Link information-based autonomous meter reading link fault detection method and system

Abstract

This invention relates to the field of link fault detection, and particularly to a method and system for autonomous meter reading link fault detection based on link information. The method includes: acquiring a fixed network topology and calculating the importance weight of any node; constructing a path set for the target path, and constructing a node existence matrix and a node co-occurrence matrix based on the path set; calculating node stability features; calculating link stability features; using the entropy of the node co-occurrence matrix as an entropy feature; constructing a comprehensive feature vector from the node stability features, link stability features, and entropy features; inputting the comprehensive feature vector of the communication path to be detected into a fault detection model, and outputting the fault probability of the communication path to be detected. The technical solution of this invention can improve the accuracy of link fault detection results.

Description

Technical Field

[0001] This invention relates to the field of link fault detection. In particular, it relates to a method and system for autonomous meter reading link fault detection based on link information. Background Technology

[0002] As a core infrastructure for the automatic collection of electricity consumption information in smart grids, the power line carrier communication autonomous meter reading system relies on a multi-hop communication link between the concentrator and the electricity meter for data transmission. However, in the actual complex power grid environment, this link is easily affected by dynamic factors such as node failure, channel interference and routing oscillation, which can lead to data acquisition failure. This not only reduces the meter reading success rate, but also weakens the real-time monitoring and fault response capabilities of the power grid.

[0003] Existing monitoring methods mainly employ simple heartbeat detection or fixed threshold alarm mechanisms, which can only identify obvious faults such as complete link interruption, but cannot capture early hidden dangers of gradual degradation of communication quality. Although some studies have introduced machine learning methods to improve monitoring accuracy, their feature extraction is often limited to single indicators such as communication success rate, failing to integrate multi-dimensional information such as dynamic changes in network topology (such as the strength of cooperation between nodes and path stability), resulting in low accuracy of fault detection results. Summary of the Invention

[0004] To address the aforementioned technical problems, the present invention provides solutions in the following aspects.

[0005] In the first aspect, the autonomous meter reading link fault detection method based on link information includes: acquiring a fixed network topology, calculating the betweenness centrality of any node in the network topology, and using the normalized betweenness centrality as the importance weight of any node; taking any communication path in history as the target path, acquiring the fault probability corresponding to the target path, constructing a path set of the target path, wherein the path set consists of communication paths with a preset number of consecutive occurrences, and the target path is the communication path at the end of the path set, and each communication path in the path set has the same starting point and the same ending point; constructing a node existence matrix and a node co-occurrence matrix based on the path set; calculating the participation rate of any node based on the node existence matrix, and calculating the relationship between 1 and participation. The difference in participation rate is used as the product of the participation rate and the difference to represent the volatility of any node. Node stability features are calculated based on volatility and importance weights. The weighted similarity between any two adjacent communication paths in the path set is calculated, and the mean of all weighted similarities is used as the link stability feature. The entropy of the node co-occurrence matrix is ​​used as the entropy feature. The node stability feature, link stability feature, and entropy feature are combined to construct a comprehensive feature vector. The comprehensive feature vector is used as input, and the true value of the fault probability is used as a label to train the fault detection model, which outputs the predicted value of the fault probability. The comprehensive feature vector of the communication path to be detected is input into the fault detection model, which outputs the fault probability of the communication path to be detected, thus completing the fault detection.

[0006] Preferably, the number of rows in the node matrix is ​​the total number of nodes in the network topology, and the number of columns in the node matrix is ​​the total number of communication paths in the path set; in response to the target path passing through any node, the intersection of the row of any node and the column of the target path is assigned a value of 1.

[0007] Preferably, the number of rows in the node co-occurrence matrix is ​​the total number of nodes in the network topology, and the number of columns in the node co-occurrence matrix is ​​the total number of nodes in the network topology; the node in the row containing any element in the node co-occurrence matrix is ​​taken as the row node, and the node in the column containing any element in the node co-occurrence matrix is ​​taken as the column node, and the value of any element in the node co-occurrence matrix represents the number of times the row node and the column node appear simultaneously in the same communication path.

[0008] Preferably, the calculation of the participation rate of any node includes: for any node, taking the average value of all elements in the row where any node is located in the node existence matrix as the participation rate of any node.

[0009] Preferably, the computation node stability feature includes: taking all nodes in the path set as a node set; for any node, calculating the product of the fluctuation degree of any node and the importance weight, and calculating the cumulative sum of the product of all nodes in the node set; calculating the cumulative weight of the importance weight of all nodes in the node set; calculating the ratio of the cumulative product to the cumulative weight, and taking the difference between 1 and the ratio as the node stability feature.

[0010] Preferably, obtaining the weighted similarity includes: for any two adjacent communication paths in the path set, taking the intersection of the nodes along the two communication paths as a first set, and taking the union of the nodes along the two communication paths as a second set; calculating a first cumulative value of the importance weights of all nodes in the first set, and calculating a second cumulative value of the importance weights of all nodes in the second set; and taking the ratio of the first cumulative value to the second cumulative value as the weighted similarity.

[0011] Preferably, the fault detection model is a BP network, and the loss function of the BP network is cross-entropy loss.

[0012] Secondly, an autonomous meter reading link fault detection system based on link information includes: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the autonomous meter reading link fault detection method based on link information described in any one of the above embodiments is implemented.

[0013] The present invention has the following effects:

[0014] This invention constructs a node existence matrix and a node co-occurrence matrix, integrating comprehensive indicators from three dimensions: node stability characteristics, link stability characteristics, and entropy characteristics. This overcomes the limitations of traditional methods that rely solely on a single communication success rate indicator. It can not only accurately detect explicit faults such as complete link interruption, but also effectively identify early hidden dangers of gradual degradation in communication quality. By introducing node betweenness centrality as an importance weight, fault detection is more focused on key nodes and paths. Combined with the nonlinear modeling capabilities of BP neural networks and cross-entropy loss functions, the accuracy and timeliness of fault prediction are significantly improved, enhancing real-time link monitoring and fault response capabilities. Attached Figure Description

[0015] Figure 1 This is a flowchart of the autonomous meter reading link fault detection method based on link information according to an embodiment of the present invention. Detailed Implementation

[0016] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0017] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0018] Reference Figure 1 The autonomous meter reading link fault detection method based on link information includes steps S1-S4, as follows:

[0019] S1: Obtain a fixed network topology, calculate the betweenness centrality of any node in the network topology, and use the normalized betweenness centrality as the importance weight of any node.

[0020] In one embodiment, based on a fixed network topology, the betweenness centrality of each node is calculated. This metric reflects the criticality of a node as an information flow "hub" by quantifying the frequency of its occurrence in all shortest paths of the network. Calculating the betweenness centrality of any node is a well-known technique to those skilled in the art and will not be elaborated upon here. The minimum and maximum values ​​of the betweenness centrality of all nodes in the network topology are calculated. For the betweenness centrality of any node, a min-max normalization method is used, and the normalized betweenness centrality value is directly defined as the importance weight of that node. This weight not only quantifies the impact of a node on overall communication efficiency (e.g., the failure of a high-weight node can easily lead to network segmentation), but also provides a data-driven basis for fault early warning, resource optimization, or robustness enhancement of smart grids. For example, in an autonomous meter reading system, high-weight nodes are prioritized for protection to improve link stability.

[0021] It should be noted that the network topology is obtained through the network management platform, and this structure accurately depicts the physical connections between concentrators, relay nodes, and electricity meters. Crucially, after the system is deployed and enters a stable operating phase, this topology remains static. At this point, aging electricity meters requiring replacement, or relay node failures requiring repair / replacement, will not affect the betweenness centrality of the nodes. This stability stems from the determinism of the hardware installation; for example, the locations of electricity meters are fixed, and relay node deployments are optimized and no longer adjusted. This provides a reliable foundation for subsequent network analysis, avoiding noise introduced by dynamic changes. It should be explained that when there are situations such as grid expansion adding new nodes (concentrators / electricity meters) or equipment migration due to changes in user electricity addresses, the betweenness centrality of each node is recalculated when the system simulates a new topology.

[0022] S2: Take any communication path in history as the target path, obtain the failure probability corresponding to the target path, construct the path set of the target path, the path set is the communication path with a preset number of consecutive occurrences, the target path is the communication path at the end of the path set, each communication path in the path set has the same starting point and the same ending point, and construct the node existence matrix and node co-occurrence matrix based on the path set.

[0023] In one embodiment, in the fault analysis of a power line carrier communication autonomous meter reading system, for a specific communication task, such as data acquisition from a concentrator to a specific energy meter, any communication path in the historical record is selected as the target path. A path set is constructed around the target path, containing a predetermined number of consecutive communication paths traced back from the target path. All paths share the same starting and ending points, but may exhibit different intermediate nodes due to network dynamics (such as relay node switching or channel interference), meaning different communication paths pass through different nodes. For example, any communication path in the historical record is selected as the target path, and four communication paths are traced back from the target path. These four traced communication paths and the target path together form the path set of the target path. The starting point of these five communication paths is the concentrator, and the ending point is the same energy meter. The number of times a fault occurred in the path set is obtained from the historical fault log, and the ratio of the number of faults to the total number of communication paths in the path set is used as the fault probability of the target path.

[0024] Based on the path set, construct a node existence matrix and a node co-occurrence matrix. The number of rows in the node existence matrix is ​​the total number of nodes in the network topology, and the number of columns in the node existence matrix is ​​the total number of communication paths in the path set. In response to any communication path in the path set passing through any node, the intersection of the row of any node and the column of any communication path in the path set is assigned a value of 1.

[0025] The number of rows in the node co-occurrence matrix is ​​the total number of nodes in the network topology, and the number of columns in the node co-occurrence matrix is ​​the total number of nodes in the network topology. The node in the row containing any element in the node co-occurrence matrix is ​​taken as the row node, and the node in the column containing any element in the node co-occurrence matrix is ​​taken as the column node. The value of any element in the node co-occurrence matrix represents the number of times that the row node and the column node appear in the same communication path.

[0026] It needs to be explained that the node existence matrix is... take The matrix, where This represents the total number of nodes in the network topology. This represents the total number of communication paths in the path set. The rule for assigning values ​​to elements within the matrix for nodes is as follows: when the node in the path set... Each communication path through nodes Then the node exists in the matrix at the th position. Line number The value of column 1; when the value of column 1 in the path set is 1. The communication path did not pass through any node. Then the node exists in the matrix at the th position. Line number The value of the column is 0.

[0027] It needs to be explained that the node co-occurrence matrix is take The matrix, This represents the total number of nodes in the network topology. The node co-occurrence matrix contains the node with the [missing information - likely a specific node number]. Line number The column values ​​represent the nodes. With nodes The number of times they appear on the same communication path. It should be noted that the number of times they appear simultaneously in the node co-occurrence matrix is ​​not considered. and Values ​​that are the same.

[0028] S3: Calculate the participation rate of any node based on the node existence matrix, calculate the difference between 1 and the participation rate, and use the product of the participation rate and the difference as the volatility of any node. Calculate the node stability feature based on the volatility and importance weights. Calculate the weighted similarity of any two adjacent communication paths in the path set, and use the mean of all weighted similarities as the link stability feature. Use the entropy of the node co-occurrence matrix as the entropy feature. Construct a comprehensive feature vector from the node stability feature, link stability feature, and entropy feature.

[0029] In one embodiment, based on the node existence matrix given in step S2, for any given node, the arithmetic mean of all element values ​​in the row containing that node is calculated, and this mean is defined as the participation rate of that node, quantifying its overall participation, activity, or influence in the network. A higher participation rate indicates a greater probability that any node will appear in a predetermined number of consecutive communication paths. Furthermore, since the node existence matrix is ​​essentially a 0-1 matrix, the participation rate ranges from 0 to 1.

[0030] Calculate the difference between 1 and the participation rate. Use the product of the participation rate and the difference as the volatility of any node. When the participation rate is 0.5, the volatility reaches its maximum value of 0.25, indicating that the node participation is the most unstable. When the participation rate is 0 or 1, the volatility is 0, which indicates that the node behavior is stable, that is, it either never appears or always appears.

[0031] Take all nodes in the path set as a node set. For any node, calculate the product of the fluctuation degree of any node and the importance weight, and calculate the cumulative sum of the product of all nodes in the node set. Calculate the cumulative weight of the importance weight of all nodes in the node set. Calculate the ratio of the cumulative product to the cumulative weight, and take the difference between 1 and the ratio as the node stability feature.

[0032] Node stability characteristics range from 0 to 1. The closer the value is to 1, the higher the overall stability of the node set in the path, meaning the more stable the participation patterns of each node, and the more reliably important nodes participate in communication. Conversely, the closer the node stability characteristic value is to 0, the greater the fluctuation in node participation patterns, the more frequently important nodes are lost, the poorer the link stability, and the higher the probability of failure.

[0033] For any two adjacent communication paths in the path set, the intersection of the nodes along the two paths is taken as the first set, and the union of the nodes along the two paths is taken as the second set. The first cumulative value of the importance weights of all nodes in the first set is calculated, and the second cumulative value of the importance weights of all nodes in the second set is calculated. The ratio of the first cumulative value to the second cumulative value is taken as the weighted similarity. The larger the weighted similarity value, the higher the overlap of nodes between the two adjacent communication paths.

[0034] Two adjacent communication paths are considered as adjacent path pairs. The weighted similarity values ​​of all adjacent path pairs are calculated by taking the arithmetic mean to obtain the link stability feature. The link stability feature ranges from 0 to 1 and is used to quantify the similarity, continuity, and consistency of routing selection over a short time scale. The closer the link stability feature value is to 1, the more overlapping the adjacent paths are and the more stable the participation of key nodes, indicating that the routing strategy is consistent and reliable. Conversely, the closer the link stability feature value is to 0, the more significant the differences in the path node set are, which is often caused by frequent path switching due to routing oscillations, dynamic load balancing, or external interference. This provides a sensitive indicator for network monitoring, effectively warns of communication instability risks, and optimizes routing protocol design to improve network robustness.

[0035] Entropy features aim to quantify the uncertainty and disorder of node cooperation patterns. By analyzing the distribution characteristics of cooperation relationships between nodes, it effectively correlates faults such as cooperation breakdown and routing disorder. Specifically, based on the node co-occurrence matrix, all off-diagonal elements (i.e., elements whose row and column indices are not equal) are extracted to form a set. The value of each element in the set (i.e., co-occurrence frequency) is divided by the sum of the values ​​of all off-diagonal elements, thereby normalizing it into a discrete probability distribution. The Shannon entropy of the probability distribution is calculated; this entropy value directly reflects the uniformity of the frequency of node cooperation pairs: the higher the entropy feature value, the more dispersed and disordered the distribution of cooperation relationships is, and the lack of a dominant pattern. The network is more susceptible to random disturbances, and the risk of cooperation breakdown is significantly increased. Conversely, the lower the entropy feature value, the more concentrated the cooperation pattern is on a few key node pairs, forming a stable and predictable communication backbone, thereby enhancing the robustness and reliability of the link.

[0036] The node stability features, link stability features, and entropy features are used to construct a comprehensive feature vector.

[0037] S4: Train the fault detection model using the comprehensive feature vector as input and the true value of the fault probability as label, and output the predicted value of the fault probability; input the comprehensive feature vector of the communication path to be detected into the fault detection model, and output the fault probability of the communication path to be detected to complete the fault detection.

[0038] In one embodiment, in a fault detection system, a comprehensive feature vector integrating node stability features, link stability features, and entropy features is used as input data. The true values ​​of historical fault probabilities are used as supervision labels to train a Back Propagation (BP) network. This network learns the complex mapping relationship between features and faults through nonlinear transformations in the hidden layers. The output layer uses a sigmoid activation function to generate fault probability predictions between 0 and 1. During training, a cross-entropy loss function is used as the optimization objective. This loss function effectively handles class imbalance and enhances the model's sensitivity to probability calibration by minimizing the KL divergence between the predicted probability distribution and the true label distribution. Finally, the trained model can receive real-time network feature inputs and output high-precision fault probability predictions. The BP network adopts a classic three-layer feedforward structure. The input layer has 3 nodes, strictly corresponding to the dimension of the comprehensive feature vector, receiving the three features representing the link state. The hidden layer is a single layer containing 12 nodes. This number is determined based on the input dimension and through experimental optimization, effectively preventing overfitting while ensuring the model's nonlinear expressive power. This hidden layer uses the ReLU activation function to accelerate training convergence. The output layer is a single node, using the Sigmoid activation function to map the network output to a continuous value between 0 and 1, directly representing the probability of link failure. To further improve model robustness, a Dropout layer with a dropout rate of 0.3 is introduced after the hidden layers. Model training employs the Adam optimization algorithm with an initial learning rate of 0.001, a cross-entropy loss function, and a training batch size of 64.

[0039] The comprehensive feature vector of the communication path to be detected is input into the fault detection model, which outputs the fault probability of the communication path to be detected. If the fault probability of the communication path to be detected is greater than a preset fault threshold, the link is determined to be faulty, and an alarm is triggered. For example, the preset fault threshold is set to 0.8.

[0040] The system includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the autonomous meter reading link fault detection method based on link information according to the first aspect of the present invention.

[0041] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0042] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for autonomous meter reading link fault detection based on link information, characterized in that, include: Obtain a fixed network topology, calculate the betweenness centrality of any node in the network topology, and use the normalized betweenness centrality as the importance weight of any node. Take any communication path in history as the target path, obtain the failure probability corresponding to the target path, construct the path set of the target path, the path set contains the target path and the communication path that traces back from the target path a consecutive preset number of times, each communication path in the path set has the same starting point and the same ending point, and construct the node existence matrix and the node co-occurrence matrix based on the path set. The participation rate of any node is calculated based on the node existence matrix. The difference between 1 and the participation rate is calculated. The product of the participation rate and the difference is taken as the volatility of any node. The stability characteristics of the node are calculated based on the volatility and importance weight. The weighted similarity between any two adjacent communication paths in the path set is calculated, and the mean of all weighted similarities is used as the link stability feature; the entropy of the node co-occurrence matrix is ​​used as the entropy feature; the node stability feature, link stability feature, and entropy feature are constructed into a comprehensive feature vector. The comprehensive feature vector is used as input and the true value of the fault probability is used as label to train the fault detection model, which is used to output the predicted value of the fault probability; the comprehensive feature vector of the communication path to be detected is input into the fault detection model, and the fault probability of the communication path to be detected is output to complete the fault detection. The number of rows in the matrix represents the total number of nodes in the network topology, and the number of columns in the matrix represents the total number of communication paths in the path set. In response to the target path passing through any node, the intersection of the row of any node and the column of the target path is assigned a value of 1. The number of rows in the node co-occurrence matrix is ​​the total number of nodes in the network topology, and the number of columns in the node co-occurrence matrix is ​​the total number of nodes in the network topology. The node in the row containing any element in the node co-occurrence matrix is ​​taken as the row node, and the node in the column containing any element in the node co-occurrence matrix is ​​taken as the column node. The value of any element in the node co-occurrence matrix represents the number of times that the row node and the column node appear in the same communication path. The calculation of the participation rate of any node includes: for any node, taking the average value of all elements in the row where any node is located in the node existence matrix as the participation rate of any node; the calculation of the node stability feature includes: taking all nodes in the path set as the node set, for any node, calculating the product of the fluctuation degree of any node and the importance weight, and calculating the cumulative sum of the product of all nodes in the node set; calculating the cumulative weight of the importance weight of all nodes in the node set; calculating the ratio of the cumulative product to the cumulative weight, and taking the difference between 1 and the ratio as the node stability feature; Using the entropy of the node co-occurrence matrix as an entropy feature includes: based on the node co-occurrence matrix, extracting all off-diagonal elements in the node co-occurrence matrix to form a set, dividing the value of each element in the set by the sum of the values ​​of all off-diagonal elements to obtain a discrete probability distribution, and calculating the Shannon entropy of the probability distribution as an entropy feature.

2. The autonomous meter reading link fault detection method based on link information according to claim 1, characterized in that, The number of rows in the matrix where a node exists represents the total number of nodes in the network topology, and the number of columns in the matrix where a node exists represents the total number of communication paths in the path set. In response to any node along the target path, the intersection of the row of that node and the column of the target path is assigned a value of 1.

3. The autonomous meter reading link fault detection method based on link information according to claim 1, characterized in that, The number of rows in the node co-occurrence matrix is ​​the total number of nodes in the network topology, and the number of columns in the node co-occurrence matrix is ​​the total number of nodes in the network topology; The node in the row containing any element of the node co-occurrence matrix is ​​taken as the row node, and the node in the column containing any element of the node co-occurrence matrix is ​​taken as the column node. The value of any element in the node co-occurrence matrix represents the number of times the row node and the column node appear simultaneously in the same communication path.

4. The autonomous meter reading link fault detection method based on link information according to claim 1, characterized in that, Obtaining the weighted similarity includes: For any two adjacent communication paths in the path set, the intersection of the nodes along the two communication paths is taken as the first set, and the union of the nodes along the two communication paths is taken as the second set. Calculate the first cumulative value of the importance weights of all nodes in the first set, and calculate the second cumulative value of the importance weights of all nodes in the second set; The ratio of the first accumulated value to the second accumulated value is used as the weighted similarity.

5. The autonomous meter reading link fault detection method based on link information according to claim 1, characterized in that, The fault detection model is a BP network, and the loss function of the BP network is cross-entropy loss.

6. An autonomous meter reading link fault detection system based on link information, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement the autonomous meter reading link fault detection method based on link information according to any one of claims 1-5.

Images