A method and system for identifying operating conditions of a low-voltage cabinet

Abstract

The present application relates to the field of data processing technology, and more particularly to a method and system for identifying the operating conditions of low-voltage cabinets. The method comprises the following steps: obtaining historical operating condition data of the low-voltage cabinets, randomly sampling and constructing a preset number of isolated trees; calculating the detection advantage of each isolated tree, and selecting isolated trees based on the detection advantage to construct an isolation forest model; and collecting low-voltage cabinet operating data in real time, inputting the data into the isolation forest model to monitor the operating conditions of the low-voltage cabinets. The present application improves the timeliness of identifying the operating conditions of the low-voltage cabinets.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method and system for identifying the operating conditions of a low-voltage cabinet. Background Art

[0002] Low-voltage cabinets refer to low-voltage distribution cabinets. These are complete sets of equipment used for power management and distribution in power systems. They include components such as circuit breakers, contactors, relays, and transformers, playing a critical role in powering and maintaining data for underlying equipment. As crucial power distribution and control equipment at the end of the power system, their operating status directly impacts the stability and safety of the entire power grid. Real-time monitoring of low-voltage cabinet operating conditions can effectively identify abnormal conditions (such as overload, short circuit, and excessive temperature rise), providing early warning of faults and preventing safety incidents such as equipment damage and electrical fires. Furthermore, operating data can be used to conduct intelligent operations and maintenance, condition assessment, and energy-saving management, improving system efficiency and facilitating the digitalization and intelligent upgrade of industrial systems. Currently, anomaly detection algorithms are primarily used to identify and monitor the operating conditions of low-voltage cabinets. For example, the Isolation Forest (iForest) algorithm randomly partitions sample data along randomly selected feature dimensions to construct multiple isolation trees. The length of the path from a sample to a leaf node in each tree reflects the degree of its isolation. The shorter the path, the more likely it is an outlier.

[0003] In real-world low-voltage cabinet operating condition monitoring scenarios, data is typically high-dimensional (including electrical and environmental parameters). This data creates an isolation forest with a high number of isolated trees and a redundant structure. However, with increasing demands for timely real-time monitoring, the overall efficiency of traditional isolation forest algorithms is insufficient to meet the requirements for timely identification of low-voltage cabinet operating conditions. Summary of the Invention

[0004] In order to solve the problem that the traditional method for identifying the operating conditions of low-voltage cabinets is inefficient and difficult to meet the requirements of low-voltage cabinets for timely identification and monitoring, the present application provides a method and system for identifying the operating conditions of low-voltage cabinets.

[0005] In a first aspect, the present application provides a method for identifying the operating condition of a low-voltage cabinet, which adopts the following technical solution:

[0006] A method for identifying the operating condition of a low-voltage cabinet comprises the following steps: obtaining historical operating condition data of the low-voltage cabinet, randomly sampling and constructing a preset number of isolated trees; calculating the detection advantage of each isolated tree, and selecting isolated trees based on the detection advantage to construct an isolation forest model; collecting low-voltage cabinet operating data in real time, and inputting the data into the isolation forest model to monitor the operating condition of the low-voltage cabinet;

[0007] Among them, the steps of calculating the detection advantage of each isolated tree include: for each isolated tree, analyzing the data corresponding to the isolated tree node and constructing a hierarchical segmentation index; based on the depth of the isolated tree, determining the weight of the hierarchical segmentation index corresponding to each level of the isolated tree; taking the result of the weighted summation of each level as the segmentation gain; analyzing the sample set corresponding to the isolated tree, calculating the sample representativeness of the sample set on behalf of the overall data set; and taking the product of the segmentation gain and the sample representativeness as the detection advantage.

[0008] During the construction of traditional isolation trees, the data set is randomly split to isolate abnormal data. Among the constructed isolation trees, some may have poor recognition performance on abnormal data. Therefore, in this application, the monitoring advantage of each isolation tree is calculated. The recognition performance of the isolation tree on abnormal data is then evaluated based on the detection advantage. Isolated trees with good abnormal data detection performance are selected to construct an isolation forest. Isolation tree models with poor abnormal data monitoring performance are discarded, reducing redundant structures and improving computational efficiency, thereby increasing the timeliness of identifying the operating conditions of the low-voltage cabinet.

[0009] The monitoring advantage of each isolated tree is mainly obtained by constructing the sample quality of the isolated tree and the hierarchical segmentation index of each level in the isolated tree, thereby improving the accuracy and robustness of obtaining the detection advantage of the isolated tree, thereby avoiding the impact of abandoning the isolated tree on the monitoring accuracy.

[0010] Optionally, the step of constructing a hierarchical segmentation index includes: for each level, calculating the segmentation quality index of the nodes in the level; taking two child nodes with the same parent node in the same level as a collateral point pair, and obtaining the absolute difference in the segmentation quality index of the two child nodes in the collateral point pair; taking the average of the absolute differences of multiple collateral point pairs as the distribution difference, obtaining the child-level data fluctuation of the child node of the level and the parent-level data fluctuation of the corresponding parent node, and obtaining the fluctuation difference based on the difference between the parent-level data fluctuation and the child-level data fluctuation; and taking the product of the distribution difference and the fluctuation difference as the hierarchical segmentation index.

[0011] In an isolation tree, child nodes are formed by segmenting the data in a dataset. If the segmentation is effective at a level, the data distribution in the dataset corresponding to the child nodes at that level will be significantly different from the data distribution of the parent nodes. Furthermore, because abnormal data in the data is discrete, if the abnormal data is segmented and isolated at a level, the fluctuations in the data between the child nodes and the parent nodes will also vary significantly. Therefore, this application calculates the hierarchical segmentation index by obtaining the distribution difference based on the child node segmentation index and the parent node segmentation index, and obtaining the fluctuation difference based on the difference between the child-level data fluctuation and the parent-level data fluctuation.

[0012] Optionally, the standard deviation of the data in the corresponding data set of all child nodes in each level is used as the child data fluctuation of each level, and the standard deviation of the data in the corresponding data set of the parent node corresponding to the child node is used as the parent data fluctuation of each level.

[0013] Optionally, the step of calculating the segmentation quality index of the nodes in the hierarchy includes: for any child node, obtaining the value range interval of the data in the child node and the value range interval of the data in the parent node corresponding to the child node, and obtaining the interval change degree based on the two value range intervals; obtaining the data density of each node according to the data distribution in the isolated tree node, and obtaining the density change degree according to the data density of the child node and the data density of the parent node; and taking the product of the density change degree and the interval change degree as the segmentation quality index.

[0014] The dataset corresponding to the parent node is split into child nodes, with each node corresponding to a single dataset. The distribution of the dataset corresponding to the child node changes after the split. Therefore, the degree of interval change is determined based on the value range of the data in the child node's dataset. The data density is then calculated based on the distance between the data in the child node's dataset, thereby determining the degree of density change from the parent node to the child node.

[0015] Optionally, the ratio of the length of the value range interval corresponding to the parent node to the length of the value range interval corresponding to the child node is used as the interval change degree.

[0016] Compare the left and right endpoints of the value range of the data in the corresponding data set of the child node and the parent node to obtain the change in data distribution from the parent node to the child node, and then reflect the degree of interval change.

[0017] Optionally, the step of obtaining the data density of the subnode according to the data distribution in the subnode includes: any two data points in the subnode constitute a reference point pair, and obtaining the sum of the reciprocals of the Euclidean distances of all reference point pairs as the data density of the subnode.

[0018] Traverse the distance between any two data points in the data set to obtain the data density that can reflect the data distribution in the data set.

[0019] Optionally, the ratio of the data density of the child node to the data density of the parent node is used as the density change degree.

[0020] Optionally, based on the depth of the isolated tree, the step of determining the weight of the hierarchical segmentation index corresponding to each level of the isolated tree includes: for any level, obtaining the difference between the depth of the isolated tree and the depth of the level; and using the ratio of the difference to the depth of the isolated tree as the weight of the hierarchical segmentation index corresponding to the level.

[0021] In an isolation tree structure, if the abnormal data is quickly segmented and isolated, the isolation tree has a better recognition effect on abnormal data. Therefore, the hierarchical segmentation index of each level is weighted based on the depth of each level to improve the accuracy of the final calculated detection advantage.

[0022] Optionally, the step of analyzing the sample set corresponding to the isolated tree and calculating the sample representativeness of the sample set for the entire data set includes: using a Kolmogorov-Smirnov algorithm to test a test statistic between the sample set and the entire data set, and obtaining the sample representativeness based on the test statistic.

[0023] In a second aspect, the present application provides an operating condition identification system for a low-voltage cabinet, which adopts the following technical solutions:

[0024] A system for identifying operating conditions of a low-voltage cabinet comprises a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, the system implements the above-mentioned method for identifying operating conditions of a low-voltage cabinet.

[0025] The beneficial effect is: the above-mentioned method for identifying the operating condition of a low-voltage cabinet is generated into a computer program and stored in a memory so as to be loaded and executed by a processor, thereby making a system based on the memory and the processor for easy use.

[0026] This application has the following technical effects:

[0027] Calculate the detection advantages of isolated trees, evaluate the ability of isolated trees to identify abnormal data, screen out isolated trees with strong ability to identify abnormal data, eliminate isolated trees with poor ability to identify abnormal data, reduce the amount of calculation in the subsequent real-time monitoring process, and improve the timeliness of low-voltage cabinet operating condition identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a method flow chart of a method for identifying the operating condition of a low-voltage cabinet in an embodiment of the present application.

[0029] Figure 2 This is a method flow chart of step S2 of a method for identifying the operating condition of a low-voltage cabinet in the present application.

[0030] Figure 3 This is a method flow chart of step S21 of a method for identifying the operating condition of a low-voltage cabinet in the present application. DETAILED DESCRIPTION

[0031] The embodiment of the present application discloses a method for identifying the operating conditions of a low-voltage cabinet, which obtains historical operating data of the low-voltage cabinet, constructs a preset number of isolated trees based on the historical operating data, analyzes the detection advantage of the isolated trees in identifying abnormal data, and selects isolated trees based on the detection advantage to construct an isolation forest model. The data generated during the real-time operation of the low-voltage cabinet is monitored by the isolation forest model to achieve identification and monitoring of the operating conditions of the low-voltage cabinet. In the isolation forest model of this method, isolated trees with higher detection advantages are selected to form the isolation forest model, and isolated trees with lower abnormal data identification capabilities are eliminated, thereby reducing redundant structures, simplifying calculations in the subsequent monitoring process, improving the efficiency of real-time detection, and thus improving the timeliness of real-time monitoring.

[0032] Reference Figure 1 , a method for identifying the operating condition of a low-voltage cabinet, including steps S1 to S3.

[0033] S1: Obtain historical operating condition data of the low-voltage cabinet, and randomly sample to construct a preset number of isolated trees.

[0034] The low-voltage cabinet is a complete set of equipment that integrates multiple electrical components. In order to determine the operating conditions of the low-voltage cabinet during operation, it is usually necessary to collect and analyze information from different dimensions during the operation of the low-voltage cabinet, including electrical information during the operation of the low-voltage cabinet and external environment information. The data collected in this embodiment includes: current data, voltage data, and main circuit temperature data of the low-voltage cabinet collected through current transformers, voltage transformers, and temperature sensors; and also includes: ambient temperature and humidity sensors to collect temperature data and humidity data in the working environment of the low-voltage cabinet. In this embodiment, the operating condition data is collected once per second, and in other embodiments, it can be adjusted according to actual conditions.

[0035] During the above-mentioned operating condition data collection process, the collected data may contain anomalies such as missing values ​​due to factors such as electromagnetic interference. To avoid affecting the subsequent calculation results, this application uses a regression filling method to fill in the missing values. At the same time, to avoid the inability to calculate due to different dimensions between data of different dimensions, this application uses a maximum and minimum normalization method to normalize the collected data. The normalized data is collectively referred to as operating condition data. The processing procedures of the regression filling method and the maximum and minimum normalization method are well-known technologies and will not be repeated here.

[0036] An isolation tree is constructed using the traditional isolation forest algorithm for the low-voltage cabinet operating condition data. During the isolation tree construction process, each moment of the operating condition data can be considered as a data point containing multidimensional data (current data, voltage data, main circuit temperature, temperature data, and humidity data). During the isolation tree construction process, the data points are randomly segmented by randomly selecting dimensions and segmentation points until all data points are isolated, completing the construction of the isolation tree. This step is conventional in the art and will not be described in detail here. In this application, the number of isolation trees is preset to 200, the amount of pre-collected historical low-voltage cabinet operating condition data is 3000, the depth of the isolation tree is 15, and the number of randomly selected sample moments during isolation tree construction is 256. The relevant parameters in the above algorithm can be selected according to the specific circumstances. Thus, 200 isolation trees are constructed using the traditional isolation forest algorithm. Of course, during the isolation tree construction process, the preset number of isolation trees, the number of collected operating condition data points, and the depth of the isolation trees can all be customized.

[0037] S2: Calculate the detection advantage of each isolated tree, select isolated trees based on the detection advantage to build an isolation forest model.

[0038] Traditional isolation forests consist of multiple isolated trees constructed based on random samples. Different isolated trees have varying abilities to identify abnormal data, and some may not be able to effectively detect abnormal data. These isolated trees can affect the accuracy of anomaly detection and computational efficiency during subsequent real-time monitoring. Therefore, this embodiment calculates the detection advantage of each isolated tree and selects high-quality isolated trees based on this advantage to form an isolation forest detection model, thereby improving the timeliness of real-time monitoring.

[0039] Reference Figure 2 The step of calculating the monitoring advantage of each isolated tree includes step S21 and step S22.

[0040] S21: For each isolated tree, analyze the data corresponding to the isolated tree node and construct a hierarchical segmentation index.

[0041] When a low-voltage cabinet experiences abnormal operating conditions, the relevant operating condition data will exceed the normal operating range. For example, during the operation of the low-voltage cabinet, the current may suddenly increase due to line short circuits, internal equipment faults, insulation damage, etc., far exceeding the normal value, and the temperature may continue to rise, exceeding the normal operating range of the equipment; or due to poor contact, load fluctuations, etc., the current may fluctuate intermittently, and the temperature may change significantly.

[0042] Reference Figure 3 , step S21 includes step S211-step S212.

[0043] S211: For each level, calculate the segmentation quality index of the nodes in the level.

[0044] For an isolated tree structure, which includes multiple levels, each level corresponds to multiple child nodes and leaf nodes. Therefore, in this method, the segmentation quality index of the child nodes and leaf nodes is first determined based on the data changes between the child nodes and their corresponding parent nodes. Then, the level segmentation index is determined based on the segmentation quality index of the child nodes and leaf nodes.

[0045] Specifically, for any child node, the value range interval of the data in the child node and the value range interval of the data in the parent node corresponding to the child node are obtained, and the interval change degree is obtained based on the two value range intervals.

[0046] In an isolation tree, a parent node represents the node immediately above a node, and the nodes immediately below the parent node are the child nodes of the parent node. During the isolation tree construction process, a sample set is first obtained from the entire dataset. A dimension and a split point are then randomly selected to partition the sample set. For example, consider a sample set containing 200 data points. Each data point contains information from two dimensions: current and voltage. First, a split dimension and a split point are randomly selected. For example, the split dimension is voltage, and the split point is randomly selected within the sample set's value range. For example, a value of 200 is chosen as the split point. This splits the sample set into two groups: one with a voltage greater than 200 and another with a voltage less than 200. Two child nodes are also generated. For any child node, the node is used as the corresponding dataset, and a new split dimension and split point are randomly selected. The above steps are repeated to complete the isolation tree construction. As can be seen, in an isolation tree structure, each node corresponds to a dataset, and a split dimension is associated between the parent and child nodes.

[0047] First, for any child node, obtain the value range interval of the data in the child node and the value range interval of the data in the parent node corresponding to the child node, and obtain the interval change degree based on the two value range intervals.

[0048] Based on the above analysis, for any node (child or parent), the value range of the data in its corresponding dataset can be obtained. The data in the dataset corresponding to the parent node is split to form child nodes, and the value range of the data in the dataset corresponding to the child node changes. Therefore, based on this change, the degree of change between the parent and child nodes can be determined.

[0049] Specifically, the value range interval of the data in the data set corresponding to the child node and the value range interval corresponding to the corresponding parent node are obtained, and the interval length of the value range interval corresponding to the parent node and the interval length of the value range interval corresponding to the child node are used as the interval change degree.

[0050] Then, the data density of each node is obtained according to the data distribution in the isolated tree node, and the density change degree is obtained according to the data density of the child node and the data density of the parent node; the product of the density change degree and the interval change degree is used as the segmentation quality index.

[0051] For the data density of each node, each node corresponds to a data set, which includes multiple data. Any two data points of the multiple data constitute a reference point pair. The sum of the reciprocals of the Euclidean distances of all reference point pairs is obtained as the data density of the child node.

[0052] The larger the Euclidean distance between two data points, the more dispersed the data points are. A larger sum of the reciprocals of the Euclidean distances between any two data points indicates a more dispersed distribution of the data set overall, and thus a lower data density. Normally, the data density corresponding to an outlier reference point is low. If the data density varies significantly from that of the parent node, this indicates that the hierarchical segmentation is effective and can distinguish outliers.

[0053] After obtaining the data density of the child node and the parent node, the ratio of the data density of the child node to the data density of the parent node is used as the density change degree, and then the product of the density change degree and the interval change degree is used as the segmentation quality index.

[0054] Specifically, the calculation of the segmentation quality index can also be expressed by the following formula:

[0055] Where, Indicates the segmentation quality index of the child node; Indicates the length of the data value range of the data set corresponding to the parent node; Indicates the length of the data value range of the data set corresponding to the child node; Indicates the data density of the child node; Indicates the data density of the parent node.

[0056] Indicates the degree of interval change, reflecting the change in the value range of the data set before and after segmentation. Indicates the degree of density change.

[0057] For example: Assume that in the third layer, node A3 is divided into two child nodes; the voltage data of the first child node corresponds to the value range interval: [215, 222], and the value range interval of the parent node A3 is [210, 230]; the interval length of the value range interval of the parent node is ; The length of the value range interval corresponding to the child node is 7; the maximum value is the degree of interval change: ; Assume that the child node density = for example, the density is 0.8, the parent node density is 0.5, and finally .

[0058] S212: Take two child nodes with the same parent node in the same level as a collateral point pair, and obtain the absolute difference of the segmentation quality index of the two child nodes in the collateral point pair; take the average of the absolute differences of multiple collateral point pairs as the distribution difference, obtain the child data fluctuation of the child node of the level and the parent data fluctuation of the corresponding parent node, and obtain the fluctuation difference based on the difference between the parent data fluctuation and the child data fluctuation; take the product of the distribution difference and the fluctuation difference as the level segmentation index.

[0059] The calculation of the hierarchical segmentation index can be expressed by the following formula:

[0060] Where, Represents the isolated tree Hierarchical segmentation index of the layer; Indicates the Tier The segmentation quality index of the first child node of the collateral point pair; Indicates the Tier The split quality index of the second child node of the collateral point pair, child nodes and child nodes belong to the same parent node; Indicates the Data volatility of sub-nodes in the layer; Indicates the The data fluctuation of the parent node corresponding to the child node in the layer; Indicates the number of collateral point pairs.

[0061] For example, assume three collateral point pairs, where the segmentation quality indices of the two sub-nodes in the first collateral point are 1.5 and 2.0 respectively, the segmentation quality indices of the two sub-nodes in the second collateral point are 1.2 and 1.6 respectively, and the segmentation quality indices of the two sub-nodes in the third collateral point are 2.1 and 1.7 respectively. Substitute this part of the formula into get

[0062] ; Assume that this part Equal to 0.55, the final hierarchical segmentation index is 0.433*0.55=0.238.

[0063] In this embodiment, the standard deviation of the data in the data set corresponding to all child nodes of the level is used as the data fluctuation of the child nodes; the standard deviation of the data in the data set corresponding to all parent nodes of the level is used as the data fluctuation of the parent nodes.

[0064] When identifying abnormal operating condition data of low-voltage cabinets through isolation trees, since normal operating condition data is distributed relatively evenly in the data space and has certain similarities and regularities, the changes in operating condition data in various dimensions are relatively smooth. However, abnormal operating condition data will break the original uniform distribution due to deviation from the distribution of normal data, making the data distribution within the node uneven, resulting in a larger calculated standard deviation. If a certain level in the isolation tree has a better recognition effect on abnormal data, it can segment the abnormal operating condition data into leaf nodes, thereby reducing the standard deviation of the operating condition data in the child nodes. At the same time, the data distribution difference between the child nodes and leaf nodes in the same layer will be greater, that is, the difference in the feature segmentation quality index between the child nodes and leaf nodes - The bigger.

[0065] S22: Based on the depth of the isolated tree, determine the weights of the hierarchical segmentation index corresponding to each level of the isolated tree; use the weighted sum of each level as the segmentation gain; analyze the sample set corresponding to the isolated tree, and calculate the sample representativeness of the sample set on behalf of the entire data set; use the product of the segmentation gain and the sample representativeness as the detection advantage.

[0066] Specifically, the calculation formula for the detection advantage of the isolated tree can be expressed as:

[0067] Where, represents the detection advantage of isolated trees; Indicates the depth of the isolation tree, which is mainly used to normalize the molecular part. Based on the above content, the depth of the isolation tree in this application is 15. It should be noted that the depth of the isolation tree does not include the level where the root node is located; Indicates sample representativeness; Indicates the current level depth; Represents the isolated tree The hierarchical segmentation index corresponding to the layer.

[0068] Indicates sample representativeness, mainly used to evaluate whether the sample set used to construct the isolation tree can effectively represent the entire data set. In this embodiment, the mean of the test statistics obtained after the Kolmogorov-Smirnov test on each dimension between the sample set and the total sample set is used as the sample representativeness.

[0069] is the weight of the hierarchical segmentation index of each level; in the isolation tree, if the abnormal data can be isolated faster, it means that the isolation tree has a better recognition effect on abnormal data. Based on this, the hierarchical segmentation index of each level is weighted. The smaller the level depth, the greater the weight.

[0070] This part represents the result of weighted summation of the level segmentation index, representing the segmentation gain of the level.

[0071] After the detection advantage calculation is completed, the isolation forest model is constructed by screening isolated trees based on the detection advantage.

[0072] The steps of isolated screening in this application include: obtaining the mean of the detection advantages of all isolated trees, and extracting isolated trees with detection advantages greater than the mean to form an isolation forest model.

[0073] In other embodiments, a detection threshold may be preset, and isolated trees with a detection advantage greater than the detection threshold are selected to form an isolation forest model.

[0074] S3: Collect low-voltage cabinet operating data in real time and input it into the isolation forest model to monitor the operating conditions of the low-voltage cabinet.

[0075] The real-time data points are sequentially fed into all retained isolation tree models. The path length from each tree to the leaf node is calculated for each point. The path lengths for each data point across all isolation trees are summed up and combined with the isolation forest's statistical scoring function to calculate a final anomaly score. Anomaly scores above a preset threshold are considered abnormal operating conditions.

[0076] An embodiment of the present application also discloses an operating condition identification system for a low-voltage cabinet, including a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, an operating condition identification method for a low-voltage cabinet according to the present application is implemented.

[0077] The above system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface. The configuration and functions of these components are known in the art and will not be described in detail here.

[0078] The above are all preferred embodiments of the present application, and are not intended to limit the scope of protection of the present application. Therefore, any equivalent changes made based on the structure, shape, and principle of the present application should be included in the scope of protection of the present application.

Claims

1. A method for identifying the operating condition of a low-voltage cabinet, characterized in that: The method includes the following steps: obtaining historical operating condition data of the low-voltage cabinet, randomly sampling and constructing a preset number of isolated trees; calculating the detection advantage of each isolated tree, and selecting isolated trees based on the detection advantage to construct an isolation forest model; collecting low-voltage cabinet operating data in real time, and inputting the data into the isolation forest model to monitor the operating condition of the low-voltage cabinet; The step of calculating the detection advantage of each isolated tree includes: for each isolated tree, analyzing the data corresponding to the isolated tree node and constructing a hierarchical segmentation index; The steps of constructing a hierarchical segmentation index include: for each level, calculating the segmentation quality index of the nodes in the level; treating two child nodes with the same parent node in the same level as a collateral point pair, obtaining the absolute difference of the segmentation quality index of the two child nodes in the collateral point pair; using the average of the absolute differences of multiple collateral point pairs as the distribution difference, obtaining the child data fluctuation of the child node in the level and the parent data fluctuation of the corresponding parent node, and obtaining the fluctuation difference based on the difference between the parent data fluctuation and the child data fluctuation; and using the product of the distribution difference and the fluctuation difference as the hierarchical segmentation index; The standard deviation of the data in the corresponding data set of all child nodes in each level is used as the child data fluctuation of each level, and the standard deviation of the data in the corresponding data set of the parent node corresponding to the child node is used as the parent data fluctuation of each level; The steps of calculating the segmentation quality index of the nodes in the hierarchy include: for any child node, obtaining a value range interval of the data in the child node and a value range interval of the data in the parent node corresponding to the child node, and obtaining an interval variation degree based on the two value range intervals; obtaining a data density of each node based on the data distribution in the isolated tree node, and obtaining a density variation degree based on the data density of the child node and the data density of the parent node; and taking the product of the density variation degree and the interval variation degree as the segmentation quality index; Based on the depth of the isolation tree, the weights of the hierarchical segmentation index corresponding to each level of the isolation tree are determined; the weighted sum of each level is used as the segmentation gain; the sample set corresponding to the isolation tree is analyzed, and the sample representativeness of the sample set on behalf of the entire data set is calculated; the product of the segmentation gain and the sample representativeness is used as the detection advantage.

2. The method for identifying the operating condition of a low-voltage cabinet according to claim 1, characterized in that: The ratio of the length of the value range interval corresponding to the parent node to the length of the value range interval corresponding to the child node is used as the interval change degree.

3. The method for identifying the operating condition of a low-voltage cabinet according to claim 1, characterized in that: The step of obtaining the data density of the subnode according to the data distribution in the subnode includes: any two data points in the subnode constitute a reference point pair, and obtaining the sum of the reciprocals of the Euclidean distances of all reference point pairs as the data density of the subnode.

4. The method for identifying the operating condition of a low-voltage cabinet according to claim 1, characterized in that: The ratio of the data density of the child node to the data density of the parent node is taken as the density change degree.

5. The method for identifying the operating condition of a low-voltage cabinet according to claim 1, characterized in that: Based on the depth of the isolated tree, the step of determining the weight of the level segmentation index corresponding to each level of the isolated tree includes: for any level, obtaining the difference between the depth of the isolated tree and the depth of the level; and using the ratio of the difference to the depth of the isolated tree as the weight of the level segmentation index corresponding to the level.

6. The method for identifying the operating condition of a low-voltage cabinet according to claim 1, characterized in that: The steps of analyzing the sample set corresponding to the isolated tree and calculating the sample representativeness of the sample set on behalf of the entire data set include: using the Kolmogorov-Smirnov algorithm to test a test statistic between the sample set and the entire data set, and obtaining the sample representativeness based on the test statistic.

7. An operating condition identification system for a low-voltage cabinet, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, an operating condition identification method for a low-voltage cabinet according to any one of claims 1 to 6 is implemented.

Images