A power grid multi-dimensional loss fine management method, system, device and medium based on a CIM standard

Abstract

The application discloses a kind of based on CIM standard's power grid multidimensional net loss fine management method, system, equipment and medium, belong to distribution network net loss management field, the method of the present application includes: generating power grid digital twin topological structure;Quantitative measurement data are carried out space-time alignment preprocessing, and dataset is constructed;Theoretical line loss estimation value and equipment loss contribution degree sorting of each branch are generated by using graph structure learning network to carry out deep feature cross fusion;Line loss estimation value is input into multidimensional attribution analysis engine, and equipment health characteristics, load composition characteristics, topological structure characteristics are fused, main cause category is located and loss reduction strategy is generated;It is transformed into control instruction and is executed, and after the execution, the measured data are collected to calculate loss reduction effect, and the result is fed back to the feature extraction link, and the feature weight distribution is corrected.The management accuracy and adaptability of the system are improved persistently when power grid structure and load characteristics change, and the evolution of distribution network net loss management to intelligent, precision and self-direction is promoted.

Description

Technical Field

[0001] This invention relates to the field of power distribution network loss management technology, specifically to a method, system, equipment, and medium for refined management of multi-dimensional power distribution network losses based on the CIM standard. Background Technology

[0002] In the field of distribution network loss management, traditional methods mainly rely on the difference in active power to calculate line losses, simply attributing the difference between total power supply and total power sales to line losses, lacking a refined consideration of deeper factors such as equipment status, load characteristics, and topology. With the expansion of distribution network scale and the diversification of loads, this "black box estimation" model can no longer meet the needs of precise management. While the Unified Information Model for Power Grid (CIM) standard, as a standardized semantic framework for power system equipment, topology, and measurement data, has been applied in some main grid systems, its potential in distribution network loss management has not yet been fully explored.

[0003] While existing technologies attempt to incorporate information technology, they suffer from limited feature dimensions, failing to construct a comprehensive system covering equipment operating status. Furthermore, they lack a feedback mechanism after strategy execution, hindering the automatic optimization of feature weights and attribution models based on actual loss reduction effects. Consequently, management strategies become rigid and ineffective, preventing self-evolution and reducing the accuracy of distribution network loss control. Therefore, improvements are needed. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by this invention is: how to solve the problems in the existing distribution network loss management, such as the ambiguity in cause identification due to single characteristics, the insufficient control accuracy due to rigid strategies, and the inability of the system to evolve on its own due to the lack of feedback mechanisms.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a method for refined management of multi-dimensional network losses in power grids based on the CIM standard, comprising, Semantic modeling of power transmission and distribution equipment is performed based on the power grid information model standard to generate a digital twin topology of the power grid with spatiotemporal attributes. Spatiotemporal alignment preprocessing is performed on equipment measurement data, temporal features are extracted through sliding time windows, topological correlation features are extracted through spatial neighborhood analysis, and a multi-dimensional feature dataset is constructed. A graph-structured learning network is used to perform deep feature cross-fusion on the feature dataset to generate theoretical line loss estimates and equipment loss contribution rankings for each branch. Input the estimated line loss into the multidimensional attribution analysis engine, integrate equipment health characteristics, load composition characteristics, and topology characteristics, locate the main cause category, and generate loss reduction strategies. The loss reduction strategy is transformed into control commands for execution. Measurement data after execution is collected to calculate the loss reduction effect, and the effect evaluation results are fed back to the feature extraction stage to correct the feature weight allocation.

[0007] As a preferred embodiment of the power grid multi-dimensional refined management method for network losses based on the CIM standard described in this invention, the step of semantically modeling transmission and distribution equipment based on the power grid information model standard to generate a digital twin topology structure of the power grid with spatiotemporal attributes includes: The physical parameters and electrical connections of power transmission and distribution equipment are mapped into standardized digital model components based on the power grid information model standard. By integrating the geospatial coordinates of the equipment with the electrical connections of the digital model components, nodes and edges with spatiotemporal attributes are formed. Based on nodes and edges, a dynamic topology graph is constructed that can fully represent the electrical connections and spatial distribution of the power grid.

[0008] As a preferred embodiment of the CIM standard-based multi-dimensional refined management method for power grid losses described in this invention, the method involves: performing spatiotemporal alignment preprocessing on equipment measurement data, extracting temporal features through a sliding time window, extracting topological correlation features through spatial neighborhood analysis, and constructing a multi-dimensional feature dataset. Calculate the information gain ratio of data from each measurement channel and identify key characteristic channels that significantly affect line loss. When the data missing rate of the measurement channel is detected to exceed the preset missing threshold, the feature reconstruction algorithm based on spatiotemporal interpolation is automatically activated to fill in the missing feature values ​​by utilizing the temporal correlation of adjacent nodes. The reconstructed features are evaluated for quality, feature confidence labels are generated, and the labels are used as the basis for adjusting feature weights to reduce the contribution of low-quality features in online loss calculation.

[0009] As a preferred embodiment of the CIM standard-based multi-dimensional refined management method for power grid losses described in this invention, the step of using a graph structure learning network to perform deep feature cross-fusion on the feature dataset to generate theoretical line loss estimates and equipment loss contribution rankings for each branch includes: The equipment nameplate parameters are transformed into the equipment rated loss feature vector, which is used as the prior attribute input of the nodes of the graph structure learning network to constrain the theoretical line loss calculation result to be no less than the rated no-load loss. The topological electrical distance characteristics are transformed into edge weight attenuation factors, so that the line loss calculation results of long-distance branches are nonlinearly amplified as the distance increases; Using ambient temperature characteristics as a dynamic correction coefficient, the theoretical line loss value is compensated for temperature drift, thus eliminating calculation deviations caused by seasonal temperature differences.

[0010] This invention embeds multiple physical constraints and correction mechanisms, including rated equipment loss, electrical distance, and ambient temperature, into a graph-structured learning network. This ensures that theoretical line loss estimation not only relies on data correlation but also strictly adheres to the fundamental physical laws of power equipment. By using rated loss as a baseline, this invention guarantees the physical rationality of the estimation results and eliminates underestimations that defy common sense. The introduction of electrical distance as a nonlinear attenuation factor allows the calculation results to accurately reflect the inherent loss characteristics of long-distance power supply lines. Furthermore, dynamic compensation for ambient temperature eliminates the interference of seasonal environmental changes on line loss estimation, improving the model's computational accuracy and stability across all weather conditions and scenarios.

[0011] As a preferred embodiment of the CIM standard-based multi-dimensional refined management method for power grid losses described in this invention, the step of inputting the estimated line loss value into a multi-dimensional attribution analysis engine, integrating equipment health characteristics, load composition characteristics, and topology characteristics, locating the main cause category, and generating loss reduction strategies includes: Construct a health status degradation characteristic curve for equipment, and use the time-series change rate of insulation parameters and dielectric loss as health status characteristics. When the health characteristics show an accelerated degradation trend, increase the attribution weight of the main cause of equipment aging. Extract load imbalance characteristics, calculate the proportion of zero-sequence and negative-sequence components of three-phase current, and trigger the reactive power compensation strategy priority execution queue when the imbalance characteristics exceed the preset missing threshold. Identify topology bottleneck characteristics, calculate the spatial matching degree between power supply radius and load density, and when the matching degree is lower than the preset matching threshold, increase the weight of the main cause of structural network loss and generate load transfer suggestions.

[0012] This invention utilizes a multi-dimensional attribution analysis engine that combines design rules and feature criteria to achieve intelligent diagnosis of the root causes of line losses and generate differentiated strategies. By continuously monitoring the degradation trends of equipment health characteristics, the engine can proactively identify potential equipment aging issues and use this as a key basis for formulating replacement or maintenance plans. Through real-time calculation of load imbalance, it can automatically trigger rapid adjustment commands such as reactive power compensation when anomalies occur, achieving real-time loss suppression. By analyzing the spatial matching degree between power supply radius and load density, it can accurately locate network structural weaknesses and generate optimization suggestions such as load shifting.

[0013] As a preferred embodiment of the CIM standard-based multi-dimensional refined management method for power grid losses described in this invention, the steps of converting loss reduction strategies into control commands for execution, collecting measurement data after execution to calculate the loss reduction effect, and feeding back the effect evaluation results to the feature extraction stage for correcting feature weight allocation include: Calculate the rate of change of contribution of each dimension feature before and after loss reduction, and identify the driving features that affect the loss reduction effect; If the change rate of feature contribution is lower than the preset effective threshold, the feature corresponding to the current change rate of feature contribution is marked as an inert feature, and the inert feature is optimized in the next round of feature extraction. The contribution ranking results of the driving features are fed back to the multidimensional attribution analysis engine to optimize the initial weight allocation of each feature channel when locating the main cause category.

[0014] As a preferred embodiment of the CIM standard-based multi-dimensional refined management method for power grid losses described in this invention, the steps of performing spatiotemporal alignment preprocessing on equipment measurement data, extracting temporal features through a sliding time window, extracting topological correlation features through spatial neighborhood analysis, and constructing a multi-dimensional feature dataset further include: The raw measurement data is classified at the source, and the data source type, collection time period attribute and quality level are marked to generate a labeled raw data pool; Cross-category association analysis is performed on the categorized data pool to identify the implicit correlations between different categories of data and to construct a data association mapping table; Based on the aforementioned association mapping table, data consistency comparison is performed. When a logical conflict is found between categories, a correction process is automatically triggered to resample the abnormal category data and adjust the weights of the corresponding data according to the degree of conflict.

[0015] This invention provides a refined management system for multi-dimensional network loss based on the CIM standard.

[0016] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a multi-dimensional refined management system for power grid losses based on the CIM standard, comprising: a digital twin topology generation module, a multi-dimensional feature dataset construction module, a loss calculation module, a loss reduction strategy generation module, and a loss reduction effect calculation module; The digital twin topology generation module is used to perform semantic modeling of power transmission and distribution equipment based on the power grid information model standard, and generate a power grid digital twin topology with spatiotemporal attributes. The multi-dimensional feature dataset construction module is used to perform spatiotemporal alignment preprocessing on device measurement data, extract temporal features through a sliding time window, extract topological correlation features through spatial neighborhood analysis, and construct a multi-dimensional feature dataset. The loss calculation module is used to perform deep feature cross-fusion on the feature dataset using a graph structure learning network to generate theoretical line loss estimates and equipment loss contribution rankings for each branch. The loss reduction strategy generation module is used to input the line loss estimate into the multidimensional attribution analysis engine, integrate equipment health characteristics, load composition characteristics, and topology characteristics, locate the main cause category, and generate a loss reduction strategy. The loss reduction effect calculation module is used to convert the loss reduction strategy into control commands for execution, collect measurement data after execution to calculate the loss reduction effect, and feed the effect evaluation results back to the feature extraction stage to correct the feature weight allocation.

[0017] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the aforementioned method for refined management of multi-dimensional network losses in power grids based on the CIM standard.

[0018] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the aforementioned method for refined management of multi-dimensional network losses in a power grid based on the CIM standard.

[0019] The beneficial effects of this invention are as follows: Based on the CIM standard, this invention constructs a digital twin topology for the power grid, providing a unified semantic foundation for all network equipment and measurement data. This transforms line loss analysis from traditional black-box estimation to transparent analysis based on a unified spatiotemporal benchmark. By extracting and fusing the temporal operating characteristics of equipment with topological association characteristics, and utilizing graph structure learning to deeply characterize the loss transmission relationships between equipment, the accuracy of theoretical line loss calculation is improved, and key loss-prone equipment and paths can be automatically identified.

[0020] This invention's multi-dimensional attribution analysis engine comprehensively analyzes multi-source information such as equipment health, load characteristics, and network structure to accurately pinpoint the root causes of line losses and generate targeted loss reduction strategies, achieving precise decision-making from experience-driven to data-driven approaches. By feeding back the actual effects of strategy execution to the front-end feature extraction and model analysis stages, this invention forms a self-evolving closed loop capable of dynamically evaluating feature value and continuously optimizing weight allocation and attribution logic. This ensures a sustained improvement in the system's management accuracy and adaptability as the power grid structure and load characteristics change, comprehensively promoting the evolution of distribution network loss management towards intelligence, precision, and autonomy. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 The following is an overall flowchart of a method for refined management of multi-dimensional network losses in power grids based on the CIM standard, provided as an embodiment of the present invention. Detailed Implementation

[0023] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0024] Example 1, referring to Figure 1 This is one embodiment of the present invention, which provides a method for refined management of multi-dimensional network losses in power grids based on the CIM standard, including: This paper addresses three major shortcomings in current distribution network loss management: reliance on single power statistics leading to blind spots in causal analysis, difficulty in quantifying the impact of topological correlations through discrete feature processing, and a disconnect between strategy execution and effect evaluation, preventing the system from evolving.

[0025] This invention proposes a multi-dimensional refined management method for power grid losses based on the Computational Information Model (CIM) standard. This method fundamentally changes the traditional "black box estimation" model by constructing a standardized digital twin foundation, integrating graph structure learning and multi-dimensional attribution analysis, and establishing a closed-loop feedback optimization mechanism. This systematically achieves interpretability, quantifiability, and self-optimization of line loss management.

[0026] S1. Based on the power grid information model standard, semantic modeling of power transmission and distribution equipment is performed to generate a digital twin topology of the power grid with spatiotemporal attributes; S2. Perform spatiotemporal alignment preprocessing on the equipment measurement data, extract temporal features through a sliding time window, extract topological correlation features through spatial neighborhood analysis, and construct a multi-dimensional feature dataset. S3. A graph structure learning network is used to perform deep feature cross-fusion on the feature dataset to generate theoretical line loss estimates and equipment loss contribution rankings for each branch. S4. Input the estimated line loss into the multidimensional attribution analysis engine, integrate equipment health characteristics, load composition characteristics, and topology characteristics, locate the main cause category, and generate loss reduction strategies. S5. Transform the loss reduction strategy into control commands and issue them for execution. Collect the measurement data after execution to calculate the loss reduction effect, and feed the effect evaluation results back to the feature extraction stage to correct the feature weight allocation.

[0027] Example 2, an embodiment of the present invention, provides a method for refined management of multi-dimensional network losses in power grids based on the CIM standard, based on the previous embodiment, including: S1. Based on the power grid information model standard, semantic modeling of power transmission and distribution equipment is performed to generate a digital twin topology of the power grid with spatiotemporal attributes, including the following steps: S11. Map the physical parameters and electrical connection relationships of power transmission and distribution equipment into standardized digital model components according to the power grid information model standard; S12. The geospatial coordinates of the fusion equipment and the electrical connection relationship of the digital model components are used to form nodes and edges with spatiotemporal attributes. S13. Based on nodes and edges, construct a dynamic topology graph that can fully represent the electrical connections and spatial distribution of the power grid.

[0028] Specifically, for the power distribution network of large commercial complexes, semantic modeling is carried out on power transmission and distribution equipment such as dry-type transformers, medium-voltage switchgear, and low-voltage outgoing cables based on the unified information model standard.

[0029] The nameplate parameters of the transformer, such as rated capacity, short-circuit impedance, and no-load loss, the node connection relationship and protection settings of the switchgear, and the physical information such as cable type, length, and laying method are mapped into a digital model in a unified format.

[0030] By integrating geographic information system coordinates with electrical connection relationships, a digital twin topology of the power grid with spatiotemporal attributes is generated, presenting the full-link topology relationship from the incoming line to the terminal merchant distribution box.

[0031] This structure provides a unified benchmark for subsequent line loss calculations, avoiding calculation deviations caused by outdated drawing versions or wiring errors when traditionally drawing topology diagrams manually, and improving the accuracy and traceability of basic data for line loss management.

[0032] It should be noted that this step is based on the CIM standard to perform semantic modeling of power transmission and distribution equipment (transformers, switchgear, cables, etc.), mapping the physical parameters, connection relationships, and measurement data of the equipment into a unified information model, generating a digital twin topology of the power grid with spatiotemporal attributes. The CIM standard provides a standardized equipment description language and topology expression framework for the entire solution, serving as the data foundation for all subsequent feature extraction, line loss calculation, and attribution analysis.

[0033] S2. Perform spatiotemporal alignment preprocessing on the equipment measurement data, extract temporal features through a sliding time window, extract topological correlation features through spatial neighborhood analysis, and construct a multi-dimensional feature dataset, including the following steps: S21. Calculate the information gain ratio of the data from each measurement channel and identify the key characteristic channels that have a significant impact on line loss.

[0034] Specifically, in the multi-dimensional feature dataset of the commercial complex's power distribution network, the system calculates the information gain rate of each measurement channel and finds that the three-phase current imbalance feature has the strongest explanatory power for line loss changes, while the cable sheath temperature feature has a weak impact on the results. Based on this, the system marks this three-phase imbalance feature as a key feature channel, assigning it higher weight in subsequent line loss calculations, while reducing the participation of temperature features. This mechanism automatically selects the feature dimensions with the greatest value for line loss management, avoids redundant features interfering with calculation accuracy, makes attribution analysis more focused on core issues, and reduces unnecessary computational overhead.

[0035] S22. When the data missing rate of the measurement channel is detected to exceed the preset missing threshold, the feature reconstruction algorithm based on spatiotemporal interpolation is automatically activated to fill in the missing feature values ​​by utilizing the temporal correlation of adjacent nodes.

[0036] S23. Evaluate the quality of the reconstructed features, generate feature confidence labels, and use the labels as the basis for adjusting feature weights to reduce the contribution of low-quality features in online loss calculation.

[0037] Specifically, the system performs a quality assessment on the reconstructed current characteristics of S202, generating a "Reconstructed - Medium Confidence" label. This label is used as the basis for adjusting feature weights, automatically reducing the contribution of this feature in subsequent line loss calculations. If a feature from the original data collection is assessed as "Original - High Confidence," its weight is increased accordingly. This mechanism enables the line loss calculation model to distinguish data quality, avoid low-confidence features from misleading calculation results, significantly improve overall estimation accuracy and result reliability, and provide high-quality data support for precise loss reduction strategies.

[0038] In one optional implementation, the spatiotemporal alignment preprocessing of S2 is achieved based on a precise time synchronization protocol and real-time querying of a graph database. Specifically, clock synchronization modules deployed on each measurement terminal (such as those based on GPS or IEEE 1588 protocol) are used to achieve microsecond-level time standardization, ensuring precise alignment of data time stamps. In the spatial dimension, the topological relationships exported from CIM are imported into a real-time graph database. Through efficient neighborhood traversal queries, electrical connections and geographical proximity relationships between devices are dynamically bound, and related features are extracted based on this. In addition, an adaptive sliding window mechanism is introduced to dynamically adjust the window size according to the load change rate to capture the best time series features. At the same time, an anomaly detection model based on streaming processing is used to clean and correct the raw data in real time.

[0039] In another optional implementation, the spatiotemporal alignment preprocessing of S2 employs a strategy of edge computing collaboration and lightweight spatiotemporal coding. This approach deploys preprocessing units close to the data source at the edge. Initial time alignment is achieved first using local clock synchronization and post-injection time stamping techniques. Spatial relationships are quickly matched using pre-defined device topology configuration files and lightweight spatial indexes, eliminating the need for continuous reliance on a centralized CIM model. The feature extraction stage uses a fixed-step sliding window and predefined feature calculation templates to reduce computational complexity. Simultaneously, differential coding and compressed transmission techniques are used to upload only feature vectors and necessary spatiotemporal association identifiers to the central platform, significantly reducing communication and processing overhead while ensuring data spatiotemporal consistency.

[0040] In this embodiment, the feature reconstruction algorithm S22 is specific: due to a communication failure, the current transformer in a branch of the power distribution room of a commercial complex experiences a measurement data loss rate exceeding a preset loss threshold. The system automatically activates a feature reconstruction algorithm based on spatiotemporal interpolation, using load data (time-dependent) from the upstream transformer of this branch and current data (spatial-dependent) from adjacent parallel branches to complete the current feature values ​​at the missing time through weighted interpolation. The completed feature values ​​are smoothed, and the deviation from the true values ​​is controlled within an acceptable range, ensuring that line loss calculation is not interrupted due to single-point data loss. This mechanism ensures data integrity and improves the system's robustness and continuous operation capability under sensor failure or communication interruption environments.

[0041] Specifically, this application embodiment further explains the preset missing threshold. In the power distribution network of a commercial complex, the system preset missing threshold is 40%. When the data missing rate of a current transformer in a branch of the power distribution room reaches 42% (exceeding the 40% threshold) within 30 minutes due to a communication failure, the system automatically triggers a feature reconstruction algorithm based on spatiotemporal interpolation.

[0042] Preset Scheme Explanation: The setting of this 40% threshold is based on the following engineering practice: For missing values ​​below 30%, linear interpolation can be used directly to meet the required accuracy, eliminating the need to initiate complex spatiotemporal interpolation algorithms and saving computational resources.

[0043] 30%-40% missing: This is a critical range. The system marks it as a "warning" but will not reconstruct it for the time being, waiting for data recovery.

[0044] A data loss exceeding 40% is considered a persistent sensor failure, necessitating a refactoring process to ensure the continuity of line loss calculations and prevent the entire feeder line loss calculation from failing due to a single point of data interruption. This threshold balances data integrity with computational overhead, ensuring system robustness in sensor failure scenarios. It should be noted that in other embodiments, the preset missing threshold value can be different.

[0045] In one alternative implementation, the feature reconstruction algorithm of S22 is an autoencoder completion algorithm based on graph neural networks.

[0046] The entire distribution network topology, along with its historical measurement data, is treated as a dynamic graph sequence. During the training phase, a graph autoencoder model is trained using complete historical data, enabling it to learn the compressed representation and reconstruction of the grid's operating state in a low-dimensional latent space. In the online completion phase, when data loss is detected, the current graph data containing missing values ​​(with missing positions filled with zeros or the mean) is input into the trained encoder to obtain its latent space representation. The decoder then reconstructs the complete node features. This model can non-linearly capture complex spatiotemporal relationships, and is particularly adept at handling concurrent missing values ​​across multiple nodes and consecutive time periods, maintaining global spatiotemporal consistency in the completion results.

[0047] In another alternative implementation, the feature reconstruction algorithm of S22 is an iterative completion algorithm based on multivariate time series prediction.

[0048] The problem is transformed into time-series prediction of missing channels. It utilizes CIM topology to identify several "critical associated devices" with the strongest electrical coupling and highest correlation to the load curve of the missing device. Using complete measurement data of these critical associated devices before and after the missing period as input features, a lightweight time-series prediction model (such as a temporal convolutional network or gradient boosting tree model) is trained to directly predict the feature values ​​of the missing device at the corresponding time. For long-term missing periods, iterative prediction and confidence feedback are used for multi-step completion. This algorithm is computationally efficient and suitable for edge deployment scenarios with stringent real-time requirements or limited computing resources.

[0049] This invention employs a unified framework of CIM digital twin topology for rigorous spatiotemporal alignment, accurately binding measurement data from different devices, acquisition frequencies, and even timescales to specific physical devices and their electrical neighborhood relationships. This fundamentally eliminates analytical biases caused by spatiotemporal misalignment of data. This ensures that subsequently extracted time-series features (such as volatility) and topological correlation features (such as upstream and downstream influences) possess genuine physical meaning and consistency, overcoming the limitation of traditional methods that can only handle isolated time-series data.

[0050] Furthermore, this invention automatically identifies and focuses on key feature channels that significantly impact line loss by calculating the information gain rate, effectively filtering out redundant data dimensions with low information content. This ensures that the system concentrates limited computing and storage resources on high-value information, avoiding the resource waste and analysis noise caused by the traditional full-feature, equal-processing mode.

[0051] By introducing intelligent feature reconstruction algorithms and a closed-loop quality assessment system, the system possesses strong fault tolerance and self-recovery capabilities. When faced with missing data from some measurement channels, it can automatically complete the data based on spatiotemporal correlation, ensuring uninterrupted analysis. By evaluating the confidence level of reconstructed features and all data and dynamically adjusting their contribution weights, the system can automatically reduce the impact of low-quality and unreliable data on the calculation results. This overcomes the heavy reliance of traditional methods on data integrity and quality, enhancing the system's practicality in complex real-world operating environments.

[0052] Further explanation should be given regarding the spatiotemporal alignment preprocessing of equipment measurement data, including extracting temporal features through a sliding time window, extracting topological correlation features through spatial neighborhood analysis, and constructing a multi-dimensional feature dataset, which also includes: S24. Classify the raw measurement data by source, mark the data source type, collection time period attribute and quality level, and generate a labeled raw data pool.

[0053] Specifically, in the power distribution network of commercial complexes, the raw measurement data comes from diverse sources: data from the distribution boxes in the catering area comes from the meters of catering merchants, with collection periods covering the lunch and evening peak hours; data from the shopping area comes from the floor master meters, with collection periods during business hours; and data from the parking area comes from the charging pile management system, collected 24 / 7. The system categorizes these data at their source, labeling the source type, collection period attributes, and quality level, generating a tagged raw data pool. For example, catering area data is labeled "Merchant Level - Lunch and Evening Peak - Medium Quality," shopping area data is labeled "Floor Level - Business Hours - High Quality," and parking area data is labeled "Equipment Level - All Hours - High Quality." This categorization allows subsequent analysis to clearly identify data characteristics, avoids misjudgments due to source confusion, and improves the orderliness and traceability of data management.

[0054] S25. Perform cross-category association analysis on the categorized data pool to identify implicit correlations between different categories of data and construct a data association mapping table.

[0055] Specifically, the system performs cross-category correlation analysis on the categorized data pool: it discovers a strong correlation between the load data of the catering area and the air conditioning system data, meaning that when the load of the catering area increases, the current of the air conditioning system increases synchronously, indicating that the heat generated by kitchen equipment drives the cooling demand of the air conditioning. Simultaneously, it identifies a positive correlation between the pedestrian flow data and the lighting load data in the shopping area, with lighting automatically brightening during peak pedestrian hours. Based on these implicit correlations, a data association mapping table is constructed to clarify the relationships such as "catering load - air conditioning current" and "pedestrian flow - lighting load." This analysis reveals the inherent connections between the data, providing a logical basis for consistency comparison and enabling network loss attribution analysis to trace cross-system impacts.

[0056] S26. Based on the association mapping table, perform data consistency comparison. When a data logic conflict is found between categories, automatically trigger the correction process, resample the abnormal category data, and correct the weight of the corresponding data according to the degree of conflict.

[0057] Based on the association mapping table, the system performs data consistency comparison: it finds that the load data of the catering area is high during a certain period, but the current data of the air conditioning system is abnormally low, which conflicts with the strong correlation logic in the mapping table. An automatic correction process is triggered, determining that the air conditioning current data may be distorted due to sensor failure. Resampling is initiated for this abnormal data category, and the current value is reconstructed using the association model between catering load data and historical air conditioning current. Simultaneously, the reconstructed data is weighted and adjusted according to the degree of conflict, reducing its contribution to line loss calculation. This mechanism automatically identifies and corrects logically conflicting data, ensuring the internal consistency of the multi-dimensional feature dataset, preventing abnormal data from contaminating the line loss calculation results, and significantly improving the reliability of subsequent attribution analysis and strategy generation.

[0058] S3. Using a graph-based learning network to perform deep feature cross-fusion on the feature dataset, generating theoretical line loss estimates and equipment loss contribution rankings for each branch, includes the following steps: S31. Convert the equipment nameplate parameters into the equipment rated loss feature vector, and use it as the prior attribute input of the graph structure learning network node to constrain the theoretical line loss calculation result to be no less than the rated no-load loss.

[0059] Specifically, in the power distribution room of the commercial complex, the CIM model established by S1 includes nameplate no-load loss parameters. During the S3 graph structure learning network calculation, these parameters are transformed into a device rated loss feature vector, serving as prior attribute input for the transformer nodes. When calculating the line loss of heavy-load cables in the catering area, if the initial estimate is lower than the transformer no-load loss due to measurement errors, the feature constraint mechanism is automatically triggered, raising the line loss estimate to a level no lower than the no-load loss. This mechanism ensures that the calculation results always conform to the physical characteristics of the equipment, eliminating the logical paradox of "calculated line loss being smaller than no-load loss," and significantly improving the credibility and physical rationality of the line loss estimate.

[0060] S32. The topological electrical distance characteristics are transformed into edge weight attenuation factors, so that the line loss calculation results of long-distance branches are nonlinearly amplified as the distance increases.

[0061] Specifically, in commercial complexes, the power supply cable for the catering area is far from the transformer in the distribution room, resulting in a long electrical distance for this path in the CIM topology model. S3 transforms the electrical distance characteristic into an edge weight attenuation factor, causing the estimated line loss of this cable to amplify non-linearly with increasing power supply distance. During graph network calculations, due to the distance attenuation factor, the estimated line loss of this cable is significantly higher than that of nearby short-distance cables, accurately reflecting the additional losses caused by long-distance power supply. This design avoids errors caused by estimating cables of varying distances using the same standard, making the line loss distribution closer to actual power supply conditions and providing a basis for accurately locating long-distance heavy-load lines.

[0062] S33. Using the ambient temperature characteristics as a dynamic correction coefficient, temperature drift compensation is applied to the theoretical line loss value to eliminate calculation deviations caused by seasonal temperature differences.

[0063] Specifically, in summer, the temperature in the power distribution rooms of commercial complexes is high, increasing the impact of temperature on transformer and cable losses. S3 uses the ambient temperature characteristics associated with the CIM model as a dynamic correction coefficient to compensate for temperature drift in the theoretical line loss values. During calculation, if the system detects that the power distribution room temperature is higher than the baseline value, it automatically adjusts the estimated line loss value upwards to reflect the impact of the high-temperature environment on equipment losses; conversely, it adjusts it downwards accordingly when the temperature decreases in winter. This compensation eliminates calculation biases caused by seasonal temperature differences, ensuring the comparability of line loss estimates across different seasons and supporting accurate analysis and strategy formulation of network loss trends throughout the year.

[0064] In the embodiment of this application, the deep feature cross-fusion of S3 is based on the digital twin topology of the power grid. The device node attributes and the weights of the connecting edges are input into the graph structure learning network. Through multi-layer iterative calculations within the network, the node features are transmitted and interacted along the topological edges, thereby achieving deep fusion calculation of the device's own attributes, operating sequence status and complex topological relationships.

[0065] In one alternative implementation, the deep feature cross-fusion of S3 is achieved by aggregating the feature information of all associated device nodes of the target device node within a multi-level electrical neighborhood in the graph structure learning network, and combining the electrical relationship attributes of the connecting edges to perform comprehensive calculations to achieve feature fusion.

[0066] In another alternative implementation, the deep feature cross-fusion of S3 is to assign different fusion weights to the features of neighboring nodes based on the electrical coupling strength of the connecting edges (such as the apparent power flow magnitude) during the feature propagation process of the graph structure learning network, and then perform weighted aggregation calculation.

[0067] In the embodiment of this application, the dynamic correction coefficient of S33 is a dimensionless multiplier factor obtained by querying or calculating based on the real-time collected ambient temperature value, according to the pre-established corresponding relationship data reflecting the change law of equipment loss with ambient temperature. It is used to scale the preliminary theoretical line loss value by per-unit value.

[0068] In one alternative implementation, the dynamic correction factor of S33 is determined in real time by linear interpolation based on the temperature-loss correction curve specified in the rated parameter manual or technical specifications provided by the equipment manufacturer.

[0069] In another optional implementation, the dynamic correction coefficient of S33 is dynamically generated based on the statistical ratio between the actual measured loss data and the baseline theoretical loss data corresponding to different ambient temperature periods under comparable historical operating conditions of the same equipment, through rolling calculation and smoothing processing.

[0070] The rated loss constraint in S301 of this invention ensures that the calculation results are always within the physically feasible region, eliminating logical contradictions; the electrical distance attenuation factor in S302 enables accurate quantification of long-distance power supply losses, avoiding estimation distortion; and the environmental temperature compensation in S303 eliminates seasonal biases and improves calculation stability. The synergy of these three elements makes line loss estimation closer to actual operating conditions, providing highly reliable input for multi-dimensional attribution in S4, supporting precise implementation of loss reduction strategies, and promoting the transformation of distribution network loss management from experience-based estimation to data-driven refined calculation.

[0071] S4. Input the estimated line loss value into the multidimensional attribution analysis engine, integrate equipment health characteristics, load composition characteristics, and topology characteristics, locate the main cause category, and generate a loss reduction strategy, including the following steps: S41. Construct a health status degradation characteristic curve for equipment, and use the time-series change rate of insulation parameters and dielectric loss as health status characteristics. When the health characteristics show an accelerated degradation trend, increase the attribution weight of the main cause of equipment aging.

[0072] Specifically, in the power distribution room of the commercial complex in this application embodiment, the CIM model established in S1 includes historical insulation resistance and dielectric loss test records of the dry-type transformer. In S4, a health status degradation characteristic curve is constructed, and the annual decreasing trend of insulation resistance and the increasing rate of dielectric loss are input into the attribution analysis engine as health status characteristics. The engine identifies that the transformer's health characteristics show an accelerated degradation trend, determining that its internal insulation material is nearing the end of its lifespan. Therefore, the attribution weight of the main cause of equipment aging is increased to the highest level. Based on this judgment, the generated loss reduction strategy prioritizes arranging for the transformer to be taken out of operation for maintenance and replacement during non-business hours, rather than blindly adjusting the load distribution. This mechanism avoids misjudging line losses caused by aging as load problems, allowing the loss reduction strategy to directly address the root cause, reducing ineffective operations, and improving the accuracy and economy of equipment operation and maintenance.

[0073] S42 extracts load imbalance characteristics, calculates the proportion of zero-sequence and negative-sequence components of the three-phase current, and triggers the reactive power compensation strategy priority execution queue when the imbalance characteristics exceed the preset missing threshold.

[0074] Specifically, the CIM topology model identified a power supply cable in the shopping area of ​​a commercial complex with a long power supply radius and a large number of lighting and display loads connected to its end, resulting in high load density and a spatial matching degree lower than the preset matching threshold. The attribution analysis engine increased the weight of structural network loss as the primary cause, determining that this long-distance power supply was the root cause of the high line loss, and generated a load transfer suggestion: to switch some of the lighting loads at the end to a nearby short-distance cable via a tie switch to achieve load balancing. This suggestion is based on topological bottleneck characteristics rather than human experience, accurately locating the shortcomings of the network structure, avoiding the blind expansion of new lines, and achieving loss reduction through optimized allocation of existing resources.

[0075] S43 identifies topology bottleneck characteristics, calculates the spatial matching degree between power supply radius and load density, and when the matching degree is lower than the preset matching threshold, it increases the weight of the main cause of structural network loss and generates load transfer suggestions.

[0076] Specifically, S4, through the CIM topology model, identified a power supply cable in the shopping area of ​​a commercial complex with a long power supply radius and a large number of lighting and display loads connected to its end, resulting in high load density and a spatial matching degree lower than the preset matching threshold. The attribution analysis engine increased the weight of structural network loss as the primary cause, determining that this long-distance power supply was the root cause of the high line loss, and generated a load transfer suggestion: switch some of the lighting loads at the end to nearby short-distance cables via tie switches to achieve load balancing. This suggestion, based on topological bottleneck characteristics rather than human experience, accurately locates the shortcomings in the network structure, avoids blindly expanding new lines, and achieves loss reduction through optimized allocation of existing resources, resulting in significant investment benefits.

[0077] Furthermore, in this embodiment, the system preset matching threshold is 75% (in this application, the preset matching threshold is a pre-set confidence matching threshold). When the current characteristic reconstructed by S202 is evaluated for quality, if its consistency with historical real data (calculated through cross-validation) is 68%, it is marked as "reconstruction - medium confidence"; if the consistency reaches 82%, it is marked as "reconstruction - high confidence".

[0078] Regarding the preset scheme of this application: the 75% threshold is set using a dynamic grading strategy: When the confidence level is greater than or equal to 85%, it is marked as "high confidence level", the reconstructed feature weight remains at 100%, and it fully participates in the line loss calculation.

[0079] When the confidence level is between 75% and 85%, it is marked as "medium confidence level" and its weight is reduced to 60% to decrease its impact on the calculation results while retaining its reference value.

[0080] When the confidence level is less than 75%, it is marked as "low confidence level", the weight is reduced to 20%, and a sensor maintenance work order is triggered to avoid low-quality data from polluting the calculation results.

[0081] This threshold setting is based on historical data statistics: the average consistency between the reconstructed data and the true value is 78%, and the standard deviation is 8%. Therefore, 75% is used as the dividing line between "usable" and "use with caution," ensuring that line loss calculations maintain engineering-acceptable accuracy even in scenarios with missing data, while suppressing interference from unreliable data. The preset matching threshold can be replaced with other matching thresholds in other schemes, subject to revision according to the scheme's rules.

[0082] In summary, by explicitly integrating equipment health, load composition, and topology characteristics into attribution analysis, we achieve a leap from "symptom manifestation" to "root cause localization." Degradation characteristic curves enable accurate identification of aging equipment, avoiding misjudgments; imbalance characteristics trigger reactive power compensation priority queues, achieving rapid closed-loop loss reduction; and topology bottleneck characteristics guide load transfer, optimizing the network structure. The synergy of these three elements allows loss reduction strategies to directly address the root causes, avoiding the high costs of blindly replacing equipment in the past, and driving the transformation of network loss management from passive response to proactive prevention.

[0083] S5. Convert the loss reduction strategy into control commands and issue them for execution. Collect the measurement data after execution to calculate the loss reduction effect, and feed the effect evaluation results back to the feature extraction stage to correct the feature weight allocation. This includes the following steps: S51. Calculate the rate of change of contribution of each dimension feature before and after loss reduction, and identify the driving features that affect the loss reduction effect.

[0084] Specifically, after implementing loss reduction strategies in the distribution network of commercial complexes, the system calculated the rate of change of contribution of each dimension of characteristics: it was found that the three-phase imbalance characteristic made a significant contribution to the reduction of line losses after the reactive power compensation was switched on, and the equipment health status characteristic made a significant contribution after transformer maintenance; while the rate of change of contribution of cable length characteristics was weak. This calculation identified the high-value characteristics that truly drive the loss reduction effect, pointing the way for subsequent optimization, enabling management resources to focus on key influencing factors, and avoiding wasting calculation and analysis efforts on low-value characteristics.

[0085] S52. If the change rate of feature contribution is lower than the preset effective threshold, the feature corresponding to the current change rate of feature contribution is marked as an inert feature, and the inert feature is optimized in the next round of feature extraction. Specifically, for cable length features whose contribution rate of change is below a preset effective threshold, the system marks them as inert features. In the next round of feature extraction, the sampling frequency of this feature is automatically reduced, and it is temporarily removed from the feature set of the attribution analysis engine. This allows the line loss calculation and attribution process to focus more on high-contribution features such as three-phase imbalance and equipment health status. This mechanism continuously purifies the feature set, improves overall computational efficiency and attribution accuracy, avoids inert features interfering with the determination of the main cause, and achieves automatic optimization and simplification of the feature space.

[0086] S53. Feed back the contribution ranking results of the driving features to the multidimensional attribution analysis engine to optimize the initial weight allocation of each feature channel when locating the main cause category.

[0087] Specifically, the driving characteristics (three-phase imbalance, equipment health status) identified by S501 are sorted by contribution and fed back to the multi-dimensional attribution analysis engine. When locating the main cause category in the next cycle, the engine automatically increases the initial weight of these driving characteristic channels and decreases the weight of inert characteristics that have been eliminated, making the attribution results more accurately focus on high-value characteristics. After continuous iteration of this feedback loop, the attribution engine's sensitivity to line losses caused by three-phase imbalance is significantly improved. It can automatically adapt to changes in load characteristics in different areas of the commercial complex without manual parameter tuning, driving network loss management towards an adaptive and self-optimizing intelligent mode.

[0088] In summary, this application, through the data-driven feature feedback closed loop constructed by S501-S503, achieves a leap from "static feature engineering" to "dynamic feature optimization" in network loss management. The system continuously identifies and focuses on high-contribution driving features, automatically eliminating inert features, ensuring that attribution analysis and loss reduction strategies always revolve around core influencing factors, significantly improving computational efficiency and decision-making accuracy. This mechanism reduces the workload of manual feature tuning, enhances the system's adaptability to different scenarios (such as commercial areas, catering areas, and parking areas), and ultimately promotes the transformation of distribution network loss management towards a feature-driven, continuously evolving intelligent paradigm.

[0089] Example 3 is an embodiment of the present invention. This embodiment provides a multi-dimensional refined management system for power grid losses based on the CIM standard, including a digital twin topology generation module, a multi-dimensional feature dataset construction module, a loss calculation module, a loss reduction strategy generation module, and a loss reduction effect calculation module. The digital twin topology generation module is used to perform semantic modeling of power transmission and distribution equipment based on the power grid information model standard, and generate a power grid digital twin topology with spatiotemporal attributes. The multi-dimensional feature dataset construction module is used to perform spatiotemporal alignment preprocessing on device measurement data, extract temporal features through a sliding time window, extract topological correlation features through spatial neighborhood analysis, and construct a multi-dimensional feature dataset. The loss calculation module is used to perform deep feature cross-fusion on the feature dataset using a graph structure learning network to generate theoretical line loss estimates and equipment loss contribution rankings for each branch. The loss reduction strategy generation module is used to input the line loss estimate into the multidimensional attribution analysis engine, integrate equipment health characteristics, load composition characteristics, and topology characteristics, locate the main cause category, and generate a loss reduction strategy. The loss reduction effect calculation module is used to convert the loss reduction strategy into control commands for execution, collect measurement data after execution to calculate the loss reduction effect, and feed the effect evaluation results back to the feature extraction stage to correct the feature weight allocation.

[0090] This embodiment also provides an electronic device applicable to a CIM-based multi-dimensional network loss refined management method for power grids, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the CIM-based multi-dimensional network loss refined management method for power grids as proposed in the above embodiment.

[0091] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a multi-dimensional refined management method for power grid losses based on the CIM standard as proposed in the above embodiment.

[0092] The storage medium proposed in this embodiment belongs to the same inventive concept as the method for implementing a multi-dimensional refined management of power grid losses based on the CIM standard proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0093] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0094] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for refined management of power grid losses based on CIM standards, characterized in that: The steps are as follows: Semantic modeling of power transmission and distribution equipment is performed based on the power grid information model standard to generate a digital twin topology of the power grid with spatiotemporal attributes. Spatiotemporal alignment preprocessing is performed on equipment measurement data, temporal features are extracted through sliding time windows, topological correlation features are extracted through spatial neighborhood analysis, and a multi-dimensional feature dataset is constructed. A graph-structured learning network is used to perform deep feature cross-fusion on the feature dataset to generate theoretical line loss estimates and equipment loss contribution rankings for each branch. Input the estimated line loss into the multidimensional attribution analysis engine, integrate equipment health characteristics, load composition characteristics, and topology characteristics, locate the main cause category, and generate loss reduction strategies. The loss reduction strategy is transformed into control commands for execution. Measurement data after execution is collected to calculate the loss reduction effect, and the effect evaluation results are fed back to the feature extraction stage to correct the feature weight allocation.

2. The method for refined management of power grid losses based on CIM standards as described in claim 1, characterized in that: The step of semantically modeling power transmission and distribution equipment based on the power grid information model standard to generate a digital twin topology of the power grid with spatiotemporal attributes includes the following steps: The physical parameters and electrical connections of power transmission and distribution equipment are mapped into standardized digital model components based on the power grid information model standard. By integrating the geospatial coordinates of the equipment with the electrical connections of the digital model components, nodes and edges with spatiotemporal attributes are formed. Based on nodes and edges, a dynamic topology graph is constructed that can fully represent the electrical connections and spatial distribution of the power grid.

3. The method for refined management of multi-dimensional network losses in power grids based on CIM standards as described in claim 2, characterized in that: The step of performing spatiotemporal alignment preprocessing on the device measurement data, extracting temporal features through a sliding time window, extracting topological correlation features through spatial neighborhood analysis, and constructing a multi-dimensional feature dataset includes the following steps. Calculate the information gain ratio of data from each measurement channel and identify key characteristic channels that significantly affect line loss. When the data missing rate of the measurement channel is detected to exceed the preset missing threshold, the feature reconstruction algorithm based on spatiotemporal interpolation is automatically activated to fill in the missing feature values ​​by utilizing the temporal correlation of adjacent nodes. The reconstructed features are evaluated for quality, feature confidence labels are generated, and the labels are used as the basis for adjusting feature weights to reduce the contribution of low-quality features in online loss calculation.

4. The method for refined management of power grid losses based on CIM standards as described in claim 3, characterized in that: After the step of using a graph-structured learning network to perform deep feature cross-fusion on the feature dataset to generate theoretical line loss estimates and equipment loss contribution rankings for each branch, the following steps are included. The equipment nameplate parameters are transformed into the equipment rated loss feature vector, which is used as the prior attribute input of the nodes of the graph structure learning network to constrain the theoretical line loss calculation result to be no less than the rated no-load loss. The topological electrical distance characteristics are transformed into edge weight attenuation factors, so that the line loss calculation results of long-distance branches are nonlinearly amplified as the distance increases; Using ambient temperature characteristics as a dynamic correction coefficient, the theoretical line loss value is compensated for temperature drift, thus eliminating calculation deviations caused by seasonal temperature differences.

5. The method for refined management of multi-dimensional network losses in power grids based on CIM standards as described in claim 4, characterized in that: The step of inputting the estimated line loss into the multidimensional attribution analysis engine, integrating equipment health characteristics, load composition characteristics, and topology characteristics, locating the main cause category, and generating a loss reduction strategy includes the following steps: Construct a health status degradation characteristic curve for equipment, and use the time-series change rate of insulation parameters and dielectric loss as health status characteristics. When the health characteristics show an accelerated degradation trend, increase the attribution weight of the main cause of equipment aging. Extract load imbalance characteristics, calculate the proportion of zero-sequence and negative-sequence components of three-phase current, and trigger the reactive power compensation strategy priority execution queue when the imbalance characteristics exceed the preset missing threshold. Identify topology bottleneck characteristics, calculate the spatial matching degree between power supply radius and load density, and when the matching degree is lower than the preset matching threshold, increase the weight of the main cause of structural network loss and generate load transfer suggestions.

6. The method for refined management of multi-dimensional network losses in power grids based on CIM standards as described in claim 5, characterized in that: The steps involved in converting the loss reduction strategy into control commands for execution, collecting measurement data after execution to calculate the loss reduction effect, and feeding the effect evaluation results back to the feature extraction stage for revising feature weight allocation include the following steps. Calculate the rate of change of contribution of each dimension feature before and after loss reduction, and identify the driving features that affect the loss reduction effect; If the change rate of feature contribution is lower than the preset effective threshold, the feature corresponding to the current change rate of feature contribution is marked as an inert feature, and the inert feature is optimized in the next round of feature extraction. The contribution ranking results of the driving features are fed back to the multidimensional attribution analysis engine to optimize the initial weight allocation of each feature channel when locating the main cause category.

7. The method for refined management of multi-dimensional network losses in power grids based on CIM standards as described in claim 6, characterized in that: Following the steps of performing spatiotemporal alignment preprocessing on the device measurement data, extracting temporal features through a sliding time window, extracting topological correlation features through spatial neighborhood analysis, and constructing a multi-dimensional feature dataset, the following steps are included. The raw measurement data is classified at the source, and the data source type, collection time period attribute and quality level are marked to generate a labeled raw data pool; Cross-category association analysis is performed on the categorized data pool to identify the implicit correlations between different categories of data and to construct a data association mapping table; Based on the aforementioned association mapping table, data consistency comparison is performed. When a logical conflict is found between categories, a correction process is automatically triggered to resample the abnormal category data and adjust the weights of the corresponding data according to the degree of conflict.

8. A power grid multi-dimensional network loss refined management system based on the CIM standard, employing the power grid multi-dimensional network loss refined management method based on the CIM standard as described in any one of claims 1 to 7, characterized in that, include: The module includes a digital twin topology generation module, a multi-dimensional feature dataset construction module, a loss calculation module, a loss reduction strategy generation module, and a loss reduction effect calculation module. The digital twin topology generation module is used to perform semantic modeling of power transmission and distribution equipment based on the power grid information model standard, and generate a power grid digital twin topology with spatiotemporal attributes. The multi-dimensional feature dataset construction module is used to perform spatiotemporal alignment preprocessing on device measurement data, extract temporal features through a sliding time window, extract topological correlation features through spatial neighborhood analysis, and construct a multi-dimensional feature dataset. The loss calculation module is used to perform deep feature cross-fusion on the feature dataset using a graph structure learning network to generate theoretical line loss estimates and equipment loss contribution rankings for each branch. The loss reduction strategy generation module is used to input the line loss estimate into the multidimensional attribution analysis engine, integrate equipment health characteristics, load composition characteristics, and topology characteristics, locate the main cause category, and generate a loss reduction strategy. The loss reduction effect calculation module is used to convert the loss reduction strategy into control commands for execution, collect measurement data after execution to calculate the loss reduction effect, and feed the effect evaluation results back to the feature extraction stage to correct the feature weight allocation.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method for refined management of multi-dimensional network losses in power grids based on the CIM standard, as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for refined management of multi-dimensional network losses in power grids based on the CIM standard, as described in any one of claims 1 to 7.

Images