Transformer area topological structure estimation method, system and equipment based on multi-dimensional power utilization characteristics and medium

Abstract

The invention discloses a transformer area topological structure estimation method, system, equipment and medium based on multi-dimensional power utilization characteristics, and belongs to the technical field of power distribution transformer area topologies, and the method comprises the steps: collecting multi-source data, carrying out dynamic weight distribution and abnormal value correction, carrying out the characteristic extraction according to the collected multi-source data, and generating a high-dimensional characteristic vector; clustering user nodes, dividing cluster labels, performing topological modeling according to a cluster division result, generating a topological graph, and performing anomaly verification through multi-dimensional anomaly scoring and abnormal power utilization detection; and performing incremental model parameter correction and multi-objective optimization according to anomaly verification feedback, and generating dynamic topological graph rendering and multi-dimensional decision suggestions by integrating topological modeling, anomaly verification and optimization results. According to the method, accurate estimation of a topological structure is realized through dynamic clustering and hidden node recognition, real-time diagnosis of abnormal nodes and self-correction of a model are realized by means of a multi-dimensional abnormal scoring and self-adaptive optimization mechanism, and the accuracy and operation and maintenance efficiency of transformer area management are improved.

Description

Technical Field

[0001] This invention relates to the field of distribution transformer area topology technology, specifically to a method, system, device, and medium for estimating transformer area topology based on multi-dimensional power consumption characteristics. Background Technology

[0002] With the development of smart grids and the large-scale integration of electric vehicle charging stations, the need for dynamic identification of distribution transformer topology is becoming increasingly urgent. However, traditional distribution transformer topology identification technologies have significant shortcomings in both ordinary and renewable energy integration scenarios. In ordinary distribution transformers, existing methods mainly rely on manual record maintenance or static impedance matrix analysis. The former is inefficient and susceptible to human error, while the latter, due to the lack of integration of multi-dimensional data such as user behavior and environmental characteristics, results in rigid models and insufficient dynamic adaptability, making it difficult to cope with sudden load changes such as peak electricity consumption during holidays.

[0003] Meanwhile, traditional methods have weak ability to identify hidden nodes and lack dynamic correction mechanisms based on physical characteristics such as phase angle differences and sensitivity matrices, further exacerbating topology distortion. After the charging piles are connected, the power consumption characteristics of the transformer area change fundamentally: the random access and intermittent charging behavior of electric vehicles cause highly nonlinear fluctuations in the load. Traditional impedance matrix analysis, which relies on fixed parameter modeling, cannot capture dynamic changes, and the topology identification error increases significantly. The flexible deployment of mobile charging piles leads to frequent migration of user nodes. Clustering algorithms such as K-Means are limited by the preset number of clusters and fixed distance thresholds, making it difficult to adapt to dynamic adjustment requirements. The cluster partitioning results deviate significantly from the actual physical connection relationship.

[0004] Furthermore, the impact of unique abnormal power consumption patterns and environmental interference on the impedance matrix of charging piles has not been effectively modeled, resulting in high false alarm and false negative rates in the anomaly detection mechanism. There is an urgent need to introduce multi-dimensional feature fusion and dynamic compensation technologies to improve the accuracy and robustness of topology recognition. Summary of the Invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, the purpose of this invention is to solve the problems of static and rigid models, difficulty in detecting hidden nodes, and lag in anomaly feedback in existing transformer area topology identification. By multi-source data fusion and dynamic feature extraction, improved clustering and hidden node identification, and closed-loop optimization of anomaly verification, the invention achieves dynamic and accurate estimation of topology structure and real-time intelligent diagnosis of anomalies such as abnormal power consumption.

[0007] To address the aforementioned technical problems, this invention provides the following technical solution: a method for estimating the topology of transformer substations based on multi-dimensional electricity consumption characteristics, comprising, Collect multi-source data, perform dynamic weight allocation and outlier correction, extract features based on the collected multi-source data, and generate high-dimensional feature vectors; User nodes are clustered and labeled, and topology modeling is performed based on the clustering results to generate a topology graph. Anomalies are verified through multi-dimensional anomaly scoring and abnormal power consumption detection. Incremental model parameter correction and multi-objective optimization are performed based on anomaly verification feedback. By integrating topology modeling, anomaly verification and optimization results, dynamic topology graph rendering and multi-dimensional decision suggestions are generated.

[0008] As a preferred embodiment of the transformer area topology estimation method based on multi-dimensional electricity consumption characteristics described in this invention, the multi-source data collection includes collecting original multi-source data and user behavior characteristics, and performing dynamic weight allocation and outlier correction. Time-stamp alignment is performed on the original multi-source data, missing values ​​are filled in, non-numerical data is converted into categorical codes, feature weights are assigned, and statistics are calculated. Identify outliers that exceed the threshold and determine the type of abnormal fault, then perform different preprocessing steps based on the different types of abnormal faults.

[0009] As a preferred embodiment of the transformer area topology estimation method based on multi-dimensional electricity consumption characteristics described in this invention, the generation of high-dimensional feature vectors includes: aligning preprocessed multi-source data according to time steps to form a unified time series matrix, calculating the transformer area topology parameters of each user node, and extracting time series features. Extract the dominant frequency component of the electricity consumption cycle, calculate the cycle intensity, calculate the moving average and moving standard deviation, analyze long-term trend changes, extract electrical features, identify abrupt change points, mark them as potential abnormal events, and retain the original data.

[0010] As a preferred embodiment of the transformer area topology estimation method based on multi-dimensional power consumption characteristics described in this invention, the generation of high-dimensional feature vectors further includes calculating the voltage sensitivity coefficient to current based on preprocessed multi-source data, and generating a dynamic sensitivity matrix. Calculate the difference in voltage phase angle between adjacent nodes, identify the phase consistency of the same branch node, calculate the unbalance of three-phase load data, and extract the three-phase load distribution pattern on the user side; Calculate the temperature sensitivity coefficient and generate seasonal feature vectors by dividing electricity consumption patterns according to the season.

[0011] As a preferred embodiment of the transformer area topology estimation method based on multi-dimensional electricity consumption characteristics described in this invention, the generation of high-dimensional feature vectors further includes: dynamically adjusting feature weights according to user type, training a feature importance model using historical data, and dynamically correcting the sensitivity matrix based on real-time data using an online learning strategy to compensate for model bias caused by load changes. in, This represents the dynamic weight of feature i at time t. Indicates the weighting fusion ratio. Indicates the steepness parameter. This represents the average volatility of feature i within the time window t. Indicates the sensitivity threshold. Indicates the user type coefficient; The time-series features, electrical features, and seasonal features are concatenated into a high-dimensional vector to generate the final high-dimensional feature vector set.

[0012] The beneficial effects of the preferred technical solution in the embodiments of the present invention are as follows: by dynamically weighting and extracting multi-dimensional features, the accuracy and adaptability of electricity consumption feature representation are improved, thereby enhancing the ability to capture user behavior and environmental changes.

[0013] As a preferred embodiment of the transformer area topology estimation method based on multi-dimensional electricity consumption characteristics described in this invention, the topology modeling includes receiving a weighted high-dimensional feature vector and calculating the weighted Euclidean distance between user nodes based on the dynamic weight matrix of the feature engineering module. The clustering parameters are adjusted according to the real-time data distribution. If the current data concentration decreases, the number of clusters is dynamically increased. If the overlap between clusters increases, the distance threshold is decreased. Cluster labels are assigned based on an improved hierarchical clustering algorithm and a weighted distance matrix. The cluster affiliation of each user node is output. Based on the cluster assignment results and cluster labels, the impedance similarity of nodes within a cluster is calculated to identify abnormal nodes within the cluster. By comparing the phase angle differences of different cluster nodes, potential hidden nodes are identified. Based on the sensitivity matrix of the nodes within a cluster, an initial impedance matrix is ​​constructed. The mean phase angle is calculated for each cluster node, and the phase angle difference between clusters is calculated. The existence of hidden nodes is determined by combining the phase angle difference and impedance similarity. in, This is an indicator for identifying hidden nodes. The phase angle difference between node i and node j For reference phase angle value, For impedance similarity between node i and node k, For reference impedance similarity; when If node k is identified as a hidden node, the impedance values ​​of abnormal nodes in the impedance matrix are adjusted based on the hidden node identification results. Combined with environmental characteristics, the impedance deviation caused by load changes is compensated. Based on the corrected impedance matrix, the connection relationship between nodes is determined. The connectivity and stability of the topology graph are optimized through the minimum spanning tree algorithm to generate the final topology graph.

[0014] The beneficial effects of the preferred technical solution in the embodiments of the present invention are as follows: by improving the clustering algorithm and hidden node identification, the accuracy and robustness of topology estimation are improved, and abnormal nodes and hidden nodes can be effectively identified.

[0015] As a preferred embodiment of the transformer area topology estimation method based on multi-dimensional power consumption characteristics described in this invention, the anomaly verification includes receiving real-time collected electrical parameters, constructing the topology matrix of the power network, and initializing the impedance parameters of each node by combining historical impedance baseline values ​​and real-time load data. Starting from the power supply node, the voltage, current and power of each node are calculated layer by layer using Ohm's law and Kirchhoff's equations to generate the impedance prediction value. The predicted values ​​are compared with real-time collected data to generate an error vector. Based on the absolute value of the error vector and the weight of the user type, a multi-dimensional anomaly score is calculated. When any dimension of the error vector exceeds the preset threshold, local parameter correction is triggered: for the impedance matrix sub-block corresponding to the abnormal node, the least squares method is used to correct its impedance parameters, giving priority to preserving physical rationality constraints. When the overall root mean square error exceeds the dynamic threshold N times consecutively, global parameter recalibration is triggered: the overall parameter range of the impedance matrix is ​​dynamically adjusted by combining historical data distribution and environmental changes. Based on error statistics within the sliding window, the scoring threshold is automatically updated to adapt to load changes or environmental noise. The multi-dimensional scoring results are matched with the preset abnormal power consumption feature library to identify high-probability abnormal nodes. Based on the distribution of historical data, abnormal labels and abnormal power consumption suspicion levels are output. Based on the anomaly feedback and the predicted-measured error data of the impedance matrix, the key parameters of the model are first updated in small steps; then, the parameters are individually corrected for the impedance matrix sub-blocks corresponding to the anomaly nodes; next, Pareto optimal solution sets are generated through particle swarm optimization, and the optimal solution is selected from the Pareto front based on the current system state; finally, the impedance matrix parameters are corrected based on the optimal solution. Local corrections are performed on the impedance matrix sub-blocks corresponding to abnormal nodes. By adjusting local parameters and performing multi-objective optimization, the computational resource consumption and model optimization effect are balanced.

[0016] The beneficial effects of the preferred technical solution in the embodiments of the present invention are as follows: by using multi-dimensional anomaly scoring and incremental parameter optimization, the efficiency and reliability of anomaly detection are improved, and adaptive model correction and abnormal power consumption suspicion assessment are realized.

[0017] Another objective of this invention is to provide a transformer substation topology estimation system based on multi-dimensional power consumption characteristics.

[0018] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a transformer area topology estimation system based on multi-dimensional electricity consumption characteristics, comprising: a data acquisition and preprocessing module, a feature engineering module, a cluster analysis module, a topology modeling module, a verification and detection module, a correction and optimization module, and a visualization module; The data acquisition and preprocessing module collects multi-source data on voltage, current, and power, as well as user behavior characteristics, and performs dynamic weight allocation and outlier correction. The feature engineering module extracts time-series, electrical, and environmental features and models dynamic sensitivity based on multi-source data from the data acquisition module, generating high-dimensional feature vectors. The clustering analysis module performs clustering and labeling of user nodes by weighted Euclidean distance and dynamic parameter adjustment; The topology modeling module performs hidden node identification and impedance matrix self-correction based on the clustering results of the clustering module and the impedance sensitivity data of the feature engineering module, and generates a topology map. The verification and detection module performs multi-dimensional anomaly scoring and abnormal power consumption detection by comparing real-time data with the predicted values ​​of the impedance matrix. The correction and optimization module performs incremental model parameter correction and multi-objective optimization based on the abnormal feedback from the verification module and the impedance matrix error. The visualization module integrates topology modeling, anomaly detection, and optimization results to perform dynamic topology graph rendering and generate multi-dimensional decision suggestions.

[0019] 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 method for estimating the topology of transformer substations based on multi-dimensional power consumption characteristics.

[0020] 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 method for estimating the topology of transformer substations based on multi-dimensional power consumption characteristics.

[0021] The beneficial effects of this invention are as follows: This invention, through weighted Euclidean distance and dynamically adjusted parameter clustering, combined with an improved hierarchical clustering algorithm, can flexibly adapt to changes in real-time data distribution, accurately identify potential grouping patterns of user nodes, effectively improve the robustness of cluster partitioning, provide a more reliable node correlation foundation for topology modeling, and achieve accurate location and hierarchical early warning of abnormal nodes through a multi-dimensional anomaly scoring and abnormal power consumption feature database matching mechanism. Dynamic threshold adjustment can effectively cope with seasonal load fluctuations and reduce false alarm rates.

[0022] This invention constructs a topology structure that more closely resembles that of actual power networks by identifying hidden nodes and self-calibrating the impedance matrix, combined with cluster labels, phase angle differences, and sensitivity matrices. It introduces a minimum spanning tree algorithm to optimize connectivity and stability, solving the problem of topology distortion caused by missing hidden nodes or impedance errors. This significantly improves the physical rationality and dynamic adaptability of the topology graph. Real-time updating and global optimization of model parameters are achieved through incremental parameter correction and multi-objective optimization. The local correction strategy for abnormal nodes avoids the overhead of retraining the entire model. At the same time, it balances error correction, stability, and computational efficiency, ensuring the system's continuous optimization capability under dynamic load changes and improving overall operational performance. Attached Figure Description

[0023] 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.

[0024] Figure 1 The operation flow of the transformer area topology estimation method based on multi-dimensional electricity consumption characteristics provided in one embodiment of the present invention Figure 1 .

[0025] Figure 2 The operation flow of the transformer area topology estimation method based on multi-dimensional electricity consumption characteristics provided in one embodiment of the present invention Figure 2 .

[0026] Figure 3 This is a schematic diagram of a transformer area topology estimation system based on multi-dimensional power consumption characteristics, provided as an embodiment of the present invention. Detailed Implementation

[0027] 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.

[0028] Example 1, referring to Figures 1-2 This is one embodiment of the present invention, which provides a method for estimating the topology of transformer substations based on multi-dimensional electricity consumption characteristics, including: S100: Collect multi-source data, perform dynamic weight allocation and outlier correction, extract features based on the collected multi-source data, and generate high-dimensional feature vectors; S200: Cluster user nodes and divide them into cluster labels. Based on the cluster division results, perform topology modeling and generate a topology graph. Verify anomalies through multi-dimensional anomaly scoring and abnormal power consumption detection. S300: Based on the feedback from anomaly verification, perform incremental model parameter correction and multi-objective optimization. By integrating the results of topology modeling, anomaly verification, and optimization, generate dynamic topology graph rendering and multi-dimensional decision suggestions. It should be noted that existing technologies for transformer area topology estimation and anomaly detection typically have the following drawbacks: reliance on a single data source or static model, resulting in incomplete feature representation and poor adaptability to load changes and user behavior; and insufficient ability of topology modeling methods to identify hidden nodes and abnormal connections, making them susceptible to noise interference. Therefore, to address the aforementioned problems, steps S100-S300 are used to achieve dynamic and accurate estimation of transformer topology and intelligent anomaly diagnosis based on multi-dimensional electricity consumption characteristics. Specifically, multi-source data fusion and dynamic feature engineering are used to solve the problems of insufficient data utilization and model rigidity. Improved clustering and hidden node identification algorithms are used to enhance the accuracy and robustness of topology modeling. By introducing anomaly verification feedback closed loop and multi-objective optimization mechanisms, online self-correction of model parameters and real-time decision support are achieved.

[0029] Example 2, refer to Figures 1-2 This is one embodiment of the present invention, which provides a method for estimating the topology of transformer substations based on multi-dimensional electricity consumption characteristics, including: In this embodiment of the invention, step S100 involves collecting multi-source data, performing dynamic weight allocation and outlier correction, and extracting features based on the collected multi-source data to generate a high-dimensional feature vector, including the following steps S101-S103: S101. Collect multi-source data on voltage, current, and power, as well as user behavior characteristics, and perform dynamic weight allocation and outlier correction. By using smart meters, sensors, and user behavior databases, voltage, current, power, phase angle electrical quantity data, as well as user type, electricity consumption time series, weather conditions, behavior, and environmental characteristics are collected synchronously. The raw data is time-stamped to eliminate misalignment problems caused by inconsistent sampling frequencies. In an embodiment of the present invention, missing value completion includes the following steps A1-A2: A1. Remove missing values; A2. Complete the function using linear interpolation or forward padding; In an optional embodiment, missing value completion in S101 can be done using a moving average method. For each missing value, the moving average of its adjacent data points before and after it is calculated, and the missing value is filled with this average. The position of the missing value is determined, and the arithmetic mean of the two data points before and after the missing value is calculated. If there are not enough data points in the window, the window size is adjusted or the available points are used for calculation.

[0030] In another optional embodiment, missing value completion in S101 can also be done by directly filling in neighboring points. For each missing value, its nearest non-missing value is used directly for filling, with priority given to the value of the previous time point (forward filling). If the previous value is not available, the value of the next time point is used (backward filling), ensuring that the filling value comes from a continuous time series of the same user or sensor.

[0031] In an embodiment of the present invention, converting non-numerical data into categorical codes for dynamic weight allocation includes the following step B1: B1. Convert non-numerical data into category codes and assign feature weights according to user type and electricity consumption pattern; In an optional embodiment, the dynamic weight allocation in S101 can be based on the dynamic weight allocation of feature variability. The coefficient of variation (the ratio of standard deviation to mean) of each numerical feature is calculated as a measure of feature volatility. For non-numerical features, they are first converted into class codes, and the dispersion of their class frequencies is calculated as an indicator of similar variability. The weights are adjusted according to user type and electricity consumption pattern: for example, for industrial users, the variability weight of power features is higher; for residential users, the variability weight of time series features is higher. Finally, the weights are normalized to ensure that the sum is 1.

[0032] In another optional embodiment, the dynamic weight allocation in S101 can also be a dynamic weight allocation based on feature contribution. A linear regression model is trained using historical data to predict electricity consumption reference indicators, and the absolute value of the coefficient of each feature is obtained as the initial contribution. For non-numerical features, they are first converted into category codes and then included in the model to calculate the contribution. The weights are fine-tuned according to user type and electricity consumption pattern: for example, for commercial users, the contribution weight of time features is higher. The final weights are allocated in combination with the contribution and are dynamically updated.

[0033] A sliding window is used to calculate statistics, and the IQR method is used to identify outliers exceeding the threshold. The anomalies are then combined with data from adjacent nodes to determine whether they are due to sensor malfunctions. For sensor noise: replace outliers with the median of the nearest values ​​within the window; For mutations caused by abnormal power consumption or power outages: mark them as events to be verified and retain the original data.

[0034] S102. Based on the multi-source data from the data acquisition module, extract time-series, electrical, and environmental features and perform dynamic sensitivity modeling to generate high-dimensional feature vectors; Extracting temporal features: Receive standardized data output from the data acquisition and preprocessing module, perform time-scale alignment, missing value completion, and outlier marking, align multi-source data according to time steps, and form a unified time series matrix; Extracting electrical features: Calculate the mean, variance, peak, valley and volatility of electricity consumption and power for each user node. Extract the main frequency components of daily and weekly electricity consumption cycles through Fourier transform, calculate the cycle intensity, and calculate the moving average and moving standard deviation through a sliding window to capture long-term trend changes.

[0035] Extracting environmental features: Based on time-series data of voltage and current, calculate the voltage sensitivity coefficient to current and generate a dynamic sensitivity matrix; Calculate the difference in voltage phase angle between adjacent nodes, identify the phase consistency of the same branch node, calculate the unbalance of three-phase load data, and extract the three-phase load distribution pattern on the user side; By combining weather data with user electricity consumption, the temperature sensitivity coefficient is calculated through linear regression or mutual information, and electricity consumption patterns are divided according to the season to generate a seasonal feature vector.

[0036] S103. Dynamically adjust feature weights based on user type, train a feature importance model using historical data, and dynamically correct the sensitivity matrix based on real-time data using an online learning strategy to compensate for model bias caused by load changes. The formula is as follows: in, This represents the dynamic weight of feature i at time t. Indicates the weighting fusion ratio. Indicates the steepness parameter. This represents the average volatility of feature i within the time window t. Indicates the sensitivity threshold. Indicates the user type coefficient; The temporal features, electrical features, and environmental features are concatenated into a high-dimensional vector to generate the final high-dimensional feature vector set.

[0037] In this embodiment of the invention, S200 involves clustering user nodes and assigning cluster labels, performing topology modeling based on the cluster division results, generating a topology graph, and verifying anomalies through multi-dimensional anomaly scoring and abnormal power consumption detection, including the following steps S201-S203: S201. User nodes are clustered and labeled using weighted Euclidean distance and dynamic parameter adjustment. Receive the weighted high-dimensional feature vector output by the feature engineering module, which includes time, electrical, and environmental dimensions. Based on the dynamic weight matrix of the feature engineering module, calculate the weighted Euclidean distance between user nodes. The clustering parameters are adjusted based on the real-time data distribution; if the centrality of the current data decreases, the number of clusters is dynamically increased. If the overlap between clusters increases, the distance threshold is reduced to enhance separation. Cluster labels are assigned based on the improved hierarchical clustering algorithm and weighted distance matrix, and the cluster affiliation result of each user node is output.

[0038] S202. Based on the clustering results of the clustering module and the sensitivity coefficient of the feature engineering module, perform hidden node identification and impedance matrix self-correction to generate a topology map. Receive the cluster partitioning results output by the clustering analysis module and the dynamic sensitivity matrix provided by the feature engineering module. Based on the cluster labels, calculate the impedance similarity of nodes within the cluster and identify abnormal nodes within the cluster. By comparing the phase angle differences between different cluster nodes, potential hidden nodes can be identified. Based on the sensitivity matrix of the nodes within the cluster, an initial impedance matrix is ​​constructed, the mean phase angle is calculated for each cluster node, and the phase angle difference between clusters is calculated. The presence of hidden nodes is determined by combining phase angle differences and impedance similarities. in, This is an indicator for identifying hidden nodes. The phase angle difference between node i and node j For reference phase angle value, For impedance similarity between node i and node k, For reference impedance similarity; when If node k is identified as a hidden node, the impedance values ​​of abnormal nodes in the impedance matrix are adjusted based on the hidden node identification results. By combining environmental characteristics, the impedance deviation caused by load changes is compensated. Based on the adjusted impedance matrix, the connection relationship between nodes is determined. The connectivity and stability of the topology graph are optimized by the minimum spanning tree algorithm to generate the final topology graph.

[0039] S202. By comparing real-time data with the predicted values ​​of the impedance matrix, multi-dimensional anomaly scoring and abnormal power consumption detection are performed. It receives real-time collected electrical parameters such as voltage, current, and power, as well as the impedance matrix prediction value generated by the topology modeling module. Based on the node connection relationship provided by the topology modeling module, it constructs the topology matrix of the power network. By combining historical impedance baseline values ​​and real-time load data, the impedance parameters of each node are initialized. Starting from the power supply node, the voltage, current and power of each node are calculated layer by layer using Ohm's law and Kirchhoff's equations to generate the impedance prediction value. The predicted values ​​are compared with real-time collected data to generate an error vector. Based on the absolute value of the error vector and the weight of the user type, a multi-dimensional anomaly score is calculated. When any dimension of the error vector exceeds the preset threshold, local parameter correction is triggered: for the impedance matrix sub-block corresponding to the abnormal node, the least squares method is used to correct its impedance parameters, giving priority to preserving physical rationality constraints. When the overall root mean square error exceeds the dynamic threshold N times consecutively, global parameter recalibration is triggered: the overall parameter range of the impedance matrix is ​​dynamically adjusted by combining historical data distribution and environmental changes. Every 24 hours, based on error statistics within a sliding window, the scoring threshold is automatically updated to adapt to load changes or environmental noise. The multi-dimensional scoring results are matched with a preset abnormal power consumption feature library to identify high-probability abnormal nodes, and abnormal labels and abnormal power consumption suspicion levels are output according to historical data distribution.

[0040] S203. Based on the abnormal feedback from the verification module and the impedance matrix error, perform incremental model parameter correction and multi-objective optimization. The system receives anomaly feedback and predicted-measured error data of the impedance matrix from the verification and detection module. First, it updates the key parameters of the model in small steps. Then, it corrects the parameters of the impedance matrix sub-blocks corresponding to the anomaly nodes individually. Next, it generates a Pareto optimal solution set through particle swarm optimization and selects the optimal solution from the Pareto front based on the current system state. Finally, it corrects the impedance matrix parameters based on the optimal solution. Local corrections are performed on the impedance matrix sub-blocks corresponding to abnormal nodes. By adjusting local parameters and performing multi-objective optimization, the computational resource consumption and model optimization effect are balanced.

[0041] In an embodiment of the present invention, S300 involves incremental model parameter correction and multi-objective optimization based on anomaly verification feedback. By integrating topology modeling, anomaly verification, and optimization results, dynamic topology graph rendering and multi-dimensional decision suggestions are generated, including the following steps S301: S301. By integrating topology modeling, anomaly detection, and optimization results, dynamic topology graph rendering and multi-dimensional decision suggestions are generated. In an embodiment of the present invention, generating a dynamic topology graph rendering includes the following steps C1-C2: C1. Receive the real-time network structure from the topology modeling module, the abnormal node labels and scores from the verification and detection module, and the parameter adjustment records from the correction and optimization module, and display the power network nodes and connection relationships in the form of a dynamic diagram. C2. Real-time update of node status, flashing, magnification or color marking of abnormal nodes, and support for clicking to view detailed scores, error values ​​and suspected abnormal power consumption types, and overlay of layers such as impedance error distribution, power fluctuation trend, and the impact range of optimization strategies; In an optional embodiment, the dynamic topology rendering in S301 can be static topology rendering based on periodically batch updates. A static power network topology diagram is drawn based on the node connection relationships output by the topology modeling module. Nodes and connections are displayed in a fixed layout, without real-time animation effects. The latest topology structure and anomaly detection results are retrieved from the data source at fixed time intervals, and the node status and connection relationships in the topology diagram are updated in batches. During the update process, the entire topology diagram is refreshed, rather than continuously changing in real time. During each update, abnormal nodes are marked with predefined colors or icons based on the scoring results of the anomaly verification module, but dynamic flashing or zooming effects are not supported. Users can view basic attributes by clicking on nodes, but detailed information must be obtained by generating a separate report or jumping to another interface; embedded dynamic rendering is not provided. Only a core data layer is provided, displayed as a static overlay, and real-time switching or multi-layer dynamic overlay is not supported. The layer content is refreshed with periodic updates.

[0042] In this embodiment, in the data acquisition and preprocessing module, the system aligns multi-source data through a 5-minute sliding window, significantly improving data synchronization by reducing time misalignment. For missing value completion, linear interpolation and moving average methods are used to reduce the standard deviation of the completed data, ensuring data integrity. The dynamic weight allocation strategy adjusts the weights of power, time series, and environmental features according to user type, improving anomaly detection accuracy in industrial and residential scenarios. The feature engineering module extracts time-series features through a 24-hour / 7-day sliding window and combines it with moving average methods to capture long-term trends, improving trend recognition accuracy. The phase angle difference threshold is set to 7°, significantly improving the accuracy of hidden node identification. Temperature sensitivity modeling reduces topology errors in areas with seasonal load fluctuations, verifying the role of environmental features in improving model performance.

[0043] In this embodiment, the clustering analysis module improves clustering stability in scenarios with centralized charging pile access by employing an improved hierarchical clustering algorithm, initial cluster number, and dynamic parameter adjustment strategies. It also incorporates hidden node identification based on phase angle differences and impedance similarity. The impedance matrix self-correction correction coefficient α=0.3 significantly enhances physical rationality. A preset dynamic threshold for anomaly scoring reduces the false alarm rate. In actual deployment, the system dynamically adjusts the parameter range by real-time monitoring of load changes and environmental interference, ensuring the model's robustness in complex scenarios.

[0044] In another optional embodiment, the dynamic topology rendering in S301 can also be a lightweight topology rendering based on simplified interaction, displaying only key nodes (such as power points and abnormal nodes) and main connections, ignoring secondary nodes to reduce visualization complexity. The topology is generated using a fixed template, and when the states of multiple nodes change, batch updates are performed centrally (such as every 15 minutes or when a certain number of abnormal events accumulate). During the update, the entire topology is re-rendered, without providing real-time node-by-node updates. Basic color coding (green, yellow, red) is used to represent the abnormality level, without dynamic effects. Abnormal nodes are only distinguished by color, without additional markings. Users can view basic node information by hovering the mouse, but click interaction is not supported. Detailed data needs to be obtained by exporting logs or querying separately. Only a single layer is provided, statically overlaid on the topology. The layer content cannot be customized or adjusted in real time, and only changes with batch updates.

[0045] In an embodiment of the present invention, generating multidimensional decision recommendations includes the following steps D1-D2: D1. Based on the score and the scope of impact, generate a list of abnormal nodes that need to be prioritized and mark the recommended measures. Transform the parameter adjustment results of the correction and optimization module into readable suggestions. D2. Combining historical trends and current status, simulate the potential impact of different optimization schemes, generate interactive visualization reports, and send user operation feedback back to the verification, detection, and correction optimization modules to form a closed-loop optimization.

[0046] In an optional embodiment, the generation of multi-dimensional decision recommendations in S301 can be based on priority ranking and a static rule base. A set of fixed rules is pre-defined based on historical anomaly cases and expert experience to associate anomaly scores, impact range, and user types, directly mapping them to recommended measures. Based on the anomaly scores and impact range output by the verification and detection module, anomaly nodes are automatically sorted according to the rule base to generate a static priority list, with standardized recommended measures attached. The priority list and measures are converted into a text report, which does not support simulated optimization of impact, only providing simple visualization of historical data comparison. User operation feedback is only used for recording and does not update the rule base in real time; rules are only manually adjusted periodically. However, the static rule base cannot dynamically adapt to network changes and new anomaly patterns, leading to outdated or inaccurate recommendations.

[0047] In another optional embodiment, the multidimensional decision recommendations generated in S301 can also be decision recommendations based on clustering and template matching. Using the output of the clustering analysis module, abnormal nodes are clustered into groups according to scores and electrical characteristics. Predefined template measures are assigned to each cluster. Based on the clustering results, the templates are directly matched to generate a list of abnormal nodes and measures. No complex simulations are performed. Only a simple impact range is calculated based on the current cluster centers. A static chart based on the clustering results is generated to display the abnormal distribution. Interactive simulation is not supported. Only basic historical trend overlays are provided. User feedback is only used to adjust clustering parameters and does not involve global model optimization. The feedback delay is relatively high. However, template matching depends on the clustering quality. When the abnormal patterns are complex or overlapping, the recommendation effect decreases.

[0048] Example 3, referring to Figure 3 This is an embodiment of the present invention, and the above is an illustrative scheme of a transformer substation topology estimation method based on multi-dimensional electricity consumption characteristics. It should be noted that the technical solution of the transformer substation topology estimation system based on multi-dimensional electricity consumption characteristics and the technical solution of the above-described transformer substation topology estimation method based on multi-dimensional electricity consumption characteristics belong to the same concept. Details not described in detail in the technical solution of the transformer substation topology estimation system based on multi-dimensional electricity consumption characteristics in this embodiment can be found in the description of the technical solution of the above-described transformer substation topology estimation method based on multi-dimensional electricity consumption characteristics.

[0049] This embodiment provides a transformer area topology estimation system based on multi-dimensional electricity consumption characteristics, including: a data acquisition and preprocessing module, a feature engineering module, a cluster analysis module, a topology modeling module, a verification and detection module, a correction and optimization module, and a visualization module; The data acquisition and preprocessing module collects multi-source data on voltage, current, and power, as well as user behavior characteristics, and performs dynamic weight allocation and outlier correction. The feature engineering module extracts time-series, electrical, and environmental features and models dynamic sensitivity based on multi-source data from the data acquisition module, generating high-dimensional feature vectors. The clustering analysis module performs clustering and labeling of user nodes by weighted Euclidean distance and dynamic parameter adjustment; The topology modeling module performs hidden node identification and impedance matrix self-correction based on the clustering results of the clustering module and the impedance sensitivity data of the feature engineering module, and generates a topology map. The verification and detection module performs multi-dimensional anomaly scoring and abnormal power consumption detection by comparing real-time data with the predicted values ​​of the impedance matrix. The correction and optimization module performs incremental model parameter correction and multi-objective optimization based on the abnormal feedback from the verification module and the impedance matrix error. The visualization module integrates topology modeling, anomaly detection, and optimization results to perform dynamic topology graph rendering and generate multi-dimensional decision suggestions.

[0050] This embodiment also provides an electronic device applicable to the method for estimating the topology of transformer substations based on multi-dimensional power consumption characteristics, 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 implement the method for estimating the topology of transformer substations based on multi-dimensional power consumption characteristics as proposed in the above embodiment.

[0051] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the transformer area topology estimation method based on multi-dimensional power consumption characteristics proposed in the above embodiments.

[0052] The storage medium proposed in this embodiment and the method for estimating the topology of transformer substations based on multi-dimensional power consumption characteristics proposed in the above embodiments belong to the same inventive concept. 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.

[0053] 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.

[0054] 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 estimating transformer substation topology based on multi-dimensional electricity consumption characteristics, characterized in that: include, Collect multi-source data, perform dynamic weight allocation and outlier correction, extract features based on the collected multi-source data, and generate high-dimensional feature vectors; User nodes are clustered and labeled, and topology modeling is performed based on the clustering results to generate a topology graph. Anomalies are verified through multi-dimensional anomaly scoring and abnormal power consumption detection. Incremental model parameter correction and multi-objective optimization are performed based on anomaly verification feedback. By integrating topology modeling, anomaly verification and optimization results, dynamic topology graph rendering and multi-dimensional decision suggestions are generated.

2. The method for estimating transformer substation topology based on multi-dimensional electricity consumption characteristics as described in claim 1, characterized in that: The collection of multi-source data includes collecting raw multi-source data and user behavior characteristics, and performing dynamic weight allocation and outlier correction. Time-stamp alignment is performed on the original multi-source data, missing values ​​are filled in, non-numerical data is converted into categorical codes, feature weights are assigned, and statistics are calculated. Identify outliers exceeding the threshold to determine the type of abnormal fault, and perform different preprocessing based on the different types of abnormal faults.

3. The method for estimating transformer substation topology based on multi-dimensional electricity consumption characteristics as described in claim 2, characterized in that: The process of generating high-dimensional feature vectors includes aligning preprocessed multi-source data according to time steps to form a unified time series matrix, calculating the transformer topology parameters of each user node, and extracting time series features. Extract the dominant frequency component of the electricity consumption cycle, calculate the cycle intensity, calculate the moving average and moving standard deviation, analyze long-term trend changes, extract electrical features, identify abrupt change points, mark them as potential abnormal events, and retain the original data.

4. The method for estimating transformer substation topology based on multi-dimensional electricity consumption characteristics as described in claim 3, characterized in that: The generation of high-dimensional feature vectors also includes calculating the voltage sensitivity coefficient to current based on preprocessed multi-source data, and generating a dynamic sensitivity matrix. Calculate the difference in voltage phase angle between adjacent nodes, identify the phase consistency of the same branch node, calculate the unbalance of three-phase load data, and extract the three-phase load distribution pattern on the user side; Calculate the temperature sensitivity coefficient and generate seasonal feature vectors by dividing electricity consumption patterns according to the season.

5. The method for estimating transformer substation topology based on multi-dimensional electricity consumption characteristics as described in claim 4, characterized in that: The generation of high-dimensional feature vectors also includes dynamically adjusting feature weights according to user type, training a feature importance model using historical data, and dynamically correcting the sensitivity matrix based on real-time data using an online learning strategy to compensate for model bias caused by load changes. in, This represents the dynamic weight of feature i at time t. Indicates the weighting fusion ratio. Indicates the steepness parameter. This represents the average volatility of feature i within the time window t. Indicates the sensitivity threshold. Indicates the user type coefficient; The time-series features, electrical features, and seasonal features are concatenated into a high-dimensional vector to generate the final high-dimensional feature vector set.

6. The method for estimating transformer substation topology based on multi-dimensional electricity consumption characteristics as described in claim 5, characterized in that: The topology modeling includes receiving a weighted high-dimensional feature vector, calculating the weighted Euclidean distance between user nodes based on the dynamic weight matrix of the feature engineering module, and implementing it as follows: In the formula, Indicates the weighted Euclidean distance; The clustering parameters are adjusted according to the real-time data distribution. If the current data concentration decreases, the number of clusters is dynamically increased. If the overlap between clusters increases, the distance threshold is decreased. Based on the improved hierarchical clustering algorithm, cluster labels are divided according to the weighted distance matrix, and the cluster affiliation result of each user node is output. Based on the cluster division result and the cluster labels, the impedance similarity of nodes within the cluster is calculated to identify abnormal nodes within the cluster. By comparing the phase angle differences of different cluster nodes, possible hidden nodes are identified. Based on the sensitivity matrix of the nodes within a cluster, an initial impedance matrix is ​​constructed. The mean phase angle is calculated for each cluster node, and the phase angle difference between clusters is also calculated. The existence of hidden nodes is determined by combining the phase angle difference and impedance similarity. in, This is an indicator for identifying hidden nodes. The phase angle difference between node i and node j For reference phase angle value, For impedance similarity between node i and node k, For reference impedance similarity; when If node k is identified as a hidden node, the impedance values ​​of abnormal nodes in the impedance matrix are adjusted based on the hidden node identification results. Combined with environmental characteristics, the impedance deviation caused by load changes is compensated. Based on the corrected impedance matrix, the connection relationship between nodes is determined. The connectivity and stability of the topology graph are optimized through the minimum spanning tree algorithm to generate the final topology graph structure.

7. The method for estimating transformer substation topology based on multi-dimensional electricity consumption characteristics as described in claim 6, characterized in that: The anomaly verification includes receiving real-time collected electrical parameters, constructing the topology matrix of the power network, and initializing the impedance parameters of each node by combining historical impedance baseline values ​​and real-time load data. Starting from the power supply node, the voltage, current and power of each node are calculated layer by layer using Ohm's law and Kirchhoff's equations to generate the impedance prediction value. The predicted values ​​are compared with real-time collected data to generate an error vector. Based on the absolute value of the error vector and the weight of the user type, a multi-dimensional anomaly score is calculated. When any dimension of the error vector exceeds the preset threshold, local parameter correction is triggered: for the impedance matrix sub-block corresponding to the abnormal node, the least squares method is used to correct its impedance parameters, giving priority to preserving physical rationality constraints. When the overall root mean square error exceeds the dynamic threshold N times consecutively, global parameter recalibration is triggered: the overall parameter range of the impedance matrix is ​​dynamically adjusted by combining historical data distribution and environmental changes. Based on error statistics within the sliding window, the scoring threshold is automatically updated to adapt to load changes or environmental noise. The multi-dimensional scoring results are matched with the preset abnormal power consumption feature library to identify high-probability abnormal nodes. Based on the distribution of historical data, abnormal labels and abnormal power consumption suspicion levels are output. Based on the anomaly feedback and the predicted-measured error data of the impedance matrix, the key parameters of the model are updated in small steps according to the error data. For the impedance matrix sub-blocks corresponding to the anomaly nodes, the parameters are corrected separately. Pareto optimal solution set is generated through particle swarm optimization. Based on the current system state, the optimal solution is selected from the Pareto frontier. The impedance matrix parameters are corrected according to the optimal solution. Local corrections are performed on the impedance matrix sub-blocks corresponding to abnormal nodes. By adjusting local parameters and performing multi-objective optimization, the computational resource consumption and model optimization effect are balanced.

8. A transformer substation topology estimation system based on multi-dimensional electricity consumption characteristics, employing the transformer substation topology estimation method based on multi-dimensional electricity consumption characteristics as described in any one of claims 1 to 7, characterized in that, include: The system includes a data acquisition and preprocessing module, a feature engineering module, a cluster analysis module, a topology modeling module, a verification and testing module, a correction and optimization module, and a visualization module. The data acquisition and preprocessing module collects multi-source data on voltage, current, and power, as well as user behavior characteristics, and performs dynamic weight allocation and outlier correction. The feature engineering module extracts time-series, electrical, and environmental features and models dynamic sensitivity based on multi-source data from the data acquisition module, generating high-dimensional feature vectors. The clustering analysis module performs clustering and labeling of user nodes by weighted Euclidean distance and dynamic parameter adjustment; The topology modeling module performs hidden node identification and impedance matrix self-correction based on the clustering results of the clustering module and the impedance sensitivity data of the feature engineering module, and generates a topology map. The verification and detection module performs multi-dimensional anomaly scoring and abnormal power consumption detection by comparing real-time data with the predicted values ​​of the impedance matrix. The correction and optimization module performs incremental model parameter correction and multi-objective optimization based on the abnormal feedback from the verification module and the impedance matrix error. The visualization module integrates topology modeling, anomaly detection, and optimization results to perform dynamic topology graph rendering and generate multi-dimensional decision suggestions.

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 transformer area topology estimation method based on multi-dimensional electricity consumption characteristics 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 transformer area topology estimation method based on multi-dimensional electricity consumption characteristics as described in any one of claims 1 to 7.

Images