A method and system for power equipment asset management

Abstract

The present invention relates to the technical field of data processing, and particularly relates to a power equipment asset management method and system. The method includes: determining the selection probability of each dimension under the child node according to the distribution of power data in the isolated tree child nodes of the power equipment; determining the necessity of splitting the child node according to the temporal regularity, selection probability, and the number of power data in the child node; when the necessity of splitting is greater than the splitting threshold, randomly selecting a dimension with unequal probability according to the selection probability to divide the child node; continuously updating the isolated forest according to the newly generated power data; identifying the current state of the power equipment according to the anomaly score of the current power data; and real-time synchronizing the current state of the power equipment to the digital twin model of the power equipment, so as to realize the monitoring and management of power equipment assets. The present invention accurately and efficiently identifies the state of power equipment, and improves the real-time performance of power equipment asset monitoring and management.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a power equipment asset management method and system. Background Art

[0002] Power equipment asset management is a key link to ensure the stable, safe and efficient operation of the power system. With the continuous growth of power demand and the complexity of the power system, the importance of power equipment asset management has become increasingly prominent. Traditional power equipment management methods mainly rely on manual records and simple maintenance plans, often facing problems such as inaccurate equipment fault prediction, unreasonable maintenance resource allocation, and low asset utilization efficiency, which affect the reliability and economy of the power system.

[0003] With the development of information technology and data analysis technology, modern power equipment asset management is gradually transforming towards intelligence and data-driven directions, and real-time monitoring and predictive maintenance of equipment status are carried out through sensor technology. For example, the patent document with the publication number CN117556347A discloses a power equipment fault prediction and health management method based on industrial big data, including: S1: constructing a data set of operation parameters containing fault information; S2: performing outlier and missing value processing and realizing normalization; S3: performing feature selection and selecting a feature group with strong correlation with equipment operation index parameters; S4: dividing the data of the feature group into a training set, a test set and a validation set; S5: creating a normal operation parameter prediction model of MSA-BLSTM for prediction; S6: using the validation set to verify the prediction model, taking the predicted operation parameters output by MSA-BLSTM as input, and using the isolation forest method model to perform fault determination to achieve equipment fault prediction; S7: performing equipment fault early warning and health status assessment according to the fault prediction results.

[0004] The above method can achieve power equipment fault prediction to a certain extent. However, since the power data of power equipment is streaming data, in order to realize real-time monitoring of the power equipment status, it is necessary to continuously construct new isolation forests based on newly generated power data, which is inefficient and occupies a large amount of computing resources. At the same time, it affects the real-time performance of power equipment status monitoring and is not conducive to power equipment asset monitoring and management. Summary of the Invention

[0005] In order to solve the problems of continuously constructing new isolation forests according to newly generated power data, which is inefficient and occupies a large amount of computing resources, affects the real-time performance of power equipment status monitoring, and is not conducive to power equipment asset monitoring and management, the present invention provides a power equipment asset management method and system.

[0006] In the first aspect, the present invention provides a power equipment asset management method, adopting the following technical solution:

[0007] A power equipment asset management method, comprising the steps of:

[0008] Constructing an isolation forest based on historical power data of power equipment; during the construction of the isolation forest: for each child node in the isolation tree, determining the selection probability of each dimension under the child node according to the distribution of the power data in each dimension of the child node, the selection probability being positively correlated with the size of the fluctuation range of the power data in each dimension of the child node and negatively correlated with the information entropy of the power data in each dimension of the child node; determining the necessity of splitting the child node, the necessity of splitting being negatively correlated with the temporal regularity of the power data in the child node and positively correlated with the maximum value among the selection probabilities of each dimension under the child node and the number of power data included in the child node; in response to the necessity of splitting being greater than a preset splitting threshold, randomly selecting a dimension with unequal probabilities according to the selection probabilities of each dimension under the child node as the splitting dimension of the child node, and using the splitting dimension to divide the power data in the child node into two new child nodes; adding the newly generated power data currently to the isolation forest and updating the isolation forest; determining the anomaly score of the newly generated power data currently according to the position of the newly generated power data currently in each isolation tree of the isolation forest, and identifying the current state of the power equipment according to the size of the anomaly score; and synchronizing the current state of the power equipment to the digital twin model of the power equipment in real time to achieve the monitoring and management of power equipment assets.

[0009] The effects are as follows: The present invention constructs an isolation forest based on historical power data, and on the basis of the constructed isolation forest, updates the isolation forest according to newly generated power data, avoiding continuously reconstructing the isolation forest, saving computing resources, improving the efficiency of abnormal identification of power equipment, and further improving the real-time performance of power equipment asset monitoring and management; The present invention calculates the selection probability of each dimension when dividing a node according to the distribution of each dimension of the power data in the node, and randomly selects a dimension to divide the node with unequal probability according to the selection probability, enhancing the probability that the dimension that may contain anomalies is selected, avoiding the situation in traditional isolation forests where dimensions are randomly selected with equal probability for division, which may lead to the dimension containing anomalies being selected relatively late, improving the iterative efficiency of the isolation forest, making abnormal power data as likely as possible to be included in the nodes in the shallower layers of the isolation tree, and enhancing the recognition accuracy of abnormal power data; The power data is time-series data, and abnormal power data usually appears continuously. The present invention determines the necessity of dividing a child node according to the time-series regularity of the power data in the child node, the selection probability of each dimension under the child node, and the number of power data included in the child node, and divides the child node according to the necessity of division, avoiding normal power data with periodic regular distribution or abnormal power data continuously distributed in time series from being divided into multiple child nodes, improving the division efficiency of the isolation tree and the recognition efficiency of abnormal power data, enhancing the accuracy and real-time performance of power equipment state recognition, and further improving the real-time performance of power equipment asset monitoring and management.

[0010] Preferably, the selection probabilities of each dimension under the child node satisfy the expression:

[0011] ; where is the selection probability of the th dimension under the child node; is the maximum value of the th dimension of all power data in the child node; is the minimum value of the th dimension of all power data in the child node; is the number of dimensions of the power data; is the information entropy of the th dimension of all power data in the child node; is the number of different numerical types in the th dimension of all power data in the child node; is the maximum value function.

[0012] The effect is as follows: Therefore, according to the distribution of each dimension of the power data in the node, the present invention calculates the selection probability of each dimension when dividing the node, so as to randomly select a dimension for division with unequal probabilities according to the selection probability, thereby enhancing the probability that the dimension that may contain anomalies is selected, enabling the dimension that may contain anomalies to be selected as early as possible, and making the abnormal power data as much as possible contained in the nodes in the shallower layer of the isolation tree, thereby improving the recognition accuracy of abnormal power data and the construction efficiency of the isolation forest.

[0013] Preferably, the time series regularity of the power data in the child node satisfies the expression:

[0014] ; where is the time series regularity of the power data in the child node; is the length of the th power data segment in the child node; is the number of power data segments in the child node; is the average length of all power data segments in the child node; represents the th power data segment and the th power data segment in the child node The time interval between them.

[0015] The effect is as follows: The present invention takes into account the periodic fluctuation law of power data, measures whether there may be anomalies in the power data by calculating the time series regularity of the power data in the child node, determines the necessity of dividing the child node according to the time series regularity, improves the division accuracy of abnormal power data, and further improves the recognition accuracy of anomalies.

[0016] Preferably, the necessity of dividing the child node satisfies the expression: ; where is the necessity of dividing the child node; is the time series regularity of the power data in the child node; is the selection probability of the th dimension under the child node; is the number of dimensions of the power data; is the number of power data contained in the child node; is the number of all power data; is the natural exponential function; is the maximum value function.

[0017] Preferably, adding the newly generated power data into the isolation forest and updating the isolation forest includes: for the newly generated power data, determining the leaf nodes to which the power data belongs in each isolation tree, and adding the power data into the corresponding leaf nodes; obtaining the splitting necessity of the corresponding leaf nodes after adding the power data; in response to the splitting necessity of the leaf nodes being greater than a preset splitting threshold, randomly selecting a dimension with unequal selection probabilities according to each dimension under the leaf nodes as the splitting dimension of the leaf nodes, and dividing the power data in the leaf nodes into two new child nodes by using the splitting dimension; repeatedly performing splitting judgment on the new child nodes until the splitting necessity of all new child nodes is not greater than the preset splitting threshold, and then stopping the iteration to complete the update of the isolation forest.

[0018] The effect is that: based on the constructed isolation forest, the present invention updates the isolation forest according to the newly generated power data, avoiding continuously reconstructing the isolation forest, saving computing resources, improving the efficiency of abnormal identification of power equipment, and further improving the real-time performance of power equipment asset monitoring and management.

[0019] Preferably, the anomaly score of the newly generated power data satisfies the expression: ; where represents the anomaly score of the newly generated power data; represents the depth of the leaf node where the newly generated power data is located in the th isolation tree; represents the height of the th isolation tree; represents the number of power data included in the leaf node where the newly generated power data is located in the th isolation tree; represents the mean value of the number of power data included in all leaf nodes in the th isolation tree; represents the number of power data segments in the leaf node where the newly generated power data is located in the th isolation tree; represents the mean value of the number of power data segments in all leaf nodes in the th isolation tree.

[0020] The effect is as follows: Considering that the abnormal data of power equipment changes continuously, this invention determines whether the power data in the leaf node where the currently newly generated power data is located in the isolation forest is continuous in time series according to the number of power data segments in the leaf node where the currently newly generated power data is located in the isolation forest. If the power data contained in the leaf node where the currently newly generated power data is located is more continuous in time series, it indicates that the difference between the power data in the recent period and the previous power data is greater, and at this time, the power equipment is more likely to have an abnormality. This invention combines the depth of the leaf node where the currently newly generated power data is located in the isolation forest, the number of power data, and the number of power data segments in the leaf node to obtain an abnormal score, and the result is more accurate.

[0021] Preferably, the method for obtaining the power data segment is: dividing all the power data in the child node into multiple power data segments according to time series, and all the power data in each power data segment is continuous in time series, and the time series between different power data segments is not continuous.

[0022] Preferably, identifying the current state of the power equipment according to the size of the abnormal score includes: in response to the abnormal score of the currently newly generated power data being greater than the abnormal threshold, the state of the power equipment is abnormal; otherwise, the state of the power equipment is normal.

[0023] In a second aspect, this invention provides a power equipment asset management system, adopting the following technical solution:

[0024] A power equipment asset management system includes: a processor and a memory, and the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned power equipment asset management method is implemented.

[0025] By adopting the above technical solution, the above-mentioned power equipment asset management method is generated into a computer program and stored in the memory to be loaded and executed by the processor, so as to manufacture a terminal device according to the memory and the processor, which is convenient to use.

[0026] This invention has the following technical effects: Based on the constructed isolation forest, this invention updates the isolation forest according to the newly generated power data, improves the efficiency of power equipment abnormality recognition, and further improves the real-time performance of power equipment asset monitoring and management; This invention randomly selects a dimension to divide the node with unequal probabilities, enhances the probability of the dimension that may contain abnormalities being selected, improves the iteration efficiency of the isolation forest, and enhances the recognition accuracy of abnormal power data; This invention divides the child node according to the necessity of segmentation, improves the segmentation efficiency of the isolation tree and the recognition efficiency of abnormal power data, improves the accuracy and real-time performance of power equipment state recognition, and further improves the real-time performance of power equipment asset monitoring and management. Brief Description of the Drawings

[0027] By referring to the detailed description below with reference to the accompanying drawings, the above and other objects, features, and advantages of the exemplary embodiments of the present invention will become readily understood. In the drawings, several embodiments of the present invention are shown in an exemplary rather than restrictive manner, and the same or corresponding reference numerals denote the same or corresponding parts.

[0028] Figure 1 is a flowchart of a power equipment asset management method according to an embodiment of the present invention;

[0029] Figure 2 is a schematic diagram of the periodic change of current;

[0030] Figure 3 is a schematic diagram of abnormal temperature distribution;

[0031] Figure 4 is a schematic diagram of the current distribution of all power data in the child nodes during the normal periodic change of current;

[0032] Figure 5 is a schematic diagram of the abnormal periodic change of current data;

[0033] Figure 6 is a schematic diagram of the current distribution of all power data in the child nodes during the abnormal periodic change of current. Detailed Embodiments

[0034] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are 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 those skilled in the art without creative efforts fall within the scope of protection of the present invention.

[0035] It should be understood that when the claims, specifications, and drawings of the present invention use terms such as "first" and "second", they are only used to distinguish different objects, rather than to describe a specific order. The terms "including" and "comprising" used in the specifications and claims of the present invention indicate the existence of the described features, wholes, steps, operations, elements, and / or components, but do not exclude the existence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.

[0036] An embodiment of the present invention discloses a power equipment asset management method, referring to Figure 1 , including steps S1 - S5:

[0037] S1: Collect power data of power equipment.

[0038] For each power device, power data of the power device is collected. The power data includes multiple dimensions, such as current, voltage, temperature, etc. In the embodiments of the present invention, the collection interval of the power data is once per second, and the implementer can set the collection interval according to the actual implementation situation.

[0039] Thus, the collection of the power data of the power device is realized.

[0040] S2: Construct an isolation forest according to the historical power data of the power device.

[0041] Step S2 includes steps S21 - S25. Specifically:

[0042] S21: The power data of the power device over a period of history is formed into a sequence as the root node.

[0043] It should be noted that the implementer can set the length of the historical period according to the actual implementation situation. For example, the power data of the power device in the past day.

[0044] S22: Determine the selection probability of each dimension under the root node according to the distribution of each dimension of the power data in the root node. Randomly select a dimension with unequal probability according to the selection probability of each dimension under the root node as the splitting dimension of the root node. Use the splitting dimension of the root node to divide all the power data in the root node to obtain two child nodes.

[0045] It should be noted that in the traditional isolation forest algorithm, when dividing a node, a dimension is randomly selected with equal probability from all dimensions for division. The data of the power data may have very small differences in a certain dimension and large differences in a certain dimension. The abnormality of the power data is more likely to be reflected in the dimension with large data differences. Randomly selecting a dimension with equal probability for division may cause the dimension containing abnormalities to be selected relatively late, affecting the iteration efficiency of the isolation forest. At the same time, it may cause the abnormal power data to be included in the leaf nodes deeper in the isolation tree. Since the isolation forest algorithm usually calculates the abnormality score of a data point according to the depth of the leaf node where the data point is located, the smaller the depth, the larger the abnormality score. When the abnormal power data is included in the leaf nodes deeper in the isolation tree, its corresponding abnormality score is smaller, which may cause the abnormal power data to not be recognized. Therefore, in the present invention, according to the distribution of each dimension of the power data in the node, the selection probability of each dimension when splitting the node is calculated, so as to randomly select a dimension with unequal probability according to the selection probability, thereby increasing the probability that the dimension that may contain abnormalities is selected.

[0046] Specifically, the determination of the selection probability of each dimension under the root node according to the distribution of each dimension of the power data in the root node includes:

[0047] ;

[0048] wherein, is the selection probability of the th dimension under the root node; is the maximum value of the th dimension of all power data in the root node; is the minimum value of the th dimension of all power data in the root node; is the number of dimensions of the power data; is the information entropy of the th dimension of all power data in the root node; is the number of different numerical types in the th dimension of all power data in the root node; is the maximum value function, is to prevent the denominator from being 0, is to prevent the exponent from being 0.

[0049] When is larger, the fluctuation range of the data of the th dimension of all power data in the root node is larger, is used to perform normalization. When the fluctuation range of the data of the th dimension is larger, the th dimension is more likely to contain anomalies, and then the selection probability of the th dimension is larger.

[0050] Under normal circumstances, the data of some dimensions of the power data may change periodically. For example, when the power equipment is alternating current, the current and voltage change periodically. Figure 2 is a schematic diagram of the periodic change of the current. At this time, the fluctuation of the data of this dimension within its fluctuation range may be a normal fluctuation. When the frequencies of the different values that appear within the fluctuation range of this dimension are basically the same, the information entropy of this dimension is larger, and the data of this dimension is more likely to change periodically, and the necessity of selecting this dimension for division is smaller; when the values that appear within the fluctuation range of this dimension are mostly concentrated in one place, the information entropy of this dimension is smaller, and the values that are not concentrated in this dimension are more likely to be anomalies. Figure 3 is a schematic diagram of the abnormal distribution of the temperature. At this time, the necessity of selecting this dimension for division is larger. Therefore, the present invention corrects according to the information entropy of the th dimension of all power data in the root node. When the frequencies of the different values that appear in the th dimension of all power data in the root node are the same, the frequency of each value is , at this time, the information entropy of all power data in the root node for the th dimension reaches the maximum, which is . Therefore, the present invention uses to normalize the information entropy . When the information entropy is smaller, is smaller, and at this time, the degree of increase in is greater, thereby increasing the selection probability of the th dimension. When the information entropy is larger, is larger and approaches 1, and at this time, the degree of increase in is smaller.

[0051] The method of using the splitting dimension of the root node to divide all power data in the root node to obtain two child nodes includes:

[0052] Obtain the maximum value and the minimum value of the splitting dimension of all power data in the root node, and randomly select a value between the maximum value and the minimum value as the splitting threshold of the root node. Divide all power data in the root node with a splitting dimension greater than the splitting threshold into one category as one child node of the root node, and divide all power data in the root node with a splitting dimension not greater than the splitting threshold into one category as the other child node of the root node.

[0053] S23: For any one child node, calculate the splitting necessity of the child node according to the temporal regularity of the power data in the child node and the selection probability of each dimension under the child node. In response to the splitting necessity of the child node being greater than a preset splitting threshold, randomly select a dimension with unequal probabilities according to the selection probability of each dimension under the child node as the splitting dimension of the child node, and use the splitting dimension of the child node to divide all power data in the child node to obtain two child nodes.

[0054] Specifically, the method for obtaining the temporal regularity of the power data in the child node is:

[0055] Divide all the power data in the child node into multiple power data segments according to time sequence. All the power data in each power data segment are continuous in time sequence, and are discontinuous in time sequence between different power data segments. For any two power data segments, the time difference between the last power data in the previous power data segment and the first power data in the next power data segment is used as the time interval between these two power data segments. For example, when the power data at the 1st, 2nd, 3rd, 8th, 9th, 10th, 15th, 16th, and 17th moments are included in the child node, the power data at the 1st, 2nd, and 3rd moments are continuous in time sequence, so the power data at the 1st, 2nd, and 3rd moments constitute a power data segment. Similarly, the power data at the 8th, 9th, and 10th moments constitute a power data segment, and the power data at the 15th, 16th, and 17th moments constitute a power data segment. The time interval between the power data segment composed of the power data at the 1st, 2nd, and 3rd moments and the power data segment composed of the power data at the 15th, 16th, and 17th moments is 12.

[0056] Determine the time sequence regularity of the power data in the child node according to the length of the power data segments included in the child node and the time interval between adjacent power data segments:

[0057] ;

[0058] where, is the time sequence regularity of the power data in the child node; is the length of the th power data segment in the child node; is the number of power data segments in the child node; is the average value of the lengths of all power data segments in the child node; represents the time interval between the th power data segment and the th power data segment in the child node.

[0059] is the average difference between the length of each power data segment in the child node and the average value of the lengths of all power data segments, is the average difference between the time interval between power data segments in the child node and the average time interval. When and are smaller, the time sequence regularity of the power data in the child node is stronger, the power data corresponding to the segmentation dimension of the parent node of the child node is more likely to be periodically changed, and the power data in the child node is regularly distributed in the segmentation dimension corresponding to the parent node of the child node. For example, when the segmentation dimension corresponding to the parent node of the child node is current and there is no abnormality in the current, the schematic diagram of its periodic change is shown in Figure 2 , the schematic diagram of the current distribution of all power data in the child node is shown in Figure 4 . When or The greater the value is, the weaker the temporal regularity of the power data in the child node is, and the power data in the child node is irregularly distributed in the segmentation dimension corresponding to the parent node of the child node. For example, if the segmentation dimension corresponding to the parent node of the child node is current and there are abnormalities in the current data, see the schematic diagram of its periodic change in Figure 5 When, see the schematic diagram of the current distribution of all power data in the child node in Figure 6 .

[0060] The selection probability of each dimension under the child node is calculated in the same way as the selection probability of each dimension under the root node in S22, which will not be elaborated here.

[0061] It should be noted that the power data is temporal data, and abnormal power data usually appears continuously. Therefore, the present invention determines the necessity of segmenting the child node according to the temporal regularity of the power data in the child node, the selection probability of each dimension under the child node, and the number of power data included in the child node, so as to segment the child node according to the necessity of segmentation, and avoid normal power data with periodic regular distribution or abnormal power data continuously distributed in time series from being segmented into multiple child nodes, which affects the accuracy of abnormal power data recognition.

[0062] Specifically, calculating the necessity of segmenting the child node according to the temporal regularity of the power data in the child node, the selection probability of each dimension under the child node, and the number of power data included in the child node includes:

[0063] ;

[0064] wherein, is the necessity of segmenting the child node; is the temporal regularity of the power data in the child node; is the selection probability of the th dimension under the child node; is the number of dimensions of the power data; is the number of power data included in the child node; is the number of all power data; is the natural exponential function; is the maximum value function.

[0065] When the temporal regularity of the power data in a child node is stronger, the possibility that the power data in the child node contains anomalies is smaller, and at this time, the necessity of splitting the child node is smaller; when the selection probabilities of all dimensions under a child node are relatively small, it indicates that the fluctuation ranges of all dimensions of the power data in the child node are small, and the data of all dimensions are regularly distributed, and at this time, the necessity of splitting the child node is smaller; when the amount of power data contained in a child node is less, the necessity of splitting the child node is smaller; conversely, when the temporal regularity of the power data in a child node is weaker, or the selection probability of a certain dimension under the child node is larger, or the amount of power data contained in the child node is more, the necessity of splitting the child node is greater.

[0066] The splitting threshold is set by the implementer according to the actual implementation situation. Since the value range of the splitting necessity is [0, 1], the set splitting threshold also needs to be between [0, 1]. For example, the splitting threshold is 0.5.

[0067] The use of the splitting dimension of the child node to divide all the power data in the child node to obtain two child nodes includes:

[0068] Obtain the maximum value and the minimum value of the splitting dimension of all the power data in the child node, and randomly select a value between the maximum value and the minimum value as the splitting threshold of the child node. Divide all the power data in the child node whose splitting dimension is greater than the splitting threshold into one category as one child node of the child node, and divide all the power data in the child node whose splitting dimension is not greater than the splitting threshold into one category as another child node of the child node.

[0069] S24: Repeat S23 until the splitting necessity of all child nodes is not greater than the preset splitting threshold, and then stop the iteration to complete the construction of the isolation tree.

[0070] S25: Repeat S21 - S24 to construct multiple isolation trees to form an isolation forest.

[0071] Specifically, set the number of repetitions , and repeat the execution of steps S21 - S24 times to complete the construction of isolation trees. Since when constructing each isolation tree, the splitting dimension of each node is randomly selected according to the selection probabilities of all dimensions under each node, and the splitting threshold of each node is also randomly selected, the constructed isolation trees are all different from each other. These isolation trees form an isolation forest. Among them, is set by the implementer according to the actual implementation situation, and no specific limit is made. For example, .

[0072] So far, the construction of the isolation forest has been completed.

[0073] S3: Add the newly generated power data to the Isolation Forest and update the Isolation Forest.

[0074] It should be noted that the power data of power equipment is streaming data and is generated in real time. In the traditional Isolation Forest, to implement anomaly detection for newly generated power data, it is necessary to reconstruct the Isolation Forest based on the newly generated power data and historical power data. Therefore, to achieve real-time anomaly detection of power equipment, every time a new power data is generated, the Isolation Forest needs to be reconstructed once, which is inefficient and consumes a large amount of computing resources. Therefore, in the present invention, every time a new power data is generated, based on the existing Isolation Forest, the Isolation Forest is updated according to the newly generated power data, thereby saving computing resources and improving the efficiency of anomaly detection.

[0075] Step S3 includes steps S21 - S35, specifically:

[0076] S31: For the newly generated power data, according to the splitting dimension and splitting threshold of each node in the isolation tree, obtain the leaf node to which the power data belongs in the isolation tree, and add the power data to the corresponding leaf node.

[0077] Specifically, for the newly generated power data, according to the splitting dimension and splitting threshold of each node in the isolation tree, obtaining the leaf node to which the power data belongs in the isolation tree includes:

[0078] Determine whether the value of the newly generated power data under the splitting dimension of the root node of the isolation tree is greater than the splitting threshold of the root node. When it is greater than the splitting threshold of the root node, use the child nodes corresponding to all power data with a splitting dimension greater than the splitting threshold in the root node as the first target child nodes of the newly generated power data. Otherwise, use the child nodes corresponding to all power data with a splitting dimension not greater than the splitting threshold in the root node as the first target child nodes of the newly generated power data. When the first target child node is not a leaf node, determine whether the value of the newly generated power data under the splitting dimension of the first target child node is greater than the splitting threshold of the first target child node. When it is greater than the splitting threshold of the first target child node, use the child nodes corresponding to all power data with a splitting dimension greater than the splitting threshold in the first target child node as the second target child nodes of the newly generated power data. Otherwise, use the child nodes corresponding to all power data with a splitting dimension not greater than the splitting threshold in the first target child node as the second target child nodes of the newly generated power data. When the second target child node is not a leaf node, determine whether the value of the newly generated power data under the splitting dimension of the second target child node is greater than the splitting threshold of the second target child node. When it is greater than the splitting threshold of the second target child node, use the child nodes corresponding to all power data with a splitting dimension greater than the splitting threshold in the second target child node as the third target child nodes of the newly generated power data. Otherwise, use the child nodes corresponding to all power data with a splitting dimension not greater than the splitting threshold in the second target child node as the third target child nodes of the newly generated power data. And so on, until the obtained target child node is a leaf node, then stop the iteration. This target child node is the leaf node to which the newly generated power data belongs in the isolation tree.

[0079] S32: Use the method in step S23 to obtain the splitting necessity of the corresponding leaf node after adding the newly generated power data. In response to the splitting necessity of the leaf node being greater than the preset splitting threshold, randomly select a dimension with unequal probabilities according to the selection probabilities of each dimension under the leaf node as the splitting dimension of the leaf node, and use the splitting dimension of the leaf node to divide all the power data in the leaf node to obtain two child nodes.

[0080] S33: For any newly obtained child node, calculate the splitting necessity of the child node according to the temporal regularity of the power data in the child node and the selection probabilities of each dimension under the child node. In response to the splitting necessity of the child node being greater than the preset splitting threshold, randomly select a dimension with unequal probabilities according to the selection probabilities of each dimension under the child node as the splitting dimension of the child node, and use the splitting dimension of the child node to divide all the power data in the child node to obtain two child nodes.

[0081] S34: Repeat S33 until the segmentation necessity of all newly obtained child nodes is not greater than a preset segmentation threshold, and then stop the iteration to complete the update of the isolation tree.

[0082] S35: Execute steps S31 to S34 for each isolation tree in the isolation forest to update each isolation tree in the isolation forest, thereby completing the update of the isolation forest.

[0083] Thus, the update of the isolation forest is completed.

[0084] S4: Determine the anomaly score of the currently newly generated power data, and identify the current state of the power equipment based on the size of the anomaly score.

[0085] It should be noted that the currently newly generated power data can reflect the current state of the power equipment. Therefore, in the present invention, the anomaly score of the currently newly generated power data is determined based on the leaf nodes where the currently newly generated power data is located in each isolation tree in the isolation forest, thereby identifying the current state of the power equipment.

[0086] Specifically, according to the depth of the leaf node where the currently newly generated power data is located in each isolation tree in the isolation forest, the number of power data included in the leaf node, and the number of power data segments included in the leaf node, determine the anomaly score of the currently newly generated power data:

[0087] ;

[0088] where represents the anomaly score of the currently newly generated power data; represents the depth of the leaf node where the currently newly generated power data is located in the th isolation tree; represents the th isolation tree height; represents the number of power data included in the leaf node where the currently newly generated power data is located in the th isolation tree; represents the th isolation tree mean value of the number of power data included in all leaf nodes; represents the number of power data segments in the leaf node where the currently newly generated power data is located in the th isolation tree; represents the th isolation tree mean value of the number of power data segments included in all leaf nodes.

[0089] When is smaller, it indicates that the currently newly generated power data in the Among the isolated trees, the earlier it is isolated, the greater the difference between the power data contained in the leaf node where the currently newly generated power data is located and the rest of the power data. At this time, the currently newly generated power data is more abnormal; when is smaller, it indicates that the difference between the currently newly generated power data and the rest of the power data is greater. At this time, the currently newly generated power data is more abnormal; since the power data of power equipment changes continuously, when is smaller, it indicates that the power data contained in the leaf node where the currently newly generated power data is located is more continuous in time series. Furthermore, it indicates that the difference between the power data in the recent period and the previous power data is greater. At this time, the power equipment is more likely to have an abnormality. Therefore, when 、 、 is smaller, the abnormal score of the currently newly generated power data in the th isolated tree is greater. The present invention synthesizes the abnormal scores of the currently newly generated power data in all the isolated trees of the isolated forest to obtain the abnormal score of the currently newly generated power data.

[0090] In response to the abnormal score of the currently newly generated power data being greater than the abnormal threshold, it is considered that the power equipment has an abnormality. Otherwise, it is considered that the state of the power equipment is normal. Among them, the abnormal threshold is set by the implementer according to the actual implementation situation. Since the value range of the abnormal score is [0, 1], the set abnormal threshold also needs to be between [0, 1]. The empirical value of the abnormal threshold is 0.9.

[0091] S5: Real-time synchronize the current state of the power equipment to the digital twin model of the power equipment to achieve asset monitoring and management of the power equipment.

[0092] Construct a digital twin model of the power equipment to achieve 3D visualization display of all power equipment.

[0093] Real-time synchronize the current state of each power equipment to the digital twin model of the power equipment, and highlight the abnormal power equipment in the digital twin model to remind the staff to repair the abnormal power equipment, so as to achieve asset monitoring and management of the power equipment.

[0094] The embodiment of the present invention also discloses a power equipment asset management system, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, a power equipment asset management method according to the present invention is implemented.

[0095] The above system also includes other components well-known to those skilled in the art such as a communication bus and a communication interface. Their settings and functions are known in the art, so they will not be elaborated here.

[0096] In the present invention, the aforementioned memory may be any tangible medium that contains or stores a program, which can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium may be any suitable magnetic storage medium or magneto-optical storage medium, or any other medium that can be used to store the required information and can be accessed by an application program, module, or both. Any such computer storage medium may be part of the device or accessible or connectable to the device.

[0097] Although this specification has shown and described multiple embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many changes, alterations, and alternative forms will occur to those skilled in the art without departing from the spirit and scope of the present invention. It should be understood that various alternatives to the embodiments of the present invention described herein may be employed in practicing the present invention.

[0098] The above are all preferred embodiments of the present invention, and the protection scope of the present invention is not limited thereby. Therefore, all equivalent changes made according to the structure, shape, and principle of the present invention shall be covered within the protection scope of the present invention.

Claims

1. A method for power equipment asset management, characterized in that including: Construct an isolation forest based on the historical power data of power equipment; During the construction of the isolation forest: for each child node in the isolation tree, determine the selection probability of each dimension under the child node according to the distribution of the power data in the child node in each dimension. The selection probability is positively correlated with the fluctuation range of the power data in each dimension in the child node and negatively correlated with the information entropy of the power data in each dimension in the child node; determine the necessity of splitting the child node, and the necessity of splitting is negatively correlated with the temporal regularity of the power data in the child node, and positively correlated with the maximum value of the selection probabilities of each dimension under the child node and the number of power data included in the child node; in response to the necessity of splitting being greater than a preset splitting threshold, randomly select a dimension according to the selection probabilities of each dimension under the child node as the splitting dimension of the child node, and the selection probabilities of each dimension under the child node are not equal; use the splitting dimension to divide the power data in the child node into two new child nodes; the selection probabilities of each dimension under the child node satisfy the expression: ； Among them, is the selection probability of the th dimension under the child node; is the maximum value of the th dimension of all power data in the child node; is the minimum value of the th dimension of all power data in the child node; is the number of dimensions of the power data; is the information entropy of the th dimension of all power data in the child node; is the number of different value types in the th dimension of all power data in the child node; is the maximum value function; Add the newly generated power data to the isolation forest and update the isolation forest; determine the anomaly score of the newly generated power data according to its position in each isolation tree of the isolation forest, and identify the current state of the power equipment according to the size of the anomaly score; Synchronize the current state of the power equipment to the digital twin model of the power equipment in real time to realize the monitoring and management of power equipment assets.

2. The power equipment asset management method according to claim 1, characterized in that, The temporal regularity of the power data in the child node satisfies the expression: ； Among them, is the temporal regularity of the power data in the child node; is the length of the th power data segment in the child node; is the number of power data segments in the child node; is the average length of all power data segments in the child node; represents the time interval between the th power data segment and the th power data segment in the child node.

3. The method for managing power equipment assets according to claim 1, wherein The necessity of splitting the child node satisfies the expression: ； Among them, is the necessity of splitting of the child node; is the timing regularity of the power data in the child node; is the selection probability of the th dimension under the child node; is the number of dimensions of the power data; is the number of power data included in the child node; is the number of all power data; is the natural exponential function; is the maximum value function.

4. A method for power equipment asset management according to claim 1, characterized in that, The adding the newly generated power data to the isolation forest and updating the isolation forest includes: For the newly generated power data, determine the leaf node to which the power data belongs in each isolation tree, and add the power data to the corresponding leaf node; obtain the necessity of splitting the corresponding leaf node after adding the power data; in response to the necessity of splitting the leaf node being greater than a preset splitting threshold, randomly select a dimension according to the selection probabilities of each dimension under the leaf node as the splitting dimension of the leaf node, and the selection probabilities of each dimension under the leaf node are not equal; use the splitting dimension to divide the power data in the leaf node into two new child nodes; repeat the splitting judgment on the new child nodes until the necessity of splitting of all new child nodes is not greater than the preset splitting threshold, and then stop the iteration to realize the update of the isolation forest.

5. A power equipment asset management method according to claim 1, characterized in that, The anomaly score of the newly generated power data satisfies the expression: ； Among them, represents the anomaly score of the currently newly generated power data; represents the depth of the leaf node where the currently newly generated power data is located in the th isolation tree; represents the height of the th isolation tree; represents the number of power data contained in the leaf node where the currently newly generated power data is located in the th isolation tree; represents the mean value of the number of power data contained in all leaf nodes in the th isolation tree; represents the number of power data segments in the leaf node where the currently newly generated power data is located in the th isolation tree; represents the mean value of the number of power data segments in all leaf nodes in the th isolation tree.

6. The method for managing power equipment assets according to claim 2 or 5, characterized in that, The method for obtaining the power data segment is: Divide all the power data in the child node into multiple power data segments according to time sequence. All the power data in each power data segment are continuous in time sequence, and the time sequences between different power data segments are not continuous.

7. A power equipment asset management method according to any one of claims 1, characterized in that The identifying the current state of the power equipment according to the size of the anomaly score includes: In response to the anomaly score of the newly generated power data being greater than the anomaly threshold, the state of the power equipment is abnormal, otherwise, the state of the power equipment is normal.

8. An electric power equipment asset management system, characterized in that, including: A processor and a memory, where the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a power equipment asset management method according to any one of claims 1-7 is implemented.

Images