Research On Systematic Biases Registration Method Based On Optimal Weight Network

In the multi-sensor monitoring system,information such as the position and speed of the target is measured through sensor.As the technology advances,the performance of sensors has also been greatly improved.However,in the long-term use of the sensor,it is inevitable that the measurement information will has biases due to the aging of its internal components and the influence of the external environment,which will affect the subsequent monitoring performance.In actual engineering applications,the source of the sensor's systematic biases is often unknown,making it difficult to construct its systematic biases model.Therefore,this paper takes the sensor systematic biases registration with unknown systematic biases prior model as the research background,takes the weight network(WN)as the research object,uses the optimization algorithm(or filtering algorithm)as the technical means to optimize WN,and develops sensors research on systematic biases registration.This paper has carried out related research work on some problems in systematic biases registration,including the following aspects:Unscented Kalman filtering and traditional network are combined to carry out systematic biases registration,aiming at the problem of system divergence caused by the negative determination of state covariance,this paper proposes to use the square root unscented Kalman filter algorithm(SRUKF)to optimize WN perform systematic biases registration——SRUKF-WN.First,the SRUKF-WN algorithm constructs a WN systematic biases estimation model including network connection weights,sensor measurements,and systematic biases.Second,the state space model of the SRUKF-WN algorithm is established,which seem the WN network connection weights as the state quantity of the SRUKF algorithm,and the output of WN,that is,the systematic biases as the observation.It updates the estimated WN network connection weights through the SRUKF algorithm,and then estimate the sensor systematic biases.Finally,the systematic biases estimates are used to register the sensor measurements.The SRUKF-WN algorithm transfers the square root matrix of the state covariance,avoiding the situation of negative definite state covariance.The simulation results show that proposed algorithm has no divergence,and it has faster convergence speed and shorter algorithm running time.Aiming at the problem that the gradient descent algorithm will cause the phenomenon of "sawtooth",this paper proposes to use particle swarm optimization(PSO)to optimize WN and then perform systematic biases registration——PSO-WN.The PSO-WN algorithm establishes a mapping relationship between the network connection weights of WN and the individual particles of the PSO algorithm.It updates and optimizes the network connection weights of WN through the PSO algorithm to obtain the optimal estimate of network connection weight and then the optimal estimate of the systematic biases.Then the systematic biases estimates are used to register the sensor measurements.The simulation results show that the proposed algorithm utilizes past empirical knowledge when solving problems and does not depend on gradient information,thereby avoiding the "sawtooth" phenomenon of systematic biases estimation and improving the estimation accuracy.Aiming at the problem of PSO-WN algorithm's decrease in population diversity in the late evolution,based on the combination of PSO and DE algorithm(PSO-DE),this paper proposes to use improved PSO-DE algorithm to perform systematic biases registration for WN optimization——OSPSO-DE-WN.In the OSPSO-DE-WN algorithm,an optimization selection strategy is added to reduce unnecessary information exchange error.The algorithm also uses the improved PSO-DE algorithm to update and optimize the WN network connection weights to obtain the optimal estimate of network connection weight,apply it to WN to obtain the optimal systematic biases estimate,and use it to register the sensor measurements.The simulation results show that proposed algorithm not only increases population diversity,but also improves algorithm accuracy and convergence speed.