CN-121978616-A - Unknown correlation multi-sensor state and noise covariance joint estimation method

CN121978616ACN 121978616 ACN121978616 ACN 121978616ACN-121978616-A

Abstract

The invention discloses an unknown related multi-sensor state and noise covariance joint estimation method which comprises the following steps of S1, initializing a multi-sensor system model and parameters of local Kalman filters of all sensors, S2, predicting and updating target states of all sensors based on the initialized system model and the parameters of the filters to generate and accumulate an innovation sequence, S3, respectively estimating and obtaining local process noise covariance and local measurement noise covariance based on a sliding time window with a preset length based on the accumulated innovation sequence, S4, distributing fusion weights for all the sensors based on the estimated local noise covariance, and S5, carrying out weighted fusion on the local process noise covariance and the local measurement noise covariance based on the fusion weights distributed by S4 to obtain global process noise covariance estimation and global measurement noise covariance estimation.

Inventors

LI JIAHONG
ZHENG YING

Assignees

北京联合大学

Dates

Publication Date: 20260505
Application Date: 20260123

Claims (8)

1. The method for jointly estimating the state of the unknown related multisensor and the covariance of the noise is characterized by comprising the following steps of: S1, initializing a multi-sensor system model and parameters of a local Kalman filter of each sensor; s2, based on the initialized system model and filter parameters, each sensor predicts and updates a target state, and generates and accumulates an innovation sequence; S3, based on the accumulated innovation sequence, each sensor respectively estimates and obtains a local process noise covariance and a local measurement noise covariance based on a sliding time window with a preset length; S4, distributing fusion weights for the sensors based on the estimated local noise covariance; and S5, carrying out weighted fusion on the local process noise covariance and the local measurement noise covariance based on the fusion weight distributed in the S4, and obtaining a global process noise covariance estimation and a global measurement noise covariance estimation.
2. The method for combined estimation of unknown correlated multisensor states and noise covariance according to claim 1, wherein S1 comprises: Establishing a state space model and a measurement model of target motion; Initializing state estimation values and an estimation covariance matrix of a local Kalman filter of each sensor; an initial process noise covariance matrix and an initial measurement noise covariance matrix are set for each sensor local Kalman filter.
3. The method for estimating the state and the noise covariance of the unknown relevant multiple sensors according to claim 1, wherein the step S2 comprises the steps of obtaining the observation value of the current moment by each sensor, predicting the target state and the predicted observation by utilizing a time updating step of a local Kalman filter according to a state space model constructed in the step S1, calculating the observation information, executing a measurement updating step of the local Kalman filter based on the observation information to obtain corrected state estimation and updated estimation covariance, and adding the observation information into an information sequence of the local Kalman filter to accumulate.
4. The method for combined estimation of unknown correlated multisensor states and noise covariance according to claim 1, wherein S3 comprises: The method comprises the steps of collecting an innovation sequence in a sliding time window for each sensor, calculating a sample autocovariance set of the innovation sequence, constructing a linear equation set of a process noise covariance unknown quantity and a measurement noise covariance unknown quantity based on a state transition matrix, an observation matrix and a current Kalman gain matrix of a system, and solving the linear equation set to obtain local process noise covariance estimation and local measurement noise covariance estimation of the sensor.
5. The method for jointly estimating unknown correlated multi-sensor states and noise covariances according to claim 1, wherein S4 comprises calculating an estimated variance upper bound based on residual statistics of local noise covariances of the sensors, and assigning fusion weights to the corresponding sensors according to the ratio of normalized reciprocals of the estimated variance upper bounds.
6. The method for combined estimation of unknown correlated multisensor states and noise covariance according to claim 1, wherein S5 comprises: taking as input the local process noise covariance estimate of each sensor and its corresponding uncertainty measure, and the local measured noise covariance estimate of each sensor and its corresponding uncertainty measure; and (3) applying a batch covariance cross fusion formula, and calculating to obtain global process noise covariance estimation and uncertainty thereof and global measurement noise covariance estimation and uncertainty thereof by combining fusion weights distributed in the step (S4).
7. The method of claim 1, further comprising using the global process noise covariance estimate and the global measured noise covariance estimate obtained in step S5 to update corresponding noise covariance parameters in each sensor' S local Kalman filter, respectively.
8. The method of claim 7, further comprising adaptively adjusting an image segmentation threshold in a target detection segment based on the updated global process noise covariance estimate, wherein the segmentation threshold is positively correlated with a norm or weighted sum of process noise covariance.

Description

Unknown correlation multi-sensor state and noise covariance joint estimation method Technical Field The invention relates to the field of signal processing, in particular to a method for jointly estimating the state of multiple unknown related sensors and the covariance of noise. Background The kalman filter has optimal performance in the state estimation of a linear gaussian system, however in many practical systems, such as low cost integrated navigation positioning systems, energy direction-based target positioning, fault tolerant control systems, etc., the covariance of the process noise and measurement noise is often unknown or varies with time, and offline calibrated noise parameters may also fail. Incorrect setting of the noise covariance can significantly impair the filtering effect, leading to increased state estimation bias and even filter divergence when severe. In order to estimate unknown noise covariance online, a variety of algorithms using historical data are proposed in academia, including bayesian estimation methods, maximum likelihood estimation methods, covariance matching methods, minimum and maximum methods, subspace identification methods, correlation function methods, and the like. The correlation function method is widely focused because of moderate computational complexity and no special assumption on a noise model. The correlation function method is presented as a three-step noise statistic estimation scheme based on the autocorrelation of the innovation sequence at the earliest, and is then reduced to a single-step completion estimation autocovariance least-squares method. On the basis, a series of improvements such as noise covariance and state disturbance simultaneous estimation, noise variance estimation uniqueness conditions, minimum variance optimal weight selection, an innovation difference method of unbiased estimation under limited data and the like are developed. However, when the available data is less, for example, the sensor sampling sequence is shorter, the estimation result is easy to have larger variance, and the reliability is reduced. Along with the rise of a networked sensing system, multi-sensor information fusion becomes an important means for improving estimation accuracy. Fusing the innovation sequences of multiple sensors can equivalently increase the effective data volume, thereby reducing the estimation variance to some extent. In the traditional centralized fusion method, although global optimal estimation can be obtained by utilizing data of all sensors, the problems of high communication cost, poor expandability, high single-point failure risk and the like exist, and compared with the traditional centralized fusion method, the distributed fusion method can reduce communication and calculation burden and has fault tolerance capability, but the influence caused by correlation among new information of all sensors needs to be solved. When the cross covariance between the multi-sensor measurement data is known, the optimal linear fusion estimation formula can be obtained through the existing fusion deduction, and the method is popularized to any number of estimation fusion. However, in practical application, there is often unknown correlation between the information of different sensors, and the covariance intersection algorithm is a common means. The method can ensure consistency of fusion estimation through conservation weighting without priori knowing the correlation degree between sensor estimation, is initially used for conservation fusion of two estimation, is popularized to batch fusion of a plurality of estimation, and improves an acceleration sequence fusion scheme, a diffusion fusion scheme and the like. Research has shown that covariance cross-fusion has robust optimal performance in the dual sensor case, but becomes suboptimal in the multi-sensor case due to the conservation based on Minkowski and the set. It should be noted that the covariance intersection method is mainly used for the fusion of state estimation, and there is no direct solution to the problem of synchronous estimation of the joint state and the noise covariance. In view of this, although there is a study on distributed auto-covariance least-squares noise covariance estimation of a multi-sensor system in the prior art, there is a lack of a systematic method that can implement state and noise covariance combined high-precision estimation and fusion under unknown sensor correlation conditions. Disclosure of Invention In order to overcome the technical problems in the background, the invention aims to overcome the defects of the existing noise covariance estimation method that the accuracy is reduced and the fusion is difficult when the correlation of the output of the sensor is unknown in a multi-sensor environment, and provides a method for jointly estimating and fusing the state and the noise covariance in a multi-sensor system. The method aims to solve the technical problems th