Intelligent equipment integrated management system for power grid distribution network inspection

Abstract

The invention provides an intelligent equipment integrated management system for power grid distribution network inspection, and the system comprises a data preprocessing module, a distributed anomaly detection module, a self-adaptive threshold adjustment module, a fault positioning and propagation analysis module, and an alarm and visualization module. The data preprocessing module carries out hierarchical clustering on sensors and constructs a tree structure, each clustering node is provided with a local anomaly detector, the distributed anomaly detection module uses an RASA algorithm to calculate an anomaly score of data collected by bottom layer nodes, and the intermediate nodes gather bottom layer leaf node information and carry out local anomaly detection. The top-layer node summarizes the information of the intermediate nodes and performs global anomaly detection; the adaptive threshold adjustment module dynamically adjusts an abnormal threshold by using an ATA algorithm; when the fault positioning and propagation analysis module detects an anomaly, the GCN-FPA module is activated, and a fault source and a possible propagation path are analyzed; through the system, a fault source can be quickly positioned, and a fault propagation path can be predicted.

Description

Technical Field

[0001] The present invention proposes an intelligent integrated management system for power distribution network inspection, which relates to the technical field of power distribution network control. Background Art

[0002] With the rapid development of smart grid technology, a large number of smart sensors and monitoring equipment have been deployed in power distribution network systems to monitor the operation status of the power grid in real time, detect faults and abnormal conditions. However, the traditional power distribution network inspection and equipment management methods have the following major problems when facing massive data and complex network topology: Traditional methods usually use a single central processing system, which needs to process the data of each sensor one by one. With the sharp increase in the number of sensors, this method can no longer meet the needs of real-time monitoring, resulting in data processing delays and slow system response; Most existing systems rely only on simple threshold comparisons or single statistical methods to detect anomalies. This method is easily affected by noise, resulting in high false alarm rates or missing important anomalies. The power grid operation environment is complex and changeable, and fixed detection thresholds and rules are difficult to adapt to different scenarios and seasonal changes, affecting the reliability of anomaly detection; Traditional systems can often only detect single-point anomalies, making it difficult to analyze the root cause of the fault and the possible propagation path, which greatly increases the difficulty of fault diagnosis and processing. Summary of the invention

[0003] The present invention provides an intelligent integrated management system for power distribution network inspection to solve the above-mentioned problems:

[0004] The present invention proposes an intelligent integrated management system for power distribution network inspection, the system comprising: a data preprocessing module, a distributed anomaly detection module, an adaptive threshold adjustment module, a fault location and propagation analysis module and an alarm and visualization module;

[0005] The data preprocessing module hierarchically clusters the sensors according to geographic location and functional type to construct a tree structure, each clustering node is equipped with a local anomaly detector, and the tree structure includes bottom leaf nodes, middle nodes and top nodes;

[0006] The distributed anomaly detection module uses the RASA algorithm to calculate the anomaly score of the data collected by the bottom node, the middle node summarizes the bottom leaf node information to perform local anomaly detection, and the top node summarizes the middle node information to perform global anomaly detection;

[0007] The adaptive threshold adjustment module uses the ATA algorithm to dynamically adjust the abnormal threshold;

[0008] When the fault location and propagation analysis module detects an abnormality, it activates the GCN-FPA module to analyze the fault source and possible propagation path;

[0009] The alarm and visualization module generates anomaly reports and visual fault diagrams, and sends alarms to relevant personnel.

[0010] Furthermore, the data preprocessing module cleans and standardizes the data collected by the original sensor.

[0011] Furthermore, the distributed anomaly detection module uses the RASA algorithm to calculate the anomaly score of the data collected by the underlying nodes, including using the Z-score anomaly model to detect statistically significant outliers.

[0012] Specifically, the Z-score abnormality model is:

[0013]

[0014] Where x(s, t) is the reading of sensor s at time t, μ(s) and σ(s) are the historical mean and standard deviation, respectively.

[0015] Use the ARIMA model to predict the current value and calculate the forecast deviation:

[0016]

[0017] in, Represents the predicted value.

[0018] Furthermore, the distributed anomaly detection module uses the RASA algorithm to calculate the anomaly score of the data collected by the bottom node, and further includes calculating the correlation deviation with the adjacent sensor through a context correlation anomaly degree model. Specifically, the context correlation anomaly degree model is:

[0019]

[0020] Where N(s) is the neighbor set of sensor s, corr(s, i, t) is the current correlation, is the historical average correlation,

[0021] The total abnormal model is calculated using the RASA algorithm. Specifically, the RASA algorithm is:

[0022] A(s,t)=α·Z(s,t)+β·T(s,t)+γ·C(s,t)

[0023] Among them, α, β and γ are weight coefficients, T(s, t) is the time series prediction deviation, C(s, t) is the contextual relevance anomaly, and Z(s, t) is the Z-score anomaly.

[0024] Furthermore, the adaptive threshold adjustment module uses the ATA algorithm to dynamically adjust the abnormal threshold, including:

[0025] Specifically, the ATA algorithm is:

[0026] Threshold(t)=λ·Threshold(t-1)+(1-λ)·MedianScire(t)

[0027] Where λ is the smoothing factor and MedianScore(t) is the median anomaly score of all sensors in the current time window.

[0028] Furthermore, the GCN-FPA module constructs a power grid topology graph G=(V, E), where V is a node set including sensors and power equipment, and E is an edge set.

[0029] Furthermore, the GCN-FPA module initializes the node features:

[0030] H (0) =[A(s,t),Z(s,t),T(s,t),C(s,t)]

[0031] Furthermore, the GCN-FPA module uses a graph convolutional neural network to simulate fault propagation:

[0032]

[0033] Where A is the adjacency matrix with self-loops added, is the degree matrix, H (l) is the node feature of the lth layer, W (l) is a learnable weight matrix.

[0034] Furthermore, the GCN-FPA module predicts the failure probability:

[0035] P(fault|s)=σ(W·H(L)+b)

[0036] Among them, σ is the sigmoid activation function, W is the learnable weight matrix, H(L) is the node feature of the last layer of the graph convolutional network, i.e. the Lth layer, and b is the bias term.

[0037] Furthermore, when the predicted fault probability exceeds the preset threshold, the attention mechanism is used to calculate the importance of the edge, and the edge with the highest importance is selected as the possible fault propagation path:

[0038] e ij =LeakyReLU(a T [Wh i ||Wh j ])

[0039]

[0040] Among them, eij represents the unnormalized importance score of the edge from node i to node j, w is a learnable weight matrix, hi and hj are the feature vectors of node i and node j respectively, Whi and Whj are the results of linear transformation of node features, a is a learnable attention vector, LeakyReLU is an activation function that introduces nonlinearity and allows small negative values ​​to pass, αij is the normalized attention weight of the edge from node i to node j, Ni represents the neighbor set of node i, and eik represents the unnormalized importance score of the edge from node i to node k. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a system structure block diagram of the present invention. DETAILED DESCRIPTION

[0042] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0043] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. The embodiments described are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0045] One embodiment of the present invention is a comprehensive management system for intelligent equipment for power distribution network inspection, characterized in that the system includes: a data preprocessing module, a distributed anomaly detection module, an adaptive threshold adjustment module, a fault location and propagation analysis module, and an alarm and visualization module;

[0046] The data preprocessing module hierarchically clusters the sensors according to geographic location and functional type to construct a tree structure, each clustering node is equipped with a local anomaly detector, and the tree structure includes bottom leaf nodes, middle nodes and top nodes;

[0047] The distributed anomaly detection module uses the RASA algorithm to calculate the anomaly score of the data collected by the bottom node, the middle node summarizes the bottom leaf node information to perform local anomaly detection, and the top node summarizes the middle node information to perform global anomaly detection;

[0048] The adaptive threshold adjustment module uses the ATA algorithm to dynamically adjust the abnormal threshold;

[0049] When the fault location and propagation analysis module detects an abnormality, it activates the GCN-FPA module to analyze the fault source and possible propagation path;

[0050] The alarm and visualization module generates anomaly reports and visual fault diagrams, and sends alarms to relevant personnel.

[0051] The working principle and effect of the above technical solution are as follows: the data preprocessing module hierarchically clusters a large amount of sensor data according to geographical location and functional type to form a tree structure. This structure includes bottom leaf nodes that process raw data, intermediate nodes that summarize information, and top nodes for global monitoring. Each node is equipped with a local anomaly detector to achieve distributed processing. The bottom leaf nodes use RASA (real-time anomaly scoring algorithm) to calculate anomaly scores, the intermediate nodes summarize subordinate information and perform local anomaly detection, and the top nodes perform global anomaly detection. The entire network status is comprehensively analyzed, and the ATA (adaptive threshold adjustment) algorithm is used to dynamically adjust the anomaly threshold according to real-time data; this ensures that the system can adapt to different operating environments and changes in conditions; when an anomaly is detected, the GCN-FPA (graph convolutional network fault propagation analysis) module is activated, and the graph convolutional neural network is used to analyze the power grid topology, identify the source of the fault and predict the possible propagation path; the alarm and visualization module generates detailed anomaly reports and intuitive fault maps, and sends timely alarm information to relevant personnel. Through hierarchical clustering and distributed processing, the system can efficiently process massive sensor data and greatly improve the data processing speed; combining multiple anomaly detection methods (RASA algorithm) improves the accuracy of anomaly detection and reduces false positives and false negatives; dynamic threshold adjustment enables the system to adapt to different operating environments and seasonal changes, improving the stability and reliability of the system; using graph convolutional networks for fault propagation analysis can not only quickly locate the source of the fault, but also predict the possible propagation path of the fault, providing a basis for timely intervention; by generating visual fault maps and abnormal reports, operation and maintenance personnel can intuitively understand the fault situation and accelerate the decision-making process; sending timely alerts to relevant personnel to ensure rapid Respond to potential problems; the distributed architecture makes the system easy to expand and can adapt to the growing scale of the power grid and the number of equipment; preventive maintenance, through early anomaly detection and fault propagation analysis, the system can support preventive maintenance strategies, reduce unexpected power outages and equipment damage; improve operation and maintenance efficiency, the comprehensive management system integrates multiple functional modules, provides a one-stop solution for operation and maintenance personnel, and greatly improves work efficiency; through fast and accurate fault detection and intelligent analysis, the system helps to improve the reliability and stability of the entire power grid. Through innovative algorithms and architectural design, the system realizes intelligent supervision of the power distribution network, providing a strong guarantee for the safe and stable operation of the power system.

[0052] In one embodiment of the present invention, the data preprocessing module cleans and standardizes the data collected by the original sensor.

[0053] In one embodiment of the present invention, the distributed anomaly detection module uses the RASA algorithm to calculate the anomaly score of the data collected by the bottom node, including using the Z-score anomaly model to detect statistically significant outliers.

[0054] Specifically, the Z-score abnormality model is:

[0055]

[0056] Where x(s, t) is the reading of sensor s at time t, μ(s) and σ(s) are the historical mean and standard deviation, respectively.

[0057] Use the ARIMA model to predict the current value and calculate the forecast deviation:

[0058]

[0059] in, Represents the predicted value.

[0060] In one embodiment of the present invention, the distributed anomaly detection module uses the RASA algorithm to calculate the anomaly score of the data collected by the bottom node, and further includes calculating the correlation deviation with the adjacent sensor through a context correlation anomaly degree model. Specifically, the context correlation anomaly degree model is:

[0061]

[0062] Where N(s) is the neighbor set of sensor s, corr(s, i, t) is the current correlation, is the historical average correlation,

[0063] The total abnormal model is calculated using the RASA algorithm. Specifically, the RASA algorithm is:

[0064] A(s,t)=α·Z(s,t)+β·T(s,t)+γ·C(s,t)

[0065] Among them, α, β and γ are weight coefficients, T(s, t) is the time series prediction deviation, C(s, t) is the contextual relevance anomaly, and Z(s, t) is the Z-score anomaly.

[0066] In one embodiment of the present invention, the adaptive threshold adjustment module uses the ATA algorithm to dynamically adjust the abnormal threshold, including:

[0067] Specifically, the ATA algorithm is:

[0068] Threshold(t)=λ·Threshold(t-1)+(1-λ)·MedianScore(t)

[0069] Where λ is the smoothing factor and MedianScore(t) is the median anomaly score of all sensors in the current time window.

[0070] In one embodiment of the present invention, the GCN-FPA module constructs a power grid topology graph G=(V, E), where V is a node set including sensors and power equipment, and E is an edge set.

[0071] In one embodiment of the present invention, the GCN-FPA module initializes the node features: (0) =[A(s,t),Z(s,t),T(s,t),C(s,t)]

[0072] In one embodiment of the present invention, the GCN-FPA module uses a graph convolutional neural network to simulate fault propagation:

[0073]

[0074] Where A is the adjacency matrix with self-loops added, is the degree matrix, H (l) is the node feature of the lth layer, W (l) is a learnable weight matrix.

[0075] In one embodiment of the present invention, the GCN-FPA module predicts the probability of failure:

[0076] P(fault|s)=σ(W·H(L)+b)

[0077] Among them, σ is the sigmoid activation function, W is the learnable weight matrix, H(L) is the node feature of the last layer of the graph convolutional network, i.e. the Lth layer, and b is the bias term.

[0078] One embodiment of the present invention is a comprehensive management system for intelligent equipment for power distribution network inspection. When the predicted fault probability exceeds a preset threshold, an attention mechanism is used to calculate the importance of edges and the edge with the highest importance is selected as a possible fault propagation path:

[0079] e ij =LeakyReLU(a T [Wh i ||Wh j ])

[0080]

[0081] Among them, eij represents the unnormalized importance score of the edge from node i to node j, w is a learnable weight matrix, hi and hj are the feature vectors of node i and node j respectively, Whi and Whj are the results of linear transformation of node features, a is a learnable attention vector, LeakyReLU is an activation function that introduces nonlinearity and allows small negative values ​​to pass, αij is the normalized attention weight of the edge from node i to node j, Ni represents the neighbor set of node i, and eik represents the unnormalized importance score of the edge from node i to node k.

[0082] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. An intelligent integrated management system for power distribution network inspection, characterized in that: The system includes: a data preprocessing module, a distributed anomaly detection module, an adaptive threshold adjustment module, a fault location and propagation analysis module, and an alarm and visualization module; The data preprocessing module hierarchically clusters the sensors according to geographic location and functional type to construct a tree structure, each clustering node is equipped with a local anomaly detector, and the tree structure includes bottom leaf nodes, middle nodes and top nodes; The distributed anomaly detection module uses the RASA algorithm to calculate the anomaly score of the data collected by the bottom node, the middle node summarizes the bottom leaf node information to perform local anomaly detection, and the top node summarizes the middle node information to perform global anomaly detection; The adaptive threshold adjustment module uses the ATA algorithm to dynamically adjust the abnormal threshold; When the fault location and propagation analysis module detects an abnormality, it activates the GCN-FPA module to analyze the fault source and possible propagation path; The alarm and visualization module generates anomaly reports and visual fault diagrams, and sends alarms to relevant personnel.

2. According to claim 1, a comprehensive management system for intelligent equipment for power distribution network inspection, characterized in that: The data preprocessing module cleans and standardizes the data collected by the original sensor.

3. According to claim 1, a comprehensive management system for intelligent equipment for power distribution network inspection, characterized in that: The distributed anomaly detection module uses the RASA algorithm to calculate the anomaly score of the data collected by the underlying nodes, including using the Z-score anomaly model to detect statistically significant anomalies. Specifically, the Z-score abnormality model is: Where x(s, t) is the reading of sensor s at time t, μ(s) and σ(s) are the historical mean and standard deviation, respectively. Use the ARIMA model to predict the current value and calculate the forecast deviation: in, Represents the predicted value.

4. According to claim 3, a comprehensive management system for intelligent equipment for power distribution network inspection, characterized in that: The distributed anomaly detection module uses the RASA algorithm to calculate the anomaly score of the data collected by the bottom node, and further includes calculating the correlation deviation with the adjacent sensor through the context correlation anomaly degree model. Specifically, the context correlation anomaly degree model is: Where N(s) is the neighbor set of sensor s, corr(s, i, t) is the current correlation, is the historical average correlation, The total abnormal model is calculated using the RASA algorithm. Specifically, the RASA algorithm is: A(s,t)=α·Z(s,t)+β·T(s,t)+γ·C(s,t) Among them, α, β and γ are weight coefficients, T(s, t) is the time series prediction deviation, C(s, t) is the contextual relevance anomaly, and Z(s, t) is the Z-score anomaly.

5. According to claim 1, a comprehensive management system for intelligent equipment for power distribution network inspection, characterized in that: The adaptive threshold adjustment module uses the ATA algorithm to dynamically adjust the abnormal threshold, including: Specifically, the ATA algorithm is: Threshold(t)=λ·Threshold(t-1)+(1-λ)·MedianScore(t) Where λ is the smoothing factor and MedianScore(t) is the median anomaly score of all sensors in the current time window.

6. An intelligent integrated management system for power distribution network inspection, characterized in that: The GCN-FPA module constructs a power grid topology graph G=(V, E), where V is a node set including sensors and power equipment, and E is an edge set.

7. According to claim 6, an intelligent integrated management system for power distribution network inspection, characterized in that: The GCN-FPA module initializes the node features: H (0) =[A(s,t),Z(s,t),T(s,t),C(s,t)]。 8. According to claim 7, a comprehensive intelligent management system for power distribution network inspection, characterized in that: The GCN-FPA module uses a graph convolutional neural network to simulate fault propagation: Where A is the adjacency matrix with self-loops added, is the degree matrix, H (l) is the node feature of the lth layer, W (l) is a learnable weight matrix.

9. According to claim 8, a comprehensive management system for intelligent equipment for power distribution network inspection, characterized in that: The GCN-FPA module predicts the failure probability: P(fault|s)=σ(W·H(L)+b) Among them, σ is the sigmoid activation function, W is the learnable weight matrix, H(L) is the node feature of the last layer of the graph convolutional network, i.e. the Lth layer, and b is the bias term.

10. The intelligent integrated management system for power distribution network inspection according to claim 9, characterized in that: When the predicted fault probability exceeds the preset threshold, the attention mechanism is used to calculate the importance of the edge and select the edge with the highest importance as the possible fault propagation path: e ij =LeakyReLU(a T [Wh i ||Wh j ]) Among them, eij represents the unnormalized importance score of the edge from node i to node j, w is a learnable weight matrix, hi and hj are the feature vectors of node i and node j respectively, Whi and Whj are the results of linear transformation of node features, a is a learnable attention vector, LeakyReLU is an activation function that introduces nonlinearity and allows small negative values ​​to pass, αij is the normalized attention weight of the edge from node i to node j, Ni represents the neighbor set of node i, and eik represents the unnormalized importance score of the edge from node i to node k.

Images