Self-tuning Fusion Estimation Of Multi-sensor Networked Stochastic Uncertain System

With the development of network communication technology and automatic control technology,the mutual penetration between computers,communications and control,networked control systems have emerged at the historic moment,and have broad application prospects.Due to the expansion of the system scale,the control system has become more and more complicated,and the limited network resources have brought a lot of uncertainty to the networked system.In an actual networked system,uncertainties such as random delay,attenuation or loss appear in control and measurement information due to the unreliability of network communication.In addition,due to unmodeled dynamic and unknown environmental interference,the system model parameters and noise statistics are also often unknown.The uncertainty existing in networked systems must be identified,and a self-tuning estimation theory is proposed.This paper mainly studies the self-tuning fusion estimation of multi-sensor networked systems with unknown model parameters,unknown noise variance,unknown packet dropouts rate and unknown data transmission delay rate.The main research contents are as follows:A matrix-weighted self-tuning distributed fusion state estimation algorithm is proposed for multi-sensor networked linear stochastic discrete systems with unknown packet dropouts,noise variances and model parameters.The Bernoulli random variable is used to describe the packet dropout phenomenon,and the problem of identifying the packet dropout is transformed into the problem of identifying the model parameters through state augmentation.The recursive extend least squares(RELS)algorithm was used to identify both the unknown model parameters and packet dropout rate,and the correlation function method was used to identify the unknown noise variances.Finally the identified packet dropouts,model parameters and noise variances are substituted into the matrix weighted optimal distributed fusion filter to obtain the corresponding self-tuning distributed fusion filter.Using analysis method of dynamic error system and dynamic variance error system,self-tuning filters converge to the corresponding optimal filters,i.e.,self-tuning filters have the asymptotic optimality.For multi-sensor linear discrete stochastic system with unknown one-step delay rate and model parameter,a matrix-weighted self-tuning distributed fusion state estimation algorithm is proposed.The original system is transformed into a system with random parameter in measurement matrix through augmenting state,and further into a system with unknown parameter matrix and virtual additive noise with unknown variance.The least squares method and the correlation function method were used to identify the model parameter and the one-step delay rate,and the identified results were substituted into the optimal filtering algorithm to obtain the corresponding self-tuning distributed fufion filter.Finally,asymptotic optimality of self-tuning filtering algorithms is analyzed.For state or/and measurement stochastic uncertain systems with multiplicative noise,a self-tuning fusion estimation algorithm is proposed.Discrete system with multiplicative noise and unknown variance in the state matrix and additive process noise with unknown variance are transformed into system with known state matrix and vitual process noise with unknown variance through model transformation.The variance of the innovation and the sampled innovation are used to identify the variance of the vitual process noise.Furthermore,a local self-tuning state filter is obtained,and covariance intersection(CI)method and matrix weighted fusion methods are used to obtain self-tuning state fusion filter.Discrete system with measurement multiplicative noise and unknown measurement noise variance and unknown measurement noise variance are transformed into system with known measurement matrix and virtual measurement noise with unknown variance through mosel transformation.The correlation function method is used to identify the variance of virtual measurement noise,and then the corresponding local and fusion self-tuning state filters are obtained,which proves that the self-tuning estimation algorithm converges to the optimal estimation algorithm.For the discrete system with multiplicative noise and additive noise in state and measurement matrices and unknown noise variance,a self-tuning state estimate is obtained by combining the two methods described above.For the multi-sensor asynchronous uniform sampling system with unknown packet dropout and model parameters,first the state iteration method is used to synchronize the asynchronous sampling system.Based on the synchronization model,the local filters at the measurement sampling moment and the state update moment are given.The least square method is used to identify the unknown parameter,and the correlation function method is used to identify unknown packet dropout.The identified model parameters and packet dropouts are substituted into the optimal filtering algorithm to obtain the corresponding self-tuning local filters and self-tuning fusion filter.Finally,asymptotic optimality of self-tuning filtering algorithms is analyzed.