Multiple fading factors-based strong tracking variational Bayesian adaptive Kalman filter
Introduction
For the linear Gaussian state space model whose observation noise and system noise are stationary, the Kalman filter (KF) is a recursive optimal algorithm that employs the minimum mean square error (MMSE) as the estimation criterion and has been widely used in parameter estimation [1], [2]. The KF adopts the estimated state vector at previous moment and the measurements at current moment to obtain the current state, which is essentially a process of continuous prediction and correction [3], and it does not need to store a large amount of historical observation data. Due to its excellent parameter estimation performance, the KF has been widely used in various dynamic systems, such as target location and tracking, fault diagnosis and detection, navigation and guidance, risk index assessment and so on [4], [5], [6], [7], [8].
However, in practical applications, the noise statistical characteristics of the state space model are not always stationary [9], [10], [11], [12], [13]. The time-varying process noise covariance matrix (PNCM) and measurement noise covariance matrix (MNCM) usually affect the accuracy of the prior information obtained from noise statistics, and the wrong prior information will cause a large number of estimation errors or even filtering divergence [14], [15], accordingly diminish the filtering performance. Aiming at these problems, scholars put forward numbers of adaptive Kalman filters (AKF), e.g., innovation-based AKF (IAKF), fading AKF (FAKF), variational Bayesian-based AKF (VBAKF), strong tracking-based variational Bayesian AKF (ST-VBAKF) and so on [16], [17], [18], [19]. The IAKF solves the problem of imperfect prior information through a filtering learning process based on an innovative sequence, which has a significant improvement in performance over fixed filters [16]. However, the IAKF requires a fairly large data window to obtain a reliable estimation of the MNCM, which makes it unsuitable for rapidly changing MNCM. The FAKF decreases the weights of current observations by increasing the one-step prediction error covariances. However, the calculation process of the scalar fading factor is more cumbersome and it has the same adjustment ability for each filtering channel, which is not conducive to improve the stability and accuracy of the filter [20]. The VBAKF selects the inverse Wishart prior to model the measurement noise, and it uses the variational Bayesian (VB) approach to obtain the suboptimal estimations of the state vector and the slowly varying MNCM [18]. However, the PNCM is set as a fixed value which is not consistent with reality [21], thus, the filtering performance of VBAKF will decrease due to the inaccurate PNCM. On the basis of VBAKF, Huang et al. [15] proposed a novel variational Bayesian-based adaptive Kalman filter (N-VBAKF), which can estimate not only MNCM but also predicted error covariance matrix (PECM) in the process of variational iterative recursion, and good results have been achieved in the target tracking problem where both PECM and MNCM are slowly varying. However, the N-VBAKF also has the disadvantages of high computational complexity and time consumption. Tan [19] introduced the suboptimal fading factor into VBAKF and proposed a ST-VBAKF algorithm, which could adaptively track the MNCM in a linear Gaussian system with time-varying noise, and effectively overcome the influence of time-varying PNCM. The convergence speed and accuracy of the results are improved. However, as with FAKF, the scalar fading factor has regulatory limitations in ST-VBAKF.
In complex dynamic systems, the estimation accuracies of state variables represented by diagonal elements of PECM are different. Therefore, conventional FAKF can no longer meet the demand of accuracy since the scalar fading factor has the same ability to adjust the state variables. Zhou et al. [22] proposed an extended Kalman filter (EKF) with multiple suboptimal fading factors, which determined the weights corresponding to the fading factors through prior knowledge, and provided a new idea for the establishment of multiple fading factors. Geng et al. [23] scaled the PECM based on an assumption that the prediction residual vectors follow a chi-square distribution, and proposed a novel AKF with multiple fading factors for GPS/INS integrated navigation. The characteristics of this filter overcome the lack of traditional KF in robust estimation. Furthermore, scholars have designed a variety of multiple fading factors-based filters according to the characteristics of different dynamic systems to make the filters more applicable [24], [25], [26].
In order to further improve the performance of ST-VBAKF, a multiple fading factors-based strong tracking variational Bayesian adaptive filter (MST-VBAKF) is proposed. Compared with ST-VBAKF, the improvements of proposed method are reflected in two aspects. Firstly, an exponential weighting method based on fading memory is introduced to improve the utilization weight of current observations, thus the innovation covariance matrix is estimated more accurately. Secondly, the multiple fading factors are employed to promote the tracking ability for PECM and the filter robustness is enhanced. Simulation results show that the estimation accuracy of the new algorithm is better than ST-VBAKF and the other filters.
The rest of this paper is organized as follows. Section 2 formulates the problem and gives the KF solution. Section 3 introduces the formation process of the MST-VBAKF in detail, including the selection of prior distributions for process and measurement noise, the construction of multiple fading factors and the derivation of variational measurement update. Section 4 performs simulations to verify the proposed filtering algorithm, in which the influences of weakening factor and forgetting factor on MST-VBAKF and the robustness of different nominal noise covariance settings are analyzed, respectively, and then the proposed filtering algorithm is compared with the existing filters. Section 5 summarizes the performance of the new algorithm.
Section snippets
Problem formulation and KF solution
The state space model of discrete linear stochastic system includes state equation and measurement equation, which is defined as [27], [28]where and are the state and measurement vectors, respectively; is the system transition matrix; is the design matrix; and are the process and measurement noise vectors, respectively. It should be noted that the initial state vector is assumed to be the Gaussian distribution of mean vector and
Strong tracking variational Bayesian adaptive Kalman filter based on multiple fading factors
In this section, the multiple fading factors are introduced into VBAKF and a novel filtering method named MST-VBAKF is proposed, which can adjust the PECM and inaccurate MNCM simultaneously. The selection of prior distribution, the construction of multiple fading factors and the variational measurement update process of MST-VBAKF are introduced in detail, respectively.
The experimental simulation
Like many existing researches [15], [18], [19], this paper adopts the continuous white noise acceleration model in two-dimensional (2D) Cartesian coordinates to verify the performance of the proposed MST-VBAKF algorithm, where the PNCM and MNCM of the target change slowly with time. It is assumed that the motion process is monitored in real time by sensors such as the global navigation satellite systems (GNSS). The position and velocity of target are respectively denoted as and
Conclusions
When the PNCM and MNCM of the dynamic system are time-varying, the KF is not ideal for parameter estimation accuracy, while existing filter algorithms are not able to weaken the influence of inaccurate PNCM on filtering performance or have limited ability to weaken. On the basis of existing researches, this paper proposes a strong tracking variational Bayesian adaptive filtering algorithm with multiple fading factors, namely MST-VBAKF, which can adjust the PECM and MNCM simultaneously.
CRediT authorship contribution statement
Cheng Pan: Conceptualization, Methodology, Formal analysis, Writing - original draft. Jingxiang Gao: Supervision, Project administration, Funding acquisition. Zengke Li: Software, Visualization. Nijia Qian: Investigation, Writing - review & editing. Fangchao Li: Investigation, Writing - review & editing.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
This work is supported by the National Natural Science Foundation of China (grant numbers 41974026, 41874006), the China University of Mining and Technology (grant number 2020WLJCRCZL052), and the Jiangsu Education Department (grant number KYCX20_2058). The authors are grateful for the three anonymous reviewers for their constructive comments on the manuscript.
References (37)
- et al.
Maximum correntropy Kalman filter
Automatica
(2017) - et al.
ML Estimation of Process Noise Variance in Dynamic Systems
IFAC Proc.
(2011) - et al.
Variational Bayesian adaptation of noise covariances in multiple target tracking problems
Measurement
(2018) - et al.
Moving-window extended Kalman filter for structural damage detection with unknown process and measurement noises
Measurement
(2016) - et al.
Multiple Fading Factors Kalman Filter for SINS Static Alignment Application
Chinese J. Aeronaut.
(2011) - et al.
The Kalman filter as the optimal linear minimum mean-squared error multiuser CDMA detector
IEEE Trans. Inf. Theory.
(2000) Kalman filtering implementation with Matlab
(2004)- et al.
Target Tracking Using Machine Learning and Kalman Filter in Wireless Sensor Networks
IEEE Sens. J.
(2014) - et al.
A New Adaptive Extended Kalman Filter for Cooperative Localization
IEEE Trans. Aerosp. Electron. Syst.
(2018) - et al.
Experimental Test of a Two-Stage Kalman Filter for Actuator Fault Detection and Diagnosis of an Unmanned Quadrotor Helicopter
J. Intell. Robot. Syst.
(2013)
Underground risk index assessment and prediction using a simplified hierarchical fuzzy logic model and Kalman filter
Processes.
A Variational Bayesian-Based Robust Adaptive Filtering for Precise Point Positioning Using Undifferenced and Uncombined Observations
Adv. Sp. Res.
INS/GPS Integration System using Adaptive Filter for Estimating Measurement Noise Variance
IEEE Trans. Aerosp. Electron. Syst.
Estimating the Process Noise Variance for Vehicle Motion Models
2015 IEEE 18th Int Conf. Intell. Transp. Syst.
Target motion analysis with unknown measurement noise variance
2017 20th Int. Conf. Inf. Fusion
An Improved Alignment Method for the Strapdown Inertial Navigation System (SINS)
Sensors.
A Novel Adaptive Kalman Filter With Inaccurate Process and Measurement Noise Covariance Matrices
IEEE Trans. Automat. Contr.
Adaptive Kalman filtering for INS/GPS
J. Geod.
Cited by (36)
A new smooth variable structure Tobit filter for systems with censored measurements and model parameters uncertainty
2024, Journal of the Franklin InstituteVariational Bayesian-based robust adaptive filtering for GNSS/INS tightly coupled positioning in urban environments
2023, Measurement: Journal of the International Measurement ConfederationMultiple similarity measure-based maximum correntropy criterion Kalman filter with adaptive kernel width for GPS/INS integration navigation
2023, Measurement: Journal of the International Measurement ConfederationA modified federated Student's t-based variational adaptive Kalman filter for multi-sensor information fusion
2023, Measurement: Journal of the International Measurement ConfederationA novel variational robust filter with Gaussian mixture model for unknown non-Gaussian noises
2023, Measurement: Journal of the International Measurement Confederation