state estimation,and expounds the significance of this study;Secondly,this dissertation summarizes the state estimation framework based on GPF and APF,and put forward to the PF-DRGED mechanism for data correction of nonlinear system.The algorithm design idea is expounded and verified by the experimental cases.And the limitations for the coexistance of different types of gross errors are pointed out,which laid a foundation for further research;Thirdly,a robust particle filter GEDI-MC-RPF algorithm which fusing the error identification and measurement compensation is proposed to solve the problem of non-random significant errors such as outliers,systematic biases and drifts caused by measurement uncertain systems.Firstly,the abnormal measurement is detected based on the residuals of the observed functions,and then the different error mathematical feature are used to identify the error types and estimate the magnitude.Then,through the targeted compensation scheme,the influence is eliminated and its weight is updated.Finally,the updated state is deduced.In order to verify the accuracy and validity of the proposedalgorithm,a progress industrial example is used to simulate and compare it with the traditional GPF and APF algorithm results.The results show that GEDI-MC-RPF has better robustness to the system state estimation with uncertain measurement errors such as outliers,systematic biases and drifts;Fourthly,this dissertation analyzes the existing problems in the field of state parameter estimation based on particle filter,and the problems existing in Artificial Evolution and Kernel Smoothing: it is impossible to detect instantly and track changes in model parameters due to the change of system operating conditions or error occurred.An improved RPF-SPE method for robust parameter estimation based on particle filter for uncertain system is proposed.The tracking parameter is identified by the test criterion based on measurement error,and the variance of the particle is modified according to the changing parameter,and finally realizing the reliable SPE by particle iteration.The effectiveness of the proposed method is verified by a typical nonlinear dynamic system with variable parameters,which setting device and sensor faults respectively,and compared with the traditional methods for targeted.The results show that the proposed RPF-SPE method can effectively track the change of parameters when a system fault or operating condition changes,so as to achieve robust estimation to state and parameter accuracy;Lastly,the problem of fault diagnosis of measurement and parameter uncertain systems based on state estimation is studied.A robust fault diagnosis strategy based on the improved particle filter algorithm RPF-FDD is proposed,and the complex distributed data collection network based on wireless sensor network is designed,a method to construct distributed intelligent fault diagnosis system is put forward,which is applied to the hybrid tire brake energy feedback braking system,the effectiveness of the proposed robust fault diagnosis strategy is verified by setting the corresponding simulation.The experimental results show that the strategy can effectively identify the parameters and the system fault types under uncertain measurement.In this dissertation,we focus on several key problems of state estimation and fault diagnosis based on particle filter,and proposed a robust particle filter GEDI-MC-RPF algorithm for measurement uncertain systems,and the RPF-SPE method for robust parameter estimation of uncertain parameter systems.The effects of uncertain significant errors anddynamic changing parameters on the iterative process of particle filter and the estimation results are solved.A distributed data acquisition network is designed,and the robust fault diagnosis strategy RPF-FDD based on particle filter is proposed.The experimental simulation results show that the proposed method achieves the expected goals and requirements.