Target tracking method based on trust-based distributed fault-tolerant bulk information fusion filtering

Abstract

This invention discloses a trust-based distributed fault-tolerant volumetric information fusion filtering target tracking method, belonging to the field of wireless sensor networks and information fusion technology. The method includes: Step 1, each sensor node introduces a Bernoulli random variable to model observation anomalies and uses a volumetric information filtering method to calculate the local posterior estimate of the target state; Step 2, each sensor node exchanges its estimated target state information with its neighboring nodes; Step 3, each sensor node classifies the exchanged information, calculates the difference between the neighboring node's estimate and its own prior estimate, and uses a K-means clustering method to identify the set of trust nodes; Step 4, each sensor node performs adaptive weighted consensus fusion of the trust node information based on the set of trust nodes to obtain the fused estimate of the target state. This invention has a simple structure and reasonable design, and can effectively improve target tracking performance even when sensor nodes are faulty or observation anomalies exist.

Description

Technical Field

[0001] This invention relates to the field of wireless sensor networks and information fusion technology. Specifically, it relates to a trust-based distributed fault-tolerant volumetric information fusion filtering target tracking method suitable for complex environments where there are observational anomalies and node failures caused by sensor malfunctions. Background Technology

[0002] With the increasing demand for highly reliable and continuous environmental sensing in fields such as smart grids, smart transportation, and smart mines, distributed sensing systems in complex environments face severe challenges. Wireless sensor networks (WSNs), as a distributed system composed of a large number of micro-nodes, are widely used.

[0003] Moving target tracking, as one of the key technologies supporting numerous applications of WSNs, directly impacts the stability and reliability of the system. However, in resource-constrained and complex deployment environments, the inherent limitations of sensor nodes' computational and communication resources, along with potential perception anomalies, are major factors significantly restricting the improvement of target tracking accuracy. In existing technologies, volumetric information filtering exhibits good performance in strongly nonlinear systems. However, traditional distributed filtering methods have limitations when dealing with real-world complex environments. On one hand, sensor failures often lead to observational anomalies, disrupting the foundation for continuous filter updates. On the other hand, some nodes may generate persistent erroneous data due to failures; if this is not effectively isolated during distributed fusion, it will severely contaminate the global estimation results.

[0004] While existing research has identified faulty nodes through consensus fusion or cluster analysis, these approaches often address single problems in isolation. In real-world scenarios, anomalies in observations caused by sensor failures and persistent node failures frequently occur. Existing methods are ill-equipped to handle such complex and non-ideal conditions and are prone to misjudgment even without attacks, resulting in insufficient fault tolerance mechanisms.

[0005] Therefore, how to design an integrated fault-tolerant mechanism within the distributed filtering framework that can collaboratively address observational anomalies and persistent node failures caused by faults is an urgent problem to be solved. Summary of the Invention

[0006] This invention aims to address the technical problems of insufficient fault tolerance and decreased estimation accuracy in existing distributed target tracking methods when faced with complex scenarios involving intertwined observational anomalies caused by sensor failures and persistent node malfunctions. Its structure is simple and rationally designed. First, each sensor node (including faulty nodes) uses a volumetric information filtering method to calculate local estimation information of the target state. Second, based on information exchange only with neighboring nodes, a K-means-based dimensionality reduction two-cluster clustering method is used to identify trusted nodes. Finally, adaptive weighted consensus fusion is performed on the trusted node information, and the fusion result is used as the target state estimate at the current moment, thereby effectively improving target tracking performance even when sensor node malfunctions exist.

[0007] To achieve the above objectives, this invention provides a trust-based distributed fault-tolerant and volumetric information fusion filtering method.

[0008] The target tracking method includes the following steps:

[0009] S1. Bernoulli random variables are introduced into each sensing node to model observed anomalies, and the volumetric information filtering method is used to calculate...

[0010] Local posterior estimation information of the target state

[0011] S101. According to , Calculate the prior information matrix and prior information vector ,in The dimension representing the target state. To propagate volume points, For prior state estimation, For process noise covariance;

[0012] S102. According to , Calculate the posterior information matrix and posterior information vector Information contribution matrix , The cross covariance matrix is... For measuring noise, the information contribution vector , For the observed values, To predict the measured value.

[0013] S2. Each sensing node exchanges estimated target state information with its neighboring nodes;

[0014] S3. Each node divides the information it exchanges into trusted node information and untrusted node information: After determining whether there are any abnormal nodes, if there are no abnormal nodes, then... Otherwise execute Dimensionality reduction and two-class clustering yield the set of trusted nodes. , For nodes And its single-hop neighbor set.

[0015] S4. Each node performs adaptive weight consensus on the trusted node information to obtain the fusion estimation information of the target state.

[0016] The aforementioned trust-based distributed fault-tolerant volumetric information fusion filtering target tracking method is characterized in that: in step S1, the observation data is described by the following observation model:

[0017] ;

[0018] in Indicates the first Each node The observed value at time, Let be a Bernoulli random variable, if This means that the i-th sensor is fault-free at time t, otherwise it means that the sensor node is a faulty node at that time. Represents the observation matrix. Indicates observation noise. Indicates the target is State estimation at time 10:00 This represents observed anomaly information.

[0019] Further, step S2 specifically includes: according to , Calculate the posterior covariance matrix and posterior state estimation reuse Calculate the nodes Self-prior estimation With local posterior estimation dissimilarity .

[0020] Furthermore, step S2 specifically includes: if This represents a set of trustworthy nodes with relatively low dissimilarity. Otherwise, execute the K-means dimensionality reduction two-cluster clustering algorithm to classify all clusters. Divide into two clusters and set the cluster center value The nodes corresponding to the smaller clusters are identified as trustworthy nodes, forming a set of trustworthy neighbor nodes. ;in This represents the largest eigenvalue of the matrix.

[0021] Furthermore, in step S4, the dynamic adjustment of the fusion weights is based on the following steps: initializing the weight matrix based on the network topology. Then based on the set of trusted nodes Generate trust instructions The updated weight matrix is Finally, through the formula The final weight matrix is ​​obtained by performing normalization. ,in for Inside the first OK Column elements.

[0022] Further, in step S4, according to the formula right and conduct In the next iteration, when and The fused information matrix is ​​obtained upon convergence. Information vector ,in Finally passed , Calculate the fused covariance matrix and state estimation .

[0023] Compared with the prior art, the present invention has the following advantages:

[0024] 1. The present invention has a simple structure, a reasonable design, and is easy to apply.

[0025] 2. This invention adopts a distributed architecture, which estimates the target motion state through multiple sensors, and has stronger anti-interference capabilities compared to a centralized system where only a central node processes information.

[0026] 3. This invention uses volumetric information filtering for local estimation, and at the same time has the advantages of high-precision estimation capability for processing nonlinear systems and computational advantage of information additivity.

[0027] 4. This invention introduces K-means dimensionality reduction two-cluster clustering into volumetric information filtering and designs a trust-adaptive K-means dimensionality reduction two-cluster clustering consensus fusion mechanism, which not only realizes the identification and fusion of local estimates of trust nodes, but also separates faulty nodes and their local estimates.

[0028] 5. This invention only requires data exchange between neighboring nodes. Although it adds a K-means dimensionality reduction two-cluster clustering step, it only trusts node estimation to participate in the calculation in the consensus fusion step. While keeping the communication topology unchanged, it maintains a computational complexity similar to that of distributed volumetric information filtering.

[0029] In summary, this invention features a simple structure and a reasonable design. First, each sensing node (including faulty nodes) uses a volumetric information filtering method to calculate local estimates of the target state. Second, based on information exchange only with neighboring nodes, a K-means-based dimensionality-reduced two-cluster clustering method is used to identify trusted nodes. Finally, the trusted node information is subjected to adaptive weighted consensus fusion, and the fusion result is used as the target state estimate at the current moment, thereby effectively improving target tracking performance even when sensing nodes are faulty.

[0030] Simulation results show that, under conditions of observational anomalies and node failures caused by sensor malfunctions, this invention significantly improves tracking accuracy and stability compared to methods such as distributed voluminous Kalman filtering, distributed voluminous information filtering, and cluster-based distributed voluminous information filtering, demonstrating its value in achieving high-reliability target tracking in complex wireless sensor network environments. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0032] Figure 1 This is a flowchart of the target tracking method steps of the present invention;

[0033] Figure 2 The diagram shows the trajectory of the uniform turning motion and a comparison of various algorithms.

[0034] Figure 2 (a) in the figure is the trajectory diagram of uniform turning motion;

[0035] Figure 2 (b) Uniform turning motion (Root Mean Square Error) Comparison Chart;

[0036] Figure 3 The diagram shows the trajectory of uniform linear motion tracking and a comparison of various algorithms, including:

[0037] Figure 3 (a) in the figure is the trajectory diagram of uniform linear motion;

[0038] Figure 3 (b) Uniform linear motion (Root Mean Square Error) Comparison Chart;

[0039] Figure 4 For uniform turning motion and uniform linear motion, different node failure probabilities (Mean root mean square error) comparison chart, where:

[0040] Figure 4 (a) represents the uniform turning motion under different node failure probabilities. Comparison chart;

[0041] Figure 4 (b) represents uniform linear motion under different node failure probabilities. Comparison chart. Detailed Implementation

[0042] like Figure 1 As shown, this embodiment provides a trust-based distributed fault-tolerant volumetric information fusion filtering method for target tracking. This method is primarily applied to target tracking in Wireless Sensor Networks (WSNs), particularly in challenging environments such as decreased detection accuracy due to sensor node failures and abnormal observation data. It effectively improves the accuracy and robustness of state estimation and possesses better fault tolerance. The method in this embodiment introduces Bernoulli random variables to model observation anomalies and combines them with a K-means random evaluation mechanism to achieve accurate identification and adaptive weighted fusion of anomalous nodes.

[0043] The method in this embodiment is first based on a precise description of the target motion model and the sensor observation model. In wireless sensor networks, it is assumed that there exists Several sensor nodes are used to collaboratively monitor a moving target. To describe the target's dynamic behavior, this embodiment uses a state-space equation to represent the target's motion. Specifically, the target... state of time and state of time The relationship between them is described by the state transition equation, namely In this equation, This represents the state transition matrix, which determines the evolution of the target state over time. (Target state estimation) It includes the target's position and velocity information, specifically defined as... ,in and They represent Always the goal is shaft and Position coordinates on the axis and They represent Always the goal is shaft and The velocity component along the axial direction. In the equation... This represents process noise, used to simulate uncertainties in the target motion model, and is assumed to have a mean of 0 and a covariance of . The model uses a Gaussian distribution. This modeling approach can cover various common target motion patterns, such as uniform linear motion and uniform turning motion.

[0044] In terms of observation, targeting the first in the network sensor nodes ( This embodiment considers the scenario of sensor failure. The observation data is described by the following observation model:

[0045] ;

[0046] in, Indicates the first Each node The observed value at time, Let be a Bernoulli random variable, when This indicates that the i-th sensor is fault-free at time t, otherwise it indicates that the sensor is faulty. Represents the observation matrix (nonlinear observation function). Indicates observation noise. Indicates the target is State estimation at time 10:00 This represents observed anomalies. The introduction of this mathematical model can realistically simulate situations where data is missing or abnormal, thus laying the foundation for subsequent fault-tolerant processing.

[0047] Based on the above model, the specific process of this embodiment is as follows: Figure 1 As shown, it includes the following steps:

[0048] S1. Each sensor node detects its distance and azimuth angle from the target and estimates the target's state information based on the observation data. This step is performed independently on each sensor node, using a volumetric information filtering framework to handle nonlinear state estimation. First, a time update is performed. Node Based on the posterior state estimate from the previous time step and posterior covariance matrix To generate sampling points, one can use... Decomposition method for the posterior covariance matrix Decompose it to obtain its square root factor matrix. , making Subsequently, using this square root factor matrix Generate a set of volume points These points are then substituted into the state transition model for nonlinear propagation to obtain the propagated volume points. Based on this, the prior covariance matrix and information matrix are calculated according to the following formula. and prior information vector :

[0049] ;

[0050] ;

[0051] Among them, prior state estimation , The dimension representing the target state. For process noise covariance;

[0052] Next, the measurement is updated, and the measurement propagation volume point is calculated. ,as follows:

[0053] ;

[0054] in: , .

[0055] Calculate the predicted measurement value Its covariance matrix :

[0056] ;

[0057] ;

[0058] Calculate the cross-covariance matrix between the prior estimate and the predicted measurement. :

[0059]

[0060] Finally, the local posterior state estimate is calculated. With covariance matrix :

[0061] ;

[0062] ;

[0063] Information contribution matrix , For measuring noise, the information contribution vector , For the observed values, To predict the measured value.

[0064] S2. Each node exchanges estimated target state information with its neighboring nodes. Based on... , Calculate the posterior covariance matrix and posterior state estimation redefining For nodes And its set of single-hop communication neighbor nodes, and finally using the formula Calculate the nodes Self-prior estimation With each neighboring node posterior estimation Dissimilarity values ​​between Used to characterize the degree of consistency between the information provided by neighboring nodes and the local prediction.

[0065] S3. Each node divides the information it exchanges into trusted and untrusted node information. This step uses a clustering discrimination mechanism to identify whether there are node failures. First, the maximum eigenvalue is used as a threshold to determine whether there are abnormal nodes, with the following discrimination criteria:

[0066] ;

[0067] in, This represents the largest eigenvalue of the corresponding matrix, used to measure the uncertainty boundary of the estimate. If the above condition holds, it means that there are no faulty nodes deviating from the normal range in the set. Trust Node Set Otherwise, execute the K-means dimensionality reduction two-cluster algorithm, as follows:

[0068] ;

[0069] in, It is the largest eigenvalue; The cluster center point. express Does it belong to the characteristic function of cluster g? If it belongs to cluster g, but ,otherwise .

[0070] right Update according to the following rules:

[0071] ;

[0072] Repeat the appeal steps until the cluster assignment results converge and no longer change. Nodes within the set are considered trusted nodes, and this set of trusted nodes is regarded as... .

[0073] S4. Each node performs adaptive weight consensus on the trusted node information to obtain the fusion estimation information of the target state. After obtaining the trusted set... Then, the abnormal information is isolated by adaptively adjusting the weight matrix:

[0074] 1. Weight Generation: Initialize the weight matrix based on the network topology. Define a trust indicator variable. Specifically .

[0075] 2. Weight Normalization: Calculate the final weight matrix. Then, through the formula Perform normalization to ensure nodes The sum of the weights of all trusted neighbors is 1.

[0076] 3. Consensus Iteration and Fusion: The following consensus iteration formula is executed using normalized weights:

[0077] ;

[0078] in:

[0079] ;

[0080] After L iterations, a consensus fusion information pair is obtained. Finally, the information matrix is ​​obtained. Information vector Estimating the covariance matrix and state estimation :

[0081] ;

[0082] ;

[0083] ;

[0084] .

[0085] Simulation experiment analysis:

[0086] like Figures 2 to 4 As shown, this embodiment achieves this through constant speed turning ( Figure 2 ) and uniform linear motion ( Figure 3 The algorithm's performance was verified in two scenarios. With a 30% probability of sensor node anomalies, the tracking trajectory of this method was closest to the true trajectory, and the root mean square error was [not specified]. Smaller. By Figure 4 As can be seen, as the probability of observation failure increases from 0% to 60%, the average root mean square error of this method ( ) The fact that the level of [something] was consistently kept at a minimum demonstrates the strong robustness of the method under complex fault environments. ;

[0087] ;

[0088] in, Represents the number of experiments. This represents the simulation time.

[0089] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A trust-based distributed fault-tolerant volumetric information fusion filtering target tracking method, characterized in that, Including the following step: S1. Bernoulli random variables are introduced into each sensing node to model observed anomalies, and the volumetric information filtering method is used to calculate... Local posterior estimation information of the target state; S101. According to , Calculate the prior information matrix and prior information vector ,in The dimension representing the target state. To propagate volume points, For prior state estimation, For process noise covariance; S102. According to , Calculate the posterior information matrix and posterior information vector Information contribution matrix , The cross covariance matrix is... For measuring noise, the information contribution vector , For the observed values, To predict the measured value; S2. Each sensing node exchanges estimated target state information with its neighboring nodes; S3. Each node divides the information it exchanges into trusted node information and untrusted node information: After determining whether there are any abnormal nodes, if there are no abnormal nodes, then... Otherwise execute Dimensionality reduction and two-class clustering yield the set of trusted nodes. , For nodes and its single-hop neighbor set; S4. Each node performs adaptive weight consensus on the trusted node information to obtain the fusion estimation information of the target state.

2. The target tracking method based on trust-based distributed fault-tolerant volumetric information fusion filtering according to claim 1, characterized in that, In step S1, the observation data is described by the following observation model: ； in Indicates the first Each node The observed value at time, Let be a Bernoulli random variable, if This means that the i-th sensor is fault-free at time t; otherwise, it means that the sensor node is a faulty node at that time. Represents the observation matrix. Indicates observation noise. Indicates the target is State estimation at time 10:00 This represents observed anomaly information.

3. The target tracking method based on trust-tolerant distributed fault-tolerant volumetric information fusion filtering according to claim 1, characterized in that, Step S2 specifically includes: according to , Calculate the posterior covariance matrix and posterior state estimation reuse Calculate the nodes Self-prior estimation With local posterior estimation dissimilarity .

4. The target tracking method based on trust-tolerant distributed fault-tolerant volumetric information fusion filtering according to claim 1, characterized in that, Step S2 specifically includes: If This represents a set of trustworthy nodes with relatively low dissimilarity. Otherwise, execute the K-means dimensionality reduction two-cluster clustering algorithm to classify all clusters. Divide into two clusters and set the cluster center value The nodes corresponding to the smaller clusters are identified as trustworthy nodes, forming a set of trustworthy neighbor nodes. ;in This represents the largest eigenvalue of the matrix.

5. The target tracking method based on trust-tolerant distributed fault-tolerant volumetric information fusion filtering according to claim 1, characterized in that, In step S4, the adaptive adjustment of the fusion weights is based on the following steps: initializing the weight matrix based on the network topology. Then based on the set of trusted nodes Generate trust instructions The updated weight matrix is Finally, through the formula The final weight matrix is ​​obtained by performing normalization. ,in for Inside the first OK Column elements.

6. The target tracking method based on trust-tolerant distributed fault-tolerant volumetric information fusion filtering according to claim 1, characterized in that, In step S4, the consensus fusion step: according to the formula right and conduct In the next iteration, when and The fused information matrix is ​​obtained upon convergence. Information vector ,in Finally passed , Calculate the estimated covariance matrix after fusion and state estimation .

Images