Improved Multiple Model State Estimation Algorithm And Application

To obtain the accurate observation of the system is an important approach to ensure the stability and reliability of the complicated system and the accurate execution of tasks.However,the system state can be observed only under some specific noise conditions,and the online sensors for the special variables are generally expensive.Therefore,it is important to obtain the estimated value of the system state by processing the signal with noise.One of the most powerful method is multiple model estimation,it is an adaptive state estimation method with superior performance.It uses multiple models that match the different modes of the system to describe the possible operating mode changes or structural changes of the system.Each model has a matching filter,and the overall system state estimation is obtained by fusing the output of each sub-filter.Therefore,the multiple model algorithm is particularly suitable for solving state estimation problems with unknown parameters or changing modes.The researches on multiple model state estimation relate to many fields such as system monitoring,maneuvering target tracking,image recognition and fault diagnosis.Based on the multiple model state estimation theory,this article mainly focuses on the improvement of model set selection and estimation algorithm as well as its practical application,and the main innovations are as follows:(1)Firstly,aiming at the selection of model set in the multi-model state estimation method,considering that when the model set of the system is large or the modal space of the actual process is unknown,there may be models in the model set determined by the priori information that are very different from the operating state of the system,this situation will lead to the performance degradation of the multi-model estimation algorithm.A multiple model state estimation method based on simplified model set is proposed.Based on the Bayesian principle,the actual modal space of the system is estimated by the historical observation values,and the corresponding model sets are obtained.Then,the simplified model set is used to estimate the system state,so as to reduce the interference caused by the models that do not match the actual operating state of the system,and improve the estimation accuracy.(2)Secondly,aiming at the detection problem of toxin concentration in effluent of sewage treatment process,the changes in the number of living cells exposed to the effluent samples of the sewage treatment process is considered as an indirect indicator to expand the existing cytotoxicity model.Different models are used to describe the cytotoxic process under various water qualities,and these models constitute the model set describing the process dynamics.Each model corresponds to a mode,and different modal transitions obey the Markov process,which constitutes a stochastic hybrid model of cytotoxicity.This model can not only describe a variety of possible dynamic processes caused by unclear mechanisms and difficult experimental tests in the process of cytotoxicity,but also help to obtain data that are difficult to measure in experiments to improve the scope and accuracy of the model.(3)Finally,based on the established stochastic hybrid model of cytotoxicity,the interactive multiple model particle filter algorithm is used to obtain the unmeasurable intracellular and extracellular toxin concentration from the noisy observation(cell number)to realize the estimation of the toxin concentration in the environment.As a multiple model state estimation method,the interactive multiple model particle filter algorithm can ensure good estimation effect,high accuracy and speed under the condition of stochastic changes of process modes.Compared with the estimation results by using a single model shows that the proposed method in this paper can adapt well to the stochastic changes of external conditions and realize the online estimation of the toxin concentration of the effluent of the sewage treatment process in real time.