Charging facility portrait and abnormal mode detection method and system

Abstract

The invention relates to the technical field of data analysis, and discloses a charging facility portrait and abnormal mode detection method and system, and the method comprises the following steps: data collection and preprocessing: carrying out the high-frequency current sampling of a charging facility, obtaining time series data through sliding window segmentation, and synchronizing an equipment log and a power grid state timestamp; multi-dimensional feature extraction and fusion: extracting 23-dimensional time sequence features including a current fluctuation entropy value, a plugging behavior slope and a harmonic distortion rate, and generating a standardized feature vector through Z-score normalization; graph neural network modeling: a charging facility association graph is constructed, nodes are charging equipment (including power and geographical location attributes), edges are power grid homology or physical proximity, and the anomaly propagation probability is quantified through a GNN message transmission mechanism; and generating a dynamic maintenance strategy. The method makes remarkable technical progress in the aspect of abnormal detection and maintenance of the charging facility, and has wide application prospects and market value.

Description

Technical Field

[0001] The present invention relates to the field of data analysis technology, and in particular to a charging facility profiling and abnormal pattern detection method and system. Background Art

[0002] As the demand for electric vehicle charging continues to increase, the stability and reliability of charging piles, as an important part of electric vehicle infrastructure, have become particularly critical. However, existing charging pile fault diagnosis and maintenance methods still have many problems, mainly reflected in the identification of complex abnormal patterns, quantitative analysis of the impact of correlations between devices, and accurate scheduling of maintenance resources. The following are several major shortcomings of existing technologies:

[0003] Traditional charging pile fault diagnosis methods rely on set thresholds for current and voltage to trigger alarms. For example, if the current exceeds 10A for 30 seconds, the system will trigger an alarm. However, this threshold-based detection method cannot effectively identify complex abnormal patterns, such as intermittent overloads and poor contact. These abnormalities may not immediately reach the threshold to trigger an alarm, but can still affect normal operation of the equipment.

[0004] Currently, the coverage rate of threshold-based alarm-style anomaly detection is only 76%, which means that more than 24% of anomalies are not identified in time (missing reports).

[0005] Because the system relies on fixed thresholds, it has a high false alarm rate. Slight changes, such as normal current fluctuations, can be misjudged as abnormalities, with a false alarm rate as high as 18%-22%.

[0006] Existing charging pile monitoring systems often rely on data from isolated devices for fault analysis, ignoring the interdependencies among groups of charging piles. For example, power grid fluctuations can trigger cascading anomalies across multiple charging piles. Existing technologies are unable to effectively analyze the interdependencies between these devices, making it impossible to predict or identify cluster failures in advance.

[0007] In the existing technology, such as patent CN112926111A, although some methods are used to analyze the data of a single device in isolation, they still cannot solve the problem of correlation analysis between charging pile groups and it is difficult to capture group behavior under complex abnormal patterns.

[0008] Existing maintenance strategies primarily rely on manual experience and traditional maintenance methods. Due to the lack of accurate equipment health assessments and the inability to precisely quantify the operating status of each device, maintenance personnel frequently conduct ineffective inspections, wasting significant O&M resources. Statistics show that over 30% of maintenance work orders are ineffective inspections that fail to effectively resolve actual equipment issues.

[0009] Existing charging pile fault diagnosis and maintenance methods leave significant room for improvement, particularly in the areas of automatically identifying complex anomaly patterns, analyzing inter-device correlations, and precisely scheduling operations and maintenance resources. Therefore, a new technical solution is urgently needed that can overcome these shortcomings through intelligent means, improve the accuracy and efficiency of charging pile fault diagnosis, reduce maintenance costs, and optimize resource scheduling. Summary of the Invention

[0010] The purpose of the present invention is to provide a charging facility profiling and abnormal pattern detection method and system to solve the problems raised in the above background technology.

[0011] To achieve the above object, the present invention provides the following technical solutions:

[0012] A charging facility profiling and abnormal pattern detection method and system technical solution includes the following steps:

[0013] Data acquisition and preprocessing: High-frequency current sampling (1kHz) is performed on charging facilities. Time series data is acquired using sliding window segmentation (60-second windows with 50% overlap). The device logs are synchronized with the grid status timestamps (error < 100ms).

[0014] Multi-dimensional feature extraction and fusion: Extracts 23-dimensional time series features, including current fluctuation entropy, plugging and unplugging behavior slope, and harmonic distortion rate, and generates standardized feature vectors through Z-score normalization.

[0015] Graph neural network modeling: Construct a charging facility association graph, where nodes are charging devices (including power and geographic location attributes) and edges are grid homology or physical proximity (distance < 50m). The probability of anomaly propagation is quantified through the GNN message passing mechanism.

[0016] Dynamic maintenance strategy generation: Based on the health score model (formula: Among them, W i is the weight S i is the abnormal feature score)

[0017] Calculate the health of the equipment and combine the equipment value and risk level to generate the maintenance priority: P = H × V × R

[0018] Output precise scheduling instructions.

[0019] As a preferred technical solution, the calculation of the current fluctuation entropy value includes:

[0020] The approximate entropy (ApEn) is calculated for the current waveform within a 1-minute window using the formula:

[0021] ApEn(m,r,N)=φ m (r)-φ m+1 (r)

[0022] Where m is the pattern dimension, r is the similarity tolerance, and N is the data length. When ApEn>0.65, it is determined to be an abnormal fluctuation.

[0023] As a preferred technical solution, the calculation of the plugging behavior slope includes:

[0024] The change rate of plug-in and plug-out times per hour is calculated using the following formula:

[0025] Slope = ΔN / Δt

[0026] When the slope is greater than 3 times / minute, it is judged as abnormally high-frequency use.

[0027] As a preferred technical solution, the graph neural network modeling adopts a 3-layer GraphSAGE structure and is configured with a 2-layer clustering head. The hidden layer dimension is 256, and the model inference delay is less than 150ms / device.

[0028] As a preferred technical solution, the dynamic maintenance strategy further includes:

[0029] When a group abnormality is detected, if:

[0030]

[0031] Where M is the total number of associated devices, Iabnormal(k) is the device, and k is the abnormal indication function, then the grid side maintenance work order is triggered first.

[0032] A technical solution for a charging facility profiling and abnormal pattern detection system, including:

[0033] Data acquisition and preprocessing module for high-frequency current sampling and timestamp synchronization;

[0034] A feature extraction and fusion module, configured to calculate multi-dimensional time series features and perform normalization;

[0035] Graph neural network modeling module, used to build device association graphs and analyze abnormal propagation paths;

[0036] Dynamic maintenance strategy module, which generates scheduling instructions based on health score and priority algorithm;

[0037] The model training and deployment module uses TensorRT to accelerate inference and supports real-time data processing of more than 100,000 charging piles.

[0038] As a preferred technical solution, the graph neural network modeling module further includes:

[0039] The visualization unit is used to present the abnormal propagation path and the device association topology, where the edge weight is calculated by the formula:

[0040] w ij =softmax(α ij ).

[0041] As a preferred technical solution, the dynamic maintenance strategy module integrates the grid status data and detects the grid fluctuation coefficient. Automatically generate grid-based maintenance plans.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] This invention provides a method and system for profiling and detecting abnormal patterns in charging facilities. Through intelligent means, it achieves high-precision anomaly detection and prediction for charging facilities. First, by sampling the high-frequency current of the charging facilities and analyzing time-series data, it captures more detailed and comprehensive information about the equipment's operating status, effectively improving the accuracy of anomaly detection. Compared to traditional threshold-based detection methods, this solution can identify more complex anomaly patterns, such as intermittent overloads and poor contact, significantly reducing the rate of missed detections.

[0044] Secondly, the present invention utilizes graph neural network modeling technology to construct a charging facility association diagram and quantify the probability of anomaly propagation. This innovation enables the system to deeply analyze the interdependence between charging pile groups, predicting or identifying group failures in advance, thereby improving the efficiency and effectiveness of operations and maintenance. By presenting the anomaly propagation path and the device association topology through visualization units, operations and maintenance personnel can more intuitively understand the mutual influence between devices, providing strong support for precise scheduling.

[0045] Furthermore, this invention proposes a dynamic maintenance strategy that generates scheduling instructions based on equipment health scores and a priority algorithm. This strategy not only considers the equipment's operating status but also incorporates its value and risk level, enabling precise scheduling of maintenance resources. By integrating grid status data, the system can automatically generate a grid-based maintenance plan when grid fluctuations are detected, further improving the timeliness and effectiveness of maintenance.

[0046] The present invention has achieved significant technological progress in the detection and maintenance of abnormalities in charging facilities and has broad application prospects and market value. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 The module diagram of a charging facility profiling and abnormal pattern detection method and system. DETAILED DESCRIPTION

[0048] The features and exemplary embodiments of various aspects of the present invention will be described in detail below. In order to make the objects, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. For those skilled in the art, the present invention can be implemented without some of these specific details. The following description of the embodiments is merely intended to provide a better understanding of the present invention by illustrating examples of the present invention.

[0049] like Figure 1 As shown, the present invention provides a charging facility profiling and abnormal pattern detection method technical solution, which includes the following steps:

[0050] Data acquisition and preprocessing: High-frequency current sampling (1kHz) is performed on charging facilities. Time series data is acquired using sliding window segmentation (60-second windows with 50% overlap). The device logs are synchronized with the grid status timestamps (error < 100ms).

[0051] Multi-dimensional feature extraction and fusion: Extracts 23-dimensional time series features, including current fluctuation entropy, plugging and unplugging behavior slope, and harmonic distortion rate, and generates standardized feature vectors through Z-score normalization.

[0052] Graph neural network modeling: Construct a charging facility association graph, where nodes are charging devices (including power and geographic location attributes) and edges are grid homology or physical proximity (distance < 50m). The probability of anomaly propagation is quantified through the GNN message passing mechanism.

[0053] Dynamic maintenance strategy generation: Based on the health score model (formula: Among them, Wi is the weight and Si is the abnormal feature score)

[0054] Calculate the health of the equipment and combine the equipment value and risk level to generate the maintenance priority: P = H × V × R

[0055] Output precise scheduling instructions.

[0056] The calculation of the current fluctuation entropy value includes:

[0057] The approximate entropy (ApEn) is calculated for the current waveform within a 1-minute window using the formula:

[0058] ApEn(m,r,N)=φ m (r)-φ m+1 (r)

[0059] Where m is the pattern dimension, r is the similarity tolerance, and N is the data length. When ApEn>0.65, it is determined to be an abnormal fluctuation.

[0060] The calculation of the plugging behavior slope includes:

[0061] The change rate of plug-in and plug-out times per hour is calculated using the following formula:

[0062] Slope = ΔN / Δt

[0063] When the slope is greater than 3 times / minute, it is judged as abnormally high-frequency use.

[0064] The graph neural network modeling adopts a 3-layer GraphSAGE structure and is configured with a 2-layer clustering head. The hidden layer dimension is 256, and the model inference delay is less than 150ms / device.

[0065] The dynamic maintenance strategy further includes:

[0066] When a group abnormality is detected, if:

[0067]

[0068] Where M is the total number of associated devices, Iabnormal(k) is the device, and k is the abnormal indication function, then the grid side maintenance work order is triggered first.

[0069] A technical solution for a charging facility profiling and abnormal pattern detection system, including:

[0070] Data acquisition and preprocessing module for high-frequency current sampling and timestamp synchronization;

[0071] A feature extraction and fusion module, configured to calculate multi-dimensional time series features and perform normalization;

[0072] Graph neural network modeling module, used to build device association graphs and analyze abnormal propagation paths;

[0073] Dynamic maintenance strategy module, which generates scheduling instructions based on health score and priority algorithm;

[0074] The model training and deployment module uses TensorRT to accelerate inference and supports real-time data processing of more than 100,000 charging piles.

[0075] The graph neural network modeling module further includes:

[0076] The visualization unit is used to present the abnormal propagation path and the device association topology, where the edge weight is calculated by the formula:

[0077] w ij =softmax(α ij ).

[0078] The dynamic maintenance strategy module integrates the grid status data and detects the grid fluctuation coefficient. Automatically generate grid-based maintenance plans.

[0079] Comparison between existing technology and the present invention:

[0080]

[0081] The working principle and usage process of this invention are based on key steps: high-frequency current sampling, multi-dimensional feature extraction, graph neural network modeling, and dynamic maintenance strategy generation. First, through the data acquisition and preprocessing module, the system samples high-frequency current from charging facilities and uses sliding window segmentation technology to obtain time series data. It also synchronizes device logs with grid status timestamps to ensure data accuracy and timeliness.

[0082] Subsequently, the multidimensional feature extraction and fusion module extracts 23-dimensional time series features from the collected time series data, including current fluctuation entropy, plug-in behavior slope, harmonic distortion rate, etc., and generates standardized feature vectors through the Z-score normalization method, providing a basis for subsequent analysis.

[0083] During the graph neural network modeling phase, the system constructs a charging facility association graph, treating charging equipment as nodes. Edge connections are established based on grid homology or physical proximity, and the probability of anomaly propagation is quantified through the GNN message passing mechanism. This step enables the system to deeply analyze the impact of associations between charging pile groups and predict or identify group failures in advance.

[0084] Finally, the dynamic maintenance strategy generation module calculates equipment health based on a health scoring model, combines equipment value and risk level to generate maintenance priorities, and outputs precise dispatch instructions. When a cluster of anomalies is detected, the system prioritizes grid-side maintenance work orders, further improving the timeliness and effectiveness of maintenance.

[0085] During use, the present invention leverages TensorRT to accelerate inference through the model training and deployment module, supporting real-time data processing for a large number of charging piles and ensuring efficient system operation. Furthermore, the visualization unit within the graph neural network modeling module can present abnormal propagation paths and device-related topology structures, providing operators with intuitive fault analysis tools and facilitating precise scheduling and maintenance decision-making.

[0086] In summary, the present invention achieves high-precision anomaly detection and prediction of charging facilities through intelligent means, effectively improving operation and maintenance efficiency and effectiveness, and has broad application prospects and market value.

[0087] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

[0088] While the embodiments of the present invention are described above, these embodiments do not exhaustively describe all details, nor do they limit the present invention to specific embodiments. Clearly, many modifications and variations can be made based on the above description. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to make good use of the present invention and its modifications and uses. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A charging facility profiling and abnormal pattern detection method, characterized in that: The following steps are involved: Data acquisition and preprocessing: High-frequency current sampling (1kHz) is performed on charging facilities. Time series data is acquired using sliding window segmentation (60-second windows with 50% overlap). The device logs are synchronized with the grid status timestamps (error < 100ms). Multi-dimensional feature extraction and fusion: Extracts 23-dimensional time series features, including current fluctuation entropy, plugging and unplugging behavior slope, and harmonic distortion rate, and generates standardized feature vectors through Z-score normalization. Graph neural network modeling: Construct a charging facility association graph, where nodes are charging devices (including power and geographic location attributes) and edges are grid homology or physical proximity (distance < 50m). The probability of anomaly propagation is quantified through the GNN message passing mechanism. Dynamic maintenance strategy generation: Based on the health score model (formula: Among them, W i is the weight S i is the abnormal feature score) Calculate the health of the equipment and combine the equipment value and risk level to generate the maintenance priority: P = H × V × R Output precise scheduling instructions.

2. The charging facility profiling and abnormal pattern detection method according to claim 1, characterized in that: The calculation of the current fluctuation entropy value includes: The approximate entropy (ApEn) is calculated for the current waveform within a 1-minute window using the formula: ApEn(m, r, N)=φ m (r)-φ m+1 (r) Where m is the pattern dimension, r is the similarity tolerance, and N is the data length. When ApEn>0.65, it is determined to be an abnormal fluctuation.

3. The charging facility profiling and abnormal pattern detection method according to claim 1, characterized in that: The calculation of the plugging behavior slope includes: The change rate of plug-in and plug-out times per hour is calculated using the following formula: Slope = ΔN / Δt When the slope is greater than 3 times / minute, it is judged as abnormally high-frequency use.

4. The charging facility profiling and abnormal pattern detection method according to claim 1, characterized in that: The graph neural network modeling adopts a 3-layer GraphSAGE structure and is configured with a 2-layer clustering head. The hidden layer dimension is 256, and the model inference delay is less than 150ms / device.

5. The charging facility profiling and abnormal pattern detection method according to claim 1, characterized in that: The dynamic maintenance strategy further includes: When a group abnormality is detected, if: Where M is the total number of associated devices, Iabnormal(k) is the device, and k is the abnormal indication function, then the grid side maintenance work order is triggered first.

6. A charging facility profiling and abnormal pattern detection system according to any one of claims 1 to 5, characterized in that: include: Data acquisition and preprocessing module for high-frequency current sampling and timestamp synchronization; A feature extraction and fusion module, configured to calculate multi-dimensional time series features and perform normalization; Graph neural network modeling module, used to build device association graphs and analyze abnormal propagation paths; Dynamic maintenance strategy module, which generates scheduling instructions based on health score and priority algorithm; The model training and deployment module uses TensorRT to accelerate inference and supports real-time data processing of more than 100,000 charging piles.

7. The system according to claim 6, characterized in that The graph neural network modeling module further includes: The visualization unit is used to present the abnormal propagation path and the device association topology, where the edge weight is calculated by the formula: w ij =softmax(a ij )。 8. The system according to claim 6, wherein: The dynamic maintenance strategy module integrates the grid status data and detects the grid fluctuation coefficient. Automatically generate grid-based maintenance plans.

Images