Distributed sensor network fault detection method based on Bayesian inference

Abstract

The invention belongs to the field of fault detection, and discloses a distributed sensor network fault detection method based on Bayesian inference. Firstly, each sensor node calculates an initial posterior fault probability of the sensor node according to priori knowledge, then each sensor node calculates and sends a message to a neighbor node according to a current fault state, a coupling relation between the nodes and a message received from the neighbor node at the last moment, and a message transmission process is completed. Each sensor node updates its posterior failure probability using a message received from a neighbor node. The updated posterior fault probability is compared with a preset threshold value, if the posterior fault probability is larger than the preset threshold value, it is judged that the sensor node breaks down, an automatic fault response program is started, and an expert diagnosis platform is notified to conduct deep analysis and fault removal. According to the invention, high fault detection efficiency can be maintained under the condition of effectively reducing the dependence on the central node, and the communication bottleneck problem of the centralized control system is relieved.

Description

Technical Field

[0001] The present invention belongs to the field of fault detection, and more specifically, relates to a distributed sensor network fault detection method based on Bayesian inference. Background Art

[0002] With the rapid development of the Internet of Things and intelligent control systems, sensor networks, as one of the key technologies, have been widely used in smart homes, smart grids, industrial control systems and other fields. A sensor network is a network system composed of multiple sensor nodes. Each sensor node in the system is interconnected through wireless communication technology and works together to complete the monitoring, data collection and information processing tasks of a physical process.

[0003] By monitoring and analyzing the operating status of the system, timely discovering and identifying anomalies or faults in the system can significantly improve the reliability and stability of the sensor network. However, as the scale and complexity of the network increase, fault detection in sensor networks becomes increasingly difficult.

[0004] Traditional centralized fault detection methods require periodic collection of sensor measurement and status information to the central node. A large amount of data communication may lead to faster energy consumption of the central node, especially in large-scale centralized control systems, which are easily limited by data transmission rate or bandwidth, resulting in limited data communication performance, seriously affecting the real-time and overall performance of the system, that is, forming a communication bottleneck in the system. In addition, the failure of the central node will seriously affect the overall performance of the system, and may even cause the system to crash and shorten the life cycle of the sensor network. The delay caused by centralized data processing is also difficult to meet the real-time fault detection requirements in dynamic environments. Therefore, a distributed fault detection method is needed, so that each node can perform efficient fault detection and isolation based on local information and information of neighboring nodes, thereby improving the reliability and stability of the sensor network. Summary of the invention

[0005] In view of the limitations of traditional centralized fault detection methods in large-scale distributed sensor networks, the purpose of the present invention is to propose a distributed sensor network fault detection method, which reduces the dependence on central nodes, reduces the risk of communication bottlenecks and single point failures, and improves the overall fault detection accuracy of the system by combining localized computing, dynamic Bayesian inference, distributed message passing and a flexible fault judgment mechanism.

[0006] The technical solution of the present invention is as follows:

[0007] The specific steps include:

[0008] Step 1: Each sensor node calculates its own initial a posteriori fault probability based on prior knowledge;

[0009] In the Bayesian inference model of distributed sensor networks, each sensor node i has a fault state value F i With the observed value z i , each sensor node i distributed in the distributed sensor network collects local data observation value z i , and calculates its own initial a posteriori failure probability:

[0010]

[0011] In the formula, P(z i |F i ) indicates that the fault state value F i The next observation z i The conditional probability of i ) indicates the fault status value F i This value can be set based on historical data, expert knowledge, or an understanding of the system failure characteristics.

[0012] Step 2: Each sensor node transmits messages to its neighbor nodes;

[0013] Sensor node i is based on its current fault state value F i and the message received from the neighbor node at the last moment, calculate and send the message m i→j (F j ) (t+1) To neighbor node j:

[0014]

[0015] In the formula, ψ ij (F i ,F j ) is the interaction function between nodes, which quantifies the degree of mutual influence between the states of nodes i and j, and can select the appropriate form according to different system requirements. For example, when there is noise or uncertainty in the system, and the difference in node states has a significant impact on the interaction, the interaction function between nodes is a Gaussian function, that is, Where, is the standard deviation of σ; when the behavior of the system can be described by a linear relationship, the interaction function between nodes is a linear function, that is, ψ ij (F i ,F j )=a·F i +b·F j +c, where a, b, and c are linear coefficients; when there is no direct interaction between nodes, the interaction function between nodes is zero interaction, that is, ψ ij (F i ,F j )=1. m j→i(F i ) (t) is the message received from the neighbor node at the last moment. In particular, m j→i (F i ) (0) It is a priori knowledge and is set according to the system model and historical data.

[0016] Step 3: Update the sensor posterior fault probability;

[0017]

[0018] In the formula, represents the set of neighbor nodes of node i. Sensor node i uses the message m received from neighbor nodes j→i (F i ) (t+1) Update its own posterior failure probability P(F i |z i ) (t+1) .

[0019] Step 4: Fault determination;

[0020] The posterior failure probability P(F i |z i ) (t+1) Compared with the preset threshold τ, if P(F i |z i ) (t+1) >τ, the sensor node i is judged to be faulty, and the fault handling procedure is initiated to conduct in-depth analysis to repair / eliminate the fault.

[0021] The beneficial effects of the present invention are as follows: First, by adopting a distributed architecture, each node operates independently, reducing dependence on the central node and improving the robustness and scalability of the system. Second, it has high real-time performance. Through continuous information exchange and Bayesian inference, the detection results can be quickly updated to cope with dynamic environmental changes. Third, it has high efficiency. The information exchange and computational complexity between nodes are low, which is suitable for resource-constrained sensor networks. Fourth, it has high flexibility. The method can adjust the prior probability and conditional probability models according to different application scenarios to adapt to various complex network environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a diagram of the distributed sensor network structure;

[0023] Figure 2 It is a flow chart of a distributed sensor network fault detection method based on Bayesian inference; DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, other embodiments obtained by ordinary technicians in this field without creative work are all within the scope of protection of the present invention.

[0025] In a specific implementation of the present invention, there are three sensor nodes in the target sensor network, each of which can independently process and calculate its own information, and the three sensor nodes can communicate with each other. Each sensor node distributed in the network first calculates its own initial a posteriori fault probability based on prior knowledge. Subsequently, each sensor node calculates and sends messages to its neighbor nodes for message transmission based on its current fault state, the coupling relationship between nodes, and the message received from the neighbor node at the last moment. After the message transmission is completed, each sensor node uses the message received from the neighbor node to update its own a posteriori fault probability. Finally, the a posteriori fault probability is compared with the preset threshold. If the a posteriori fault probability is greater than the preset threshold, it is determined that the sensor node has a fault, thereby starting the fault handling program for in-depth analysis to repair / eliminate the fault. The following takes the initialization, calculation and fault determination process of sensor node 1 as an example to introduce a specific implementation process of the present invention.

[0026] The specific implementation steps are as follows:

[0027] Step 1: Sensor node 1 calculates its own initial a posteriori fault probability based on prior knowledge;

[0028] In the Bayesian inference model of distributed sensor networks, sensor node 1 has a fault state value F1 and an observation value z1. Sensor node 1 distributed in the distributed sensor network collects local data observation value z1 and calculates its own initial posterior fault probability:

[0029]

[0030] Based on the understanding of the system fault characteristics, the conditional probability of sensor node 1 observing the value z1 under the fault state value F1 is initialized as P(z1|F1)=0.1; the prior probability of the fault state value F1 is initialized as P(F1)=0.1. Substituting into the formula, the initial a posteriori fault probability P(F1|z1) of sensor node 1 is calculated. (0) =0.012.

[0031] Step 2: Each sensor node transmits messages to its neighbor nodes;

[0032] Each sensor node calculates and sends a message to its neighbor node based on its current fault state value and the message received from the neighbor node at the last moment. In this example, sensor node 1 receives message m from neighbor nodes 2 and 3. 2→1 (F1) (1) 、m 3→1 (F1) (1) :

[0033]

[0034] According to system requirements, the interaction function between nodes is selected as zero interaction, that is, ψ 12 (F1,F2)=1,ψ 13 (F1, F3) = 1. Initialize m according to the system model and historical data 1→2 (F2) (0) =1,m 1→3 (F3) (0) =1. Substitute into the formula and calculate m 2→1 (F1) (1) =0.012, m 2→1 (F1) (1) =0.012.

[0035] Step 3: Update the posterior fault probability of sensor node 1;

[0036]

[0037] In the formula, represents the set of neighbor nodes of node 1. Sensor node 1 uses the message m received from neighbor nodes 2 and 3 2→1 (F1) (1) 、m 3→1 (F1) (1) Update its own posterior failure probability P(F1|z1) (1) Substituting into the formula, we get P(F1|z1) (1) =0.0122.

[0038] Step 4: Fault determination;

[0039] Set the threshold τ = 0.1 and convert the posterior failure probability of sensor node 1 P(F1|z1) (1) Compared with the preset threshold τ, since P(F1|z1) (1) =0.0122<τ=0.1, it is determined that sensor node 1 has not failed, and the process returns to step 2 for the next iteration. If the posterior probability of failure of sensor node 1 at a certain moment is greater than the preset threshold, it is determined that sensor node 1 has failed, and the fault handling procedure is started to perform in-depth analysis to repair / eliminate the fault.

[0040] With the above-mentioned ideal embodiments of the present invention as inspiration, through the above-mentioned description, those skilled in the art can make various changes and modifications without departing from the technical concept of the present invention. The technical scope of the present invention is not limited to the contents of the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A distributed sensor network fault detection method based on Bayesian inference, characterized in that: The specific steps include: Step 1: Each sensor node calculates its own initial a posteriori fault probability based on prior knowledge; In the Bayesian inference model of distributed sensor networks, each sensor node i has a fault state value F i With the observed value z i , each sensor node i distributed in the distributed sensor network collects local data observation value z i , and calculates its own initial a posteriori failure probability: In the formula, P(z i |F i ) indicates that the fault state value F i The next observation z i The conditional probability of i ) indicates the fault status value F i The prior probability of , which can be set based on historical data, expert knowledge, or an understanding of the system failure characteristics; Step 2: Each sensor node transmits messages to its neighbor nodes; Sensor node i is based on its current fault state value F i and the message received from the neighbor node at the last moment, calculate and send the message m i→j (F j ) (t+1) To neighbor node j: In the formula, ψ ij (F i ,F j ) is the interaction function between nodes, which quantifies the degree of mutual influence between the states of node i and node j, and can select the appropriate form according to different system requirements: when there is noise or uncertainty in the system, and the difference in node states has a significant impact on the interaction, the interaction function between nodes is a Gaussian function, that is, Where, is the standard deviation of σ; when the behavior of the system can be described by a linear relationship, the interaction function between nodes is a linear function, that is, ψ ij (F i ,F j )=a·F i +b·F j +c, where a, b, and c are linear coefficients; when there is no direct interaction between nodes, the interaction function between nodes is zero interaction, that is, ψ ij (F i ,F j )=1;m j→i (F i ) (t) is the message received from the neighbor node at the last moment. In particular, m j→i (F i ) (0) It is a priori knowledge, set according to the system model and historical data; Step 3: Update the sensor posterior fault probability; In the formula, Represents the set of neighbor nodes of node i; sensor node i uses the message m received from the neighbor node j→i (F i ) (t+1) Update its own posterior failure probability P(F i |z i ) (t+1) ; Step 4: Fault determination; The posterior failure probability P(F i |z i ) (t+1) Compared with the preset threshold τ, if P(F i |z i ) (t+1) >τ, the sensor node i is judged to be faulty, and the fault handling procedure is initiated to conduct in-depth analysis to repair / eliminate the fault.

Images