Distribution transformer system charging load cloud side-end cooperative regulation and control method

Abstract

The invention provides a distribution transformer system charging load cloud side-end cooperative regulation and control method, which comprises the steps of obtaining operation, charging requirements, network states and meteorological environment information of a distribution transformer system, and performing data cleaning. And a regulation and control scene is determined according to the cleaned information, a regulation and control strategy model is generated based on a network topology model, cloud side constraints, side resources and a current scene, and local correction is performed in combination with edge side real-time feedback. If the current scene is different from the subsequent synchronization scene, generating a subsequent synchronization regulation and control strategy model; and if yes, linking the subsequent synchronization strategy model to the corrected current regulation and control strategy model. And determining a load aggregation model and calling a related model to execute real-time regulation and control, collecting end-side measurement information, performing closed-loop updating on the load aggregation model, and performing secondary correction on the regulation and control strategy model to complete real-time regulation and control. The system operation efficiency and stability can be improved, resource allocation and model management are optimized, and the adaptability and flexibility of the system are enhanced.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for power systems, and more specifically, to a cloud-edge-end coordinated control method for charging loads in distribution transformer systems. Background Technology

[0002] In today's power systems, with the widespread adoption of electric vehicles and the extensive application of distributed energy resources, the management of charging loads in distribution transformer systems faces unprecedented challenges. Traditional distribution transformer systems primarily rely on centralized control strategies for charging load regulation. While this strategy can manage charging loads to some extent, it has several limitations. First, centralized control strategies cannot fully consider the real-time operating status of each node in the distribution transformer system, the dynamic changes in charging demand, and the real-time feedback of network status, resulting in insufficient precision and real-time performance in regulation. Second, traditional regulation methods lack effective utilization of meteorological and environmental information, failing to flexibly adjust charging loads based on weather changes and other factors. Furthermore, centralized control is prone to computational resource bottlenecks when facing large-scale charging equipment access, affecting system response speed and stability.

[0003] While some existing technologies attempt to improve the charging load regulation of distribution transformer systems by introducing the concepts of distributed control or edge computing, most of these methods focus on a single cloud or edge side, lacking a mechanism for cloud-edge-device collaboration. For example, some studies only perform centralized scheduling through the cloud side, neglecting the real-time feedback and autonomous regulation capabilities of the edge and device sides; others focus on local optimization at the edge side but fail to effectively integrate global information from the cloud side and real-time data from the device side, resulting in limited regulation effectiveness. Furthermore, existing technologies also have shortcomings in model updates and resource management, lacking effective mechanisms for model version management, dynamic invocation, and optimized resource allocation, making it difficult to meet the demands of complex and ever-changing distribution transformer system operating environments. Summary of the Invention

[0004] This invention provides a method for coordinated cloud-edge control of charging load in a distribution transformer system, comprising:

[0005] Obtain information on the operation of the distribution transformer system, charging demand, network status, and meteorological environment;

[0006] Data cleaning is performed on the distribution transformer system operation information, charging demand information, network status information, and meteorological environment information to obtain cleaned information;

[0007] The control scenario information is determined based on the cleaned information; the control scenario information includes the current control scenario, subsequent synchronization scenario, and predicted scenario.

[0008] The current control strategy model is generated based on the network topology model, cloud-side constraint information, edge-side resource information, and the current control scenario.

[0009] Obtain real-time feedback information from the edge;

[0010] The current control strategy model is locally modified based on real-time feedback information from the edge side to generate a modified current control strategy model.

[0011] Compare the current control scenario with the subsequent synchronization scenario. If the comparison result is a difference, generate a subsequent synchronization control strategy model based on the network topology model, cloud-side constraint information, edge-side resource information, and subsequent synchronization scenario. If the comparison result is the same, link the calling function for the subsequent synchronization control strategy model to the corrected current control strategy model.

[0012] Determine the load aggregation model corresponding to the operating information of the distribution transformer system;

[0013] The modified current control strategy model, subsequent synchronous control strategy model, and load aggregation model are invoked to perform real-time control of the distribution transformer system.

[0014] During the control process, end-side measurement information is continuously collected;

[0015] The load aggregation model is updated in a closed loop based on the end-side measurement information to obtain the updated load aggregation model.

[0016] Based on the updated load aggregation model, a second modification is performed on the current control strategy model and the subsequent synchronous control strategy model to complete real-time control.

[0017] Furthermore, the invocation of the revised current control strategy model, subsequent synchronous control strategy model, and load aggregation model includes:

[0018] Based on the pre-established mapping relationship between distribution transformer system operation and strategy activation and aggregation, determine the strategy activation speed, load aggregation model switching speed, model update cycle, cache refresh cycle and redundancy removal threshold corresponding to the distribution transformer system operation information;

[0019] The load aggregation model is updated based on the switching speed of the load aggregation model.

[0020] The activation order of the current regulation strategy model and the subsequent synchronous regulation strategy model is adjusted according to the activation speed control of the strategy.

[0021] Periodic recalculation is triggered based on the model update cycle;

[0022] Update the cache content according to the cache refresh cycle;

[0023] Identify and remove redundant strategy segments based on the redundancy removal threshold;

[0024] Perform consistency arbitration for policy conflicts during the sequential activation process;

[0025] Verify the legality of the arbitration result.

[0026] Furthermore, it also includes:

[0027] Obtain the original digital twin model; the original digital twin model includes the original load aggregation model and the original network topology model;

[0028] Perform a mesh simplification operation on each original digital twin model to obtain an initial lightweight model;

[0029] Remove the internal invisible structure of the initial lightweight model to obtain the intermediate model;

[0030] Perform a topology consistency check on the intermediate model to generate a model that passes the check;

[0031] Security hardening is performed on the verified model to generate a trusted model;

[0032] Perform digital signatures on the trusted model to generate a signature model;

[0033] Versioning is performed on the signature model to generate a versioned model;

[0034] Versioned models are stored in a model repository to complete model archiving.

[0035] Furthermore, the sequential activation of the current regulatory strategy model and the subsequent synchronous regulatory strategy model after control correction includes:

[0036] Get the current system clock;

[0037] The activation sequence is determined based on the current system clock;

[0038] After the control correction, the current control strategy model and the subsequent synchronous control strategy model are activated sequentially according to the activation order.

[0039] The control strategy model whose Euclidean distance from the center of the load aggregation model is less than the first distance threshold is preferentially activated;

[0040] Discontinue control strategy models whose distance from the center of the load aggregation model is greater than the second distance threshold;

[0041] Perform resource reclamation on the decommissioned model;

[0042] The recovered resources are then redistributed.

[0043] Record activation and deactivation logs.

[0044] Furthermore, activating the regulatory strategy model includes:

[0045] The resource allocation scheme for each node is determined based on its topological position and influence weight in the control environment.

[0046] Allocate computing resources to the control strategy model according to the resource allocation scheme;

[0047] Perform redundancy checks on the allocated resources to release redundant resources;

[0048] The released resources are then redistributed to optimize resource utilization.

[0049] Record resource allocation logs for subsequent auditing;

[0050] Perform anomaly detection on the audit results;

[0051] An alert will be sent to any detected anomalies;

[0052] Alarm information is archived and stored.

[0053] Furthermore, the invocation of the revised current control strategy model, subsequent synchronous control strategy model, and load aggregation model includes:

[0054] The content to be invoked during the control process is invoked using a dynamic invocation method;

[0055] Prioritize the execution of the content to be called;

[0056] Based on the priority ranking results, the corrected current control strategy model, the subsequent synchronous control strategy model, and the load aggregation model will be called first.

[0057] Perform cache status checks on dynamically accessed content to update the cache queue;

[0058] Perform aging processing on the updated cache queue to release expired content;

[0059] Perform secure erasure on expired content after release;

[0060] Perform defragmentation on the erased space;

[0061] Record cache operation logs.

[0062] Furthermore, performing mesh simplification operations on each original digital twin model includes:

[0063] The simplification ratio is determined based on the complexity level of the original digital twin model;

[0064] Redundant meshes from the original digital twin model are removed according to a simplified ratio to obtain an initial lightweight model;

[0065] Edge-preserving detection is performed on the initial lightweight model after deletion to repair topological cracks;

[0066] Perform normal consistency correction on the repaired model to eliminate lighting errors;

[0067] The corrected model is compressed to reduce storage overhead;

[0068] Perform an integrity check on the compressed model;

[0069] Encrypt the stored data for models that pass verification.

[0070] Record metadata for the simplified process.

[0071] Furthermore, the resource allocation scheme for each node in the model to be activated is determined based on its topological position and influence weight within the control environment, including:

[0072] Calculate the node distance weight based on the distance between the node and the center of the load aggregation model;

[0073] Calculate the influence weight of each node based on its impact on the control accuracy.

[0074] The node distance weight and the node influence weight are combined to obtain a comprehensive weight;

[0075] Allocate computing resources based on comprehensive weighting;

[0076] Perform load balancing checks on the allocated computing resources to adjust the resource allocation results;

[0077] Perform a performance evaluation on the adjusted results;

[0078] The strategy will be fine-tuned based on the assessment results;

[0079] Record fine-tuning logs.

[0080] Furthermore, the dynamic invocation method for calling the content to be invoked during the control process includes:

[0081] The cycle time is dynamically adjusted based on the frequency of changes in the distribution transformer system's operating information.

[0082] Update the call priority within each dynamically adjusted beat;

[0083] Prioritize calling the modified current control strategy model, the subsequent synchronous control strategy model, and the load aggregation model that are most relevant to the current control scenario;

[0084] Perform version consistency checks on the content that is called first to synchronize the model version;

[0085] Incremental synchronization is triggered after the verification passes to maintain data consistency;

[0086] Perform an integrity check on the synchronized data;

[0087] Execute an anomaly alarm on the verification results;

[0088] Record synchronization logs.

[0089] Furthermore, determining the load aggregation model corresponding to the distribution transformer system operation information includes:

[0090] The initial aggregation scale of the load aggregation model is determined based on the distribution range of charging demand information;

[0091] The initial aggregation scale is corrected based on network state information to obtain the target aggregation scale;

[0092] Construct a load aggregation model based on the target aggregation scale;

[0093] Perform boundary consistency checks on the constructed load aggregation model to correct the aggregation boundaries;

[0094] The revised load aggregation model is dynamically updated to match real-time charging demand information;

[0095] Record update logs for use in backtracking analysis;

[0096] Perform trend prediction on the results of the retrospective analysis;

[0097] Based on the prediction results, a backup aggregation model is pre-generated.

[0098] The embodiments of the present invention have at least the following beneficial effects:

[0099] 1. By acquiring and cleaning the operation information, charging demand information, network status information and meteorological environment information of the distribution transformer system, and generating control scenarios and control strategy models based on this information, the fine-grained control of the charging load of the distribution transformer system is realized.

[0100] 2. By adopting a cloud-edge-device collaborative control mechanism, the global information from the cloud side, the real-time feedback from the edge side, and the measurement data from the device side are organically combined, realizing the transformation from centralized control to distributed collaborative control.

[0101] 3. By introducing a dynamic update and closed-loop update mechanism for the load aggregation model, and combining the prediction and simulation scenarios with the pre-generation of the backup aggregation model, dynamic adaptation and forward-looking control of the charging load of the distribution transformer system are realized. Attached Figure Description

[0102] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent from the following detailed description taken in conjunction with the accompanying drawings. Several embodiments of the invention are illustrated in the drawings by way of example and not limitation, wherein:

[0103] Figure 1 This is a flowchart illustrating a cloud-edge-end coordinated control method for charging load in a distribution transformer system, as provided in an embodiment of the present invention. Detailed Implementation

[0104] The technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0105] like Figure 1 As shown, this application proposes a cloud-edge-end coordinated control method for charging load in a distribution transformer system, comprising:

[0106] S1. Obtain information on the operation of the distribution transformer system, charging demand, network status, and meteorological environment;

[0107] S2. Perform data cleaning on the distribution transformer system operation information, charging demand information, network status information and meteorological environment information to obtain cleaned information;

[0108] S3. Determine the control scenario information based on the cleaned information; the control scenario information includes the current control scenario, subsequent synchronization scenario, and prediction scenario.

[0109] S4. Generate the current control strategy model based on the network topology model, cloud-side constraint information, edge-side resource information, and the current control scenario;

[0110] S5. Obtain real-time feedback information from the edge side;

[0111] S6. Perform local corrections on the current control strategy model based on real-time feedback information from the edge side to generate a corrected current control strategy model;

[0112] S7. Compare the current control scenario with the subsequent synchronization scenario. If the comparison result is different, generate the subsequent synchronization control strategy model based on the network topology model, cloud-side constraint information, edge-side resource information and the subsequent synchronization scenario. If the comparison result is the same, link the calling function for the subsequent synchronization control strategy model to the corrected current control strategy model.

[0113] S8. Determine the load aggregation model corresponding to the operating information of the distribution transformer system;

[0114] S9. Invoke the corrected current control strategy model, subsequent synchronous control strategy model and load aggregation model to perform real-time control of the distribution transformer system;

[0115] S10. Continuously collect end-side measurement information during the control process;

[0116] S11. Perform closed-loop update on the load aggregation model based on the end-side measurement information to obtain the updated load aggregation model;

[0117] S12. Based on the updated load aggregation model, perform a second correction on the corrected current control strategy model and the subsequent synchronous control strategy model to complete real-time control.

[0118] First, we acquire information on the distribution transformer system's operation, charging demand, network status, and meteorological conditions. This information includes the real-time operating status of the distribution transformer system, user charging needs, the power grid's network topology, and current meteorological parameters. The purpose of acquiring this information is to gain a comprehensive understanding of the distribution transformer system's operating environment and influencing factors.

[0119] Next, data cleaning is performed on the acquired information. The data cleaning process includes removing outliers, filling in missing values, and standardizing data formats to ensure the quality of data for subsequent processing. The cleaned information is more accurate and reliable, providing a high-quality data foundation for subsequent scenario determination and model generation.

[0120] Based on the cleaned information, control scenario information is determined. This information includes the current control scenario, subsequent synchronization scenarios, and predictive scenarios. The current control scenario reflects the system's immediate state, subsequent synchronization scenarios predict potential short-term changes, and predictive scenarios analyze longer-term trends. This multi-scenario division helps in developing more comprehensive and forward-looking control strategies.

[0121] Based on the network topology model, cloud-side constraint information, edge-side resource information, and the current control scenario, a current control strategy model is generated. This model comprehensively considers the network structure, global constraints in the cloud, available resources on the edge, and the current operating scenario to formulate the most suitable control strategy for the current situation.

[0122] To adapt to real-time changes, the system acquires real-time feedback information from the edge and performs local corrections to the current control strategy model based on this information, generating a revised control strategy model. This real-time correction mechanism can quickly respond to changes at the edge, improving the flexibility and accuracy of control.

[0123] The system compares the current control scenario with subsequent synchronization scenarios. If there are differences, a subsequent synchronization control strategy model is generated based on the network topology model, cloud-side constraint information, edge-side resource information, and the subsequent synchronization scenario. If the two are the same, the function call for the subsequent synchronization control strategy model is linked to the corrected current control strategy model. This mechanism can both adapt to scenario changes and avoid unnecessary resource waste.

[0124] A load aggregation model corresponding to the distribution transformer system's operating information is determined. This model reflects the aggregation characteristics of each load in the system. Then, the modified current control strategy model, subsequent synchronization control strategy model, and load aggregation model are invoked to execute real-time control of the distribution transformer system.

[0125] During the control process, the system continuously collects end-side measurement information. This information is used to perform closed-loop updates on the load aggregation model, resulting in an updated load aggregation model. The updated load aggregation model more accurately reflects the real-time load situation.

[0126] Finally, based on the updated load aggregation model, a second revision is performed on the modified current control strategy model and the subsequent synchronous control strategy model to complete real-time control. This multi-level revision mechanism ensures that the control strategy can continuously adapt to the dynamic changes of the system.

[0127] Data cleaning refers to the preprocessing of raw data to remove noise, fill in missing values, and correct errors. This can be achieved using data filtering, interpolation algorithms, and anomaly detection algorithms to ensure the reliability of the data relied upon for subsequent analysis. Control scenario information refers to different control stages defined by real-time and historical data. This can be achieved using cluster analysis, time series forecasting, and pattern recognition algorithms to dynamically adapt to the operational needs of the distribution transformer system at different time scales. The network topology model refers to structured data describing the connection relationships between devices in the distribution transformer system. This can be achieved using graph theory modeling, adjacency matrices, and node-edge relationship databases to provide a physical constraint basis for strategy generation. Real-time feedback information from the edge refers to the second- or millisecond-level operational data collected from distributed terminal devices. This can be achieved using real-time data stream processing technologies from IoT sensors and edge computing nodes to support dynamic correction of the strategy model.

[0128] Local correction refers to adjusting the parameters of the strategy model based on real-time data. This can be achieved using online learning algorithms and incremental optimization algorithms to improve the adaptability of the control strategy to sudden operating conditions. Load aggregation modeling integrates dispersed charging loads into controllable units based on their spatiotemporal characteristics. This can be achieved using load clustering algorithms, spatiotemporal correlation analysis, and distributed energy aggregation technology to reduce control complexity. Closed-loop updating refers to iteratively optimizing the model through continuous measurement data. This can be achieved using feedback control theory, recursive least squares, and model predictive control techniques to ensure consistency between the model and the actual system state.

[0129] Secondary correction refers to adjusting the strategy parameters in reverse based on the updated load model. Specifically, it can be achieved using backpropagation algorithms and multi-objective optimization techniques to realize dynamic matching between the strategy and load characteristics.

[0130] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0131] In a city's smart grid distribution transformer system, multiple data acquisition points are deployed, including distribution transformer monitoring devices, smart meters, a charging pile management system, and a weather station. These devices collect information on distribution transformer system operation, user charging demand, network status, and meteorological conditions. The data acquisition frequency is set to once every 5 minutes, and the data is transmitted to the cloud data center via a secure and encrypted channel.

[0132] The data cleaning module uses outlier detection algorithms and interpolation techniques to process the raw data. For example, abnormal voltage readings are smoothed using a moving average method; missing charging demand data is filled in using historical data from the same period. The cleaned data is stored in a distributed database for fast access.

[0133] The scene recognition module uses machine learning algorithms, such as decision trees or random forests, to determine the control scene based on cleaned data. The current control scene reflects the real-time state, the subsequent synchronous scene predicts changes within the next 30 minutes, and the predictive extrapolation scene analyzes the next 24 hours.

[0134] The regulation strategy generation module, based on a deep reinforcement learning algorithm, combines network topology model, cloud-side constraint information (such as power grid safety operation limits), edge-side resource information (such as adjustable charging pile capacity), and the current regulation scenario to generate the current regulation strategy model. This model aims to minimize distribution network losses and maximize the charging demand satisfaction rate.

[0135] Edge computing nodes collect data such as charging pile power and voltage in real time, serving as real-time feedback information from the edge. Based on this information, the cloud uses an incremental learning algorithm to locally revise the current control strategy model, generating a revised control strategy model.

[0136] The scene comparison module uses similarity calculation methods, such as cosine similarity, to compare the current control scene with subsequent synchronization scenes. If the similarity is lower than a preset threshold, it triggers the generation of a subsequent synchronization control strategy model; otherwise, it calls a function to link to the corrected current control strategy model.

[0137] The load aggregation model is based on clustering algorithms, such as K-means or DBSCAN, to cluster charging demand into different load aggregation types. This model is updated periodically to adapt to changes in charging patterns.

[0138] The real-time control execution module invokes the revised current control strategy model, subsequent synchronous control strategy model, and load aggregation model to generate specific control instructions, such as adjusting charging power and starting / stopping charging piles. These instructions are then distributed to each charging device via edge computing nodes.

[0139] The end-side measurement information includes parameters such as real-time power, voltage, and current of the charging pile, with a sampling frequency of once per second. This data is used to perform online learning and updates on the load aggregation model, resulting in an updated load aggregation model.

[0140] Finally, based on the updated load aggregation model, model fusion technology is used to perform a second correction on the revised current control strategy model and the subsequent synchronous control strategy model, completing a full real-time control cycle. The entire process is executed cyclically to ensure continuous optimization of the control strategy.

[0141] A control strategy model is generated based on the current control scenario, and local corrections are made through real-time feedback from the edge side, enhancing the real-time performance and adaptability of the control. Simultaneously, subsequent synchronization scenarios are predicted and backup models are built, improving the system's response speed to scenario changes. The dynamic update mechanism of the load aggregation model, combined with closed-loop calibration of edge-side measurement information, makes the prediction of charging demand more accurate.

[0142] The bidirectional correction mechanism between the control strategy model and the load model ensures continuous matching between control commands and real-time operating conditions, effectively avoiding control deviations caused by data lag in traditional methods. This cloud-edge-device collaborative approach significantly improves the control accuracy and response speed of the distribution transformer system to charging loads, effectively alleviates local overload problems, and improves the utilization efficiency of charging resources.

[0143] When facing weather-sensitive scenarios, this method can quickly adjust strategies, reduce voltage fluctuations, and lower the risk of power outages. By optimizing the allocation of computing resources, it solves the problem of idle edge computing resources and overloaded cloud computing in traditional methods, improving the overall system efficiency. The closed-loop update mechanism of edge measurement information ensures continuous optimization of the control model and avoids the problem of control accuracy decaying over time.

[0144] This application further proposes to determine the strategy activation speed, load aggregation model switching speed, model update cycle, cache refresh cycle, and redundancy removal threshold corresponding to the distribution transformer system operation information based on the pre-established mapping relationship between distribution transformer system operation and strategy activation and aggregation; control the load aggregation model update according to the load aggregation model switching speed; control the sequential activation of the corrected current control strategy model and the subsequent synchronous control strategy model according to the strategy activation speed; trigger periodic recalculation according to the model update cycle; update the cache content according to the cache refresh cycle; identify and remove redundant strategy fragments according to the redundancy removal threshold; perform consistency arbitration for strategy conflicts in the sequential activation process; and perform legality verification on the arbitration results.

[0145] The strategy activation speed dynamically adjusts the model activation interval through time parameters in the mapping relationship; the load aggregation model switching speed sets a switching delay threshold based on the difference between models; the model update cycle uses a fixed time window to trigger model parameter recalculation; the cache refresh cycle sets the cache expiration time according to the data update frequency; and the redundancy removal threshold is set as the strategy duplication percentage threshold. Consistency arbitration adopts a multi-strategy priority ranking mechanism, and legality verification matches strategy compliance through a preset rule base.

[0146] Specifically, a pre-established mapping relationship linearly correlates load volatility, network latency parameters, and policy activation speed in the distribution transformer system's operating information. When load volatility exceeds a threshold, the policy activation speed is increased proportionally to reduce control latency. The load aggregation model switching speed is dynamically adjusted based on the difference between models; switching is delayed when the difference is below a set threshold to avoid frequent jitter. The model update cycle is set to a millisecond-level time window, periodically triggering model parameter recalculation to match real-time data. The cache refresh cycle is dynamically adjusted based on the data update frequency, with higher-frequency updates corresponding to shorter cache expiration times. The redundancy removal threshold is set to trigger automatic cleanup when policy duplication exceeds 70%. During consistency arbitration, policy segments closest to the load aggregation model center are retained first, and conflicting policies are selectively merged after being sorted by timestamp. In the legality verification phase, the arbitrated policies must pass preset voltage fluctuation range and equipment capacity limit rules. If verification fails, the policy triggers an alarm and rolls back to the previous valid version.

[0147] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0148] Based on the pre-established mapping relationship between distribution transformer system operation and strategy activation / aggregation, the strategy activation speed, load aggregation model switching speed, model update cycle, cache refresh cycle, and redundancy removal threshold corresponding to the distribution transformer system operation information are determined. Specifically, a multi-dimensional mapping table can be established to associate the distribution transformer system's operating parameters, such as load factor, voltage deviation, and power factor, with parameters such as strategy activation speed and load aggregation model switching speed. For example, when the load factor exceeds 90%, the strategy activation speed can be set to 10 times per second, the load aggregation model switching speed to once per minute, the model update cycle to once every 5 minutes, the cache refresh cycle to once every 30 seconds, and the redundancy removal threshold to 10%.

[0149] The load aggregation model is updated based on its switching speed. For example, when the switching speed of the load aggregation model is detected to be once per minute, the system will trigger an update operation on the load aggregation model every hour on the hour to ensure that the model can reflect the latest load distribution in a timely manner.

[0150] The activation sequence of the current control strategy model and the subsequent synchronous control strategy model is controlled according to the strategy activation speed. For example, when the strategy activation speed is 10 times per second, the system will sequentially activate one modified current control strategy model and one subsequent synchronous control strategy model every 100 milliseconds, ensuring that the two models can be executed alternately to achieve a balance between real-time control and future prediction.

[0151] Periodic recalculations are triggered based on the model update cycle. For example, when the model update cycle is set to once every 5 minutes, the system will trigger a comprehensive model recalculation every 5 minutes on the hour, including the load aggregation model, the control strategy model, etc., to ensure the accuracy and timeliness of the model.

[0152] The cache content is updated according to the cache refresh cycle. For example, when the cache refresh cycle is set to once every 30 seconds, the system will perform a full update of the data in the cache every 30 seconds, including the latest distribution transformer system operating data, charging demand information, etc., to ensure that control decisions are based on the latest data.

[0153] Redundant policy fragments are identified and removed based on a redundancy removal threshold. For example, when the redundancy removal threshold is set to 10%, the system analyzes all active policies. If a policy fragment's execution frequency is less than 10%, or its contribution to the control effect is less than 10%, it is marked as a redundant policy and removed from the execution queue.

[0154] Consistency arbitration is performed to resolve policy conflicts during sequential activation. For example, when two consecutively activated policy models issue conflicting control commands to the same charging pile, the system will decide which command to execute based on preset arbitration rules (such as prioritizing grid safety).

[0155] The system verifies the legality of the arbitration results. For example, after arbitration, the system checks whether the control commands issued after arbitration comply with the safety constraints of the power grid operation and the technical parameter limitations of the charging equipment. If any illegal commands are found, an alarm will be triggered and manual intervention will be required.

[0156] This application further proposes obtaining the original digital twin model, which includes the original load aggregation model and the original network topology model; performing a mesh simplification operation on each original digital twin model to obtain an initial lightweight model; deleting the internal invisible structure of the initial lightweight model to obtain an intermediate model; performing topology consistency verification on the intermediate model to generate a verification-passed model; performing security hardening processing on the verification-passed model to generate a trusted model; performing digital signing on the trusted model to generate a signed model; performing version marking on the signed model to generate a versioned model; and storing the versioned model in a model repository to complete model archiving.

[0157] Mesh simplification involves setting a simplification ratio corresponding to the complexity level, such as reducing the number of polygonal faces to 30%-50% of the original model, removing redundant meshes while retaining key topological structures. Internal invisible structures refer to supporting frames or hidden layers within the model that do not affect the external morphology; these are identified and removed using 3D model analysis tools. Topological consistency verification uses graph theory algorithms to verify that the connectivity relationships of the simplified model are consistent with the original model, ensuring mesh closure after crack repair. Security hardening includes embedding tamper-proof watermarks and access control policies in the model file, and using asymmetric encryption algorithms to generate unique identifiers for digital signatures. Version marking generates version numbers by combining timestamps and hash values, and the model repository uses a distributed storage architecture for multi-copy backup.

[0158] Specifically, after mesh simplification, the data volume of the original digital twin model is reduced to 40% of its original size; for example, a model containing 1 million triangular faces is reduced to 400,000. Removing internal invisible structures further reduces the data volume by 15%, such as removing the internal winding support structure of a transformer. Topology consistency verification ensures an error rate below 0.5% by comparing the connectivity of simplified nodes. In security hardening, tamper-proof watermarks are embedded in the model's texture coordinates using steganography, and access control policies restrict unauthorized devices' access to the model. Digital signatures use the RSA algorithm to generate 2048-bit key pairs, and the signature information is bound to the model file for storage. Version marking uses a year-month-day-serial number format, such as 20231025-003, and each version is stored independently in the model repository with an index directory. During archiving, versioned models are divided into storage nodes by region; for example, models for East China are stored on a dedicated server cluster, and data consistency is ensured through a periodic synchronization mechanism.

[0159] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0160] Obtain the original digital twin model, including the original load aggregation model and the original network topology model. The original load aggregation model contains detailed parameter information for charging stations, distribution transformers, and distribution lines, while the original network topology model contains the complete connectivity of the distribution network.

[0161] Mesh simplification was performed on each original digital twin model to obtain an initial lightweight model. Mesh simplification included steps such as removing redundant nodes, merging adjacent edges, and reducing surface precision, reducing model complexity by 50%.

[0162] Remove the internal invisible structures of the initial lightweight model to obtain the intermediate model. Internal invisible structures include hidden auxiliary geometry, unused reference planes, etc.

[0163] Perform a topology consistency check on the intermediate model to generate a model that passes the check. The topology consistency check includes checking the model's boundary integrity, surface orientation consistency, and geometric feature continuity.

[0164] The verified model undergoes security hardening to generate a trusted model. Security hardening includes measures such as data encryption, access control, and integrity protection.

[0165] A digital signature is performed on the trusted model to generate a signature model. The digital signature uses an asymmetric encryption algorithm to ensure the authenticity and non-repudiation of the model.

[0166] Versioning is performed on the signature model to generate a versioned model. Versioning includes information such as version number, creation time, and modification history.

[0167] Versioned models are stored in a model repository to complete model archiving. The model repository adopts a distributed storage architecture, supporting version control and collaborative management.

[0168] This application further proposes the following steps: obtaining the current system clock; determining the activation sequence based on the current system clock; controlling the current control strategy model and subsequent synchronous control strategy models to be activated sequentially according to the activation sequence after control correction; prioritizing the activation of control strategy models whose Euclidean distance from the load aggregation model center is less than a first distance threshold; deactivating control strategy models whose distance from the load aggregation model center is greater than a second distance threshold; performing resource reclamation on the deactivated models; performing reallocation on the reclaimed resources; and recording activation and deactivation logs.

[0169] The activation timing is determined based on the system clock to ensure synchronization with the system's operating status. Priority activation conditions are set by using a Euclidean distance threshold to filter control strategy models with high correlation to the load aggregation model, while deactivation conditions are set by using a distance threshold to identify models with low correlation. Resource reclamation and reallocation optimize resource utilization by releasing computing resources occupied by deactivated models and reallocating them to activated models. Activation and deactivation logs are recorded for subsequent auditing and analysis.

[0170] Specifically, the system clock is acquired in real time, and an activation time sequence queue is generated based on this clock. Control strategy models are activated sequentially according to the time sequence queue, ensuring that the execution order is synchronized with the system's operating rhythm. Priority activation is determined by calculating the Euclidean distance between the model and the load aggregation model center. If the distance is less than a preset first threshold, the model is marked as high priority and activated first. Deactivation is determined by comparing the distance with a second threshold. If the distance exceeds the second threshold, the model stops running and releases its occupied computing resources. The released resources are reallocated to other activated models, improving overall resource utilization. The status information of activation and deactivation operations is recorded in a log file, which includes a timestamp, model identifier, and resource allocation details for subsequent performance analysis and anomaly troubleshooting.

[0171] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0172] Obtain the current system clock, for example, by reading the system timestamp to obtain the current time accurate to milliseconds. Determine the activation sequence based on the obtained current system clock. A time window method can be used, dividing a control cycle into multiple time windows, with each window corresponding to a set of policy models to be activated.

[0173] After control correction, the current control strategy model and subsequent synchronous control strategy models are activated sequentially according to a determined activation order. During the activation process, the Euclidean distance between each control strategy model and the center of the load aggregation model is first calculated. Control strategy models with a distance less than a first distance threshold (e.g., 10 meters) are activated first, and a multi-threaded parallel approach can be used to accelerate the activation process.

[0174] Simultaneously, control strategy models whose distance from the center of the load aggregation model is greater than the second distance threshold (e.g., 100 meters) are deactivated. Resource reclamation is performed on deactivated models, including releasing memory and shutting down related processes. The reclaimed computing and storage resources are then reallocated, using dynamic resource allocation algorithms to allocate resources to higher-priority tasks.

[0175] Finally, activation and deactivation logs are recorded, including model ID, activation / deactivation time, resource usage, and other information. The logs are stored in a distributed manner to ensure reliability and traceability.

[0176] This application further proposes a resource allocation scheme for each node in the control environment based on its topological position and influence weight within the control environment; allocates computing resources to the control strategy model according to the resource allocation scheme; performs redundancy detection on the allocated resources to release redundant resources; performs secondary allocation on the released resources to optimize resource utilization; records resource allocation logs for subsequent auditing; performs anomaly detection on the audit results; pushes alarms to detected anomalies; and archives and stores alarm information.

[0177] The node distance weight is calculated based on the Euclidean distance between the node and the load aggregation model center, while the node influence weight is determined according to the node's impact on control accuracy. The comprehensive weight is obtained by fusing the distance and influence weights and is used to calculate resource allocation. Redundancy detection identifies and releases underutilized resources, and secondary allocation redistributes these released resources to high-weight nodes. Resource allocation logs record details of each allocation, anomaly detection identifies allocation deviations through log analysis, alarm pushes send anomaly information to the management terminal, and archiving stores categorized alarm information.

[0178] Specifically, node distance weight is calculated using an inverse proportional function; for example, the weight decreases by a preset percentage for every unit increase in distance. Node influence weight is determined based on the node's impact coefficient on accuracy in historical control data; for example, nodes with high influence are assigned higher weights. The comprehensive weight combines the two weights through a weighted summation method; for example, distance weight accounts for 30%, and influence weight accounts for 70%. The resource allocation scheme allocates calculated resources proportionally based on the comprehensive weight; for example, nodes with a comprehensive weight of 0.8 receive 40% of the total resources. Redundancy detection triggers release when resource utilization falls below a set threshold; for example, CPU utilization below 20% for 5 minutes is considered redundant. Secondary allocation prioritizes the allocation of released resources to nodes ranked in the top 10% by comprehensive weight. The resource allocation log records timestamps, node identifiers, allocated resource amounts, and weight parameters. Anomaly detection triggers an alarm when the deviation between the actual and expected allocation exceeds 10%. Alarm information is stored according to severity level; for example, high-priority alarms are retained for 90 days, and low-priority alarms for 30 days.

[0179] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0180] When activating the control strategy model, the resource allocation scheme for each node is first determined based on its topological location and influence weight within the control environment. Specifically, the system calculates the Euclidean distance between each node and the center of the load aggregation model, mapping this distance to a weight value between 0 and 1, which serves as the node's distance weight. Simultaneously, the system analyzes the impact of each node on control accuracy based on historical data, quantifying this impact into a weight value between 0 and 1, which serves as the node's influence weight.

[0181] Furthermore, the system calculates a weighted average of the node distance weight and the node influence weight to obtain the overall weight for each node. For example, the overall weight can be calculated using a weighting of 0.4 for distance and 0.6 for influence. Based on the calculated overall weight, the system allocates available computing resources proportionally.

[0182] Therefore, load balancing verification is performed on the allocated computing resources. Specifically, the system will detect indicators such as CPU utilization, memory usage, and network bandwidth of each node. If the load of a node exceeds a preset threshold, such as 80%, the resource allocation will be adjusted appropriately, transferring some resources from high-load nodes to low-load nodes.

[0183] The system will perform a performance evaluation on the adjusted resource allocation results. For example, it can simulate a set of standard test cases, record metrics such as response time and throughput, and compare them with preset performance targets. Based on the evaluation results, the system will perform policy fine-tuning, such as appropriately increasing resource quotas for key nodes or optimizing task scheduling algorithms.

[0184] As a preferred implementation, the system records a complete log of resource allocation, load balancing, performance evaluation, and policy fine-tuning. These logs include timestamps, operation types, involved nodes, and parameters before and after adjustments, for subsequent auditing and optimization analysis.

[0185] This application further proposes a method of dynamically invoking the content to be invoked during the control process; performing priority sorting on the content to be invoked; invoking the corrected current control strategy model, subsequent synchronous control strategy model, and load aggregation model according to the priority sorting result; performing cache status detection on the dynamically invoked content to update the cache queue; performing aging processing on the updated cache queue to release expired content; performing secure erasure on the released expired content; performing defragmentation on the erased space; and recording cache operation logs.

[0186] The dynamic invocation method determines the cycle time based on the frequency of changes in the distribution transformer system's operating information, updates the invocation priority within each cycle time, and prioritizes the model with the highest relevance to the current control scenario. Priority sorting is based on the relevance of the model to the current scenario, ensuring that critical models are allocated computing resources first. Cache status detection identifies expired or low-priority content by periodically scanning the cache queue. Aging processing uses timestamps or access frequency thresholds to determine the validity of content and automatically removes expired data. Secure erasure ensures that sensitive information is unrecoverable through data overwriting or encryption. Defragmentation continuously reorganizes the released storage space to improve subsequent storage efficiency. Cache operation logs record the time, content, and execution results of each cache update, erasure, and defragmentation operation.

[0187] Specifically, during dynamic invocation, the system generates a dynamic adjustment cycle based on the frequency of changes in runtime information, such as updating the invocation priority every 5 seconds to ensure that model invocation is synchronized with the real-time scenario. The priority sorting module marks the current control strategy model after correction as the highest priority, followed by subsequent synchronization control strategy models, and the load aggregation model dynamically adjusts its priority according to its aggregation scale. The cache status detection module scans the cache queue every adjustment cycle to identify content that has not been accessed within a preset time; for example, model fragments that have not been invoked for more than 30 seconds are marked as expired. The aging process module automatically removes expired content based on the marking results, releasing storage resources. The secure erasure module performs three data overwrite operations on the removed content to prevent leakage of residual information. The defragmentation module merges the released storage blocks into contiguous space to reduce storage fragmentation. The cache operation log records detailed information for each operation, such as erasure time, number of overwrites, and space utilization after defragmentation, providing data support for subsequent auditing. Through the above steps, the system achieves efficient resource utilization during invocation, avoids redundant data accumulation, ensures the security of sensitive information, and improves the real-time performance and stability of overall control.

[0188] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0189] In the cloud-edge-device collaborative control of charging load in the distribution transformer system, a dynamic invocation method is used to call up content to be invoked during the control process. Specifically, the content to be invoked is first prioritized. The prioritization criteria include, but are not limited to, the timeliness, relevance, and computational complexity of the content. For example, content with high timeliness, high relevance to the current control scenario, and low computational complexity can be given higher priority.

[0190] Based on the priority ranking, the revised current control strategy model, subsequent synchronous control strategy models, and load aggregation models are invoked first. These models are considered core components in the control process, and prioritizing their invocation ensures the timeliness and accuracy of control.

[0191] Furthermore, cache status checks are performed on dynamically accessed content to update the cache queue. Cache status checks include checking metrics such as content validity, access frequency, and memory usage. Based on the check results, the cache queue is updated, retaining frequently accessed content that is still valid in the cache, while removing infrequently accessed or expired content from the cache.

[0192] Perform aging processing on the updated cache queue to release expired content. Aging processing can adopt a time-based or access frequency-based strategy, such as setting an aging threshold, and marking content as expired when its existence time exceeds the threshold or its access frequency falls below a certain value.

[0193] Perform a secure erase on the expired content after it has been released. Secure erase can employ a multi-overwrite method to ensure that sensitive data cannot be recovered. For example, random data can be used to overwrite the storage space occupied by the expired content multiple times.

[0194] Defragment the erased space. Defragmentation can employ algorithms such as contiguous allocation or best-matching to merge scattered free spaces, thereby improving storage utilization and subsequent allocation efficiency.

[0195] Finally, log the cache operations. The log content includes, but is not limited to, information such as operation type, operation time, content identifiers involved, and operation results, for subsequent auditing and performance analysis.

[0196] This application further proposes performing mesh simplification operations on each original digital twin model, including: determining a simplification ratio based on the complexity level of the original digital twin model; deleting redundant meshes from the original digital twin model according to the simplification ratio to obtain an initial lightweight model; performing edge-preserving detection on the initial lightweight model after deletion to repair topological cracks; performing normal consistency correction on the repaired model to eliminate lighting errors; performing compression encoding on the corrected model to reduce storage overhead; performing integrity verification on the compressed model; performing encrypted storage on the verified model; and recording metadata of the simplification process.

[0197] The complexity level is dynamically determined by the model mesh density and the number of topological connections. For example, when the mesh density is higher than a threshold and the number of connections exceeds a preset range, it is defined as a high complexity level, corresponding to a higher simplification ratio. Edge preservation detection uses a curvature continuity algorithm to identify and fill mesh cracks, ensuring a smooth transition between adjacent mesh boundaries. Normal consistency correction unifies the normal vector direction of adjacent faces to avoid abnormal lighting rendering caused by normal misalignment. Compression encoding uses an octree-based spatial partitioning method to merge repetitive geometric structures into a single instance. Encrypted storage uses an asymmetric encryption algorithm to protect model data. Metadata records include simplification ratio, repair region coordinates, and compression ratio parameters.

[0198] Specifically, in the simplification ratio determination stage, the complexity level is dynamically classified based on the mesh density and topological connectivity of the original model. For example, models with a mesh density exceeding 1000 faces per cubic meter are classified as high complexity and allocated a simplification ratio of 60%. When deleting redundant meshes, internal repetitive structures and flat regions with curvature below a threshold are removed first, while high-curvature edge regions are preserved. In the topological crack repair stage, the difference in the second derivative of the curvature of adjacent mesh boundaries is calculated to identify fracture edges and insert transition faces. For example, a repair mechanism is triggered when the curvature difference exceeds 0.5. In the normal consistency correction stage, a region growing algorithm is used to control the difference in normal vector angle between adjacent faces to within 5 degrees, eliminating abrupt changes in illumination. In the compression encoding stage, recurring columnar structures are merged into axial instances, reducing storage space by 70%. Integrity verification ensures that the model data has not been tampered with through hash value comparison, and elliptic curve cryptography is used to protect the model files for encrypted storage. Metadata records include operation timestamps for each stage, operator identification, and processing parameters, forming a traceable simplified process chain.

[0199] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0200] Performing mesh simplification operations on each original digital twin model includes:

[0201] The simplification ratio is determined based on the complexity level of the original digital twin model. For example, a simplification ratio of 50% is set for a high complexity level model; 30% for a medium complexity level model; and 10% for a low complexity level model.

[0202] Redundant meshes from the original digital twin model are removed according to a simplification ratio to obtain an initial lightweight model. Specifically, a quadrilateral mesh simplification algorithm is used to progressively merge adjacent meshes according to a set simplification ratio until the target simplification level is achieved.

[0203] Edge-preserving detection is performed on the initial lightweight model after deletion to repair topological cracks. An edge detection algorithm is used to identify key edge features of the model, and local mesh reconstruction is performed on detected crack regions.

[0204] Perform normal consistency correction on the repaired model to eliminate lighting errors. Calculate the normal vector for each vertex, and for vertices with significant deviations in normal direction, adjust their normal direction to align with the surrounding vertices.

[0205] The corrected model is compressed to reduce storage overhead. Lossless compression algorithms are used to encode the model data, such as Huffman coding to compress vertex coordinates and topological information.

[0206] Perform an integrity check on the compressed model. Calculate the checksum of the model data and compare it with the original model to ensure that the simplification process did not result in data loss or errors.

[0207] Encrypt the models that pass the verification process. Use the AES-256 encryption algorithm to encrypt the model data and generate an encryption key.

[0208] Record metadata about the simplification process. This includes key information such as the original model information, simplification ratio, edge preservation parameters, normal correction threshold, compression ratio, and verification results, for subsequent traceability and analysis.

[0209] This application further proposes a resource allocation scheme for each node based on its topological location and influence weight within the control environment of the model to be activated. This includes: calculating node distance weights based on the distance between the node and the center of the load aggregation model; calculating node influence weights based on the degree of influence of the node on the control accuracy; merging node distance weights and node influence weights to obtain a comprehensive weight; allocating computing resources based on the comprehensive weight; performing load balancing verification on the allocated computing resources to adjust the resource allocation results; performing performance evaluation on the adjusted results; performing strategy fine-tuning based on the evaluation results; and recording fine-tuning logs.

[0210] The node distance weight can be calculated using an inverse proportional function, where nodes closer to the load aggregation model center receive higher weight values. The node influence weight is calculated based on quantifying the contribution of node state changes to system error in historical control data, for example, by using linear regression to establish a mapping relationship between error and node state. The fusion of comprehensive weights uses a weighted summation method, where the ratio coefficients of distance weight and influence weight are dynamically adjusted according to the real-time network topology. Load balancing verification is achieved by monitoring the CPU utilization and memory usage of computing nodes; resource reallocation is triggered when node resource utilization exceeds a preset threshold. Performance evaluation uses both control response time and error rate as quantitative indicators, and strategy fine-tuning optimizes weight allocation parameters using a gradient descent algorithm.

[0211] Specifically, when determining the resource allocation scheme, the spatial coordinate data of each node in the model to be activated is first obtained, and the Euclidean distance between it and the center point of the load aggregation model is calculated. The distance values ​​are normalized and converted into distance weight coefficients within the range of 0-1. Simultaneously, the correlation data of each node with voltage fluctuations and power deviations during historical control cycles is extracted, and principal component analysis is used to calculate the influence weight coefficients. The two weights are then merged according to a preset ratio to generate a comprehensive weight matrix, which serves as the basis for resource allocation. The resource allocator allocates resources proportionally based on the comprehensive weight values, prioritizing the allocation of more CPU cores and memory capacity to high-weight nodes. After allocation, the monitoring system collects resource utilization data for each node in real time. When a load difference between nodes exceeds 15%, a resource reallocation mechanism is automatically triggered. During reallocation, a minimum migration cost algorithm is used to adjust resource distribution, ensuring that the overall system load balance remains above 90%. Detailed fine-tuning parameters and performance indicators are recorded after each resource adjustment, forming a traceable log file for subsequent optimization analysis.

[0212] As a preferred embodiment, the specific implementation of this application's solution is as follows: During the charging load control process of the distribution transformer system, when determining the resource allocation scheme for each node, firstly, the Euclidean distance of each node is calculated using real-time collected node coordinate data and the center coordinates of the load aggregation model, generating a distance weight coefficient. Further, based on the node's historical control records and real-time operating parameters, its contribution to control accuracy is analyzed, generating an influence weight coefficient. The distance weight coefficient and influence weight coefficient are merged according to a preset ratio to generate a comprehensive weight index. Based on the comprehensive weight index, computing resources are divided into different priority intervals, allocating more computing resources to nodes in higher priority intervals. After resource allocation is completed, load balancing verification is performed by real-time monitoring of the computing load and response latency of each node. If uneven load distribution is found, the resource allocation ratio is dynamically adjusted. After adjustment, simulation testing is used to evaluate the response speed and convergence of the control strategy, and the resource allocation strategy is fine-tuned based on the evaluation results. All allocation adjustment operations generate log records for subsequent auditing and analysis.

[0213] This application further proposes to determine the resource allocation scheme for each node based on its topological position and influence weight in the control environment of the model to be activated, allocate computing resources to the control strategy model according to the resource allocation scheme, perform load balancing verification on the allocated computing resources to adjust the resource allocation results, perform performance evaluation on the adjusted results, perform strategy fine-tuning based on the evaluation results, and record fine-tuning logs.

[0214] The node distance weight is calculated by measuring the Euclidean distance between the node and the load aggregation model center and applying an inverse proportional function. The node influence weight is calculated based on the node's contribution to voltage fluctuations and load balance indicators in historical control data. The comprehensive weight uses a linear weighted fusion algorithm, and the weight coefficients are dynamically adjusted according to the real-time network status. Load balancing verification triggers resource reallocation by monitoring the CPU and memory usage thresholds of computing nodes. Performance evaluation uses a dual-indicator evaluation system of control error rate and response latency. Strategy fine-tuning is achieved by adjusting the weight coefficients or reallocating the number of computing cores. Fine-tuning logs record timestamps, node identifiers, adjustment parameters, and performance indicators.

[0215] Specifically, during the activation of the control strategy model, the physical coordinates of each node in the distribution network topology are first obtained, and its Euclidean distance from the load aggregation model center is calculated. For example, when a node is less than 50 meters from the model center, a distance weighting coefficient of 0.8 is assigned; when the distance exceeds 200 meters, the coefficient drops to 0.2. Simultaneously, the impact of the node on voltage stability in historical control is analyzed. If its control action has reduced voltage fluctuations by more than 15%, an impact weighting coefficient of 0.9 is assigned. After merging the two weights in a 7:3 ratio, dual-core computing resources are prioritized for nodes with a comprehensive weight higher than 0.75. After the initial allocation, the computing load of each node is monitored in real time. When the CPU utilization of a node exceeds 85% for three consecutive sampling periods, 10% of the computing resources are automatically allocated from its neighboring nodes. After each resource adjustment, it is evaluated whether the control error has decreased by more than 5% and the response latency has shortened by more than 20 milliseconds. If the expected results are not achieved, the weight fusion ratio is dynamically adjusted to 6:4, and the parameter change records in the fine-tuning log are updated. This dynamic weight fusion and resource reallocation mechanism ensures that computing resources are always concentrated on high-impact nodes in key topological regions, thereby improving control accuracy and resource utilization efficiency.

[0216] As a preferred embodiment, the solution of this application is implemented as follows: During the operation of the distribution transformer system, power demand data and geographical location information of charging piles are first collected, and an initial aggregation scale is generated based on the charging demand density distribution. The initial aggregation scale is then dynamically adjusted according to the communication network latency and bandwidth occupancy rate, splitting the aggregation units corresponding to high-latency areas into multiple sub-units, while low-latency areas are merged into larger-scale aggregation units. When constructing the load aggregation model, a geographic grid-based clustering algorithm is used to delineate the aggregation boundaries, and model updates are triggered by real-time monitoring of charging power fluctuations. When a sudden change in charging demand in a certain area is detected to exceed a set threshold, a backup aggregation model covering three adjacent grids is automatically generated, while the original model is retained as a historical version. After each model update, the system automatically records the timestamp and changed parameters, forming a traceable version chain for subsequent analysis and retrieval.

[0217] This application further proposes a load aggregation model corresponding to the operation information of the distribution transformer system, including: determining the initial aggregation scale of the load aggregation model based on the distribution range of charging demand information; correcting the initial aggregation scale based on network status information to obtain a target aggregation scale; constructing a load aggregation model based on the target aggregation scale; performing boundary consistency checks on the constructed load aggregation model to correct the aggregation boundary; performing dynamic updates on the corrected load aggregation model to match real-time charging demand information; recording update logs for retrospective analysis; performing trend prediction on the retrospective analysis results; and pre-generating a backup aggregation model based on the prediction results.

[0218] The initial aggregation scale is determined by analyzing the geographical distribution density and capacity demand range of charging demand information, for example, by using a spatial grid partitioning method to count the total charging power demand in each region. The network status information is used to correct the initial aggregation scale by adjusting the granularity of the aggregation region division based on communication latency and bandwidth limitations, specifically by introducing network topology connectivity parameters to reduce the aggregation range of high-latency regions. The target aggregation scale is constructed using a hierarchical clustering algorithm, dividing charging nodes into multiple aggregation units according to the corrected scale. Boundary consistency detection compares the rate of change of charging demand within the aggregation unit with a preset threshold; when the rate of change exceeds the threshold, the aggregation boundary is redefined. The dynamic update process uses a sliding time window mechanism, re-collecting charging demand information and triggering aggregation model adjustments at set time intervals. The update log records the scale parameters and network status data before and after each adjustment. Trend prediction uses time series analysis methods to generate demand distribution patterns for future periods based on historical log data. The pre-generation of a backup aggregation model is achieved by copying the current model structure and injecting prediction data, stored in an independent memory area for emergency use.

[0219] Specifically, when determining the load aggregation model, the initial aggregation scale is first calculated based on the spatial distribution characteristics of charging demand information. For example, when charging demand is concentrated in a specific area, the initial aggregation scale is set to the minimum geographical area covering that area. Next, the initial scale is dynamically adjusted by combining communication quality indicators from network status information: expanding the aggregation range in areas with sufficient network bandwidth to reduce computational load, and shrinking the aggregation range in areas with high communication latency to improve response speed. The corrected target aggregation scale is input into a hierarchical clustering algorithm to generate load aggregation units with well-defined boundaries. The boundary of each aggregation unit is dynamically adjusted by monitoring the rate of change in internal charging demand in real time. When a sudden change in demand is detected in a local area, that area is automatically segmented into independent aggregation units. The update log continuously records the timestamp of each scale adjustment, the adjustment parameters, and the corresponding network status snapshot, forming a traceable data chain. Based on this data chain, the ARIMA model is used to predict the charging demand distribution trend for future periods, and a backup aggregation model is generated accordingly. The backup model maintains structural synchronization with the current model but loads predicted data as input parameters, ensuring that it can be immediately switched to the backup model in the event of a sudden surge in demand, avoiding the failure of the control strategy.

[0220] As a preferred embodiment, the solution of this application is implemented as follows: When determining the load aggregation model, firstly, the geographical distribution data and charging power demand data of electric vehicle charging piles in the target area are acquired, and high-load clustering areas are identified based on the charging demand density map. According to the coverage of the charging demand distribution, a circular area with a radius of five kilometers is set as the initial aggregation scale. Further, voltage fluctuation data and communication delay parameters of each node in the distribution transformer system are acquired. When the load rate of a node exceeds a preset threshold, the initial aggregation scale is adjusted to a fan-shaped area with a radius of three kilometers to avoid overloaded nodes. When constructing the load aggregation model using the improved K-means clustering algorithm, network topology constraints are introduced to ensure that the aggregation unit matches the partition structure of the distribution transformer network. During the model operation phase, real-time power data of charging piles is collected every five minutes. When the charging power fluctuation within a certain aggregation unit exceeds 15%, the elastic expansion mechanism of the aggregation boundary is automatically triggered. All model update operations are recorded in the blockchain log system. After identifying the pattern of charging demand changes through time series analysis, three sets of backup models with different aggregation granularities are generated in advance and stored on edge computing nodes. When extreme weather is predicted to cause a surge in charging demand, the system automatically switches to the fine-grained aggregation model to improve the accuracy of regulation.

[0221] Through the above technical solutions, this application effectively solves the problem that traditional load aggregation models are difficult to adapt to dynamic charging demands. By constructing a flexibly adjustable aggregation mechanism, it achieves rapid response to real-time operating conditions. The adaptive correction mechanism based on network state ensures topological consistency between the model and the physical system, and the dynamic update function ensures that the model continuously matches actual operating requirements. The pre-generated backup model system significantly enhances the system's ability to cope with emergencies, and provides multiple safeguards for load regulation through forward-looking trend prediction, thereby improving the robustness and flexibility of the distribution transformer system's charging load management as a whole.

[0222] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for coordinated cloud-edge-end control of charging load in a distribution transformer system, characterized in that, include: Obtain information on the operation of the distribution transformer system, charging demand, network status, and meteorological environment; Data cleaning is performed on the distribution transformer system operation information, charging demand information, network status information, and meteorological environment information to obtain cleaned information; Determine the control scenario information based on the information after cleaning; The information on control scenarios includes the current control scenario, subsequent synchronization scenarios, and predicted and extrapolated scenarios. The current control strategy model is generated based on the network topology model, cloud-side constraint information, edge-side resource information, and the current control scenario. Obtain real-time feedback information from the edge; The current control strategy model is locally modified based on real-time feedback information from the edge side to generate a modified current control strategy model. Compare the current control scenario with the subsequent synchronization scenario. If the comparison result is a difference, generate a subsequent synchronization control strategy model based on the network topology model, cloud-side constraint information, edge-side resource information, and subsequent synchronization scenario. If the comparison result is the same, link the calling function for the subsequent synchronization control strategy model to the corrected current control strategy model. Determine the load aggregation model corresponding to the operating information of the distribution transformer system; The modified current control strategy model, subsequent synchronous control strategy model, and load aggregation model are invoked to perform real-time control of the distribution transformer system. During the control process, end-side measurement information is continuously collected; The load aggregation model is updated in a closed loop based on the end-side measurement information to obtain the updated load aggregation model. Based on the updated load aggregation model, a second modification is performed on the current control strategy model and the subsequent synchronous control strategy model to complete real-time control.

2. The method for coordinated control of charging load cloud-edge-end in a distribution transformer system according to claim 1, characterized in that, The revised current control strategy model, subsequent synchronous control strategy model, and load aggregation model are invoked, including: Based on the pre-established mapping relationship between distribution transformer system operation and strategy activation and aggregation, determine the strategy activation speed, load aggregation model switching speed, model update cycle, cache refresh cycle and redundancy removal threshold corresponding to the distribution transformer system operation information; The load aggregation model is updated based on the switching speed of the load aggregation model. The activation order of the current regulation strategy model and the subsequent synchronous regulation strategy model is adjusted according to the activation speed control of the strategy. Periodic recalculation is triggered based on the model update cycle; Update the cache content according to the cache refresh cycle; Identify and remove redundant strategy segments based on the redundancy removal threshold; Perform consistency arbitration for policy conflicts during the sequential activation process; Verify the legality of the arbitration result.

3. The method for coordinated control of charging load cloud-edge-end in a distribution transformer system according to claim 1, characterized in that, Also includes: Obtain the original digital twin model; The original digital twin model includes the original load aggregation model and the original network topology model; Perform a mesh simplification operation on each original digital twin model to obtain an initial lightweight model; Remove the internal invisible structure of the initial lightweight model to obtain the intermediate model; Perform a topology consistency check on the intermediate model to generate a model that passes the check; Security hardening is performed on the verified model to generate a trusted model; Perform digital signatures on the trusted model to generate a signature model; Versioning is performed on the signature model to generate a versioned model; Versioned models are stored in a model repository to complete model archiving.

4. The method for coordinated control of charging load cloud-edge-end in a distribution transformer system according to claim 2, characterized in that, The sequential activation of the current control strategy model and the subsequent synchronous control strategy model after control correction includes: Get the current system clock; The activation sequence is determined based on the current system clock. After the control correction, the current control strategy model and the subsequent synchronous control strategy model are activated sequentially according to the activation order. The control strategy model whose Euclidean distance from the center of the load aggregation model is less than the first distance threshold is preferentially activated; Discontinue control strategy models whose distance from the center of the load aggregation model is greater than the second distance threshold; Perform resource reclamation on the decommissioned model; The recovered resources are then redistributed. Record activation and deactivation logs.

5. The method for coordinated control of charging load cloud-edge-end in a distribution transformer system according to claim 4, characterized in that, Activating the regulatory strategy model includes: The resource allocation scheme for each node is determined based on its topological position and influence weight in the control environment. Allocate computing resources to the control strategy model according to the resource allocation scheme; Perform redundancy checks on the allocated resources to release redundant resources; The released resources are then redistributed to optimize resource utilization. Record resource allocation logs for subsequent auditing; Perform anomaly detection on the audit results; An alert will be sent to any detected anomalies; Alarm information is archived and stored.

6. The method for coordinated control of charging load cloud-edge-end in a distribution transformer system according to claim 1, characterized in that, The revised current control strategy model, subsequent synchronous control strategy model, and load aggregation model are invoked, including: The content to be invoked during the control process is invoked using a dynamic invocation method; Prioritize the execution of the content to be called; Based on the priority ranking results, the corrected current control strategy model, the subsequent synchronous control strategy model, and the load aggregation model will be called first. Perform cache status checks on dynamically accessed content to update the cache queue; Perform aging processing on the updated cache queue to release expired content; Perform secure erasure on expired content after release; Perform defragmentation on the erased space; Record cache operation logs.

7. The method for coordinated control of charging load cloud-edge-end in a distribution transformer system according to claim 3, characterized in that, Performing mesh simplification operations on each original digital twin model includes: The simplification ratio is determined based on the complexity level of the original digital twin model; Redundant meshes from the original digital twin model are removed according to a simplified ratio to obtain an initial lightweight model; Edge-preserving detection is performed on the initial lightweight model after deletion to repair topological cracks; Perform normal consistency correction on the repaired model to eliminate lighting errors; The corrected model is compressed to reduce storage overhead; Perform an integrity check on the compressed model; Encrypt the stored data for models that pass verification. Record simplified process metadata.

8. The method for coordinated control of charging load cloud-edge-end in a distribution transformer system according to claim 5, characterized in that, The resource allocation scheme for each node in the model to be activated is determined based on its topological position and influence weight within the control environment, including: Calculate the node distance weight based on the distance between the node and the center of the load aggregation model; Calculate the influence weight of each node based on its impact on the control accuracy. The node distance weight and the node influence weight are combined to obtain a comprehensive weight; Allocate computing resources based on comprehensive weighting; Perform load balancing checks on the allocated computing resources to adjust the resource allocation results; Perform a performance evaluation on the adjusted results; The strategy will be fine-tuned based on the assessment results; Record fine-tuning logs.

9. The method for coordinated control of charging load cloud-edge-end in a distribution transformer system according to claim 6, characterized in that, The content to be invoked during the control process using dynamic invocation includes: The cycle time is dynamically adjusted based on the frequency of changes in the distribution transformer system's operating information. Update the call priority within each dynamically adjusted beat; Prioritize calling the modified current control strategy model, the subsequent synchronous control strategy model, and the load aggregation model that are most relevant to the current control scenario; Perform version consistency checks on the content that is called first to synchronize the model version; Incremental synchronization is triggered after the verification passes to maintain data consistency; Perform an integrity check on the synchronized data; Execute an anomaly alarm on the verification results; Record synchronization logs.

10. The method for coordinated control of charging load cloud-edge-end in a distribution transformer system according to claim 1, characterized in that, Determining the load aggregation model corresponding to the distribution transformer system operating information includes: The initial aggregation scale of the load aggregation model is determined based on the distribution range of charging demand information; The initial aggregation scale is corrected based on network state information to obtain the target aggregation scale; Construct a load aggregation model based on the target aggregation scale; Perform boundary consistency checks on the constructed load aggregation model to correct the aggregation boundaries; The revised load aggregation model is dynamically updated to match real-time charging demand information; Record update logs for use in backtracking analysis; Perform trend prediction on the backtracking analysis results; pre-generate backup aggregation models based on the prediction results.

Images