An edge optimization method and system for digital twin spatial modeling of distribution network

By constructing digital twin space modeling and applying knowledge distillation edge optimization model, the problems of edge data transmission delay and slow response speed in the distribution network are solved, and the operation efficiency and data processing accuracy of the distribution network are improved.

CN118607150BActive Publication Date: 2024-11-05STATE GRID JIANGSU ELECTRIC POWER CO LTD NANTONG POWER SUPPLY BRANCH +3
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202411075232.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-07
Publication Date
2024-11-05
Estimated Expiration
2044-08-07

AI Technical Summary

Technical Problem

There are latency and slow response speed in the distribution network, resulting in low operating efficiency.

Method used

By constructing digital twin space modeling, environmental feature information, equipment feature information and operating status information are collected, and distilled processing is combined with the knowledge distillation edge optimization model, the digital twin teacher model is optimized to improve the operating efficiency of the distribution network.

Benefits of technology

The technical effect of improving the operating efficiency of the distribution network has been achieved. By optimizing the digital twin model, the speed and accuracy of data processing are improved, and the fault assessment and early warning time is reduced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118607150B_ABST
    Figure CN118607150B_ABST
Patent Text Reader

Abstract

The present invention discloses an edge optimization method and system for digital twin space modeling of distribution network, which relates to data processing related fields. The method comprises: combining environmental feature information, equipment feature information and operating status information of distribution equipment to construct a digital twin teacher model of distribution network, obtaining a knowledge distillation edge optimization model, performing distillation processing on environmental feature information according to a first distillation optimization channel, obtaining environmental distillation results, performing distillation processing on equipment feature information according to a second distillation optimization channel, obtaining equipment distillation results, performing distillation processing on operating status information according to a third distillation optimization channel, obtaining operating distillation results, and obtaining a digital twin student model. The technical problem of delayed data transmission and slow response speed at the edge of distribution network, which leads to low operating efficiency of distribution network, is solved, and the edge optimization of distribution network is realized by constructing digital twin space modeling, so as to achieve the technical effect of improving the operating efficiency of distribution network.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to data processing related fields, and in particular to an edge optimization method and system for digital twin spatial modeling of distribution networks. Background Art

[0002] With the development of modern society, the power industry is undergoing reforms and changes, especially in the field of distribution networks. The distribution network refers to the power transmission and distribution system from power sources to end users, responsible for delivering electricity from power plants to user homes, businesses and other terminals. With the growth of power demand and the development of power systems, the planning and optimization of the edge of the distribution network has become particularly important. However, there are technical problems in the distribution network today, such as delayed edge data transmission and slow response speed, which lead to low distribution network operation efficiency. Summary of the invention

[0003] The present application provides an edge optimization method and system for digital twin spatial modeling of distribution networks, which solves the technical problems of delayed data transmission and slow response speed at the edge of distribution networks, resulting in low operating efficiency of distribution networks. The application realizes edge optimization of distribution networks by constructing digital twin spatial modeling, thereby achieving the technical effect of improving the operating efficiency of distribution networks.

[0004] The present application provides an edge optimization method for digital twin space modeling of a distribution network, the method being applied to an edge optimization system for digital twin space modeling of a distribution network, comprising: respectively collecting environmental characteristic information of the distribution network and device characteristic information of distribution equipment in the distribution network;

[0005] Combining the environmental characteristic information, the equipment characteristic information and the dynamically monitored operating status information of the power distribution equipment, a digital twin teacher model of the power distribution network is constructed;

[0006] Acquire a knowledge distillation edge optimization model, where the knowledge distillation edge optimization model is used to perform distillation optimization on the digital twin teacher model, wherein the knowledge distillation edge optimization model includes a first distillation optimization channel, a second distillation optimization channel, and a third distillation optimization channel;

[0007] The first distillation optimization channel performs distillation processing on the environmental feature information based on a predetermined relationship distillation logic to obtain an environmental distillation result;

[0008] The second distillation optimization channel performs distillation processing on the device feature information based on a predetermined feature distillation logic to obtain a device distillation result;

[0009] The third distillation optimization channel performs distillation processing on the operation status information based on a predetermined response distillation logic to obtain an operation distillation result;

[0010] A digital twin student model is obtained according to the environment distillation results, the equipment distillation results and the operation distillation results, and the digital twin student model represents the distillation optimization model of the digital twin teacher model.

[0011] In a possible implementation, the digital twin teacher model of the distribution network is constructed by combining the environmental characteristic information, the device characteristic information and the dynamically monitored operating status information of the distribution equipment, and the following processing is performed: an initial digital twin teacher model of the distribution network is constructed based on the device characteristic information and the operating status information;

[0012] Extracting the first historical record in the historical distribution network record;

[0013] Performing a correlation analysis on the first historical environment feature information and the first historical device status index in the first historical record to obtain a first historical correlation analysis result;

[0014] The first historical correlation analysis result is visualized using Manhattan distance, and the visualization result is rendered to the initial digital twin teacher model to obtain the digital twin teacher model.

[0015] In a possible implementation, the first distillation optimization channel performs distillation processing on the environmental feature information based on a predetermined relationship distillation logic to obtain an environmental distillation result, and performs the following processing: extracting a predetermined relationship similarity probability distribution in the predetermined relationship distillation logic, and the predetermined relationship similarity probability distribution is used to perform distillation processing on the environmental feature information;

[0016] The expression of the probability distribution of the predetermined relationship similarity is as follows: ;in, It is used to characterize the similarity divergence between the environmental feature information and the environmental distillation result. The environmental feature information is subjected to dimensionality reduction processing with the constraint of being close to 0. Characterize the divergence of the similarity between the environmental feature information and the environmental distillation result, Characterize the environmental feature points in the environmental feature information Status index of the power distribution equipment The probability of relationship similarity is Characterize the environmental feature points in the environmental distillation result Status index of the power distribution equipment In a possible implementation, the following processing is performed: the operation status information includes voltage parameters, current parameters and power parameters;

[0017] Performing weighted calculation on normalized processing results of the voltage parameter, the current parameter and the power parameter to obtain a real-time status index of the power distribution equipment;

[0018] The real-time status index is used to evaluate the failure of the distribution network and issue an early warning.

[0019] In a possible implementation, the second distillation optimization channel performs distillation processing on the device characteristic information based on a predetermined characteristic distillation logic to obtain a device distillation result, and performs the following processing: the device characteristic information includes a plurality of device physical characteristic parameters and a plurality of device electrical characteristic parameters;

[0020] Allocating weight coefficients for the multiple device physical characteristic parameters and the multiple device electrical characteristic parameters to obtain a device characteristic weight allocation result;

[0021] Arrange the plurality of device physical characteristic parameters and the plurality of device electrical characteristic parameters in descending order according to the device characteristic weight allocation result to obtain a device characteristic descending order list;

[0022] Based on the predetermined feature distillation logic, features with a predetermined last ratio threshold in the descending list of device features are pruned to obtain the device distillation result.

[0023] In a possible implementation, the third distillation optimization channel performs distillation processing on the operation status information based on a predetermined response distillation logic to obtain an operation distillation result, and performs the following processing: extracting a predetermined student accuracy probability in the predetermined response distillation logic;

[0024] The predetermined distillation loss function is read, and the running distillation result is obtained by combining the predetermined student accuracy probability and the real-time state index, wherein the expression of the predetermined distillation loss function is as follows: ;in, characterizing the predetermined distillation loss function, Characterize the real-time status index obtained by the digital twin teacher model The real-time student status index obtained with the digital twin student model The divergence between Characterize the temperature coefficient, Characterizing the digital twin teacher model The real-time status index In a possible implementation, the following processing is performed: obtaining the index difference between the student's real-time status index and the real-time status index;

[0025] Determining whether the index difference reaches a predetermined difference range;

[0026] If the index difference does not reach the predetermined difference range, the predetermined last ratio threshold is reduced and adjusted.

[0027] The present application also provides an edge optimization system for digital twin space modeling of a distribution network, including: an information acquisition module, the information acquisition module is used to respectively acquire environmental characteristic information of the distribution network and equipment characteristic information of distribution equipment in the distribution network;

[0028] A first model building module, wherein the first model building module is used to build a digital twin teacher model of the distribution network by combining the environmental feature information, the device feature information and the dynamically monitored operating status information of the distribution equipment;

[0029] A distillation optimization module, wherein the distillation optimization module is used to obtain a knowledge distillation edge optimization model, wherein the knowledge distillation edge optimization model is used to perform distillation optimization on the digital twin teacher model, wherein the knowledge distillation edge optimization model includes a first distillation optimization channel, a second distillation optimization channel, and a third distillation optimization channel;

[0030] A first distillation processing module, wherein the first distillation processing module is used for the first distillation optimization channel to perform distillation processing on the environmental feature information based on a predetermined relationship distillation logic to obtain an environmental distillation result;

[0031] A second distillation processing module, wherein the second distillation processing module is used for the second distillation optimization channel to perform distillation processing on the device feature information based on a predetermined feature distillation logic to obtain a device distillation result;

[0032] A third distillation processing module, wherein the third distillation processing module is used for the third distillation optimization channel to perform distillation processing on the operation status information based on a predetermined response distillation logic to obtain an operation distillation result;

[0033] The second model building module is used to obtain a digital twin student model according to the environment distillation results, the equipment distillation results and the operation distillation results, and the digital twin student model represents the distillation optimization model of the digital twin teacher model.

[0034] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0035] The present application provides an edge optimization method and system for digital twin spatial modeling of a distribution network, which relates to the field of data processing technology. It solves the technical problems of data transmission delay and slow response speed at the edge of the distribution network, which lead to low operating efficiency of the distribution network. It achieves edge optimization of the distribution network by constructing digital twin spatial modeling, thereby achieving the technical effect of improving the operating efficiency of the distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the accompanying drawings of the embodiment of the present invention will be briefly introduced below. A flow chart is used in the present application to illustrate the operations performed by the system according to the embodiment of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more operations can be removed from these processes.

[0037] Figure 1 A flow chart of an edge optimization method for spatial modeling of digital twins of distribution networks provided in an embodiment of the present application;

[0038] Figure 2 A structural schematic diagram of an edge optimization system for digital twin spatial modeling of a distribution network provided in an embodiment of the present application. DETAILED DESCRIPTION

[0039] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0040] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0041] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict, and the terms "first\second" involved are merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by technicians in the technical field of this application. The terms used herein are for the purpose of describing the embodiments of the present application only.

[0042] The embodiment of the present application provides an edge optimization method for digital twin space modeling of a distribution network, and the method is applied to an edge optimization system for digital twin space modeling of a distribution network, such as Figure 1 As shown, the method includes:

[0043] Step A100, respectively collecting and obtaining environmental characteristic information of the power distribution network and equipment characteristic information of the power distribution equipment in the power distribution network;

[0044] The climate, topography, geology, electromagnetic environment and other environmental characteristics in the distribution network are collected through various sensors and monitoring equipment, thereby obtaining the environmental characteristic information of the distribution network. Furthermore, the operating status, performance parameters, fault records and other equipment characteristics of the distribution equipment in the distribution network are collected through sensors and monitoring systems installed on the equipment, thereby obtaining the equipment characteristic information of the distribution equipment. The environmental characteristic information of the distribution network and the equipment characteristic information of the distribution equipment can then be further processed and analyzed to extract useful features and build a digital twin model to optimize the operation of the distribution network and improve the reliability and economy of power supply.

[0045] Execute step A200 to construct a digital twin teacher model of the distribution network by combining the environmental characteristic information, the device characteristic information and the dynamically monitored operating status information of the distribution equipment;

[0046] In a possible implementation, step A200 further includes step A210, constructing an initial digital twin teacher model of the distribution network based on the device feature information and the operating status information;

[0047] In a possible implementation, step A210 further includes step A211, wherein the operation status information includes voltage parameters, current parameters, and power parameters; and executing step A212, performing weighted calculation on normalized processing results of the voltage parameters, the current parameters, and the power parameters to obtain a real-time status index of the power distribution equipment;

[0048] The real-time status index is used to evaluate the failure of the distribution network and issue an early warning.

[0049] By combining the above-mentioned environmental characteristic information and equipment characteristic information with the dynamically monitored operating status information, the basic data for constructing the digital twin classroom model of the distribution network is used. The operating status information is used to evaluate the health status of the distribution network and predict potential faults. After collecting the voltage parameters, current parameters, and power parameters of the distribution equipment in the distribution network, the voltage parameters, current parameters, and power parameters are normalized. Since the units and dimensions of the voltage, current, and power parameters are different, it is necessary to convert the voltage parameters, current parameters, and power parameters to the same scale for subsequent comparison and calculation. They can be normalized by normalization methods such as Z-score normalization. After the processing, the voltage parameters, current parameters and power parameters are weighted according to their importance to the equipment status. The weighted calculation can integrate multiple parameters, namely voltage parameters, current parameters and power parameters, into a single real-time status index. The real-time status index can be used to reflect the overall operating status of the distribution equipment at a certain moment, and the real-time status index can be a dynamically changing value. Among them, the real-time status index is used to evaluate the failure of the distribution network and issue an early warning, which means that the real-time status index monitors the operating status of the distribution equipment in real time. When the real-time status index is lower than a certain threshold, it is regarded that the distribution equipment is faulty or is about to fail. At this time, the early warning mechanism can be triggered to remind the operation and maintenance personnel to conduct inspections and repairs in time.

[0050] Furthermore, the equipment characteristic information and operating status information are integrated to form a complete data set. Based on the integrated data, a digital twin model of the distribution network is constructed using appropriate modeling methods and tools. The initial digital twin model is verified by comparing it with the actual operating data and conducting simulation experiments to ensure that it can accurately simulate the operation of the distribution network, so that the initial digital twin model can accurately reflect the structure, equipment and operating status of the distribution network.

[0051] Execute step A220 to extract a first historical record from the historical distribution network record; execute step A230 to perform a correlation analysis on the first historical environmental feature information in the first historical record and the first historical device status index to obtain a first historical correlation analysis result;

[0052] First, a first historical record is extracted from the historical distribution network record database. The first historical record may include multiple historical environmental characteristic information and equipment historical status index data for subsequent analysis. The historical environmental characteristic information may include meteorological data such as temperature, humidity, wind speed, and air pressure, as well as geographical data such as topography and geology. The equipment historical status index is calculated based on the real-time operating status information, reflecting the overall operating status of the equipment at a certain historical moment. Then, the correlation between the environmental characteristic information and the equipment status index can be analyzed and quantified using methods such as the Pearson correlation coefficient and the Spearman rank correlation coefficient. The first historical correlation analysis result corresponding to the first historical record is obtained through correlation analysis, and the first historical correlation analysis result includes the correlation between the historical environmental characteristic information and the equipment historical status index, as well as the nature and strength of the correlation. For example, there are some meteorological factors (such as high temperature or high humidity) that may be positively correlated with the decrease in the distribution equipment status index, indicating that these environmental factors may have an adverse effect on the normal operation of the equipment.

[0053] Execute step A240, visualize the first historical correlation analysis result using Manhattan distance, and render the visualization result to the initial digital twin teacher model to obtain the digital twin teacher model.

[0054] Based on the Manhattan distance, the environmental feature information and equipment status index in the first historical correlation analysis result are regarded as points in a multidimensional space, and then the Manhattan distance is used to measure the distance between these points. The calculation results of the Manhattan distance are further visualized using visualization tools such as scatter plots and heat maps, so that the first historical correlation analysis results can be presented in a graphical manner, wherein the position or color of each point can represent the strength and direction of the correlation between the environmental feature information and the equipment status index, thereby obtaining a visualization representation result, and by adding corresponding visualization layers or components to the initial digital twin teacher model, the visualization representation result is rendered into the initial digital twin teacher model obtained by the above construction, so that the initial digital twin teacher model can intuitively display the historical correlation analysis results, thereby combining the results of the data analysis with the physical structure of the initial digital twin teacher model to form a more complete and accurate digital twin teacher model for representation, so that the digital twin teacher model can more comprehensively reflect the actual operation of the distribution network, and provide a more accurate and effective basis for subsequent operation and maintenance management, fault prediction and decision support.

[0055] Execute step A300 to obtain a knowledge distillation edge optimization model, where the knowledge distillation edge optimization model is used to perform distillation optimization on the digital twin teacher model, wherein the knowledge distillation edge optimization model includes a first distillation optimization channel, a second distillation optimization channel, and a third distillation optimization channel;

[0056] In order to optimize the performance of the digital twin teacher model in the edge computing environment, a knowledge distillation edge optimization model can be constructed. The knowledge distillation edge optimization model includes three main distillation optimization channels: the first distillation optimization channel, the second distillation optimization channel, and the third distillation optimization channel. Each channel is based on different predetermined distillation logics, and distills the environmental feature information, device feature information, and operating status information respectively.

[0057] Executing step A400, the first distillation optimization channel performs distillation processing on the environmental feature information based on a predetermined relationship distillation logic to obtain an environmental distillation result;

[0058] In a possible implementation, step A400 further includes step A410, extracting a predetermined relationship similarity probability distribution in the predetermined relationship distillation logic, wherein the predetermined relationship similarity probability distribution is used to perform distillation processing on the environmental feature information;

[0059] Before extracting the predetermined relationship similarity probability distribution in the predetermined relationship distillation logic, it is necessary to first clarify the predetermined relationship distillation logic. The predetermined relationship distillation logic is based on the correlation between environmental feature information and device status, as well as the interaction between different environmental features, and the predetermined relationship similarity probability distribution in the predetermined relationship distillation logic according to the important environmental feature information in the digital twin teacher model and the relative relationship strength between them. The expression of the predetermined relationship similarity probability distribution is as follows: ;in, It is used to characterize the similarity divergence between the environmental feature information and the environmental distillation result. The environmental feature information is subjected to dimensionality reduction processing with the constraint of being close to 0. Characterize the divergence of the similarity between the environmental feature information and the environmental distillation result, Characterize the environmental feature points in the environmental feature information Status index of the power distribution equipment The probability of relationship similarity is Characterize the environmental feature points in the environmental distillation result Status index of the power distribution equipment The predetermined relationship similarity probability distribution in the predetermined relationship distillation logic is used for the similarity probability distribution between different environmental feature information or between environmental feature information and device status, and the predetermined relationship similarity probability distribution is applied to the distillation processing of environmental feature information. During the distillation process, this probability distribution can be used to guide the digital twin student model to focus on information similar to the key environmental features in the teacher model, ensuring that the student model can maintain feature representation and relationship structure similar to the teacher model when receiving and learning environmental feature information, thereby obtaining environmental distillation results, which can improve its performance and applicability in edge computing environments.

[0060] Next, step A500 is executed, and the second distillation optimization channel performs distillation processing on the device feature information based on the predetermined feature distillation logic to obtain the device distillation result; the predetermined feature distillation logic focuses on extracting key features such as the type, specifications, and performance parameters of the equipment, and the device feature information is distilled based on the predetermined feature distillation logic through the second distillation optimization channel.

[0061] In a possible implementation, step A500 further includes step A510, wherein the device characteristic information includes a plurality of device physical characteristic parameters and a plurality of device electrical characteristic parameters; executing step A520, assigning weight coefficients to the plurality of device physical characteristic parameters and the plurality of device electrical characteristic parameters to obtain a device characteristic weight assignment result; executing step A530, arranging the plurality of device physical characteristic parameters and the plurality of device electrical characteristic parameters in descending order according to the device characteristic weight assignment result to obtain a device characteristic descending list;

[0062] Since effective processing of device characteristic information is crucial to improving model performance, and the device characteristic information may include multiple physical characteristic parameters such as size, material, weight, and multiple electrical characteristic parameters such as voltage, current, and power, further weight coefficients are assigned to the multiple device physical characteristic parameters and multiple device electrical characteristic parameters based on the experience of domain experts and statistical analysis of historical data. Each parameter is assigned a corresponding weight value based on its importance in predicting the device operating status. The weight value reflects the relative importance of the parameter in the model. The multiple device physical characteristic parameters and multiple device electrical characteristic parameters are then arranged in descending order based on the device characteristic weight assignment results. This means putting the most important parameters at the front of the list, and arranging them in this way so that they can be better called in the subsequent model optimization and distillation process, and on this basis, a descending list of device characteristics is obtained.

[0063] Execute step A540 to perform pruning processing on the features with a predetermined last ratio threshold in the descending list of device features based on the predetermined feature distillation logic to obtain the device distillation result.

[0064] First, according to the predetermined feature distillation logic, the features to be retained and the features to be pruned are determined, wherein the predetermined feature distillation logic can be analyzed and set based on domain knowledge, expert experience or historical data to define the importance and retention criteria of the features. Then, in order to better reduce the scale of the model, reduce the computational complexity of the model, improve the generalization ability of the model, and reduce the risk of overfitting, the features of the predetermined last-place ratio threshold are set. For example, the threshold can be set to 90%, which means that the first 90% of the features with the largest weights are retained, and the remaining 10% of the features are pruned. The specific operation of the pruning process can be to directly remove these last-place features from the model, or set their weights to zero to minimize their impact in the model, thereby obtaining a streamlined device feature set, i.e., the device distillation result.

[0065] Furthermore, since unstructured pruning is efficient but easily affects accuracy, in a possible implementation, step A540 further includes step A541, obtaining the index difference between the student real-time status index and the real-time status index; executing step A542, determining whether the index difference reaches a predetermined difference range; executing step A543, if the index difference does not reach the predetermined difference range, reducing the predetermined last proportion threshold.

[0066] First, the index difference between the student real-time state index and the real-time state index is obtained. The student real-time state index and the real-time state index represent the performance of the student model and the teacher model at a specific moment, respectively. By comparing the student real-time state index and the real-time state index, the performance gap between the student model and the teacher model can be obtained, and then whether the index difference reaches the predetermined difference range is judged. The predetermined difference range is a threshold set according to actual needs to judge whether the performance gap between the student model and the teacher model is too large. If the index difference does not reach the predetermined difference range, it means that the performance of the student model is acceptable, but there is still room for optimization. Then the predetermined last-place ratio threshold is reduced and adjusted. At the same time, it is necessary to set an appropriate adjustment range according to the size of the index difference. If the index difference is small, it means that the performance of the student model is already close to the teacher model. At this time, the threshold can be appropriately reduced. If the index difference is large, it means that the performance of the student model still has a large room for improvement. At this time, the threshold can be more boldly reduced. Reducing the predetermined last-place ratio threshold is to reduce the number of features retained during pruning, so as to further simplify the model structure while retaining key features. By reducing the predetermined last-place ratio threshold, features can be screened more strictly, and features that contribute less to the performance of the student model can be removed, thereby improving the performance and generalization ability of the student model.

[0067] Executing step A600, the third distillation optimization channel performs distillation processing on the operation status information based on a predetermined response distillation logic to obtain an operation distillation result;

[0068] In a possible implementation, step A600 further includes step A610, extracting a predetermined student accuracy probability in the predetermined response distillation logic; executing step A620, reading a predetermined distillation loss function, and combining the predetermined student accuracy probability and the real-time state index to obtain the running distillation result,

[0069] First, the predetermined student accuracy probability is extracted from the predetermined response distillation logic. The predetermined student accuracy probability is set based on domain knowledge, expert experience or historical data, and can be used to measure the accuracy of the student model in response prediction, reflecting the expectations for the performance of the student model on a specific task. The predetermined distillation loss function is further read, wherein the expression of the predetermined distillation loss function is as follows: ;in, characterizing the predetermined distillation loss function, Characterize the real-time status index obtained by the digital twin teacher model The real-time student status index obtained with the digital twin student model The divergence between Characterize the temperature coefficient, Characterizing the digital twin teacher model The real-time status index A real-time status index.

[0070] The distillation loss function is an important indicator to measure the difference between the student model and the teacher model. It can be used to measure the difference between the probability distribution output by the student model of the digital twin model and the probability distribution output by the teacher model of the digital twin model, as well as the prediction loss of the digital twin student model itself. Further combining the predetermined student accuracy probability and the real-time status index and inputting them into the predetermined distillation loss function can more comprehensively evaluate the accuracy and performance of the student model in response prediction. The real-time status index is used to reflect the performance status of the student model during real-time operation, which can serve as an important reference for the performance evaluation of the student model. Finally, the running distillation result is obtained by calculating the predetermined distillation loss function.

[0071] Execute step A700 to obtain a digital twin student model according to the environment distillation result, the equipment distillation result and the operation distillation result, wherein the digital twin student model represents the distillation optimization model of the digital twin teacher model.

[0072] By analyzing the information on model performance under different environmental conditions provided by the environmental distillation results, the adaptability of the digital twin student model in different environments is obtained, and the structure and parameters of the digital twin student model are adjusted accordingly to improve its stability and accuracy in various environments. The device distillation results are then analyzed to obtain the optimization effect of the digital twin student model on specific devices, ensuring that the digital twin student model can maintain high performance on the target device while reducing computational complexity and resource consumption. The distillation results are run to evaluate the performance of the student model during real-time operation. By comparing the real-time state index of the digital twin student model with the state index of the digital twin teacher model, the accuracy and efficiency of the digital twin student model when simulating the digital twin teacher model can be obtained. The parameters and structure of the student model are further adjusted to optimize its performance in actual operation. The digital twin student model not only inherits the core knowledge of the digital twin teacher model, but is also optimized for specific environments and devices, so that it has higher accuracy and efficiency in practical applications, providing better support for practical applications.

[0073] The embodiments of the present application solve the technical problems of delayed data transmission and slow response speed at the edge of the distribution network, which lead to low operating efficiency of the distribution network. It achieves the technical effect of improving the operating efficiency of the distribution network by optimizing the edge of the distribution network through constructing digital twin space modeling.

[0074] In the above, refer to Figure 1 The present invention describes in detail an edge optimization method for digital twin space modeling of a distribution network according to an embodiment of the present invention. Figure 2 An edge optimization system for digital twin spatial modeling of a distribution network according to an embodiment of the present invention is described.

[0075] An edge optimization system for digital twin spatial modeling of a distribution network according to an embodiment of the present invention is used to solve the technical problem of delayed data transmission and slow response speed at the edge of the distribution network, which leads to low operating efficiency of the distribution network, and realizes edge optimization of the distribution network by constructing digital twin spatial modeling, thereby achieving the technical effect of improving the operating efficiency of the distribution network. An edge optimization system for digital twin spatial modeling of a distribution network includes: an information collection module 10, a first model construction module 20, a distillation optimization module 30, a first distillation processing module 40, a second distillation processing module 50, a third distillation processing module 60, and a second model construction module 70.

[0076] An information collection module 10, the information collection module 10 is used to respectively collect environmental characteristic information of the power distribution network and equipment characteristic information of the power distribution equipment in the power distribution network;

[0077] A first model building module 20, the first model building module 20 is used to build a digital twin teacher model of the distribution network by combining the environmental feature information, the device feature information and the dynamically monitored operating status information of the distribution equipment;

[0078] A distillation optimization module 30, wherein the distillation optimization module 30 is used to obtain a knowledge distillation edge optimization model, wherein the knowledge distillation edge optimization model is used to perform distillation optimization on the digital twin teacher model, wherein the knowledge distillation edge optimization model includes a first distillation optimization channel, a second distillation optimization channel, and a third distillation optimization channel;

[0079] A first distillation processing module 40, wherein the first distillation processing module 40 is used for the first distillation optimization channel to perform distillation processing on the environmental feature information based on a predetermined relationship distillation logic to obtain an environmental distillation result;

[0080] A second distillation processing module 50, wherein the second distillation processing module 50 is used for the second distillation optimization channel to perform distillation processing on the device feature information based on a predetermined feature distillation logic to obtain a device distillation result;

[0081] A third distillation processing module 60, wherein the third distillation processing module 60 is used for the third distillation optimization channel to perform distillation processing on the operation status information based on a predetermined response distillation logic to obtain an operation distillation result;

[0082] The second model building module 70 is used to obtain a digital twin student model according to the environment distillation results, the equipment distillation results and the operation distillation results, and the digital twin student model represents the distillation optimization model of the digital twin teacher model.

[0083] Below, the specific configuration of the first model construction module 20 will be described in detail. As described above, the digital twin teacher model of the distribution network is constructed in combination with the environmental characteristic information, the device characteristic information and the dynamically monitored operating status information of the distribution equipment. The first model construction module 20 may further include: a third model construction unit is used to construct an initial digital twin teacher model of the distribution network based on the device characteristic information and the operating status information; a first extraction unit is used to extract a first historical record in a historical distribution network record; a first analysis unit is used to perform a correlation analysis on the first historical environmental characteristic information in the first historical record and the first historical device status index to obtain a first historical correlation analysis result; a first rendering unit is used to visualize the first historical correlation analysis result using Manhattan distance, and render its visualization result to the initial digital twin teacher model to obtain the digital twin teacher model.

[0084] The specific configuration of the first distillation processing module 40 will be described in detail below. As described above, the first distillation optimization channel performs distillation processing on the environmental feature information based on the predetermined relationship distillation logic to obtain the environmental distillation result. The first distillation processing module 40 may further include: a second extraction unit for extracting a predetermined relationship similarity probability distribution in the predetermined relationship distillation logic, and the predetermined relationship similarity probability distribution is used to perform distillation processing on the environmental feature information;

[0085] The expression of the probability distribution of the predetermined relationship similarity is as follows: ;in, It is used to characterize the similarity divergence between the environmental feature information and the environmental distillation result. The environmental feature information is subjected to dimensionality reduction processing with the constraint of being close to 0. Characterize the divergence of the similarity between the environmental feature information and the environmental distillation result, Characterize the environmental feature points in the environmental feature information Status index of the power distribution equipment The probability of relationship similarity is Characterize the environmental feature points in the environmental distillation result Status index of the power distribution equipment The probability of relationship similarity.

[0086] The specific configuration of the third model building unit will be described in detail below. As described above, the third model building unit may further include: a first parameter acquisition unit for the operation status information including voltage parameters, current parameters and power parameters; a first calculation unit for performing weighted calculation on the normalized processing results of the voltage parameters, the current parameters and the power parameters to obtain the real-time status index of the power distribution equipment;

[0087] The real-time status index is used to evaluate the failure of the distribution network and issue an early warning.

[0088] The specific configuration of the second distillation processing module 50 will be described in detail below. As described above, the second distillation optimization channel performs distillation processing on the device feature information based on a predetermined feature distillation logic to obtain a device distillation result. The second distillation processing module 50 may further include: a second parameter acquisition unit for the device feature information including a plurality of device physical characteristic parameters and a plurality of device electrical characteristic parameters; a coefficient allocation unit for allocating weight coefficients to the plurality of device physical characteristic parameters and the plurality of device electrical characteristic parameters to obtain a device feature weight allocation result; a descending order arrangement unit for arranging the plurality of device physical characteristic parameters and the plurality of device electrical characteristic parameters in descending order according to the device feature weight allocation result to obtain a device feature descending list; a pruning processing unit for pruning features of a predetermined last ratio threshold in the device feature descending list based on the predetermined feature distillation logic to obtain the device distillation result.

[0089] The specific configuration of the third distillation processing module 60 will be described in detail below. As described above, the third distillation optimization channel performs distillation processing on the running state information based on the predetermined response distillation logic to obtain the running distillation result. The third distillation processing module 60 may further include: a third extraction unit for extracting the predetermined student accuracy probability in the predetermined response distillation logic; a first reading unit for reading a predetermined distillation loss function, and combining the predetermined student accuracy probability and the real-time state index to obtain the running distillation result, wherein the expression of the predetermined distillation loss function is as follows: ;in, characterizing the predetermined distillation loss function, Characterize the real-time status index obtained by the digital twin teacher model The real-time student status index obtained with the digital twin student model The divergence between Characterize the temperature coefficient, Characterizing the digital twin teacher model The real-time status index A real-time status index.

[0090] The specific configuration of the pruning processing unit will be described in detail below. As described above, the pruning processing unit may further include: a second calculation unit for obtaining the index difference between the student real-time status index and the real-time status index; a first judgment unit for judging whether the index difference reaches a predetermined difference range; and a reduction adjustment unit for reducing the predetermined last-place ratio threshold if the index difference does not reach the predetermined difference range.

[0091] An edge optimization system for digital twin spatial modeling of a distribution network provided by an embodiment of the present invention can execute an edge optimization method for digital twin spatial modeling of a distribution network provided by any embodiment of the present invention, and has functional modules and beneficial effects corresponding to the execution method.

[0092] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0093] The above specific implementations do not constitute a limitation on the protection scope of this application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of this application should be included in the protection scope of this application.

Claims

1. An edge optimization method for digital twin space modeling of distribution network, characterized in that: include: respectively collecting and obtaining environmental characteristic information of the power distribution network and equipment characteristic information of the power distribution equipment in the power distribution network; Building a digital twin teacher model of the distribution network by combining the environmental characteristic information, the equipment characteristic information and the dynamically monitored operating status information of the distribution equipment; Acquire a knowledge distillation edge optimization model, where the knowledge distillation edge optimization model is used to perform distillation optimization on the digital twin teacher model, wherein the knowledge distillation edge optimization model includes a first distillation optimization channel, a second distillation optimization channel, and a third distillation optimization channel; The first distillation optimization channel performs distillation processing on the environmental feature information based on a predetermined relationship distillation logic to obtain an environmental distillation result; The second distillation optimization channel performs distillation processing on the device feature information based on a predetermined feature distillation logic to obtain a device distillation result; The third distillation optimization channel performs distillation processing on the operation status information based on a predetermined response distillation logic to obtain an operation distillation result; Obtaining a digital twin student model according to the environment distillation result, the equipment distillation result and the operation distillation result, wherein the digital twin student model represents a distillation optimization model of the digital twin teacher model; Combining the environmental characteristic information, the device characteristic information and the dynamically monitored operating status information of the power distribution equipment to construct a digital twin teacher model of the power distribution network, including: Building an initial digital twin teacher model of the distribution network based on the device feature information and the operating status information; Extracting the first historical record in the historical distribution network record; Performing a correlation analysis on the first historical environment feature information and the first historical device status index in the first historical record to obtain a first historical correlation analysis result; The first historical correlation analysis result is visualized using Manhattan distance, and the visualization result is rendered to the initial digital twin teacher model to obtain the digital twin teacher model.

2. According to claim 1, an edge optimization method for distribution network digital twin space modeling is characterized in that: The first distillation optimization channel performs distillation processing on the environmental feature information based on a predetermined relationship distillation logic to obtain an environmental distillation result, including: Extracting a predetermined relationship similarity probability distribution in the predetermined relationship distillation logic, wherein the predetermined relationship similarity probability distribution is used to perform distillation processing on the environmental feature information; The expression of the probability distribution of the predetermined relationship similarity is as follows: ;in, It is used to characterize the similarity divergence between the environmental feature information and the environmental distillation result. The environmental feature information is subjected to dimensionality reduction processing with the constraint of being close to 0. Characterize the divergence of the similarity between the environmental feature information and the environmental distillation result, Characterize the environmental feature points in the environmental feature information Status index of the power distribution equipment The probability of relationship similarity is Characterize the environmental feature points in the environmental distillation result Status index of the power distribution equipment The probability of relationship similarity.

3. According to claim 2, an edge optimization method for distribution network digital twin space modeling is characterized in that: The method also includes: The operating status information includes voltage parameters, current parameters and power parameters; Performing weighted calculation on normalized processing results of the voltage parameter, the current parameter and the power parameter to obtain a real-time status index of the power distribution equipment; The real-time status index is used to evaluate the failure of the distribution network and issue an early warning.

4. According to claim 3, an edge optimization method for distribution network digital twin space modeling is characterized in that: The second distillation optimization channel performs distillation processing on the device feature information based on a predetermined feature distillation logic to obtain a device distillation result, including: The device characteristic information includes a plurality of device physical characteristic parameters and a plurality of device electrical characteristic parameters; Allocating weight coefficients for the multiple device physical characteristic parameters and the multiple device electrical characteristic parameters to obtain a device characteristic weight allocation result; Arrange the plurality of device physical characteristic parameters and the plurality of device electrical characteristic parameters in descending order according to the device characteristic weight allocation result to obtain a device characteristic descending order list; Based on the predetermined feature distillation logic, features with a predetermined last ratio threshold in the descending list of device features are pruned to obtain the device distillation result.

5. According to claim 3, an edge optimization method for distribution network digital twin space modeling is characterized in that: The third distillation optimization channel performs distillation processing on the operation status information based on a predetermined response distillation logic to obtain an operation distillation result, including: extracting a predetermined student accuracy probability in the predetermined response distillation logic; The predetermined distillation loss function is read, and the running distillation result is obtained by combining the predetermined student accuracy probability and the real-time state index, wherein the expression of the predetermined distillation loss function is as follows: ; in, characterizing the predetermined distillation loss function, Characterize the real-time status index obtained by the digital twin teacher model The real-time student status index obtained with the digital twin student model The divergence between Characterize the temperature coefficient, Characterizing the digital twin teacher model The real-time status index A real-time status index.

6. According to claim 4, an edge optimization method for distribution network digital twin space modeling is characterized in that: The method also includes: Get the index difference between the student's real-time status index and the real-time status index; Determining whether the index difference reaches a predetermined difference range; If the index difference does not reach the predetermined difference range, the predetermined last ratio threshold is reduced and adjusted.

7. An edge optimization system for digital twin spatial modeling of distribution networks, characterized in that: The system is used to implement an edge optimization method for digital twin space modeling of a distribution network according to any one of claims 1 to 6, comprising: An information collection module, the information collection module is used to respectively collect environmental characteristic information of the power distribution network and equipment characteristic information of the power distribution equipment in the power distribution network; A first model building module, wherein the first model building module is used to build a digital twin teacher model of the distribution network by combining the environmental feature information, the device feature information and the dynamically monitored operating status information of the distribution equipment; A distillation optimization module, wherein the distillation optimization module is used to obtain a knowledge distillation edge optimization model, wherein the knowledge distillation edge optimization model is used to perform distillation optimization on the digital twin teacher model, wherein the knowledge distillation edge optimization model includes a first distillation optimization channel, a second distillation optimization channel, and a third distillation optimization channel; A first distillation processing module, wherein the first distillation processing module is used for the first distillation optimization channel to perform distillation processing on the environmental feature information based on a predetermined relationship distillation logic to obtain an environmental distillation result; A second distillation processing module, wherein the second distillation processing module is used for the second distillation optimization channel to perform distillation processing on the device feature information based on a predetermined feature distillation logic to obtain a device distillation result; A third distillation processing module, wherein the third distillation processing module is used for the third distillation optimization channel to perform distillation processing on the operation status information based on a predetermined response distillation logic to obtain an operation distillation result; The second model building module is used to obtain a digital twin student model according to the environment distillation results, the equipment distillation results and the operation distillation results, and the digital twin student model represents the distillation optimization model of the digital twin teacher model.

Citation Information

Patent Citations

  • Knowledge distillation method, greening area detection method, electronic equipment and storage medium

    CN117710792A

  • Intrusion detection method under group learning architecture fused with knowledge distillation

    CN117812593A

  • Large-scale power distribution network parallel optimization method based on digital twin space

    CN118017509A

  • Twinning architecture tracking model knowledge distillation method based on hierarchical feature fusion

    CN118134966A

  • Power distribution network fault monitoring method and system based on digital twinning

    CN118381197A