Transformer area intelligent clustering method and system considering distributed resource operation characteristics

Abstract

The invention relates to the technical field of power grid planning, in particular to an intelligent transformer area clustering method considering operation characteristics of distributed resources, and the method comprises the steps: collecting operation data of multiple types of transformer areas, and building an operation characteristic parameter set of the multiple types of distributed resources; generating a characteristic fingerprint database through principal component analysis dimension reduction and clustering processing, and calculating characteristic similarity between different resource characteristics; constructing classification aggregation models of various types of resources, and fusing the classification aggregation models into a transformer area comprehensive load model; according to the topological structure of the power distribution network and the comprehensive load model, calculating an inter-station composite electrical distance index, and extracting load characteristic similarity; and performing two-dimensional clustering by taking minimization of the composite electrical distance and maximization of the load characteristic similarity as double targets, and outputting a transformer area aggregation scheme. According to the method, the problem of collaborative modeling of dynamic characteristic differences of multiple types of resources and electrical coupling strength between the transformer areas is effectively solved, and the transformer area aggregation precision and the engineering adaptability in a distributed energy scene are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of power grid planning technology, and in particular to a method and system for intelligent clustering of distribution transformer areas that takes into account the operating characteristics of distributed resources. Background Technology

[0002] With the deepening of the energy transition strategy and the rapid development of smart grid technology, distributed energy resources connected to the distribution network are showing a trend of diversification, large scale, and high penetration. The widespread integration of various types of distributed resources, such as electric vehicles, central air conditioning, distributed photovoltaics, and energy storage systems, while bringing flexible adjustment capabilities to the distribution network, also poses unprecedented challenges to traditional power grid operation and dispatching models. As the basic unit of the distribution network, the research on intelligent aggregation methods for distribution transformer areas is of great strategic significance for achieving coordinated control of distributed resources, improving system operating efficiency, and ensuring power supply reliability.

[0003] Current research on distribution network aggregation mainly focuses on building aggregation models for single-type distributed resources, lacking a comprehensive consideration of the heterogeneous characteristics of multiple types of distributed resources. Traditional aggregation methods often use simplified models, ignoring key operational characteristics such as the stochastic aggregation behavior of electric vehicle charging and discharging, the nonlinearity of central air conditioning temperature control, and the intermittency of distributed photovoltaic power generation, resulting in insufficient accuracy of aggregation models. Furthermore, existing research often focuses on equivalent modeling of resources within a distribution area, while research on collaborative aggregation between multiple distribution areas is insufficient, particularly failing to fully consider the combined effects of electrical distance and load characteristic similarity between distribution areas, making it difficult for aggregation results to accurately reflect the actual operating state of the power grid. Summary of the Invention

[0004] This invention provides a method and system for intelligent clustering of transformer substations that takes into account the characteristics of distributed resource operation, thereby effectively solving the problems pointed out in the background art.

[0005] To achieve the above objectives, this patent realizes precise and efficient management of distributed energy clusters through a two-dimensional collaborative aggregation mechanism: it constructs a multi-type resource dynamic characteristic fingerprint database to quantify the operating rules of heterogeneous devices, combines a composite electrical distance model that integrates topological connections, voltage sensitivity, and equivalent impedance to explicitly characterize the electrical coupling strength between stations, and then uses an improved clustering algorithm to achieve dual-objective collaborative optimization of characteristic similarity and electrical correlation. This fundamentally solves the aggregation distortion problem caused by traditional methods ignoring the differences in dynamic characteristics of resources and the influence of electrical interactions, and significantly improves the response accuracy and operational stability of cluster control in high-penetration distributed energy scenarios.

[0006] Specifically, the technical solution adopted in this invention is as follows:

[0007] A method for intelligent clustering of transformer substations considering the characteristics of distributed resource operation includes:

[0008] S1: Construct a dataset of runtime characteristics for multiple types of distributed resources, and calculate the similarity of characteristics between different resources, including:

[0009] Collect operational data from various types of distribution transformer areas and establish a preliminary classification parameter set for various types of distributed resources;

[0010] The preliminary classification parameter set is standardized to form a complete characteristic parameter set;

[0011] Based on the set of characteristic parameters, a characteristic fingerprint database is generated by principal component analysis for dimensionality reduction and clustering, and the characteristic similarity between different resource characteristics is calculated.

[0012] S2: Establish a comprehensive load model for the distribution network area, including:

[0013] Based on the aforementioned characteristic fingerprint database, the resource structure and topological relationship of the transformer area are identified, a classification and aggregation model of various types of resources is constructed, and a comprehensive load model of the transformer area is fused through coupling effect compensation.

[0014] S3: Intelligent aggregation of multiple transformer substations based on electrical distance and similarity, including:

[0015] Based on the distribution network topology and the comprehensive load model, the composite electrical distance index between stations is calculated by integrating topological distance, voltage sensitivity and equivalent impedance, and the load characteristic similarity is extracted based on the characteristic fingerprint database.

[0016] With the dual objectives of minimizing composite electrical distance and maximizing load characteristic similarity, the composite electrical distance index and load characteristic similarity are input into an improved K-means clustering algorithm for two-dimensional clustering, and the output is a transformer area aggregation scheme.

[0017] Furthermore, the set of characteristic parameters is subjected to multi-dimensional comprehensive evaluation processing to construct a comprehensive characteristic evaluation function, the formula of which is:

[0018]

[0019] Among them, CI i ω represents the comprehensive characteristic index of the i-th type of distributed resource. j T represents the weight coefficient of the j-th characteristic parameter. ij E ij C ij Let represent the time dimension characteristic value, electrical characteristic value, and adjustable characteristic value of j characteristic parameters of resource i, respectively. α, β, and γ are the balance coefficients of the three types of characteristics, t is the current time, and t0 is the reference time.

[0020] Furthermore, a feature fingerprint database is generated through principal component analysis dimensionality reduction and clustering, including:

[0021] Principal component analysis was used to reduce the dimensionality of high-dimensional characteristic parameters and extract the main feature dimensions.

[0022] Resources with similar characteristics are grouped based on clustering algorithms to form resource characteristic clusters;

[0023] Based on the resource characteristics clustering, the typical operating modes of each type of resource are extracted to establish the characteristic fingerprint database.

[0024] Furthermore, after generating a characteristic fingerprint database through principal component analysis dimensionality reduction and clustering, the process further includes: combining time series analysis methods to identify long-term change patterns in resource operation and obtain a time series model of resource characteristics.

[0025] Furthermore, the formula for calculating the similarity between different resource characteristics is as follows:

[0026]

[0027] Among them, D MH (x,y) represents the similarity of characteristics between resources x and y, δ k f represents the weight coefficients of the k feature functions. k (·) represents the k-th nonlinear feature transformation function, p is the distance metric parameter, T is the transpose of xy, and K is the number of nonlinear feature transformation functions.

[0028] Furthermore, the construction of classification and aggregation models for various types of resources, wherein the electric vehicle group charging power aggregation model is expressed as:

[0029]

[0030] Among them, P EV (t) represents the total charging power of the electric vehicle group at time t, NEV is the total number of electric vehicles, and P is the total charging power of the electric vehicle group. i,rated The rated charging power of the i-th electric vehicle, SOC i (t) represents the state of charge of the i electric vehicles at time t, and η i For charging efficiency, δ i (t) is the charging state indication function, φ i (t) is the charging power adjustment coefficient.

[0031] Furthermore, the integrated load model of the transformer area is expressed as follows:

[0032]

[0033] Where P total(t) represents the total load power of the transformer area at time t, where K is the total number of resource types, and ω k (t) represents the time-varying weighting coefficient of the k-th resource, P k (t) represents the aggregation power of resource type k, where M and N are the number of resource types considering coupling effects, respectively, and α ij (t) is the coupling coefficient between resource types i and j, P i (t) and P j (t) represents the aggregation power of resource types i and j, respectively, γ ij (t) is the coupling adjustment factor, used to describe the degree of mutual influence between different types of resources.

[0034] Furthermore, based on the distribution network topology and the aforementioned integrated load model, the composite electrical distance index between distribution stations is calculated by integrating topological distance, voltage sensitivity, and equivalent impedance. Load characteristic similarity is then extracted based on a characteristic fingerprint database, including:

[0035] The topological distance between the calculation stations is determined by the total number of nodes and the total length of the lines in the distribution network connection path.

[0036] Based on the power flow calculation results, a sensitivity matrix reflecting the impact of power changes in the transformer area on voltage is generated, and the equivalent impedance parameters between transformer areas are obtained.

[0037] The topological distance, voltage sensitivity norm, and equivalent impedance are linearly weighted and fused to generate a composite electrical distance index.

[0038] Based on the aforementioned characteristic parameter set and comprehensive load model, a multi-dimensional characteristic vector containing load peak-valley characteristics and fluctuation characteristics is extracted, and the load characteristic similarity between stations is calculated.

[0039] Establish a dynamic update mechanism to analyze changes in load characteristic similarity at different time scales and output similarity assessment results in real time.

[0040] Furthermore, after obtaining the equivalent impedance parameters between substations, the electrical impedance distance between substations is calculated, and a multidimensional electrical distance metric is established. This multidimensional electrical distance metric is expressed as follows:

[0041]

[0042] Where D elec (i,j) represents the combined electrical distance between transformer area i and transformer area j, D topo (i,j) represents the topological distance. The norm representing voltage sensitivity reflects the degree to which power changes in transformer area j affect the voltage in transformer area i. eq (i,j) represents the equivalent impedance, and α1, α2 and α3 are weighting coefficients.

[0043] Furthermore, the composite electrical distance index and load characteristic similarity are input into an improved K-means clustering algorithm for two-dimensional clustering. The two-dimensional clustering formula is as follows:

[0044]

[0045] Where J is the clustering objective function, K is the number of clusters, and C is the number of clusters. k Let i represent the k-th cluster, and let c represent the area belonging to cluster Ck. k Indicates clustering C k The center, D elec (i,c k ) represents the cluster area i and the cluster center c. k The electrical distance between them, Sim(I, c k ) represents the load characteristic similarity between the substation i and the cluster center ck, and β1 and β2 are weighting coefficients.

[0046] A transformer substation intelligent clustering system considering the characteristics of distributed resource operation includes:

[0047] The similarity calculation module constructs a dataset of runtime characteristics for multiple types of distributed resources and calculates the similarity between different resource characteristics, including:

[0048] Collect operational data from various types of distribution transformer areas and establish a preliminary classification parameter set for various types of distributed resources;

[0049] The preliminary classification parameter set is standardized to form a complete characteristic parameter set;

[0050] Based on the set of characteristic parameters, a characteristic fingerprint database is generated by principal component analysis for dimensionality reduction and clustering, and the characteristic similarity between different resource characteristics is calculated.

[0051] The load model construction module establishes a comprehensive load model for the distribution network area, including:

[0052] Based on the aforementioned characteristic fingerprint database, the resource structure and topological relationship of the transformer area are identified, a classification and aggregation model of various types of resources is constructed, and a comprehensive load model of the transformer area is fused through coupling effect compensation.

[0053] The intelligent aggregation module performs intelligent aggregation of multiple transformer substations based on electrical distance and similarity, including:

[0054] Based on the distribution network topology and the comprehensive load model, the composite electrical distance index between stations is calculated by integrating topological distance, voltage sensitivity and equivalent impedance, and the load characteristic similarity is extracted based on the characteristic fingerprint database.

[0055] With the dual objectives of minimizing composite electrical distance and maximizing load characteristic similarity, the composite electrical distance index and load characteristic similarity are input into an improved K-means clustering algorithm for two-dimensional clustering, and the output is a transformer area aggregation scheme.

[0056] The technical solution of this invention can achieve the following technical effects:

[0057] It effectively solves the problem of collaborative modeling of the dynamic characteristics differences of various types of resources and the electrical coupling strength between distribution stations, and significantly improves the aggregation accuracy and engineering adaptability of distribution stations in distributed energy scenarios. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is a flowchart illustrating a method for intelligent clustering of transformer substations that considers the characteristics of distributed resource operation.

[0060] Figure 2 A flowchart illustrating the process of generating a feature fingerprint library;

[0061] Figure 3 This is a flowchart illustrating the process of calculating the composite electrical distance index between stations and extracting the similarity of load characteristics. Detailed Implementation

[0062] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0063] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0064] Example 1

[0065] like Figure 1 As shown, this invention provides a method for intelligent clustering of transformer substations that considers the operational characteristics of distributed resources. The method includes:

[0066] S1: Construct a dataset of runtime characteristics for multiple types of distributed resources, and calculate the similarity of characteristics between different resources, including:

[0067] Collect operational data from various types of distribution transformer areas and establish a preliminary classification parameter set for various types of distributed resources;

[0068] The preliminary classification parameter set is standardized to form a complete characteristic parameter set;

[0069] Specifically, operational data is first collected through various monitoring devices deployed within the transformer substation area, covering typical distributed resources such as electric vehicles, central air conditioning systems, distributed photovoltaic systems, energy storage devices, and smart home appliances connected to the substation. The main data collected includes key operating parameters such as power change curves, voltage, current, frequency, device start / stop status, charging / discharging behavior, ambient temperature, and irradiance. The sampling period is preferably set to every 15 minutes or less. To ensure data consistency and comparability, various types of equipment are identified and coded before collection to ensure a clear mapping relationship between the collected data and specific equipment. The data is then divided into multiple preliminary data tables according to resource categories. After collection, obvious outliers are removed through data filtering and verification, and missing parts are filled using methods such as time interpolation. All types of data are then aligned with a unified timestamp to form a standardized raw dataset. For electric vehicles, information such as daily charging start and end times, continuous charging time, power curve trends, vehicle dwell frequency, and charging regularity can be extracted to reflect their operational characteristics over time. For example, the daily charging time for a certain type of pure electric passenger vehicle is concentrated between 6 pm and 9 pm, with the charging power maintained at around 6 kilowatts. This characteristic exhibits significant periodicity and consistency. Regarding central air conditioning, attention is focused on its start-stop frequency, load fluctuation range, and ambient temperature feedback lag. For instance, some centralized air conditioning systems show a significant increase in start-stop frequency on hot days, and the temperature control response exhibits non-linear fluctuations, demonstrating strong adjustment inertia. For distributed photovoltaic resources, the volatility and intermittency of their output power can be identified by combining the collected data on light intensity and temperature changes. For example, a power plateau occurs at midday on typical sunny days, while short-period drops occur on cloudy days. For energy storage resources, parameters such as regulation capability and response time are extracted based on the charge-discharge cycle, state of charge change trend, and discharge rate. These data collection methods ultimately form a standardized set of operating characteristic parameters covering multiple types of resources. This parameter set not only includes time-dimensional indicators reflecting equipment operating habits but also covers electrical characteristics such as power factor and voltage response, as well as controllable characteristics such as response rate and adjustment range for various resources, comprehensively characterizing the operating behavior and capability features of different resources. In practical applications, resource priorities or sensitivity labels can be set according to specific scenarios for subsequent use in feature dimensionality reduction, fingerprint modeling, and clustering calculations. This implementation method constructs a high-quality parameter set that truly reflects the behavioral characteristics of various types of distributed resources through multiple steps such as type segmentation, data cleaning, and feature extraction, providing a solid data foundation for subsequent clustering model building and intelligent regulation.

[0070] Based on the feature parameter set, a feature fingerprint database is generated through principal component analysis dimensionality reduction and clustering, and the feature similarity between different resource features is calculated.

[0071] Specifically, the various types of distributed resources connected to the distribution network often have high-dimensional and numerous operational characteristic parameters. For example, the charging start and end times, charging power, and load curve fluctuations of electric vehicles may have more than ten parameters. Principal Component Analysis (PCA) is used to extract the most representative "principal components" from these high-dimensional features. The aim is to reduce data redundancy, improve the efficiency and stability of subsequent clustering analysis, and preserve the main variability of the original information, making the model processing more physically interpretable and practical. By clustering the dimensionality-reduced resource characteristic data, distributed resources with similar operational behaviors can be automatically grouped into one category. This grouping not only helps to understand the commonalities and differences in the operational characteristics of different types of resources, but also provides a basis for building standardized resource models and formulating unified control strategies. The clustering results can also be used to identify special behaviors (such as equipment with abnormal load fluctuations), improving the robustness and early warning capabilities of the model. After establishing a characteristic fingerprint database, the similarity of operational characteristics between any two resources is further calculated. The purpose is to provide a quantitative "similarity index" for subsequent inter-distribution aggregation strategies. This index reflects the consistency of different distribution area resource combinations in terms of time characteristics, electrical characteristics, and control capabilities. It is one of the key input variables for realizing distribution area clustering. By accurately measuring these similarities, the distribution area grouping structure can be optimized, the accuracy of distribution area aggregation and the consistency of control can be improved, thereby improving the overall operating efficiency and intelligent dispatching level of the distribution network.

[0072] S2: Establish a comprehensive load model for the distribution network area, including:

[0073] Based on the aforementioned characteristic fingerprint database, the resource structure and topological relationships of the distribution area are identified, and a classification and aggregation model of various types of resources is constructed. This model is then fused into a comprehensive load model for the distribution area through coupling effect compensation. Specifically, firstly, the established resource characteristic fingerprint database is used to identify the structure and topology of various distributed resources connected to the distribution area. By combining distribution network GIS information, distribution area topology maps, and real-time data provided by smart meters and load acquisition terminals, information such as the access location of each resource node, electrical connection relationships between nodes, and branch paths is mined and reconstructed to identify the physical distribution and electrical coupling structure of various resources within the distribution area. For example, for a centralized electric vehicle charging pile group, it can be identified whether it is connected to the main line or a branch node, thereby determining its impact on the distribution area load center. For distributed photovoltaic and energy storage... For the devices, it is crucial to identify the inverter location and feeder connection method to clarify their grid connection path and parallel interaction points. After structural identification, based on resource type and operating characteristics, different types of distributed resources are constructed into classification aggregation models. Taking electric vehicles as an example, a behavioral probability modeling method is preferred to transform the charging behavior of different vehicles into a set of power curves with certain statistical regularities. Representative charging power spectra are generated based on dimensions such as charging time period, power range, and dwell frequency. For central air conditioning systems, temperature response characteristics and load change patterns are extracted through heat balance analysis, and a dynamic load adjustment model is generated accordingly. For photovoltaic equipment and energy storage devices, their power generation and energy storage behaviors are modeled based on the light-temperature model and the state of charge model, respectively, and their controllability and output stability are quantitatively evaluated. These models, while classifying and categorizing resources, retain the core characteristics of each resource in terms of temporal dynamics, electrical properties, and regulation capabilities, which helps to implement targeted classification control strategies in subsequent regulation. Subsequently, these classification and aggregation models are merged into a comprehensive load model to fully reflect the total load behavior of the entire distribution area under current operating conditions. During the fusion process, a coupling effect compensation mechanism is preferentially introduced to describe the interaction relationships between different types of resources in terms of spatial location, voltage influence, and power fluctuations. For example, during peak load periods, the increased frequency of central air conditioning operation may coincide with concentrated charging of electric vehicles, and the power superposition of these two factors will lead to aggravated voltage fluctuations. In some distribution areas, energy storage devices may actively absorb some power due to peak photovoltaic power generation output, forming negatively correlated coupling characteristics. Therefore, a coupling adjustment factor is set in the model fusion to correct various power types based on the synergy coefficients between resources, thereby enhancing the model's ability to represent coupling behavior. Specifically, when calculating the total load, not only are the independent power superposition results of various resources considered, but also the differences in influence between different resources need to be compensated based on their topological connectivity, regulation response behavior, and time dimension characteristics, thereby improving the model's accuracy in restoring and predicting the overall load fluctuation trend.This implementation method comprehensively refines the identification method of the internal resource structure of the transformer area and the construction path of the load model through the three stages of "identification-modeling-fusion". In particular, the introduction of the coupling effect compensation mechanism as a key invention point breaks through the technical limitation of traditional transformer area models that only focus on resource superposition and ignore resource interaction. It significantly improves the accuracy, stability and engineering applicability of the model, and provides a solid model support and structural foundation for subsequent intelligent aggregation and scheduling control among multiple transformer areas.

[0074] S3: Intelligent aggregation of multiple transformer substations based on electrical distance and similarity, including:

[0075] Based on the distribution network topology and integrated load model, the composite electrical distance index between stations is calculated by integrating topological distance, voltage sensitivity and equivalent impedance, and load characteristic similarity is extracted based on the characteristic fingerprint database;

[0076] Specifically, by combining the actual topology of the distribution network with the established integrated load model of the transformer substations, and starting from the electrical connection relationship, a composite electrical distance index between transformer substations is calculated by integrating three key parameters: topological distance, voltage sensitivity, and equivalent impedance. Topological distance describes the path proximity between transformer substations in the distribution network structure and is typically determined based on the number of nodes and line length. Voltage sensitivity reflects the degree to which load changes in one transformer substation affect voltage fluctuations in other substations and serves as the physical basis for measuring the strength of interaction. Equivalent impedance expresses the electrical resistance encountered when power is transmitted between different substations and is closely related to current flow direction and power distribution. By normalizing and weighting these three factors, a composite electrical distance that better reflects actual operating characteristics can be obtained to measure the tightness of electrical coupling between transformer substations. Simultaneously, to ensure consistency in the operating characteristics of aggregated transformer substations, it is also necessary to extract the load feature vector of each substation based on the aforementioned resource characteristic fingerprint database and calculate its characteristic similarity with other substations. This similarity metric considers dimensions such as the shape of the load curve, peak-to-valley variations, fluctuation amplitude, and control capability, comprehensively reflecting the consistency of load behavior across different transformer substations. By incorporating both composite electrical distance and load characteristic similarity, this step not only solves the problem of inaccurate grouping caused by relying solely on structural or behavioral information in traditional methods, but also provides structurally sound and data-rich support for subsequent bi-objective clustering, thereby improving the practical adaptability and control effectiveness of the aggregation model.

[0077] With the dual objectives of minimizing composite electrical distance and maximizing load characteristic similarity, the composite electrical distance index and load characteristic similarity are input into an improved K-means clustering algorithm for two-dimensional clustering, and the output is a transformer area aggregation scheme.

[0078] Specifically, to achieve high-precision transformer area grouping and resource aggregation, a two-dimensional clustering strategy with the dual objectives of minimizing composite electrical distance and maximizing load characteristic similarity is adopted. The composite electrical distance index between transformer areas and the load characteristic similarity calculated above are used as core input variables and integrated into an improved K-means clustering algorithm to output a transformer area aggregation scheme with structural rationality and behavioral consistency. Traditional K-means algorithms use Euclidean distance as the sole distance metric, making it difficult to simultaneously consider electrical connectivity and operational characteristic similarity, thus limiting their application in transformer substation clustering. To address this issue, this implementation expands the K-means algorithm with two-dimensional features and optimizes the objective function. This ensures that the clustering objective not only maximizes the electrical connectivity within a cluster but also considers the high consistency of load behavior between substations. In the input layer design, each substation's data is represented as a two-dimensional vector: one dimension is the normalized composite electrical distance, and the other is the load characteristic similarity calculated based on a characteristic fingerprint database. A higher similarity value indicates that the two substations are closer in their operating curves. During algorithm initialization, a density-first method is used to select cluster centers, prioritizing substations with small electrical distances and high characteristic similarity as initial cores. During iterative updates, a weighted objective function is used to jointly consider... This method considers two clustering dimensions: composite electrical distance is minimized as a constraint, and load characteristic similarity is maximized as a benefit. The weight parameters are dynamically adjusted based on user settings or historical clustering results. In practice, if the aggregation goal is short-term load regulation, the weight of load characteristic similarity can be increased; if the goal is physical-level network reconstruction or regional management, electrical distance is given priority. During the algorithm's operation, the grouping quality is evaluated using the silhouette coefficient and cluster density index, and the cluster centers and boundaries are dynamically updated to ensure that the final clustering results converge to the optimal state simultaneously in both objective dimensions. In the verification stage, this invention preferably uses a typical urban power distribution network scenario for experimentation, clustering multiple transformer areas with mixed access to photovoltaic, energy storage, and electric vehicles. The results show that this method significantly improves the consistency of load behavior within the clustered area groups while maintaining the electrical connection density between transformer areas. This method effectively solves the problem of mismatch between the structure and function of the distribution area in existing methods. It has high adaptability, interpretability and deployment feasibility, and provides a refined and hierarchical technical foundation for subsequent distributed energy scheduling, demand response control and load forecasting. It reflects the core innovative value of this invention in the design of intelligent aggregation strategy for distribution areas.

[0079] This invention effectively solves the problem of collaborative modeling of the differences in dynamic characteristics of various types of resources and the electrical coupling strength between distribution stations, and significantly improves the aggregation accuracy and engineering adaptability of distribution stations in distributed energy scenarios.

[0080] As a preferred embodiment of the above, the characteristic parameter set is subjected to multi-dimensional comprehensive evaluation processing to construct a comprehensive characteristic evaluation function, the formula of which is:

[0081]

[0082] Among them, CI i ω represents the comprehensive characteristic index of the i-th type of distributed resource. j T represents the weight coefficient of the j-th characteristic parameter. ij E ij C ij Let represent the time dimension characteristic value, electrical characteristic value, and adjustable characteristic value of j characteristic parameters of resource i, respectively. α, β, and γ are the balance coefficients of the three types of characteristics, t is the current time, and t0 is the reference time.

[0083] Specifically, to more accurately assess the operational characteristics of various types of distributed resources, a comprehensive characteristic evaluation function was constructed. This function calculates the overall performance of each type of distributed resource at a given moment, considering three main aspects: first, time-dimensional characteristics, such as equipment start-up and shutdown frequency, operating duration, and daily usage patterns; second, electrical characteristics, such as power factor, voltage fluctuation, and current variation, reflecting the electrical relationship between the resource and the power grid; and third, controllability characteristics, such as regulation response time, regulation amplitude, and regulation duration, indicating whether the resource can participate in scheduling or demand response. To reasonably combine these different types of characteristics, a weighting coefficient is set for each characteristic parameter in the function to represent the importance of that parameter in the evaluation. Furthermore, to balance the influence among the three types of characteristics, three balancing coefficients are set, corresponding to time, electrical, and controllability characteristics, respectively. These weights and coefficients can be flexibly adjusted according to actual needs. For example, in scenarios emphasizing regulation capability, the proportion of controllability characteristics can be increased, while when power grid security is prioritized, the importance of electrical characteristics can be highlighted. Furthermore, this function introduces two variables: current time and reference time, to dynamically reflect the changing trends of resource operation characteristics. For example, whether the average charging time of an electric vehicle has been earlier or later, or whether the charging power has fluctuated in recent days, can all be monitored in this way. Taking a specific example, assuming three typical parameters are selected for electric vehicle resources: average daily charging period (time dimension), average charging power (electrical dimension), and response adjustment flexibility (adjustable dimension), by setting appropriate weights and coefficients and substituting them into the above function, the comprehensive characteristic value (CI) of the electric vehicle group at the current moment can be calculated. iAs a key quantitative indicator for measuring the operational characteristics of resources, CI participates in the construction of resource characteristic vectors and feature fingerprint databases, serving as the basis for characteristic similarity calculation and used to determine the degree of similarity between resources and between transformer substations. At the same time, CI also serves as a core parameter in the load modeling and clustering objective function of transformer substations, and together with electrical distance, it drives the two-dimensional intelligent aggregation of transformer substations, thereby providing a scientific basis for differentiated control strategies.

[0084] As a preferred embodiment of the above, such as Figure 2 As shown, a feature fingerprint database is generated through principal component analysis dimensionality reduction and clustering, including:

[0085] A10: Principal component analysis is used to reduce the dimensionality of high-dimensional characteristic parameters and extract the main feature dimensions;

[0086] A20: Group resources with similar characteristics based on clustering algorithms to form resource characteristic clusters;

[0087] A30: Based on resource characteristic clustering, typical operating modes of each type of resource are extracted to establish a characteristic fingerprint database.

[0088] Specifically, firstly, principal component analysis (PCA) is used to reduce the dimensionality of the existing high-dimensional set of operational characteristic parameters. Distributed resources typically contain a dozen or even dozens of operational characteristic indicators. If all of them are included in subsequent model processing, it can easily lead to problems such as computational complexity, dimensional redundancy, and noise amplification. PCA can automatically extract a few principal feature dimensions that reflect most of the information based on the correlation between parameters, significantly reducing data dimensionality while retaining the main feature differences. For example, among the characteristic parameters of electric vehicle groups, charging power, charging time, and charging frequency have a strong correlation. PCA can integrate them into a set of comprehensive feature factors to more clearly describe the charging mode characteristics of this type of resource. Next, based on the dimensionality-reduced feature data, a clustering algorithm is used to classify distributed resources with similar operational characteristics. Based on the distance or density relationship in the principal component space, resources with similar operational characteristics are grouped together. The resources identified are divided into several feature clusters. For example, electric vehicles with charging periods concentrated at night and high power stability can be classified into one category, while vehicles with intermittent charging during the day can be classified into another. The clustering results not only improve the accuracy of resource classification but also provide a basis for building personalized models for different resource types. Finally, based on the above clustering results, representative operating modes are extracted for each resource category to establish a resource operating characteristic template library. Specifically, the operating curves and response behaviors of the resources included in each cluster are normalized and statistically analyzed to extract their common characteristics and form template entries. For example, for central air conditioning resources, average features such as start-stop cycle, load fluctuation range, and temperature control response time can be extracted to form an operating template for temperature-controlled loads; for photovoltaic resources, typical solar power generation curves and power fluctuation ranges can be extracted to establish a photovoltaic output characteristic template. The establishment of this template library not only helps to quickly identify the type of unknown resources but also provides a standardized reference for subsequent load forecasting, characteristic comparison, and cluster optimization.

[0089] As a preferred embodiment of the above, after generating a characteristic fingerprint database through principal component analysis dimensionality reduction and clustering, the method further includes: combining time series analysis methods to identify the long-term change patterns of resource operation and obtain a time series model of resource characteristics.

[0090] Specifically, the process begins by selecting representative operational characteristic parameters for each resource category, such as the average charging start time of electric vehicles, the start-stop frequency of central air conditioning, and the average daily power generation of photovoltaic equipment. Historical data on these parameters over a period of time is collected to form time-series datasets arranged by day, week, or month. By observing the trends in these data, it can be discovered that certain characteristic parameters exhibit significant periodicity or seasonality. For example, air conditioning load increases significantly in summer, photovoltaic output decreases significantly on rainy days, and electric vehicle charging behavior differs between weekdays and holidays. During the modeling process, appropriate time-series analysis methods are selected based on the regularity of characteristic changes. For characteristics with strong regularity and small fluctuations, traditional models such as moving averages or exponential smoothing can be used; while for characteristics with large fluctuations and significant influence from external factors, simple machine learning methods can be used for prediction. Finally, the trends of the main characteristics of each resource category over time will be modeled and combined with existing static feature labels to form a dynamically enhanced characteristic fingerprint database. This approach not only allows for more accurate identification of long-term resource behavior but also improves the accuracy of characteristic matching, load forecasting, and resource aggregation. It is particularly effective in responding to seasonal changes, load fluctuations, or the integration of new resources, demonstrating strong adaptability and practical application value. It serves as an important supplement and optimization to the aforementioned characteristic clustering model.

[0091] As a preferred embodiment of the above, the similarity of characteristics between different resource characteristics is calculated using the following formula:

[0092]

[0093] Among them, D MH (x,y) represents the similarity of characteristics between resources x and y, δ k f represents the weight coefficients of the k feature functions. k (·) represents the k-th nonlinear feature transformation function, p is the distance metric parameter, T is the transpose of xy, and K is the number of nonlinear feature transformation functions.

[0094] Specifically, to accurately measure the similarity of different distributed resources in terms of operational characteristics, a similarity calculation method based on improved Mahalanobis distance is proposed. The core idea of ​​this method is to perform a nonlinear transformation on the characteristic parameter vectors of the resources, and then combine weighting coefficients and distance metric parameters to calculate the "relative distance" between resources in the characteristic space. This allows for the determination of the similarity in behavioral patterns between two types of resources. Compared to traditional Euclidean distance or cosine similarity, this method better reflects the correlation between characteristic dimensions and the distribution structure of the resources themselves, making it suitable for comparative analysis of multi-type, heterogeneous distributed resources. Specifically, resources x and y each have a set of characteristic parameter vectors with the same structure. These parameters can be derived from the output of the aforementioned comprehensive characteristic evaluation function, covering multiple dimensions such as time characteristics, electrical characteristics, and controllability characteristics. First, a nonlinear feature transformation function is introduced for each dimension's parameters to map the original parameters to a space that better reflects their physical meaning or behavioral characteristics. For example, for the power generation curve of photovoltaic resources, a logarithmic function or normalization function can be used to compress its power fluctuations to a specific range; for the charging behavior of electric vehicles, a sigmoid function or piecewise function can be used to highlight the behavioral patterns of different charging stages. In the transformed feature space, the feature value of each dimension is multiplied by the corresponding weight coefficient to represent the relative importance of that feature in similarity calculation. The weight coefficient can be flexibly set according to the application scenario. For example, in load regulation applications, controllable characteristics have a higher weight; while in grid stability analysis, the importance of electrical characteristics is given priority. Subsequently, the distance between the overall characteristic vectors of the two resources is calculated using an improved Mahalanobis distance. This method can consider the covariance relationship between each feature dimension, thereby avoiding misjudgment caused by the correlation between features. In addition, to improve adaptability, a distance metric parameter is introduced to adjust the sensitivity or penalty strength of the overall distance. For example, in some scenarios, this parameter can be appropriately increased to more strictly distinguish resource types with similar differences; conversely, its impact can be reduced by relaxing clustering or matching conditions. This similarity calculation method can quantify the degree of similarity in operational behavior between different resources, providing basic data support for subsequent transformer area aggregation, control grouping, or characteristic matching.

[0095] As a preferred embodiment of the above, a classification and aggregation model for each type of resource is constructed, wherein the electric vehicle group charging power aggregation model is represented as follows:

[0096]

[0097] Among them, P EV (t) represents the total charging power of the electric vehicle group at time t, NEV is the total number of electric vehicles, and P is the total charging power of the electric vehicle group. i,ratedThe rated charging power of the i-th electric vehicle, SOC i (t) represents the state of charge of the i electric vehicles at time t, and η i For charging efficiency, δ i (t) is the charging state indication function, φ i (t) is the charging power adjustment coefficient.

[0098] Specifically, the model treats each electric vehicle as an independent sub-unit and evaluates its state at time t. First, it considers its rated charging power, representing the vehicle's maximum charging capacity under ideal conditions. Second, it determines the required charging power based on its real-time state of charge (SOC); a lower SOC indicates a higher charging demand, and vice versa. To reflect differences in energy efficiency among vehicles, an individual vehicle charging efficiency parameter is introduced to correct the actual output power. Furthermore, to determine whether each vehicle is charging, a charging status indicator function is set in the model; this value is 1 when the vehicle is charging and 0 when it is not charging, ensuring that only the actual operating load is counted. To further enhance the model's flexibility and responsiveness, a charging power adjustment coefficient is introduced in the aggregate model to express whether charging needs to be limited or adjusted under specific strategies. For example, during peak load periods in the distribution area, this coefficient can be manually lowered to control the total charging load of electric vehicles and avoid overload; while during off-peak hours at night, the coefficient can be appropriately increased to guide vehicle charging and improve off-peak electricity utilization. In practical applications, this model can access real-time load monitoring data of the distribution area to perform status tracking and charging behavior simulation for all electric vehicles connected to charging piles. For example, if a distribution area has 50 electric vehicles connected to it, and 20 of them are charging at a certain moment, with different charging efficiencies, states of charge, and rated power, this model can calculate the total charging power of the group at that moment in real time and predict its subsequent power evolution trend.

[0099] As a preferred embodiment of the above, the integrated load model of the transformer area is expressed as follows:

[0100]

[0101] Where P total (t) represents the total load power of the transformer area at time t, where K is the total number of resource types, and ω k (t) represents the time-varying weighting coefficient of the k-th resource, P k (t) represents the aggregation power of resource type k, where M and N are the number of resource types considering coupling effects, respectively, and α ij (t) is the coupling coefficient between resource types i and j, P i (t) and P j (t) represents the aggregation power of resource types i and j, respectively, γij (t) is the coupling adjustment factor, used to describe the degree of mutual influence between different types of resources.

[0102] Specifically, the total load of the distribution area consists of two parts: The first part is a power-weighted sum categorized by resource type. Each resource type (such as electric vehicles, air conditioners, and photovoltaics) has its own aggregated power and a weight that varies over time to reflect its contribution to the total load during the current period. For example, photovoltaics have a larger weight during the day when there is ample sunlight, while the impact of electric vehicle charging is more significant at night. The second part is a coupling compensation term between resources, used to describe the possible interactions between different types of equipment when they operate simultaneously, such as superposition or partial cancellation. The calculation of the coupling part considers three aspects: whether there is coupling between resource types, the magnitude of their respective aggregated power, and a factor to adjust the strength of coupling. For example, when photovoltaic output increases, the energy storage system may charge synchronously; this synergistic relationship can be represented by a positive coupling coefficient. Conversely, if concentrated charging of electric vehicles happens to overlap with high air conditioning load, it may put significant pressure on the distribution area; this situation can be reflected by a negative coupling coefficient. The coupling adjustment factor can be dynamically adjusted according to conditions such as season, weather, or electricity consumption behavior to make the model more closely reflect reality.

[0103] As a preferred embodiment of the above, such as Figure 3 As shown, in step S4, based on the distribution network topology and integrated load model, the composite electrical distance index between stations is calculated by integrating topological distance, voltage sensitivity, and equivalent impedance, and the load characteristic similarity is extracted based on the characteristic fingerprint database, including:

[0104] S41: Calculate the topological distance between distribution network stations. The topological distance is determined by the total number of nodes in the distribution network connection path and the total length of the line.

[0105] S42: Based on the power flow calculation results, generate a sensitivity matrix that reflects the degree of influence of power changes in the transformer area on voltage, and obtain the equivalent impedance parameters between transformer areas;

[0106] S43: Linearly weighted and fused topological distance, voltage sensitivity norm and equivalent impedance to generate a composite electrical distance index;

[0107] S44: Based on the characteristic parameter set and the comprehensive load model, extract a multi-dimensional characteristic vector containing load peak and valley characteristics and fluctuation characteristics, and calculate the load characteristic similarity between stations;

[0108] S45: Establish a dynamic update mechanism to analyze changes in load characteristic similarity at different time scales and output similarity assessment results in real time.

[0109] Specifically, the topological distance between distribution substations is first calculated. This distance considers not only the physical length of the lines but also the number of nodes traversed between the substations. Specifically, based on the distribution network topology, the total number of nodes and the total line length along the connection path between any two substations are counted, and then calculated using certain weighting coefficients. The smaller this distance, the closer the two substations are in terms of grid structure, and the stronger the possibility of electrical interaction. Next, in step S42, the voltage sensitivity matrix and equivalent impedance parameters are obtained through power flow analysis. Power flow calculations can reveal the degree to which changes in substation load affect the voltage of surrounding nodes, thus obtaining the voltage sensitivity between substations. Meanwhile, the equivalent impedance between any two transformer substations is calculated using the network equivalence method to reflect the ease of current transmission along the path. Both parameters can be obtained from common power distribution system simulation platforms or field measurement data. In step S43, the obtained topological distance, voltage sensitivity norm, and equivalent impedance parameters are fused in a linear weighted manner to generate a unified "composite electrical distance index." This index comprehensively reflects the electrical correlation strength between two transformer substations in multiple dimensions, including structural path, voltage interaction, and resistance influence. The weights can be set according to specific application scenarios; for example, the weight of voltage sensitivity can be increased when performing voltage stability management. Subsequently, in step S44, based on the previously constructed set of operating characteristic parameters and the comprehensive load model, multi-dimensional characteristic vectors are extracted from each transformer substation. These characteristics mainly include the peak-valley distribution, fluctuation amplitude, and change frequency of the load, used to describe the load behavior characteristics of the substation. By comparing the differences between these vectors, the load characteristic similarity between transformer substations can be calculated. The higher the similarity, the closer the operating modes of the transformer areas are, and the more suitable they are to be assigned to the same aggregation unit. Finally, in step S45, in order to improve the real-time performance and adaptability of the evaluation method, a dynamic update mechanism is established. This mechanism supports continuous tracking and analysis of load characteristic similarity at different time scales (such as daily, weekly, and monthly). When load behavior changes with seasons, time, or external environment, the characteristic vector and similarity results can be updated in real time, thereby ensuring the stability and effectiveness of the clustering model in long-term operation.

[0110] As a preferred embodiment of the above, after obtaining the equivalent impedance parameters between stations, the electrical impedance distance between stations is calculated, and a multi-dimensional electrical distance metric is established. The multi-dimensional electrical distance metric is expressed as follows:

[0111]

[0112] Where D elec (i,j) represents the combined electrical distance between transformer area i and transformer area j, D topo (i,j) represents the topological distance. The norm representing voltage sensitivity reflects the degree to which power changes in transformer area j affect the voltage in transformer area i.eq (i,j) represents the equivalent impedance, and α1, α2 and α3 are weighting coefficients.

[0113] Specifically, the comprehensive electrical distance is calculated using a linear weighted approach, taking multiple factors between transformer substations as input. First, the topological distance represents the connection path between substations in the distribution network, which can be statistically determined based on the number of nodes traversed and the total line length between the two substations; a shorter distance indicates a closer structural connection. Second, the voltage sensitivity norm measures the impact of load changes in substation j on voltage fluctuations in substation i. This indicator can be obtained through power flow calculations and reflects the coupling strength at the voltage level; a larger value indicates a more significant impact. Third, the equivalent impedance measures the ease of current transmission between the two substations; a smaller impedance indicates easier energy exchange and a closer electrical connection. These three indicators each have their own physical emphasis. To construct a unified measurement standard, three weighting coefficients are introduced for normalization and comprehensive evaluation. The weighting coefficients α1, α2, and α3 correspond to the relative importance of topological distance, voltage sensitivity, and equivalent impedance, respectively, and can be adjusted according to specific scenarios. For example, in situations where voltage stability is paramount, the weight of voltage sensitivity can be increased; while in physical structure reconstruction or topology optimization, the importance of topological distance is emphasized. Taking a typical scenario as an example, in a city's distribution network, if two transformer substations are located on the same path branch with a small node spacing, and their voltage coupling is strong and impedance is low, then the overall electrical distance will be relatively small, indicating that these two substations are structurally and electrically similar and suitable as a clustering unit. Conversely, if the topological path between two substations is long, the impedance is high, and the voltage influence is weak, then their overall distance will be large, making unified control unsuitable.

[0114] As a preferred embodiment of the above, the composite electrical distance index and load characteristic similarity are input into the improved K-means clustering algorithm for two-dimensional clustering. The two-dimensional clustering formula is as follows:

[0115]

[0116] Where J is the clustering objective function, K is the number of clusters, and C is the number of clusters. k Let i represent the k-th cluster, and let c represent the area belonging to cluster Ck. k Indicates clustering C k The center, D elec (i,c k ) represents the cluster area i and the cluster center c. k The electrical distance between them, Sim(I, c k ) represents the load characteristic similarity between the substation i and the cluster center ck, and β1 and β2 are weighting coefficients.

[0117] Specifically, the input to two-dimensional clustering is the feature values ​​of each transformer substation in two dimensions: first, the composite electrical distance, representing the degree of electrical coupling between this substation and other substations; and second, the load characteristic similarity, measuring the similarity in operating modes between this substation and other substations. During the clustering process, all substations are first initialized in a two-dimensional feature space, and the expected number of clusters K is set. Each cluster C... k Each substation has a cluster center ck, which has location coordinates in two dimensions, representing the mean electrical structure and the mean load characteristics of that substation class. The clustering objective function J consists of two parts: the sum of the electrical distances from all substations to their respective cluster centers, reflecting the electrical compactness within the cluster; and the sum of the load characteristic similarities between the substations and the cluster centers, representing the consistency of behavior within the cluster. To flexibly adjust the importance of these two objectives, weight coefficients β1 and β2 are introduced for balanced configuration. For example, in scheduling scenarios that require enhanced electrical stability, β1 can be set to a higher value; in demand response scenarios that emphasize load coordination response, the weight of β2 is increased. In practice, the algorithm operates using an iterative optimization approach: first, all substations are initially assigned to the nearest cluster center, completing the initial clustering under the combined effect of minimum electrical distance and maximum characteristic similarity; then, the new center position for each class is calculated, and the substations are reassigned; this process is repeated until the clustering results converge or the preset number of iterations is reached. In this process, the clustering effect can be evaluated using the silhouette coefficient or intra-class error index, and the clustering number K or weight coefficient configuration can be dynamically optimized accordingly. For example, in a distribution network that includes electric vehicles, air conditioners, photovoltaics and energy storage resources, multiple transformer substations are closely connected in electrical structure and also show high similarity in load operation. This method can automatically classify them into the same cluster and coordinate their control as a whole. For transformer substations that are not strongly connected in electrical structure or have large differences in operating behavior, they will be automatically classified into other categories to form independent control units.

[0118] Example 2

[0119] Based on the same inventive concept as the intelligent clustering method for transformer substations considering the operating characteristics of distributed resources in the foregoing embodiments, this invention also provides an intelligent clustering system for transformer substations considering the operating characteristics of distributed resources, the system comprising:

[0120] The similarity calculation module constructs a dataset of runtime characteristics for multiple types of distributed resources and calculates the similarity between different resource characteristics, including:

[0121] Collect operational data from various types of distribution transformer areas and establish a preliminary classification parameter set for various types of distributed resources;

[0122] The preliminary classification parameter set is standardized to form a complete characteristic parameter set;

[0123] Based on the feature parameter set, a feature fingerprint database is generated through principal component analysis dimensionality reduction and clustering, and the feature similarity between different resource features is calculated.

[0124] The load model construction module establishes a comprehensive load model for the distribution network area, including:

[0125] Based on the characteristic fingerprint database, the resource structure and topological relationship of the transformer area are identified, a classification and aggregation model of various types of resources is constructed, and the model is fused into a comprehensive load model of the transformer area through coupling effect compensation.

[0126] The intelligent aggregation module performs intelligent aggregation of multiple transformer substations based on electrical distance and similarity, including:

[0127] Based on the distribution network topology and integrated load model, the composite electrical distance index between stations is calculated by integrating topological distance, voltage sensitivity and equivalent impedance, and load characteristic similarity is extracted based on the characteristic fingerprint database;

[0128] With the dual objectives of minimizing composite electrical distance and maximizing load characteristic similarity, the composite electrical distance index and load characteristic similarity are input into an improved K-means clustering algorithm for two-dimensional clustering, and the output is a transformer area aggregation scheme.

[0129] The clustering system described above in this invention can effectively implement a smart clustering method for transformer substations that takes into account the characteristics of distributed resource operation. The technical effects it can achieve are as described in the above embodiments, and will not be repeated here.

[0130] Although this application has been described in conjunction with specific features and embodiments, it will be apparent that various modifications and combinations can be made therein without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of the application as defined herein, and are to be considered as covering any and all modifications, variations, combinations, or equivalents within the scope of this application.

[0131] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for intelligent clustering of transformer substations considering the operational characteristics of distributed resources, characterized in that, include: S1: Construct a dataset of runtime characteristics for multiple types of distributed resources, and calculate the similarity of characteristics between different resources, including: Collect operational data from various types of distribution transformer areas and establish a preliminary classification parameter set for various types of distributed resources; The preliminary classification parameter set is standardized to form a complete characteristic parameter set; Based on the set of characteristic parameters, a characteristic fingerprint database is generated by principal component analysis for dimensionality reduction and clustering, and the characteristic similarity between different resource characteristics is calculated. S2: Establish a comprehensive load model for the distribution network area, including: Based on the aforementioned characteristic fingerprint database, the resource structure and topological relationship of the transformer area are identified, a classification and aggregation model of various types of resources is constructed, and a comprehensive load model of the transformer area is fused through coupling effect compensation. S3: Intelligent aggregation of multiple transformer substations based on electrical distance and similarity, including: Based on the distribution network topology and the comprehensive load model, the composite electrical distance index between stations is calculated by integrating topological distance, voltage sensitivity and equivalent impedance, and the load characteristic similarity is extracted based on the characteristic fingerprint database. With the dual objectives of minimizing composite electrical distance and maximizing load characteristic similarity, the composite electrical distance index and load characteristic similarity are input into an improved K-means clustering algorithm for two-dimensional clustering, and the output is a transformer area aggregation scheme.

2. The intelligent clustering method for transformer substations considering the operational characteristics of distributed resources according to claim 1, characterized in that, The set of characteristic parameters is subjected to multi-dimensional comprehensive evaluation processing to construct a comprehensive characteristic evaluation function, the formula of which is: Among them, CI i ω represents the comprehensive characteristic index of the i-th type of distributed resource. j T represents the weight coefficient of the j-th characteristic parameter. ij E ij C ij Let represent the time dimension characteristic value, electrical characteristic value, and adjustable characteristic value of j characteristic parameters of resource i, respectively. α, β, and γ are the balance coefficients of the three types of characteristics, t is the current time, and t0 is the reference time.

3. The intelligent clustering method for transformer substations considering the operational characteristics of distributed resources according to claim 1, characterized in that, A feature fingerprint library is generated through principal component analysis dimensionality reduction and clustering, including: Principal component analysis was used to reduce the dimensionality of high-dimensional characteristic parameters and extract the main feature dimensions. Resources with similar characteristics are grouped based on clustering algorithms to form resource characteristic clusters; Based on the resource characteristics clustering, the typical operating modes of each type of resource are extracted to establish the characteristic fingerprint database.

4. The intelligent clustering method for transformer substations considering the operating characteristics of distributed resources according to claim 3, characterized in that, After generating a characteristic fingerprint database through principal component analysis dimensionality reduction and clustering, the process further includes: combining time series analysis methods to identify long-term change patterns in resource operation and obtain a time series model of resource characteristics.

5. The intelligent clustering method for transformer substations considering the operating characteristics of distributed resources according to claim 1, characterized in that, The formula for calculating the similarity between different resource characteristics is as follows: Among them, D MH (x,y) represents the similarity of characteristics between resources x and y, δ k f represents the weight coefficients of the k feature functions. k () represents the k-th nonlinear feature transformation function, p is the distance metric parameter, T is the transpose of xy, and K is the number of nonlinear feature transformation functions.

6. The intelligent clustering method for transformer substations considering the operating characteristics of distributed resources according to claim 1, characterized in that, The classification and aggregation model for various types of resources is constructed, wherein the electric vehicle group charging power aggregation model is represented as follows: Among them, P EV (t) represents the total charging power of the electric vehicle group at time t, N EV P represents the total number of electric vehicles. i,rated The rated charging power of the i-th electric vehicle, SOC i (t) represents the state of charge of the i electric vehicles at time t, and η i For charging efficiency, δ i (t) is the charging state indication function, φ i (t) is the charging power adjustment coefficient.

7. The intelligent clustering method for transformer substations considering the operating characteristics of distributed resources according to claim 1, characterized in that, The integrated load model for the transformer area is expressed as follows: Where P total (t) represents the total load power of the transformer area at time t, K is the total number of resource types, and ω k (t) represents the time-varying weighting coefficient of the k-th resource, P k (t) represents the aggregation power of resource type k, where M and N are the number of resource types considering coupling effects, respectively, and α ij (t) is the coupling coefficient between resource types i and j, P i (t) and P j (t) represents the aggregation power of resource types i and j, respectively, γ ij (t) is the coupling adjustment factor, used to describe the degree of mutual influence between different types of resources.

8. The intelligent clustering method for transformer substations considering the operating characteristics of distributed resources according to claim 1, characterized in that, Based on the distribution network topology and the aforementioned integrated load model, the composite electrical distance index between distribution stations is calculated by integrating topological distance, voltage sensitivity, and equivalent impedance. Furthermore, load characteristic similarity is extracted based on a characteristic fingerprint database, including: The topological distance between the calculation stations is determined by the total number of nodes and the total length of the lines in the distribution network connection path. Based on the power flow calculation results, a sensitivity matrix reflecting the impact of power changes in the transformer area on voltage is generated, and the equivalent impedance parameters between transformer areas are obtained. The topological distance, voltage sensitivity norm, and equivalent impedance are linearly weighted and fused to generate a composite electrical distance index. Based on the aforementioned characteristic parameter set and comprehensive load model, a multi-dimensional characteristic vector containing load peak-valley characteristics and fluctuation characteristics is extracted, and the load characteristic similarity between stations is calculated. Establish a dynamic update mechanism to analyze changes in load characteristic similarity at different time scales and output similarity assessment results in real time.

9. The intelligent clustering method for transformer substations considering the operating characteristics of distributed resources according to claim 8, characterized in that, After obtaining the equivalent impedance parameters between substations, the electrical impedance distance between substations is calculated, and a multidimensional electrical distance metric is established. This multidimensional electrical distance metric is expressed as follows: Where D elec (i,j) represents the combined electrical distance between transformer area i and transformer area j, D topo (i,j) represents the topological distance. The norm representing voltage sensitivity reflects the degree to which power changes in transformer area j affect the voltage in transformer area i. eq (i,j) represents the equivalent impedance, and α1, α2 and α3 are weighting coefficients.

10. The intelligent clustering method for transformer substations considering the operating characteristics of distributed resources according to claim 1, characterized in that, The composite electrical distance index and load characteristic similarity are input into an improved K-means clustering algorithm for two-dimensional clustering. The two-dimensional clustering formula is as follows: Where J is the clustering objective function, K is the number of clusters, and c k Let i represent the k-th cluster, where i indicates that it belongs to cluster C. k The area of ​​the platform, c k Indicates clustering C k The center, D elec (i,c k ) represents the cluster area i and the cluster center c. k The electrical distance between them, Sim(I, c k ) represents the load characteristic similarity between the substation i and the cluster center ck, and β1 and β2 are weighting coefficients.

11. A smart clustering system for transformer substations that considers the operational characteristics of distributed resources, characterized in that, include: The similarity calculation module constructs a dataset of runtime characteristics for multiple types of distributed resources and calculates the similarity between different resource characteristics, including: Collect operational data from various types of distribution transformer areas and establish a preliminary classification parameter set for various types of distributed resources; The preliminary classification parameter set is standardized to form a complete characteristic parameter set; Based on the set of characteristic parameters, a characteristic fingerprint database is generated by principal component analysis for dimensionality reduction and clustering, and the characteristic similarity between different resource characteristics is calculated. The load model construction module establishes a comprehensive load model for the distribution network area, including: Based on the aforementioned characteristic fingerprint database, the resource structure and topological relationship of the transformer area are identified, a classification and aggregation model of various types of resources is constructed, and a comprehensive load model of the transformer area is fused through coupling effect compensation. The intelligent aggregation module performs intelligent aggregation of multiple transformer substations based on electrical distance and similarity, including: Based on the distribution network topology and the comprehensive load model, the composite electrical distance index between stations is calculated by integrating topological distance, voltage sensitivity and equivalent impedance, and the load characteristic similarity is extracted based on the characteristic fingerprint database. With the dual objectives of minimizing composite electrical distance and maximizing load characteristic similarity, the composite electrical distance index and load characteristic similarity are input into an improved K-means clustering algorithm for two-dimensional clustering, and the output is a transformer area aggregation scheme.

Images