Multi-type reactive compensation cooperative control method

Abstract

A multi-type reactive power compensation cooperative control method comprises the following steps: step 1, performing cross-dimension heterogeneous association deconstruction on real-time operation data of a power distribution network and body state data of multi-type reactive power compensation equipment to generate a power distribution network-compensation equipment cooperative state feature topological body; step 2, based on the power distribution network-compensation equipment cooperative state feature topological body, constructing a reactive compensation cooperative optimization model under multi-target coupling constraint, and generating a cooperative control strategy sequence through compensation capacity dynamic equilibrium configuration and response time sequence cooperative matching; and step 3, according to the cooperative control strategy sequence, generating heterogeneous adaptive execution instructions of the multiple types of reactive compensation devices, performing cooperative linkage on actions of the various types of reactive compensation devices, and performing complementary adjustment on performance gains. According to the method, dynamic fine adjustment is carried out on equipment actions, and compared with one-way instruction issuing of a traditional scheme, the adaptability of instructions and the stability of control are obviously improved, and performance gain complementation of multiple types of equipment is achieved.

Description

Technical Field

[0001] This application relates to the field of reactive power compensation control technology in power distribution networks, and more specifically, to a multi-type reactive power compensation collaborative control method. Background Technology

[0002] With the large-scale grid connection of new energy power generation equipment and the diversified development of power load, the reactive power distribution of the distribution network exhibits strong spatiotemporal fluctuations and complex influencing factors, placing higher demands on the accuracy and coordination of reactive power compensation. The combined application of multiple types of reactive power compensation equipment (such as parallel capacitor banks, SVG, and energy storage reactive power compensation devices) has become an important means of voltage regulation and loss optimization in the distribution network. Its core requirement is to achieve complementary characteristics and coordinated operation of different types of compensation equipment to improve the overall efficiency of reactive power compensation.

[0003] Existing reactive power compensation control schemes typically employ a type-based independent control model. This involves setting separate control thresholds for different types of reactive power compensation equipment to achieve their respective switching or regulation. This scheme first divides control areas according to equipment type, configuring independent monitoring terminals for each type of equipment to collect operational data. Then, based on preset voltage or power factor thresholds, it triggers the switching or capacity regulation actions of individual equipment. Finally, it independently evaluates the effects of each equipment's actions, without considering the coordination and cooperation between equipment.

[0004] However, this type of independent control scheme has obvious technical defects. Due to the significant differences in the response characteristics and capacity range of different types of reactive power compensation equipment, the independent control mode cannot achieve complementary characteristics and dynamic balanced capacity configuration between equipment, which can easily lead to over- or under-compensation capacity. At the same time, the lack of a response timing coordination and matching mechanism between equipment means that timing conflicts in the actions of multiple equipment can cause secondary fluctuations in the voltage or power factor of the distribution network, making it difficult to meet the needs of refined reactive power compensation in the distribution network. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides a multi-type reactive power compensation collaborative control method to at least alleviate these problems.

[0006] A multi-type reactive power compensation coordinated control method includes: Step 1: Perform cross-dimensional heterogeneous correlation deconstruction on the real-time operation data of the distribution network and the status data of various types of reactive power compensation equipment to generate a collaborative status feature topology of the distribution network and compensation equipment. Step 2: Based on the topology of the coordinated state characteristics of the distribution network and compensation equipment, construct a reactive power compensation coordinated optimization model under multi-objective coupling constraints, and generate a coordinated control strategy sequence through dynamic balancing configuration of compensation capacity and coordinated matching of response timing. Step 3: Based on the coordinated control strategy sequence, generate heterogeneous adaptation execution instructions for various types of reactive power compensation devices, coordinate the actions of each type of reactive power compensation device, and make complementary adjustments to the performance gains.

[0007] Optionally, step 1 includes: Step 11: Perform timestamp synchronization and dimension normalization on the real-time operation data of the distribution network and the status data of various types of reactive power compensation equipment to generate a time-series unified multi-source dataset. Step 12: Based on the correlation criteria between equipment type attributes and operating conditions, perform feature clustering, divide-and-conquer, and correlation mapping modeling on the time-series collaborative multi-source dataset to generate equipment-operating condition correlation feature tensors; Step 13: Use the constructed feature enhancement fusion model to perform redundant information pruning and key information aggregation and reconstruction on the equipment-operating condition association feature tensor to generate the distribution network-compensation equipment collaborative state feature topology.

[0008] Optionally, step 11 includes: Step 111: Collect real-time operation data of the power distribution network and status data of various types of reactive power compensation equipment, and perform format compliance verification and missing value reconstruction and repair on various types of data to generate a complete original data stack. Step 112: Using the timestamp of the distribution network data acquisition terminal as the reference source, perform time axis homogeneous alignment processing on the complete original data stack, and integrate it into a dataset with a unified time granularity as a time-series aligned data cluster through interpolation reconstruction algorithm; Step 113: Perform numerical interval standardization mapping on the time-series aligned data clusters to eliminate the dimensional differences between different types of data, so as to generate a time-series unified multi-source dataset.

[0009] Optionally, step 12 includes: Step 121: Based on the technical attribute parameters of various types of reactive power compensation equipment, divide the equipment status data in the time-series unified multi-source dataset into static attribute parameters and dynamic response characteristic data to form an equipment feature classification system. Step 122: Construct an association mapping rule engine. The association mapping rule engine contains the mapping relationship between different operating conditions and equipment characteristics. Based on the rule engine, pairwise association matching modeling is performed on the distribution network operation data and equipment characteristic data in the time-series unified multi-source dataset to generate an initial association tensor. Step 123: Based on the equipment feature classification spectrum, the initial association tensor is validated and screened to remove tensor elements without actual physical association and retain effective association feature dimensions to generate equipment-operating condition association feature tensor.

[0010] Optionally, step 13 includes: Step 131: Calculate the information redundancy of each feature column in the device-operating condition correlation feature tensor using the mutual information entropy quantization layer, and mark the feature columns with redundancy higher than the preset threshold to obtain redundant feature columns with removal labels. Step 132: Use a feature dimension pruning and purification layer to prune and remove the marked redundant feature columns, sort the remaining feature columns by importance weight, and extract the core feature dimensions with the highest ranking to generate the core feature tensor. Step 133: Employ an attention mechanism layer to enhance key information in the aggregated core feature tensor, and assign weight coefficients to the operating conditions corresponding to different core features to generate a topology of the distribution network-compensation equipment collaborative state features.

[0011] Optionally, step 2 includes: Step 21: Extract voltage deviation gradient features, power factor resonance features, and equipment response delay features from the topology of the distribution network-compensation equipment cooperative state features, and construct a multi-objective optimization objective functional and constraint boundary conditions; Step 22: Based on the multi-objective optimization objective functional and constraint boundary conditions, construct a multi-type reactive power compensation collaborative optimization model. The collaborative optimization model embeds a device response characteristic differential weight adaptation mechanism. Step 23: Solve the collaborative optimization model using the constructed multi-objective particle swarm collaborative optimization model, and output the compensation capacity allocation coefficients and action timing topology tables of various types of reactive power compensation equipment to generate a collaborative control strategy sequence.

[0012] Optionally, step 21 includes: Step 211: Extract the node voltage deviation gradient value, system power factor resonance coefficient, and voltage fluctuation attenuation coefficient from the topology of the distribution network-compensation equipment cooperative state characteristics as optimization target parameters, and construct a multi-objective optimization objective functional with the objectives of minimizing voltage deviation gradient, optimizing power factor resonance, and maximizing voltage fluctuation attenuation. Step 212: Combine the rated capacity threshold, maximum response rate, and number of action threshold constraints of various types of reactive power compensation equipment with the capacity carrying capacity constraints of the distribution network lines to construct collaborative optimization constraint boundary conditions; Step 213: Standardize and normalize the multi-objective optimization objective functional and the collaborative optimization constraint boundary conditions to generate the multi-objective optimization objective functional and constraint boundary conditions.

[0013] Optionally, step 22 includes: Step 221: Based on the multi-objective optimization objective functional and constraint boundary conditions, construct a multi-type reactive power compensation collaborative optimization model; Step 222: Analyze the differences in response characteristics of various types of reactive power compensation equipment, construct a differentiated weight adaptation mechanism, assign dynamic response weight coefficients to fast response equipment, and assign capacity allocation weight coefficients to large-capacity compensation equipment. Step 223: Embed the differentiated weight adaptation mechanism into the objective functional solution process of the collaborative optimization model, so that the model solution process can adapt to the differences in technical characteristics of different types of equipment, thereby completing the construction of a multi-type reactive power compensation collaborative optimization model.

[0014] Optionally, step 23 includes: Step 231: Use a multi-objective particle swarm cooperative optimization model to solve the multi-type reactive power compensation cooperative optimization model. Initialize the population as a combination of compensation capacity and action timing solution vector, and set the iteration termination criterion. Step 232: Calculate the fitness value of each combination solution vector in the initial population using the fitness evaluation function, and perform selection, crossover, mutation and evolution operations based on the fitness value to generate a new generation population; Step 233: Repeat the iterative evolution process until the termination criterion is met. Select the Pareto optimal solution from the final population, analyze the optimal solution to obtain the compensation capacity allocation coefficient and action timing topology table of each type of reactive power compensation equipment, and generate a coordinated control strategy sequence.

[0015] Optionally, step 3 includes: Step 31: Analyze the compensation capacity allocation coefficients and action timing topology table in the collaborative control strategy sequence, and integrate them with the control interface protocol specifications of various types of reactive power compensation equipment to generate a prototype of equipment-specific control instructions; Step 32: Based on the real-time update data of the topology of the distribution network-compensation equipment collaborative state characteristics, perform feasibility verification and parameter iterative correction on the prototype of the equipment-specific control command to generate heterogeneous adaptation execution commands; Step 33: Send the heterogeneous adaptation execution command to the corresponding type of reactive power compensation equipment, and simultaneously collect the action feedback data stream of each equipment to form a feedback control link, so as to coordinate the action of each type of reactive power compensation equipment and complementarily adjust the performance gain.

[0016] Optionally, step 31 includes: Step 311: Analyze the compensation capacity allocation coefficient and action timing topology table in the collaborative control strategy sequence, determine the action parameter thresholds and execution time nodes of each type of reactive power compensation equipment, and form a list of equipment action instructions; Step 312: Retrieve the control interface protocol specifications of various types of reactive power compensation equipment. The specifications include instruction encoding format, communication baud rate, and data verification mechanism. Based on the control interface specifications, convert the parameters in the equipment action instruction list into instruction formats that can be recognized by the corresponding equipment. Step 313: Perform syntax compliance verification on the converted instruction format, correct instruction fields with abnormal formats, and generate a prototype of device-specific control instructions.

[0017] Optionally, step 32 includes: Step 321: Extract the updated data of the cooperative status feature topology of the distribution network and compensation equipment in real time to obtain the current operating status parameters of various types of reactive power compensation equipment and the real-time voltage and power data of the distribution network, and generate a real-time verification data stack accordingly. Step 322: Compare and verify the action parameters in the prototype of the device-specific control command with the current status parameters of the device in the generated real-time verification data cluster to determine whether the action parameters are within the safe operating threshold range of the device. Step 323: Iteratively correct the action parameters that exceed the safe operation threshold range so that the adjusted parameters match the equipment operation constraints and distribution network operating conditions, and generate heterogeneous adaptation execution instructions accordingly.

[0018] Optionally, step 33 includes: Step 331: Send the heterogeneous adaptation execution command to different types of reactive power compensation equipment, including parallel capacitor banks, SVG, and energy storage type reactive power compensation devices, through the corresponding communication interface; Step 332: Collect action feedback data streams of various types of reactive power compensation equipment. The feedback data streams include action completion status, actual output compensation capacity, and equipment operating thermal field data. Step 333: Compare the deviation between the action feedback data stream and the expected parameters in the collaborative control strategy sequence. If there is a deviation, fine-tune the subsequent execution instructions based on the deviation value to form a feedback control link, so as to coordinate the action of various types of reactive power compensation equipment and complementarily adjust the performance gain.

[0019] Technical advantages of the technical solution provided in this application This application's multi-type reactive power compensation collaborative control method addresses the technical shortcomings of traditional independent control schemes, such as unbalanced compensation capacity configuration and conflicting equipment action timing. It generates a collaborative state feature topology of the distribution network and compensation equipment by performing cross-dimensional heterogeneous correlation deconstruction of real-time operation data of the distribution network and the status data of various types of reactive power compensation equipment. This solves the problems of isolated multi-source data analysis and insufficient mining of equipment-operating condition correlation patterns in traditional schemes. Compared to the independent collection and simple aggregation of data in traditional schemes, this application's cross-dimensional heterogeneous correlation deconstruction achieves feature clustering and correlation mapping modeling based on timestamp synchronization and dimensional normalization of multi-source data. The feature enhancement fusion model, through redundant information pruning and key information aggregation and reconstruction, generates a collaborative state feature topology that clearly presents the correlation between distribution network operating conditions and compensation equipment states. This allows for the effective mining of the characteristic differences and complementary patterns of various types of equipment, providing more comprehensive data support for collaborative optimization and significantly improving the effectiveness of data correlation.

[0020] Based on the topology of the distribution network and compensation equipment's coordinated state characteristics, a multi-objective coupled constraint reactive power compensation collaborative optimization model is constructed. This model generates a collaborative control strategy sequence through dynamic balancing of compensation capacity and coordinated matching of response timing, solving the problems of single threshold control and low compensation efficiency in traditional schemes. Traditional schemes can only trigger the action of a single device based on a fixed threshold, failing to achieve capacity balancing and timing coordination among multiple devices. This application extracts key features from the coordinated state characteristic topology to construct a multi-objective optimization functional and constraint conditions, embedding a collaborative optimization model with a differentiated weight adaptation mechanism for device response characteristics. The Pareto optimal solution is obtained through a multi-objective particle swarm optimization algorithm. The generated collaborative control strategy sequence can achieve dynamic balancing of compensation capacity and coordinated timing matching of device actions. Compared to the independent control strategies of traditional schemes, the accuracy and coordination of reactive power compensation are significantly improved, effectively avoiding overcompensation or secondary fluctuations.

[0021] Based on a coordinated control strategy sequence, heterogeneous adaptive execution instructions are generated for various types of reactive power compensation devices. This enables coordinated linkage and complementary performance gain adjustment of the actions of different device types, solving the problems of low instruction adaptability and lack of feedback control in traditional solutions. Traditional solutions use a unified instruction format to issue instructions to different types of devices, which easily leads to instruction incompatibility and lacks a feedback mechanism to correct action deviations. In contrast, this application generates a proprietary instruction prototype by integrating device control interface protocol specifications and combines it with real-time topology data for feasibility verification and parameter correction. The generated heterogeneous adaptive execution instructions can adapt to the control requirements of different types of devices. Simultaneously, by collecting action feedback data streams to form a closed-loop control link, dynamic fine-tuning of device actions is achieved. Compared to the unidirectional instruction issuance of traditional solutions, the adaptability of instructions and the stability of control are significantly improved, realizing complementary performance gains for multiple types of devices. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating a multi-type reactive power compensation coordinated control method according to an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a multi-type reactive power compensation collaborative control device according to an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0023] like Figure 1 The diagram shown is a flowchart illustrating a multi-type reactive power compensation coordinated control method according to an embodiment of this application, including: Step 1: Perform cross-dimensional heterogeneous correlation deconstruction on the real-time operation data of the distribution network and the status data of various types of reactive power compensation equipment to generate a collaborative status feature topology of the distribution network and compensation equipment. Step 2: Based on the topology of the coordinated state characteristics of the distribution network and compensation equipment, construct a reactive power compensation coordinated optimization model under multi-objective coupling constraints, and generate a coordinated control strategy sequence through dynamic balancing configuration of compensation capacity and coordinated matching of response timing. Step 3: Based on the coordinated control strategy sequence, generate heterogeneous adaptation execution instructions for various types of reactive power compensation devices, coordinate the actions of each type of reactive power compensation device, and make complementary adjustments to the performance gains.

[0024] Optionally, step 1 includes: Step 11: Perform timestamp synchronization and dimension normalization on the real-time operation data of the distribution network and the status data of various types of reactive power compensation equipment to generate a time-series unified multi-source dataset. Step 12: Based on the correlation criteria between equipment type attributes and operating conditions, perform feature clustering, divide-and-conquer, and correlation mapping modeling on the time-series collaborative multi-source dataset to generate equipment-operating condition correlation feature tensors; Step 13: Use the constructed feature enhancement fusion model to perform redundant information pruning and key information aggregation and reconstruction on the equipment-operating condition association feature tensor to generate the distribution network-compensation equipment collaborative state feature topology.

[0025] Optionally, step 11 includes: Step 111: Collect real-time operation data of the power distribution network and status data of various types of reactive power compensation equipment, and perform format compliance verification and missing value reconstruction and repair on various types of data to generate a complete original data stack. Step 112: Using the timestamp of the distribution network data acquisition terminal as the reference source, perform time axis homogeneous alignment processing on the complete original data stack, and integrate it into a dataset with a unified time granularity as a time-series aligned data cluster through interpolation reconstruction algorithm; Step 113: Perform numerical interval standardization mapping on the time-series aligned data clusters to eliminate the dimensional differences between different types of data, so as to generate a time-series unified multi-source dataset.

[0026] Steps 111-113 construct a standardized preprocessing workflow for multi-source data of "distribution network - reactive power compensation equipment". Through progressive processing of "precise acquisition and defect repair - time-source alignment - dimensional unification elimination", the core problems of heterogeneous multi-source data formats, asynchronous time, and large differences in dimensions are solved. This workflow is deeply adapted to the scenario requirements of real-time operation of distribution network and management of reactive power compensation equipment. It innovatively adopts time-aware missing value reconstruction, benchmark-anchored time alignment, and scenario-specific standardized mapping methods to ensure that the generated time-series unified multi-source dataset has completeness, temporal consistency, and comparability, providing a high-quality data foundation for subsequent spatiotemporal dimensional fusion and feature index tree generation.

[0027] Preferably, the specific implementation process of step 111 is as follows: Drive precise acquisition of multi-source data. Two types of core data are simultaneously acquired through distributed acquisition terminals (deployed at distribution network line nodes and reactive power compensation equipment): First, real-time operation data of the distribution network, covering three-phase voltage, three-phase current, active power, reactive power, and power factor of 10kV / 0.4kV lines. The acquisition frequency is generally 20~50Hz, for example 30Hz (acquiring data approximately every 33 milliseconds). Second, status data of various types of reactive power compensation equipment, acquired differently according to equipment type (capacitor-type compensation equipment acquires capacitor temperature, insulation resistance, and switching frequency; reactor-type compensation equipment acquires winding temperature, leakage flux intensity, and cooling system status). The acquisition frequency is consistent with the operation data to ensure time-series correlation. During data acquisition, the acquisition terminal identifier, acquisition timestamp (accurate to milliseconds), and data source component (e.g., "10kV line A-phase current - terminal T01" and "capacitor compensator C1 temperature - terminal T02") are recorded simultaneously. The system implements multi-dimensional verification of format compliance, constructing a four-dimensional verification rule encompassing "field integrity, data type, value range, and format specification." Field integrity verification checks whether each data entry contains required fields (e.g., "voltage data must include phase, value, timestamp, and terminal ID"). Data type verification ensures that numeric fields (e.g., voltage, temperature) are floating-point / integer types, and identifier fields (e.g., terminal ID) are string types. Value range verification sets thresholds based on distribution network and equipment operation specifications (e.g., the normal voltage range for 10kV lines is 9.5~10.5kV; values ​​exceeding this range are marked as abnormal). Format specification verification unifies the data transmission format (e.g., converting heterogeneous formats like JSON and XML to standard binary format, and standardizing field separators to "|"). Abnormal data is marked with an anomaly type (e.g., "voltage value out of range - abnormal" or "missing terminal ID - incomplete") and isolated to an abnormal data buffer. To drive the reconstruction and repair of missing values, a time-series trend-aware reconstruction algorithm is designed for data missing values ​​after verification (such as continuous missing values ​​due to communication interruption of the acquisition terminal or random single-point missing values): For random single-point missing values, weighted interpolation of adjacent time-series data is used (the weight of the previous time-series data is 0.6, and the weight of the next time-series data is 0.4; if the next time-series data was not collected, the weight of the previous two time-series data is taken as 0.3); for continuous missing values ​​(missing duration ≤ 5 acquisition cycles, for example ≤ 165 milliseconds), reconstruction based on historical trends is used (time-series data of the same period in the past 7 days are extracted, a trend curve is fitted, and then the missing values ​​are filled); for continuous missing values ​​exceeding 5 cycles, the data is marked as "unreconstructable", and redundant data from a backup acquisition terminal is used to supplement the missing values ​​(if available). After repair, data compliance is verified again to ensure the quality of the repair.The system drives the generation of a complete raw data stack. The compliant data after verification and repair is organized into a stack-style storage structure according to the hierarchical structure of "data category-device type-acquisition terminal". Each layer of the data stack corresponds to a type of data (such as "distribution network voltage data layer" and "capacitor compensator temperature data layer"). The data within each layer is arranged in ascending order by acquisition timestamp. Each data unit contains full attribute information including "data identifier-acquisition timestamp-value-source component-verification status". At the same time, a data stack index table is generated to record the storage location, time range and data volume of each layer of data, which facilitates quick retrieval during subsequent time alignment. Finally, the complete raw data stack is output.

[0028] Preferably, in the specific technical implementation of step 112: drive the reference timestamp anchoring and homogenization processing, load the complete original data stack, extract the acquisition timestamp of each data unit, and use the timestamp of the distribution network data acquisition terminal (main terminal, selected from terminals deployed on the main line of the distribution network with high acquisition stability, such as T01) as the reference source. The timestamp deviation of other acquisition terminals (such as T02 and T03 of reactive power compensation equipment) is calibrated through a time synchronization protocol (such as SNTP protocol). The calibration logic is to calculate the difference between each terminal timestamp and the reference timestamp. If the difference is ≤ 5 milliseconds, it is directly corrected to the reference timestamp. If the difference is > 5 milliseconds, it is corrected by linear amortization of the difference (e.g., if the timestamp of terminal T02 is 8 milliseconds later than the reference, the timestamp of its four consecutive acquisition cycles is corrected by +2 milliseconds in turn, and gradually aligned to the reference), to ensure that the timestamps of all data are from the same source. Driven by the unification of time granularity and interpolation reconstruction, based on the real-time requirements of distribution network control, a unified time granularity is set (generally 10~50 milliseconds, 20 milliseconds for example). Starting from the reference timestamp, a continuous time axis is generated (e.g., "reference timestamp T0, T0+20ms, T0+40ms..."). The complete original data stack is traversed, and the timestamps of each data unit are mapped to the unified time axis. If a time axis node has no corresponding data (i.e., the original acquired data does not cover this node), a cubic spline interpolation algorithm is used to reconstruct the data: based on the two valid data points before and after the node, a smooth time series curve is fitted, and the interpolation result at the node is calculated as supplementary data. If there are insufficient valid data points before and after the node (<3), linear interpolation is used (the average of the two nearest valid data points is calculated). The interpolated data is labeled as "reconstructed data" to distinguish it from the original acquired data. The time-aligned data cluster is integrated by horizontally aggregating data of different categories and equipment types according to the node order of a unified time axis. Each time axis node corresponds to a data subset, which includes the distribution network operation data (three-phase voltage, current, etc.) and the status data of various types of reactive power compensation equipment (temperature, insulation resistance, etc.) at that moment. The data subset is organized in the order of "distribution network data - reactive power compensation equipment data", and the data of the same type of equipment is sorted by equipment number (e.g., "capacitor compensator C1 - temperature, capacitor compensator C2 - temperature"). The data subsets of all time axis nodes are integrated in chronological order to form a time-aligned data cluster. The data in the cluster is stored in the structure of "time axis node ID - data category - equipment identifier - value - data type (original / reconstructed)" and is accompanied by a time alignment log to record the timestamp correction process, interpolation nodes and interpolation algorithms used for each data unit, ensuring the traceability of time alignment.

[0029] Preferably, in one scenario, step 113 is specifically implemented as follows: drive the parsing and numerical range definition of the time-series aligned data cluster, load the time-series aligned data cluster generated in step 112, extract the numerical fields of each data type, and define the effective numerical range of each type of data (i.e., the original range before standardization) in conjunction with the power distribution network operation specifications and the technical parameters of reactive power compensation equipment; for example, the original range of three-phase voltage of 10kV line is 9.5~10.5kV, the original range of capacitor compensator temperature is -20~85℃, the original range of insulation resistance is 100~1000MΩ, and the original range of active power is 0~5000kW; for extreme outliers exceeding the effective range (such as voltage 15kV and temperature 100℃), the 3σ criterion is first used to remove them (calculate the mean μ and standard deviation σ of the data, and remove data other than μ±3σ) to avoid affecting the standardization effect. To drive scenario-specific numerical range standardization mapping, a min-max standardization algorithm adapted to multi-source data in the power distribution network is designed. This algorithm uniformly maps data with different dimensions and numerical ranges to the [0,1] interval. The mapping formula is: Standardized value = (Original value - Lower limit of original interval) / (Upper limit of original interval - Lower limit of original interval). For special data types (such as power factor, with an original interval of -1 to 1, including negative values), the mapping interval is adjusted to [-1,1], and the formula is revised as follows: Standardized value =2×(Original Value - Lower Limit of Original Range) / (Upper Limit of Original Range - Lower Limit of Original Range)-1; For example, the voltage of a 10kV line at a certain moment is 10kV, the original range is 9.5~10.5kV, and the standardized value is (10-9.5) / (10.5-9.5)=0.5; the temperature of the capacitor compensator is 25℃, the original range is -20~85℃, and the standardized value is (25-(-20)) / (85-(-20))=45 / 105≈0.43. During the standardization process, the original range and mapping formula parameters of each type of data are recorded simultaneously to generate a standardized parameter table for subsequent data inverse mapping (such as when the control command needs to be restored to the original value). The system drives the generation of a unified time-series multi-source dataset, integrating all standardized numerical data and organizing them according to the structure of "Time Axis Node ID - Distribution Network Operation Data (Standardized) - Reactive Power Compensation Equipment Status Data (Standardized) - Data Source - Original Value - Standardized Parameter Index". The "Standardized Parameter Index" points to the corresponding entry in the standardized parameter table, facilitating the tracing of original data characteristics. The dataset is arranged in ascending order according to the time axis nodes, with each node containing the standardized results of all types of data at that moment, ensuring temporal continuity and data integrity. Finally, a unified time-series multi-source dataset is output, along with a dataset quality report, including indicators such as data volume, missing rate (≤1% after repair), and standardization pass rate (≥99%), verifying that the dataset meets the requirements for subsequent spatiotemporal fusion processing.

[0030] The core of the time-series trend-aware missing value reconstruction algorithm is based on the time-series continuity and trend correlation characteristics of distribution network and reactive power compensation equipment data. For two typical missing value scenarios—random single-point missing values ​​and continuous missing values ​​(continuous time periods ≤ the threshold of the distribution network data acquisition cycle)—a progressive, generalized fitting logic is designed, consisting of "dynamic mining of time-series context, trend adaptation weight allocation, generalized trend fitting, and multi-dimensional rationality verification." By abandoning fixed numerical dependencies and adopting dynamic parameter configurations adapted to scenario characteristics, the algorithm ensures that the reconstruction logic can adapt to datasets with different distribution network architectures and different types of reactive power compensation equipment. Simultaneously, through a trend-driven differentiated processing mechanism, the algorithm improves the fit between the reconstructed data and the actual operating conditions. The specific fitting process is as follows: I. Fitting Process for Random Single-Point Missing Scenarios (Adapting to Single-Point Data Missing Caused by Instantaneous Interference at the Acquisition Terminal) Preferably, the specific implementation process of the fitting process in the scenario of random single-point missing data is as follows: Drive missing point localization and time-series context extraction; load the time-series sequences of the corresponding data types from the complete original data stack (such as "capacitor compensator body temperature time-series sequence" and "distribution network line voltage time-series sequence"); complete the missing point timestamp location based on ascending timestamp sorting; extract the time-series context data of the missing point—that is, the values ​​of several consecutive valid moments before and after the missing point (the number of valid moments is set according to the time-series correlation requirements of the distribution network data to ensure coverage of recent trend characteristics); if the data of subsequent valid moments has not been collected, the collected subsequent valid data is taken, and the insufficient part is marked as "none". Simultaneously, by calculating the time-series slope of the consecutive valid periods before the missing point, the recent trend type of the time-series sequence is determined (a positive slope indicates an upward trend, a negative slope indicates a downward trend, and a slope approaching zero indicates a stable trend), providing a trend basis for subsequent weight allocation. The system drives the construction of dynamic weight allocation rules for time series data. Based on the determined trend type, it designs differentiated general weight allocation logic. The core innovation lies in the dynamic adjustment of weight percentages according to the dominant trend direction: In an upward trend, the preceding time series data represents the dominant trend, and the weight percentage gradually increases as time approaches the missing point, while the weight percentage of subsequent non-dominant time series decreases; in a downward trend, the weight allocation logic is symmetrical to the upward trend, with subsequent time series data representing the dominant trend, and the weight percentage gradually increases as time approaches the missing point; in a stable trend, a symmetrical weight allocation logic is adopted, with the weight percentages of symmetrical time series before and after the missing point being consistent. If there is insufficient effective subsequent data, the missing subsequent weight percentages are supplemented to the corresponding preceding time series data according to the decreasing trend of preceding weights, ensuring that the sum of the weight percentages of all context data involved in the calculation is 1. The system drives weighted interpolation fitting calculations, multiplying the extracted effective context values ​​one by one with the corresponding dynamically allocated weight percentages, and obtaining the initial reconstructed value of the missing point through weighted summation. The fitting calculation process strictly follows the logic of "trend-dominant - non-dominant auxiliary," ensuring that the initial reconstructed value closely matches the recent time series trend. The process drives the rationality verification of reconstructed values. Based on the recent trend slope, an allowable range for trend deviation is set (this allowable range is determined according to the fluctuation characteristics of distribution network data to ensure the reconstructed value does not deviate from the normal trend fluctuation range). The difference between the initial reconstructed value and the value at the immediately preceding valid time point is calculated. If the difference is within the allowable range, the initial reconstructed value is retained; if the difference exceeds the allowable range, the initial reconstructed value is corrected according to the recent trend slope to ensure the corrected value conforms to the continuity of the time-series trend. After correction, the reconstructed value is further verified to see if it is within the valid operating range of the corresponding data type (the valid range is set based on distribution network operation specifications and reactive power compensation equipment technical parameters). If it is, it is used as the final reconstructed value; otherwise, it is marked as "reconstruction abnormal" and the redundant data supplementation logic of the backup acquisition terminal is activated.

[0031] II. Fitting Process for Continuous Missing Scenarios (Adapting to Continuous Missing Scenarios Caused by Short-Term Communication Interruptions at the Data Acquisition Terminal) Preferably, in the specific technical implementation of the fitting process under the continuous missing scenario: Historical time-series samples are extracted to drive the extraction of data from the same period, locate the start and end times of the continuous missing segment, and determine the duration of the continuous missing segment (the duration of the continuous missing segment does not exceed the collection cycle threshold to meet the real-time control requirements of the distribution network, ensuring the timeliness of the reconstructed data). Based on the periodic characteristics of the distribution network operation, recent historical time-series samples are extracted (the historical time span is set according to the periodic rules of the distribution network operation to ensure that the samples have trend reference value). The sample selection criteria are no missing data and data integrity meeting preset requirements. If the number of recent historical time-series samples is insufficient, the historical time range is expanded according to the same periodic rules until the number of extracted effective samples meets the trend fitting requirements. The trend features of the samples are aligned, and trend normalization is performed on each group of extracted historical time-series samples. The core innovative logic is "current time-series benchmark adaptation": First, the time-series trend slope of each group of historical samples is calculated. Then, the value of the previous effective moment of the continuous missing segment is used as the current time-series benchmark. All values ​​of each group of historical samples are shifted and adjusted according to the benchmark difference to ensure seamless connection between the initial trend of the historical samples and the trend of the current time series. Simultaneously, the average trend slope of all aligned historical samples is calculated as the baseline trend slope for subsequent fitting. This drives the construction of a segmented trend fitting model, employing a two-stage general fitting logic. The core design principle is to balance avoiding excessive deviation in the initial trend with precise fit in the later stages: the first stage, from the start to the middle of the continuous missing segment, uses the decay percentage of the baseline trend slope for fitting (the decay percentage is set according to the trend stability requirements at the beginning of the continuous missing segment, reducing the risk of trend deviation in the initial fitting); the second stage, from the middle to the end of the continuous missing segment, uses the gain percentage of the baseline trend slope for fitting (the gain percentage is set to approximate the baseline trend slope, ensuring precise fit between the later fitted trend and the average trend). If the continuous missing duration is only one collection cycle, the baseline trend slope is directly used for fitting, accommodating extreme continuous scenarios with single-point missing values. The system drives the generation of reconstructed values ​​for the continuous missing segment, using the value of the previous valid moment of the continuous missing segment as the fitting starting point. Based on the slope set by the segmented trend fitting model, the reconstructed value for each missing moment is calculated sequentially, ensuring the trend continuity of the reconstructed value sequence and the adaptability of the segmented slope. The process involves driving trend consistency verification by calculating the deviation between the overall trend slope of the reconstructed value sequence and the baseline trend slope. If the deviation is within a preset allowable range, the verification passes. Simultaneously, it verifies the connection deviation between the first and last values ​​of the reconstructed value sequence and the values ​​at the preceding and following valid times (the allowable range for connection deviation is set based on the temporal correlation of the distribution network data), ensuring seamless connection between the reconstructed value sequence and the current time series. If the verification requirements are not met, the slope ratio of the segmented fitting is dynamically adjusted and recalculated until the reconstructed value sequence meets the trend consistency and connection requirements. Finally, a complete reconstructed value sequence with continuously missing segments is output.

[0032] Optionally, step 12 includes: Step 121: Based on the technical attribute parameters of various types of reactive power compensation equipment, divide the equipment status data in the time-series unified multi-source dataset into static attribute parameters and dynamic response characteristic data to form an equipment feature classification system. Step 122: Construct an association mapping rule engine. The association mapping rule engine contains the mapping relationship between different operating conditions and equipment characteristics. Based on the rule engine, pairwise association matching modeling is performed on the distribution network operation data and equipment characteristic data in the time-series unified multi-source dataset to generate an initial association tensor. Step 123: Based on the equipment feature classification spectrum, the initial association tensor is validated and screened to remove tensor elements without actual physical association and retain effective association feature dimensions to generate equipment-operating condition association feature tensor.

[0033] Steps 121-123 construct a progressive data fusion logic adapted to the reactive power compensation management scenario of the distribution network, consisting of "equipment feature classification - operating condition association modeling - validity verification". This logic addresses issues such as generalized classification, overgeneralized associations, and lack of physical meaning in traditional multi-source data association by employing differentiated equipment feature classification, a scenario-specific association mapping rule engine, and physical principle-guided verification and filtering. This logic deeply integrates the technical attribute differences of various types of reactive power compensation equipment with the operating conditions of the distribution network, achieving precise association and coupling between distribution network operating data and equipment feature data. The generated equipment-operating condition association feature tensor can directly support subsequent spatiotemporal fusion and feature index tree construction, providing a high-quality feature data foundation for the coordinated management of distribution transformers and switches.

[0034] Preferably, the specific implementation process of step 121 is as follows: The driving data classification module loads the technical attribute parameters of multiple types of reactive power compensation equipment. The technical attribute parameters are extracted based on the differences in equipment type, covering mainstream types such as capacitive compensation equipment, reactive compensation equipment, and dynamic reactive power compensation devices. The extraction dimensions include inherent parameters of the equipment, operating response parameters, and structural characteristic parameters. Based on the requirements of data timeliness and relevance for distribution network operation and management, the equipment status data in the time-series unified multi-source dataset is divided into two categories: static attribute parameters and dynamic response characteristic data. Static attribute parameters refer to inherent parameters fixed during the equipment manufacturing stage, which do not change with the operating conditions, including rated voltage, rated capacity, rated frequency, insulation class, structural dimensions, etc. Dynamic response characteristic data refer to parameters that are dynamically adjusted during equipment operation according to the changes in distribution network conditions, including real-time temperature (capacitor core temperature, reactance winding temperature), switching response time, reactive power output amplitude, real-time insulation resistance value, cooling system operating status, etc. Based on the above classification results, a device feature classification system is constructed. This system is organized in a hierarchical structure. The top layer is the general category of various types of reactive power compensation equipment. The middle layer is divided into subcategories according to equipment type. The bottom layer further distinguishes between static attribute parameter branches and dynamic response characteristic data branches under each subcategory. Each data branch is attached with corresponding specific parameter items and parameter descriptions, data accuracy requirements, collection source identifiers and other auxiliary information, forming a device feature classification system that covers all attributes of the equipment and is suitable for subsequent association modeling.

[0035] Preferably, in the specific technical implementation of step 122: the driving rule engine construction module extracts typical operating condition types of the distribution network. Combining the characteristics of the distribution network power supply structure and the operation and control requirements, typical operating conditions include single-source radial power supply, multi-source power supply, heavy-load operation, light-load operation, maintenance transition, and fault precursor operation. Each condition clearly defines the judgment criteria (e.g., heavy-load operation is judged by the load rate exceeding a preset threshold, and multi-source power supply operation is judged by the presence of two or more power sources). Based on the extracted operating condition types and equipment feature classification system, an association mapping rule engine is constructed. This rule engine has a built-in mapping relationship library of operating conditions and equipment features. The mapping relationship is established based on the power system operation principle and the working mechanism of reactive power compensation equipment. For example, under heavy-load operation, the mapping relationship is associated with the switching response time and reactive power output amplitude in the dynamic response characteristic data of capacitor compensation equipment, as well as the rated capacity in the static attribute parameters; under multi-source power supply operation, the mapping relationship is associated with the voltage regulation accuracy and response delay time in the dynamic response characteristic data of dynamic reactive power compensation devices. Using a unified time-series multi-source dataset as the data source, distribution network operation data (including voltage, current, power factor, load rate, number of power sources, etc.) and equipment feature data (i.e., static attribute parameters and dynamic response characteristic data in the equipment feature classification system) are extracted. Based on an association mapping rule engine, pairwise association matching modeling is performed to generate an initial association tensor. This initial association tensor is a three-dimensional tensor structure. The first dimension is the acquisition time series (corresponding to the timestamp sequence of the unified time-series multi-source dataset), the second dimension is the distribution network operation data feature items (such as A-phase voltage, B-phase current, load rate, etc.), and the third dimension is the equipment feature data items (such as rated capacity, switching response time, etc.). The tensor element values ​​are the initial values ​​of the association degree between the distribution network operation data features and the equipment feature data at the corresponding time point. The initial association degree is assigned by the mapping relationship strength in the rule engine (e.g., strong association is assigned a high interval value, weak association is assigned a low interval value).

[0036] Preferably, in one scenario, step 123 is specifically implemented as follows: The drive validity verification module loads the equipment feature classification spectrum generated in step 121. Based on the hierarchical relationship and parameter attribute description in the spectrum, a correlation validity judgment criterion is established. The core of the judgment criterion is to verify whether the correlation between the distribution network operation data features and the equipment feature data in the initial correlation tensor has actual physical meaning and whether it conforms to the working principle and energy conversion law of reactive power compensation equipment in the distribution network. Based on this judgment criterion, the initial correlation tensor is screened for validity verification element by element. Tensor elements without actual physical correlation are eliminated. For example, the line wind speed data in the distribution network operation data and the rated voltage parameter in the equipment feature data have no direct physical correlation, and the corresponding tensor elements are eliminated. Tensor elements with effective correlation feature dimensions are retained. For example, the load rate data in the distribution network operation data and the reactive power output amplitude parameter in the equipment feature data have a clear physical correlation (the load rate change drives the reactive power output amplitude adjustment), and the corresponding tensor elements are retained. The filtered correlation tensors are dimensionally normalized, redundant feature dimensions are merged, and feature dimension identification information (including correlation type, physical meaning description, and data source) is added to finally generate the equipment-operating condition correlation feature tensor. This equipment-operating condition correlation feature tensor fully retains the effective correlation information between the distribution network operating conditions and the reactive power compensation equipment features, and each tensor element has a clear physical meaning and scenario adaptability, which can be directly used for subsequent hierarchical feature aggregation and the construction of the distribution transformer-switcher coordinated state feature index tree.

[0037] Optionally, step 13 includes: Step 131: Calculate the information redundancy of each feature column in the device-operating condition correlation feature tensor using the mutual information entropy quantization layer, and mark the feature columns with redundancy higher than the preset threshold to obtain redundant feature columns with removal labels. Step 132: Use a feature dimension pruning and purification layer to prune and remove the marked redundant feature columns, sort the remaining feature columns by importance weight, and extract the core feature dimensions with the highest ranking to generate the core feature tensor. Step 133: Employ an attention mechanism layer to enhance key information in the aggregated core feature tensor, and assign weight coefficients to the operating conditions corresponding to different core features to generate a topology of the distribution network-compensation equipment collaborative state features.

[0038] Steps 131-133 construct a progressive feature optimization process adapted to the collaborative management and control scenario of distribution network and reactive power compensation equipment, consisting of "redundant feature quantification, core feature purification, and key information enhancement." This process addresses the problems of high redundancy interference, low core feature recognition, and generalized key information weight allocation in traditional feature processing by designing scenario-specific information redundancy quantification logic, feature importance ranking criteria, and attention enhancement mechanisms. This deeply integrates the operating characteristics of the distribution network with the working mechanism of reactive power compensation equipment. By accurately eliminating invalid redundant features, focusing on core related features, and strengthening the representation of key information, it generates a highly recognizable and scenario-adaptable collaborative state feature topology of the distribution network and compensation equipment. This provides accurate and efficient core feature support for the subsequent construction of the distribution transformer-switch collaborative state feature index tree.

[0039] Preferably, the specific implementation process of step 131 is as follows: The mutual information entropy quantization layer loads the device-operating condition correlation feature tensor. The three dimensions of this tensor are the acquisition time series, the distribution network operation feature column, and the reactive power compensation equipment feature column, respectively. The tensor element values ​​are the correlation values ​​of the two types of features at the corresponding time points. All feature columns in the device-operating condition correlation feature tensor are extracted, and the physical meaning and data source of each feature column are clarified (e.g., the "distribution network load rate feature column" comes from distribution network operation data, and the "compensation equipment switching response time feature column" comes from equipment dynamic response characteristic data). A mutual information entropy calculation logic adapted to the distribution network time series characteristics is designed to calculate the mutual information entropy value between any two feature columns. The mutual information entropy value is used to quantify the degree of information overlap between the two feature columns. The higher the entropy value, the higher the degree of effective information overlap between the two features, and the stronger the information redundancy. In light of the requirements for feature purity in the collaborative management and control of power distribution network and compensation equipment, a preset threshold for information redundancy is set (the threshold is calibrated based on historical operation and maintenance data and feature correlation patterns to ensure that redundancy is eliminated while retaining effective correlation information). The mutual information entropy values ​​of each pair of feature columns are compared with the preset threshold. If the entropy value is higher than the preset threshold, the corresponding feature column is determined to be a redundant feature column, and a removal mark is added to it. Finally, the redundant feature columns with removal marks and the retained non-redundant feature columns are obtained, forming a marked equipment-operating condition correlation feature tensor.

[0040] Step 131 constructs a mutual information entropy quantification logic adapted to the time-series operating characteristics of the distribution network. Through a progressive processing of "tensor dimension analysis - feature scenario-based calibration - redundancy quantification of time-series constraints - threshold screening for operating condition adaptation," it addresses the problems of traditional mutual information entropy calculations failing to consider the temporal continuity of distribution network data and the weak correlation of feature physical meanings. This implementation closely integrates with the collaborative management and control scenario of the distribution network and reactive power compensation equipment, ensuring that the information redundancy quantification accurately reflects the degree of effective information overlap between feature columns while also meeting the dual requirements of real-time control of the distribution network for feature purity and temporal correlation. This provides a precise basis for redundancy labeling in subsequent feature pruning and purification.

[0041] Preferably, the specific implementation process of step 131 is as follows: The mutual information entropy quantization layer loads the device-operating condition correlation feature tensor. First, the tensor is analyzed for dimensions and its validity is verified to confirm the completeness of its three dimensions. The first dimension, the time series of data acquisition, corresponds to the time granularity of real-time control of the distribution network (based on the data acquisition frequency of the distribution network operation). Each time node uniquely corresponds to the collaborative operation data of a data acquisition time. The second dimension, the distribution network operation feature column, covers feature items directly related to the stability of the operating condition, such as line voltage, current, load rate, and power factor. The third dimension, the reactive power compensation equipment feature column, covers equipment operating status feature items such as switching response time, reactive power output amplitude, and real-time insulation resistance value. The tensor element value is the correlation value of the two types of features at the corresponding time point (the correlation value is calculated and generated based on the previous correlation mapping rule engine, representing the physical correlation strength between the two). After the verification is passed, a tensor dimension analysis report is generated to clarify the value range, feature quantity, and physical meaning of each dimension. Based on the tensor dimension parsing report, all feature columns in the equipment-operating condition correlation feature tensor are extracted to form a feature column set. Scenario-based calibration is performed on each feature column in the set: associating the equipment feature classification spectrum with the distribution network operating condition definition, clarifying the physical meaning of each feature column (e.g., "Distribution Network A-phase Voltage Feature Column" represents the real-time voltage amplitude of 10kV line A-phase, "Capacitor Compensator Switching Response Time Feature Column" represents the time interval from the issuance of the capacitor switching command to the completion of the action), data source (regional distribution network distributed acquisition terminal, reactive power compensation equipment sensor, third-party monitoring unit), and operating condition correlation attributes (marking the core adaptable operating condition corresponding to the feature, such as load rate feature associating heavy load / light load operating conditions). A scenario-based calibration table for feature columns is generated to avoid confusion between feature columns from different sources and with different meanings. The design incorporates a mutual information entropy calculation logic adapted to the time-series characteristics of the distribution network. Its core innovation lies in introducing time-series window constraints and physical correlation weight corrections. First, a time-series window is set for each feature column (the window length is set based on the time-series correlation of the distribution network data to ensure continuous coverage of feature changes). Within the time-series window, the joint probability distribution and marginal probability distribution of any two feature columns (denoted as feature column X and feature column Y) are estimated. During the estimation process, the operating condition correlation attribute weights of the two features are incorporated (the higher the attribute overlap, the greater the weight, ensuring more accurate calculation of redundancy for feature columns with similar physical meanings). An initial mutual information entropy value is calculated based on the estimated probability distribution, and then corrected using a time-series continuity correction coefficient (the correction coefficient is set based on the entropy value change amplitude of the two feature columns within adjacent time windows; the smaller the change amplitude, the closer the correction coefficient is to 1, ensuring more reliable quantification of redundancy for time-stable feature columns). Finally, the corrected mutual information entropy value is obtained, which accurately represents the comprehensive information redundancy of the two feature columns in both the time and physical dimensions.A redundancy threshold calibration rule adapted to different operating conditions is constructed, abandoning the traditional fixed threshold mode: Based on the feature purity requirements of the coordinated management and control of distribution network and compensation equipment, historical operation and maintenance data (covering multiple typical operating conditions) are selected in a recent period. Feature column combinations are extracted from the historical data, and the mutual information entropy value of each combination is calculated. Combining the impact of feature redundancy on fault identification accuracy in historical fault cases, a preset redundancy threshold is calibrated for different operating conditions (e.g., the threshold is slightly lower under heavy load conditions to ensure that more differentiated features related to heavy load are retained; the threshold is slightly higher under light load conditions to simplify redundant features and improve processing efficiency). The thresholds corresponding to different operating conditions are integrated into a threshold mapping table, and the operating condition attributes in the feature column scenario calibration table are associated to match the corresponding preset redundancy threshold for each pair of feature columns. The modified mutual information entropy values ​​of any two feature columns are compared with a preset threshold for matching. If the entropy value is higher than the preset threshold, it is determined that the pair of feature columns is redundant. A removal mark is added to the feature column with lower information contribution (the information contribution is determined based on the contribution weight of the feature to the working condition identification, and the contribution weight is extracted from the feature column scenario calibration table). After traversing all feature column combinations and completing the redundancy marking, a list of redundant feature columns with removal marks and a list of retained non-redundant feature columns are integrated. The marking information is embedded into the feature column attributes of the original equipment-working condition association feature tensor, and finally a marked equipment-working condition association feature tensor is formed. This tensor retains the original temporal dimension and correlation information, and clarifies the redundancy status of each feature column, which can be directly input into the subsequent feature dimension pruning and purification layer.

[0042] Preferably, in the specific technical implementation of step 132: the driving feature dimension pruning and purification layer loads the device-operating condition associated feature tensor with removal marks, prunes and removes redundant feature columns according to the marks, deletes all feature columns with removal marks, and obtains a pre-purified feature tensor. A feature importance weight calculation logic specific to the distribution network scenario is designed. The weight calculation is based on the degree of influence of the feature on the stability of the distribution network operating conditions, the close correlation between the feature and the safe operation of reactive power compensation equipment, and the contribution of the feature in fault precursor identification. A comprehensive importance weight value is generated by integrating the above three-dimensional indicators. Based on the comprehensive importance weight value, the remaining feature columns in the pre-purified feature tensor are sorted in descending order, and the top-ranked feature columns are selected as core feature dimensions (the number of selections is set according to the efficiency requirements and feature representation integrity requirements of the subsequent feature index tree construction, ensuring that the core features are both concise and fully cover the key information of the collaborative state). The core feature dimensions after screening are reorganized according to the time dimension and correlation dimension of the original equipment-operating condition correlation feature tensor to generate a core feature tensor. The three dimensions of the core feature tensor still retain the collection time series, the core distribution network operation feature column, and the core compensation equipment feature column. The tensor element values ​​synchronously retain the corresponding feature correlation information to ensure that the temporal correlation of the core features and the operating condition-equipment correlation attributes are not lost.

[0043] Step 132 constructs a feature purification logic specifically for the distribution network scenario, consisting of "precise redundancy removal - multi-dimensional dynamic weight calculation - operating condition adaptation screening." This addresses the problems of generalized weight calculations, fixed screening criteria, and failure to align with the collaborative management needs of distribution networks and reactive power compensation equipment in traditional feature pruning. This implementation closely revolves around three core objectives: distribution network operating condition stability, equipment safe operation, and fault precursor identification. By dynamically integrating multi-dimensional importance indicators and operating condition-differentiated screening, it ensures that the generated core feature tensors are both concise and efficient, while fully preserving key information about the collaborative operating state, providing a high-quality feature foundation for subsequent attention-enhanced aggregation.

[0044] Preferably, in the specific technical implementation of step 132: the driving feature dimension pruning and purification layer loads the device-operating condition associated feature tensor with removal marks, and first performs mark validity verification and feature column dependency analysis—comparing the previously generated feature column scenario calibration table and redundant feature column list, verifying whether the generation basis of each removal mark (mutual information entropy value, preset threshold, information contribution determination result) is complete and valid, and at the same time analyzing the dependency relationship between the feature columns to be removed and the feature columns to be retained (such as the derived dependency between "three-phase average voltage feature column of distribution network" and "single-phase voltage feature column"). To avoid compromising the integrity of core features due to the removal of redundant features, a label verification and dependency analysis report is generated. Based on this report, pruning and removal are performed on feature columns with removal labels that have no dependencies or whose dependencies can be separated. For redundant feature columns with strong dependencies, a strategy of "retaining core derivative sources + removing derivative features" is adopted (such as retaining single-phase voltage feature columns and removing three-phase average voltage feature columns). Finally, all redundant feature columns that meet the removal conditions are deleted, resulting in a preliminarily purified feature tensor. A pruning log is generated simultaneously to record the number of removed features, the reasons, and the dependency handling methods. The design incorporates a feature importance weight calculation logic specific to power distribution network scenarios, constructing a weight calculation framework of "three-dimensional indicators + dynamic adaptation of operating conditions": The first dimension is the degree of influence of features on the stability of power distribution network operating conditions, determined by quantifying the voltage deviation, frequency shift, and load rate fluctuation amplitude caused by feature fluctuations. The greater the fluctuation amplitude, the higher the weight of the influence. At the same time, the corresponding operating conditions of the feature are associated (e.g., the weight of the influence of load rate features increases under heavy load conditions). The second dimension is the close correlation between features and the safe operation of reactive power compensation equipment. Based on the equipment feature classification spectrum, the correlation strength between features and core safety indicators of equipment (e.g., insulation strength, over-temperature protection threshold, and switching mechanism life) is quantified. The correlation strength is determined through equipment fault mechanism analysis (e.g., insulation resistance features are directly related to equipment insulation breakdown faults, and the correlation closeness weight is high). The third dimension is the contribution of features in fault precursor identification. By mining historical fault case data, the frequency and amplitude of abnormal changes of features within a preset period before the fault occurs are statistically analyzed. The higher the frequency and amplitude of abnormal changes, the higher the contribution weight. A basic weighting percentage is set for each of the three dimensions of indicators, and then the percentage is dynamically adjusted based on the current real-time operating conditions of the distribution network (determined by load rate, voltage, and other features in the pre-purified feature tensor). For example, under the fault precursor condition, the weighting percentage of the fault precursor identification contribution indicator is increased; under the heavy load condition, the weighting percentage of the equipment safety operation correlation indicator is increased. The scores of the three dimensions of indicators are integrated by a weighted summation formula to generate a comprehensive importance weight value for each feature column, ensuring that the weight value covers the core objectives of collaborative management and control, and adapts to the real-time operating conditions.Based on the comprehensive importance weight value, the remaining feature columns in the initially purified feature tensor are sorted in descending order to generate a feature importance ranking table, which includes the feature name, three-dimensional index scores, comprehensive weight value, and ranking position. A dynamic screening mechanism is constructed to determine the number and criteria for screening core feature dimensions: the number of screenings is set based on the efficiency requirements of subsequent feature index tree construction (controlling the number of features to improve the index construction and retrieval speed) and the feature representation integrity requirements (ensuring coverage of the three major objectives of operating condition stability, equipment safety, and fault identification). At the same time, a comprehensive importance weight threshold is set (the threshold is determined based on historical data verification to ensure that the total comprehensive weight value of the core features after screening meets the preset requirements). Feature columns with the highest ranking and comprehensive weight values ​​higher than the threshold are selected as core feature dimensions. The effectiveness of the screened core feature dimensions is verified by calculating the representation coverage of the core features on the coordinated state of the distribution network and compensation equipment (which needs to cover the preset key state dimensions) and the mutual information entropy value between core features (ensuring no redundancy). After the verification is passed, the final core feature dimensions are determined. The selected core feature dimensions are reorganized according to the time dimension and correlation dimension of the original equipment-operating condition related feature tensor: the time granularity and node order of the original acquisition time series are retained, and the second and third dimensions are reorganized according to the classification method of "core feature column of distribution network operation + core feature column of reactive power compensation equipment". The tensor element values ​​simultaneously retain the corresponding feature correlation information (representing the physical correlation strength between core features). At the same time, the comprehensive importance weight value and operating condition adaptation attribute are embedded in the feature column attributes of the tensor to generate the core feature tensor. The core feature tensor is then subjected to a final dimensional integrity and information consistency check to confirm that its three dimensions (acquisition time series, core distribution network operation feature column, core compensation equipment feature column) are complete, and that the tensor element values ​​and feature correlation information match correctly. This ensures that the temporal correlation of the core features and the operating condition-equipment correlation attributes are not lost. After the check passes, the core feature tensor and feature dimension description document are output.

[0045] Preferably, in one scenario, step 133 is specifically implemented as follows: The attention mechanism layer loads the core feature tensor, constructing an attention scoring function adapted to multiple operating conditions of the distribution network. The input to this function is the feature value in the core feature tensor, the operating condition type identifier corresponding to the feature, and the correlation between the feature and the cooperative state. The output is the attention score of each core feature. The core design of the attention scoring function is to differentiate the influence weight of different core features on the cooperative state under various operating conditions. For example, under heavy load conditions, the attention score of "compensation equipment reactive power output amplitude feature" is higher than that under light load conditions; under fault precursor conditions, the attention score of "compensation equipment insulation resistance feature" is higher than that under normal conditions. Based on the attention score, the operating condition influence weight coefficient corresponding to each core feature is calculated. The sum of the weight coefficients is 1. The higher the score of the core feature, the larger the weight coefficient, ensuring that the information of key features is emphasized. The feature values ​​of each core feature and the corresponding operating condition influence weight coefficient are weighted and aggregated. During the aggregation process, the temporal dimension information and operating condition correlation attributes of each core feature are retained, generating a distribution network-compensation equipment cooperative state feature topology. This topology is a multi-level topology structure. The top layer is the overall representation of the cooperative state, the middle layer is divided into sub-topologies according to the operating condition type, and the bottom layer is the core features and weighted aggregated feature values ​​under the corresponding operating condition. It also includes the operating condition influence weight coefficient and correlation information of each feature, which fully represents the cooperative operation status of the distribution network and compensation equipment under different operating conditions.

[0046] Step 133 constructs an attention enhancement logic that adapts to the multi-condition characteristics of the distribution network, employing "precise condition adaptation, dynamic weight allocation, and hierarchical topology aggregation." This addresses the problems of generalized weight allocation in traditional attention mechanisms, failure to consider the characteristics of distribution network operation scenarios, and fragmented information representation after feature aggregation. This implementation closely revolves around the core requirement of coordinated operation between the distribution network and reactive power compensation equipment. By designing scenario-specific attention scoring functions, it differentiates and strengthens core features that play a crucial role in the coordinated state under different operating conditions. Then, it aggregates feature information using a hierarchical topology structure to generate a distribution network-compensation equipment coordinated state feature topology that combines temporal correlation, condition adaptability, and information completeness. This provides a highly identifiable and adaptable core feature carrier for the subsequent construction of a distribution transformer-switch coordinated state feature index tree.

[0047] Preferably, in one scenario, step 133 is specifically implemented as follows: The attention mechanism layer loads the core feature tensor, performs dimensional analysis and information extraction on the core feature tensor, clarifies the specific values ​​and physical meanings of its three dimensions (collection time series, core distribution network operation feature column, core compensation equipment feature column), and simultaneously extracts the comprehensive importance weight value, operating condition adaptation attribute, and other information attached to each core feature column to generate a core feature information list; based on the core feature information list and the feature values ​​(such as load rate, reactive power output amplitude, etc.) in the core feature tensor, the real-time operating condition type of the current distribution network is determined, and the determination is based on the preset operating condition classification rules (such as a high load rate being a heavy load condition, and abnormal fluctuations in insulation resistance feature value being a fault precursor condition), and an operating condition type identification table is generated, establishing an association between each core feature column and the corresponding operating condition type to ensure that the attention enhancement process accurately adapts to the real-time operating condition. An attention scoring function adapted to multiple operating conditions in a distribution network is constructed. The input parameters of this function include eigenvalues ​​in the core feature tensor, operating condition type identifiers from the operating condition type identifier table, the correlation between core features and cooperative states (derived from the core feature tensor element values), and the comprehensive importance weight of the core features. The output parameter is the attention score of each core feature under the current operating condition. Internally, the function employs a three-dimensional integrated logic of "operating condition adaptation factor + feature correlation enhancement factor + time series stability correction factor": the operating condition adaptation factor is dynamically adjusted according to the operating condition type. For example, under heavy load conditions, a higher operating condition adaptation factor is assigned to features related to reactive power output of compensation equipment; under fault precursor conditions... The following parameters are assigned higher operating condition adaptation factors to safety-related features such as insulation resistance and temperature, while factors are evenly distributed under normal operating conditions. The feature correlation enhancement factor is positively correlated with the correlation between the core feature and the collaborative state; the higher the correlation, the larger the factor value. The temporal stability correction factor is set based on the change amplitude of the core feature within a continuous time window; the larger the change amplitude (i.e., the lower the temporal stability), the smaller the correction factor, to avoid unstable features excessively affecting attention allocation. The three factors are integrated with the core feature value through weighted summation to obtain the attention score of each core feature, ensuring that the score can differentiate the degree of influence of the core feature on the collaborative state under the current operating condition. Based on the attention score, the operating condition influence weight coefficients corresponding to each core feature are calculated. A normalization method is used to convert the attention scores of all core features into weight coefficients that sum to 1. The higher the score of the core feature, the larger the weight coefficient. Simultaneously, a weight rationality check is performed. The check standard is whether the weight coefficient of the core feature under the critical operating condition reaches the preset range (e.g., the weight coefficient of the insulation resistance feature under the fault precursor operating condition should not be lower than a certain range). If it does not meet the standard, the factor parameters of the attention scoring function are adjusted backtracking until the weight coefficient meets the requirements. A core feature-operating condition influence weight coefficient mapping table is generated to clarify the weight value of each core feature under the corresponding operating condition.The feature values ​​of each core feature are weighted and aggregated with the corresponding operating condition influence weight coefficients at each time point. During the aggregation process, the collection timestamp information and operating condition type identifier of each core feature are retained to ensure that the temporal correlation and operating condition adaptation attributes are not lost. Based on the aggregation results, a distribution network-compensation equipment coordinated state feature topology is constructed. This topology adopts a multi-level tree topology structure: the top layer is the overall representation node of the coordinated state, which stores the overall state value after weighted aggregation of core features under all operating conditions; the middle layer is divided into sub-topology nodes according to operating condition type, and each sub-topology node corresponds to a typical operating condition (such as heavy load, light load, fault precursor, etc.), storing the aggregated feature value under that operating condition; the bottom layer is the core feature nodes under each operating condition sub-topology node, and each core feature node stores the corresponding feature value, operating condition influence weight coefficient, correlation with coordinated state, and comprehensive importance weight value; at the same time, index information is added to the topology to associate the collection node, data source, and other attributes of the core features, generating a topology structure description document. The topology of the distribution network-compensation equipment collaborative status features clearly presents the collaborative operation status of the distribution network and compensation equipment under different operating conditions through a hierarchical structure. Key feature information is precisely enhanced, which can directly support the hierarchical construction and feature retrieval of the subsequent distribution transformer-switcher collaborative status feature index tree.

[0048] Optionally, step 2 includes: Step 21: Extract voltage deviation gradient features, power factor resonance features, and equipment response delay features from the topology of the distribution network-compensation equipment cooperative state features, and construct a multi-objective optimization objective functional and constraint boundary conditions; Step 22: Based on the multi-objective optimization objective functional and constraint boundary conditions, construct a multi-type reactive power compensation collaborative optimization model. The collaborative optimization model embeds a device response characteristic differential weight adaptation mechanism. Step 23: Solve the collaborative optimization model using the constructed multi-objective particle swarm collaborative optimization model, and output the compensation capacity allocation coefficients and action timing topology tables of various types of reactive power compensation equipment to generate a collaborative control strategy sequence.

[0049] Optionally, step 21 includes: Step 211: Extract the node voltage deviation gradient value, system power factor resonance coefficient, and voltage fluctuation attenuation coefficient from the topology of the distribution network-compensation equipment cooperative state characteristics as optimization target parameters, and construct a multi-objective optimization objective functional with the objectives of minimizing voltage deviation gradient, optimizing power factor resonance, and maximizing voltage fluctuation attenuation. Step 212: Combine the rated capacity threshold, maximum response rate, and number of action threshold constraints of various types of reactive power compensation equipment with the capacity carrying capacity constraints of the distribution network lines to construct collaborative optimization constraint boundary conditions; Step 213: Standardize and normalize the multi-objective optimization objective functional and the collaborative optimization constraint boundary conditions to generate the multi-objective optimization objective functional and constraint boundary conditions.

[0050] Steps 211-213 construct a progressive multi-objective optimization foundation logic adapted to the collaborative management and control scenario of distribution network and reactive power compensation equipment. This logic involves "scenario-specific target parameter extraction - multi-dimensional collaborative constraint construction - standardized adaptation processing," addressing the problems of generalized target parameters, failure to consider the collaborative characteristics of equipment and grid in constraint boundaries, and low optimization accuracy due to inconsistent dimensions of targets and constraints in traditional multi-objective optimization. This implementation closely revolves around the core needs of distribution network voltage stability control and power balance regulation. By extracting optimization target parameters tailored to the scenario characteristics, integrating the dual constraints of multiple types of compensation equipment and distribution network lines, and then eliminating dimensional and order-of-magnitude differences through standardization and normalization processing, it provides an accurate and adaptable optimization foundation for subsequent collaborative optimization, ensuring that the optimization results not only meet the requirements of safe operation of the distribution network but also adapt to the working characteristics of multiple types of reactive power compensation equipment.

[0051] Preferably, the specific implementation process of step 211 is as follows: The driving parameter extraction module loads the distribution network-compensation equipment cooperative state feature topology, first parses the hierarchical structure of the topology, clarifies the specific relationship between the top-level cooperative state general representation, the middle-level operating condition sub-topology (such as heavy load, fault precursor, normal operating condition) and the bottom-level core feature nodes, and simultaneously extracts the original data (including node voltage time series data, system power factor time series data, voltage fluctuation time series data) related to the distribution network voltage stability and power balance in the bottom-level core feature nodes, as well as the collection timestamp and operating condition type identifier corresponding to each data, and generates the original feature data list. Based on the original feature data list, three scenario-specific optimization target parameters are extracted. The extraction logic and physical meaning of each parameter are as follows: Extraction of node voltage deviation gradient values: First, the rated voltage reference value of each node is calibrated based on the distribution network voltage level (e.g., 10kV) (for example, the rated voltage reference value of a 10kV distribution network node is 10kV ± a certain range). The real-time voltage values ​​of each node within a continuous time window are extracted from the original feature data list. The deviation value between the real-time voltage value and the rated voltage reference value of each node is calculated. Then, the gradient is calculated using the deviation values ​​of adjacent time points (gradient value = difference in deviation between adjacent time points / time interval). The average of all node gradient values ​​is taken as the final node voltage deviation gradient value. This parameter accurately characterizes the dynamic change trend of distribution network voltage stability; the smaller the gradient value, the better the voltage stability. Extraction of system power factor resonance coefficient: First, the real-time time-series data of the system power factor is extracted from the original feature data list, combined with the distribution... The optimal power factor range is determined based on the current operating condition type of the network (obtained from the sub-topology of the middle layer operating condition) (e.g., a certain range under normal operating conditions and a certain range under heavy load conditions). The deviation of the real-time power factor from the center value of the optimal range is calculated. Then, a resonant frequency correction factor is introduced (calculated based on the inductance and capacitance parameters of the distribution network lines to avoid line resonance caused by power factor adjustment). The deviation is multiplied by the correction factor to obtain the system power factor resonance coefficient. This coefficient characterizes the degree of power factor deviation from the optimal state and the risk of resonance. When the coefficient approaches 0, it is in the optimal state. The voltage fluctuation attenuation coefficient is extracted by extracting the amplitude data and attenuation time data of voltage fluctuation from the original feature data list. Based on the allowable range of voltage fluctuation in the distribution network (e.g., ± a certain range), the reference attenuation rate is calibrated. The ratio of the actual voltage fluctuation attenuation rate to the reference attenuation rate is calculated to obtain the voltage fluctuation attenuation coefficient. The larger the coefficient, the faster the voltage fluctuation attenuation and the stronger the voltage recovery stability of the distribution network.Based on the three extracted optimization target parameters, a multi-objective optimization objective functional is constructed. The functional adopts the logic of "operating condition adaptation weight + target priority integration": the weight of each objective parameter is dynamically allocated according to the current operating condition type identifier (such as increasing the weight of voltage fluctuation attenuation coefficient under heavy load condition, and increasing the weight of node voltage deviation gradient value under fault precursor condition). The objectives of minimizing node voltage deviation gradient, system power factor resonance coefficient approaching the optimal value (0), and maximizing voltage fluctuation attenuation coefficient are integrated into a unified functional expression. The input of the functional is the three optimization target parameters and the timely operating condition adaptation weight, and the output is the comprehensive optimization objective value, ensuring that the functional can accurately reflect the core requirements of the coordinated optimization of distribution network and compensation equipment under different operating conditions.

[0052] Preferably, in the specific technical implementation of step 212: the driving constraint construction module sorts out the technical parameter manuals and historical operation and maintenance data of various types of reactive power compensation equipment (such as parallel capacitor banks, static var generators, SVG), extracts the core operation constraint indicators of each equipment, and retrieves the design parameters of the distribution network line (including rated current carrying capacity, impedance, and capacity) to construct a list of constraint indicators and clarify the physical meaning and data source of each constraint. Based on the list of constraint indicators, collaborative optimization constraint boundary conditions are constructed in different dimensions. The specific construction logic of each constraint is as follows: Rated capacity threshold constraint for multiple types of reactive power compensation equipment: For each type of compensation equipment, the rated reactive power output capacity from its nameplate parameters is retrieved. Combined with the reactive power demand gap under the current operating conditions of the distribution network (extracted from the top-level general representation of the distribution network-compensation equipment collaborative state characteristic topology), the actual available capacity threshold of the equipment is calibrated (to avoid equipment overload operation; for example, the rated capacity of a certain type of parallel capacitor bank is a certain value, and the actual available capacity threshold is calibrated to a certain range based on the current reactive power gap), constraining the real-time reactive power output capacity of the equipment to not exceed this threshold; Maximum response rate constraint: Based on the dynamic response characteristic test data of each compensation equipment, the maximum response rate under different operating conditions is determined (e.g., the maximum response rate of SVG is higher than that of the parallel capacitor bank; for example, the maximum response rate of SVG is a certain value / ...). The system uses the following parameters: ms (parallel capacitor banks are defined as a certain value / ms), calibrating the upper limit of the response rate to constrain the reactive power output adjustment rate of the equipment from exceeding this limit, thus preventing excessively fast responses that could cause severe fluctuations in grid voltage; an action count threshold constraint, combining the mechanical life parameters of the compensation equipment (such as the lifespan of switching devices) and historical maintenance data, sets a maximum action count threshold per unit time (e.g., the maximum number of actions for a certain device is a certain number of times / hour), constraining the number of switching or adjustment actions of the equipment from exceeding this threshold, thus extending the equipment's lifespan; and a distribution network line capacity carrying capacity constraint, based on the rated current carrying capacity and impedance parameters of the distribution network lines, calculates the maximum reactive power that the line can carry under current operating conditions (line reactive power carrying capacity = √(rated current carrying capacity² - current active current²) × line rated voltage), calibrating the line capacity constraint threshold, and constraining the total reactive power output of all compensation equipment from exceeding this threshold, thus preventing line overload and overheating. A constraint co-validation logic is constructed to integrate the above four types of constraints into a set of co-optimization constraint boundary conditions. Through correlation validation, it is ensured that there are no logical conflicts between the constraints (such as matching the rated capacity constraint of the equipment with the line capacity constraint, and ensuring that the total output of all equipment does not exceed the line carrying capacity). The value range, validation rules and correlation with other constraints of each constraint are clarified, and a constraint boundary description document is generated to provide a clear constraint basis for the subsequent optimization process.

[0053] Preferably, in a scenario, step 213 is specifically implemented as follows: the driving standardization processing module loads the multi-objective optimization objective functional and the set of collaborative optimization constraint boundary conditions. First, the dimensions and orders of magnitude of all parameters are analyzed to identify the existing differences (e.g., the unit of the node voltage deviation gradient value is V / ms, and the unit of the compensation equipment response rate is kVar / s, which are different in dimensions; the order of magnitude of the line capacity constraint threshold and the equipment action number threshold are significantly different). The core objective of the standardization and normalization processing is to eliminate the differences in dimensions and orders of magnitude, ensure that the weights of each objective and constraint are balanced during the optimization process, and avoid a certain parameter dominating the optimization result. For different types of parameters, appropriate processing methods are adopted: For the node voltage deviation gradient value and voltage fluctuation attenuation coefficient in the multi-objective optimization objective functional, a linear normalization method is used. First, the maximum and minimum values ​​of the two parameters are statistically analyzed from historical operating data to establish a linear mapping relationship between the original values ​​and the normalized values ​​(mapped to a certain interval, such as [0,1]). The node voltage deviation gradient value adopts a reverse mapping (the larger the original value, the smaller the normalized value, which fits the minimization objective), and the voltage fluctuation attenuation coefficient adopts a forward mapping (the larger the original value, the larger the normalized value, which fits the maximization objective). For the system power factor resonance coefficient, standardization is adopted. The mean and standard deviation of this coefficient in historical data are calculated and then normalized. The formula converts the coefficient values ​​into values ​​under a standard normal distribution, making their mean 0 (fitting the goal of approaching the optimal value), and the standard deviation within a certain range (e.g., a standard deviation of 0.1), which facilitates subsequent accurate determination of the degree of deviation of the coefficients from the optimal state. For all indicators in the collaborative optimization constraint boundary conditions (such as rated capacity, response rate, number of actions, and line carrying capacity), per-unit value standardization is adopted. Based on the distribution network benchmark capacity (e.g., benchmark capacity is 100MVA) and benchmark voltage (e.g., benchmark voltage is 10kV), each constraint indicator is converted into a per-unit value (per-unit value = actual value / benchmark value), so that all constraint indicators are in the same order of magnitude (e.g., the 0-1 range), avoiding the imbalance of weights between constraints. After processing, the consistency of the standardized and normalized multi-objective optimization objective functional and constraint boundary conditions is verified. The verification includes the accuracy of parameter mapping, the logical rationality of the processed data (such as whether the per-unit value of the constraint threshold meets the actual operation requirements), and the adaptability of the objective and constraints (such as whether the normalization range of the optimization objective matches the per-unit value range of the constraints). After the verification is passed, a standardized multi-objective optimization objective functional and constraint boundary condition set is generated, along with a processing instruction document that clarifies the processing method of each parameter, the original data range, and the standardized value range, providing an accurate and unified data foundation for the subsequent collaborative optimization process.

[0054] Optionally, step 22 includes: Step 221: Based on the multi-objective optimization objective functional and constraint boundary conditions, construct a multi-type reactive power compensation collaborative optimization model; Step 222: Analyze the differences in response characteristics of various types of reactive power compensation equipment, construct a differentiated weight adaptation mechanism, assign dynamic response weight coefficients to fast response equipment, and assign capacity allocation weight coefficients to large-capacity compensation equipment. Step 223: Embed the differentiated weight adaptation mechanism into the objective functional solution process of the collaborative optimization model, so that the model solution process can adapt to the differences in technical characteristics of different types of equipment, thereby completing the construction of a multi-type reactive power compensation collaborative optimization model.

[0055] Steps 221-223 construct a progressive model building logic adapted to the collaborative management and control scenarios of multiple types of reactive power compensation equipment in the distribution network. This logic involves "scenario-based optimization model construction - equipment characteristic-differentiated weight adaptation - deep embedding of weight mechanism into the solution process." This addresses the problems of traditional reactive power compensation collaborative optimization models, such as failure to consider the differences in response characteristics of different types of equipment, generalized weight allocation, and poor adaptability between model solution and equipment technical characteristics, leading to unsatisfactory collaborative results. This implementation closely revolves around the core requirements of distribution network voltage stability control and efficient collaborative operation of multiple devices. By constructing a basic optimization model framework, it designs a differentiated weight adaptation mechanism tailored to equipment characteristics, and then deeply integrates this mechanism into the model solution process. This ensures that the model solution process accurately adapts to the technical characteristics of different types of equipment, such as fast-response and high-capacity compensation equipment. The final collaborative optimization result not only meets the real-time operation and control requirements of the distribution network but also fully leverages the advantages of each type of equipment, improving the overall collaborative compensation effect.

[0056] Preferably, the specific implementation process of step 221 is as follows: The driving model construction module loads the standardized multi-objective optimization objective functional and constraint boundary condition set. First, it performs correlation analysis on the two types of data to clarify the specific quantitative expressions of the three objectives in the multi-objective optimization objective functional: minimizing the node voltage deviation gradient, optimizing the system power factor resonance, and maximizing the voltage fluctuation attenuation. It also clarifies the value range and verification rules of the equipment rated capacity threshold, maximum response rate, action number threshold, and line capacity carrying capacity constraint in the constraint boundary condition set, generating an objective-constraint correlation list. Based on this list, the basic framework of the multi-type reactive power compensation collaborative optimization model is constructed. The framework includes four core layers: input layer, objective solution layer, constraint verification layer, and output layer. The functions and data flow logic of each layer are as follows: The input layer is responsible for receiving the standardized optimization objective parameters (node ​​voltage deviation gradient value, system power factor resonance coefficient, voltage fluctuation attenuation coefficient), and various types of reactive power compensation. The real-time operating parameters of the equipment (such as current output capacity, remaining capacity, and real-time response rate) and the real-time operating status of the distribution network are formatted and transmitted to the target calculation layer. The target calculation layer loads a multi-objective optimization objective functional to perform preliminary calculations on the optimization objective parameters transmitted from the input layer, generating an initial reactive power compensation allocation scheme (including preliminary reactive power output instructions for each device). The constraint verification layer loads a set of constraint boundary conditions to perform constraint-by-constraint verification on the initial reactive power compensation allocation scheme (such as verifying whether the output capacity of each device exceeds the rated capacity threshold, whether the total output exceeds the line carrying capacity constraint, etc.), and marks the scheme content that violates the constraints. The output layer receives the verification results from the constraint verification layer. If the scheme meets the constraints, it is directly output. If there are any content that violates the constraints, it is fed back to the target calculation layer for iterative adjustment, generating an initial multi-type reactive power compensation collaborative optimization model, and simultaneously outputting a model framework description document, clarifying the input and output parameters and data flow rules of each level.

[0057] Preferably, in the specific technical implementation of step 222: the drive characteristic analysis module retrieves the technical parameter manuals, dynamic response test data, and historical operation and maintenance records of various types of reactive power compensation equipment, extracts the core response characteristic parameters of each type of equipment, and constructs a matrix representing the differences in equipment response characteristics. The row index of this matrix is ​​the equipment type (such as parallel capacitor banks, static var generators (SVG), and controllable reactors), and the column index is the response characteristic index (response delay time, maximum response rate, rated compensation capacity, capacity adjustment accuracy, and mechanical life). The matrix element values ​​are the specific values ​​of the corresponding equipment under the corresponding characteristic index (for example, the response delay time of SVG is within a certain range of ms, and the rated compensation capacity of parallel capacitor banks is within a certain range of kVar), clearly representing the characteristic differences of different types of equipment. Based on the device response characteristic difference representation matrix, a differentiated weight adaptation mechanism is constructed. This mechanism includes a fast response weight adaptation branch and a capacity allocation weight adaptation branch, which adapt to the core characteristics of different types of devices respectively. The fast response weight adaptation branch is for fast response devices such as SVG, and constructs the calculation logic of dynamic response weight coefficients. The coefficient values ​​are positively correlated with the device's real-time response rate and the voltage fluctuation amplitude of the distribution network, and negatively correlated with the response delay time. At the same time, it is dynamically adjusted in conjunction with the real-time operating condition indicators of the distribution network (e.g., when the voltage fluctuation amplitude exceeds a certain range, the dynamic response weight coefficient is increased by a certain percentage). The coefficient value range is a certain interval (e.g., 0.6-0). .9) Ensure that fast-response equipment is prioritized for dispatch under conditions of severe voltage fluctuations; the capacity allocation weight adaptation branch targets large-capacity compensation equipment such as parallel capacitor banks, constructing the calculation logic for capacity allocation weight coefficients. The coefficient value is positively correlated with the remaining compensation capacity of the equipment and the reactive power demand gap of the distribution network, and negatively correlated with the recent number of equipment operations (to avoid excessively frequent operations affecting lifespan). It also incorporates dynamic adjustments based on operating conditions (e.g., under heavy load conditions with a large reactive power demand gap, the capacity allocation weight coefficient is increased by a certain percentage). The coefficient value range is within a certain interval (e.g., 0.5-0.8), ensuring that large-capacity equipment fully utilizes its capacity advantage under conditions of large reactive power gaps. The weight coefficient calculation logic of the two branches is associated with equipment type and operating condition identifiers to generate a differentiated weight adaptation rule set, clarifying the calculation method and value basis for the weight coefficients of different equipment types under different operating conditions.

[0058] Preferably, in one scenario, step 223 is specifically implemented as follows: The driving mechanism embedding module loads the initial multi-type reactive power compensation collaborative optimization model and the differentiated weight adaptation rule set. First, it performs structural analysis on the target solution layer of the initial model, clarifies the solution priority setting logic of its multi-objective optimization objective functional (initially equal priority), and determines the embedding position of the differentiated weight coefficients as the functional solution process of the target solution layer. The weight coefficients corresponding to the differentiated weight adaptation rule set are embedded into the functional solution expression of the target solution layer, and the solution priority of each objective is adjusted: for the optimization objective corresponding to fast-response equipment (such as maximum voltage fluctuation attenuation), a dynamic response weight coefficient is introduced to increase its solution priority; for the optimization objective corresponding to large-capacity compensation equipment (such as optimal system power factor resonance), a capacity allocation weight coefficient is introduced to increase its solution priority, so that the weight ratio of each objective can be dynamically adjusted according to the equipment characteristics and operating conditions during the functional solution process. A weight-solution collaborative verification logic is constructed to verify the model after embedding weight coefficients. Historical data from different typical operating conditions (heavy load, severe voltage fluctuation, and normal operating conditions) are selected as input for the trial calculation to obtain the reactive power compensation allocation scheme. The scheduling logic of each device in the scheme is verified to ensure it conforms to its characteristics (e.g., priority scheduling for fast-response devices during severe voltage fluctuations, and priority scheduling for large-capacity devices during heavy loads) and whether the scheme meets all constraint boundary conditions. If there are cases where the scheduling logic does not match the device characteristics or violates constraints, the calculation parameters of the differentiated weight coefficients are adjusted retrospectively (e.g., adjusting the value of the operating condition adaptation factor) until the trial calculation scheme meets the requirements. After the trial calculation is passed, the target solution layer logic with embedded differentiated weight adaptation mechanism is solidified, and the constraint verification layer of the model is updated synchronously to include the value range of the weight coefficients in the constraint verification range (to avoid solution deviation caused by abnormal weight coefficients). The construction of multi-type reactive power compensation collaborative optimization models is completed, and the final model file is generated, which includes a complete hierarchical structure, weight adaptation rules, solution logic, and constraint verification rules, and can be directly used for the generation of subsequent reactive power compensation collaborative scheduling instructions.

[0059] Optionally, step 23 includes: Step 231: Use a multi-objective particle swarm cooperative optimization model to solve the multi-type reactive power compensation cooperative optimization model. Initialize the population as a combination of compensation capacity and action timing solution vector, and set the iteration termination criterion. Step 232: Calculate the fitness value of each combination solution vector in the initial population using the fitness evaluation function, and perform selection, crossover, mutation and evolution operations based on the fitness value to generate a new generation population; Step 233: Repeat the iterative evolution process until the termination criterion is met. Select the Pareto optimal solution from the final population, analyze the optimal solution to obtain the compensation capacity allocation coefficient and action timing topology table of each type of reactive power compensation equipment, and generate a coordinated control strategy sequence.

[0060] Steps 231-233 construct a multi-objective particle swarm optimization logic adapted to various reactive power compensation collaborative control scenarios in distribution networks. This logic, which involves "scenario-based vector design, multi-objective adaptation fitness evaluation, and precise selection of Pareto optimal solutions," addresses the problems of traditional particle swarm optimization algorithms, such as solution vector generalization, fitness evaluation not being tailored to distribution network needs, and optimal solution selection not considering equipment characteristics and operating condition adaptability. This implementation closely revolves around the core objectives of distribution network voltage stability, optimal power factor, and voltage fluctuation attenuation. It combines the characteristic differences of various types of reactive power compensation equipment with customized population initialization, fitness evaluation, and iterative evolution mechanisms to ensure that the optimization process accurately adapts to distribution network operating conditions and equipment technical characteristics. The resulting collaborative control strategy sequence satisfies multi-objective optimization requirements while ensuring efficient collaborative operation of all equipment.

[0061] Preferably, the specific implementation process of step 231 is as follows: The driving optimization solution module loads the constructed multi-type reactive power compensation collaborative optimization model, first analyzes the multi-objective optimization objective functional, constraint boundary conditions and differentiated weight adaptation rules in the model, clarifies the core objective of optimization (minimizing node voltage deviation gradient, optimizing system power factor resonance, maximizing voltage fluctuation attenuation), constraint range (equipment rated capacity, response rate, number of actions and line capacity carrying constraints) and weight adaptation logic of each device, and generates an optimization parameter list. Based on the optimization parameter list, an initial combined solution vector structure for the population is designed. Each combined solution vector is a multi-dimensional vector, with the vector dimension corresponding to the sum of the number of various types of reactive power compensation devices and the number of action timing nodes. The vector elements are divided into two categories: one is the compensation capacity allocation ratio of each type of reactive power compensation device (within a certain range, e.g., 0-1, ensuring that the sum of the allocation ratios of all devices is 1), and the other is the action timing node of each device (with the value being a timestamp based on the time series collected from the distribution network, e.g., with a time granularity of 10ms, the timing node value is 0ms, 10ms, 20ms, etc.). Each combined solution vector completely represents a set of collaborative schemes of "compensation capacity allocation + action timing scheduling". An initial population is randomly generated according to a preset population size (the size is set according to the optimization efficiency and accuracy requirements, e.g., 50-100 devices). During the generation process, constraint verification is performed simultaneously to remove invalid solution vectors that exceed the device's rated capacity threshold for compensation capacity allocation ratio or violate the device's response rate constraint for action timing nodes, and to supplement valid solution vectors that meet the constraints, ensuring that the initial population consists of feasible solutions. The iteration termination criterion is set using a "dual convergence criterion + maximum iteration count fallback" logic: The first criterion is the fitness value convergence criterion, which is determined to be convergent when the change in the optimal fitness value of the population over several consecutive generations is less than a preset threshold (e.g., 0.01); the second criterion is the optimization objective achievement criterion, which is determined to be achieved when the multi-objective optimization objective values ​​corresponding to the solution vector in the population all meet the preset requirements (node ​​voltage deviation gradient is less than a certain value, system power factor resonance coefficient approaches 0, voltage fluctuation attenuation coefficient is greater than a certain value); the maximum iteration count is set to a certain range (e.g., 100-200 generations), and when the iteration count reaches the maximum value, the iteration is terminated regardless of whether convergence has occurred, and the initial population and iteration parameter configuration document are generated.

[0062] Preferably, in the specific technical implementation of step 232: the driving fitness evaluation module constructs a scenario-based fitness evaluation function based on the multi-objective optimization objective functional and differentiated weight adaptation rules. The input of this function is the combined solution vector, the real-time operating condition identifier of the distribution network, and the real-time operating parameters of the equipment. The output is the comprehensive fitness value. The function construction logic is as follows: First, the compensation capacity allocation ratio and action timing nodes in the combined solution vector are substituted into the multi-objective optimization objective functional to calculate three single objective values: node voltage deviation gradient value, system power factor resonance coefficient, and voltage fluctuation attenuation coefficient. Then, the weight coefficients corresponding to each objective under the current operating condition are retrieved from the differentiated weight adaptation rules (e.g., the weight of the voltage fluctuation attenuation coefficient is increased under severe voltage fluctuation conditions). The three single objective values ​​are weighted and summed with the corresponding weight coefficients to obtain the comprehensive fitness value. The higher the fitness value, the better the cooperative scheme corresponding to the solution vector. The constructed fitness evaluation function is used to calculate the comprehensive fitness value of each combined solution vector in the initial population, generating a population fitness ranking table. The table contains the solution vector number, each single objective value, the comprehensive fitness value, and the ranking position. Selection, crossover, and mutation evolution operations are performed based on a fitness ranking table: The selection operation uses a combination of "elite retention + roulette wheel selection" logic, prioritizing the retention of elite solution vectors with high ranking (e.g., retaining the top 20%), and the remaining solution vectors are selected using roulette wheel selection to ensure the inheritance of superior characteristics of the population; The crossover operation is performed separately for the "compensation capacity allocation segment" and the "action time segment" of the combined solution vectors. The compensation capacity allocation segment uses arithmetic crossover, and the action time segment uses staggered crossover. During the crossover process, it is ensured that the solution vectors after crossover still satisfy the compensation capacity allocation method. The sum of compensation capacity is 1, and the timing nodes meet the requirements of response rate constraints. The mutation operation adopts a dynamic mutation probability mechanism, and the mutation probability is negatively correlated with the convergence degree of the current population fitness (the mutation probability is increased when the fitness value has not converged, for example: 0.05-0.1; it is reduced to 0.01-0.03 when it converges). At the same time, it is combined with the real-time operating condition adjustment of the distribution network (the mutation probability is increased when the voltage fluctuation amplitude is large). During the mutation process, only some elements in the solution vector are slightly adjusted to avoid generating invalid solutions. Finally, a new generation of population is generated, and the optimal solution vector in the evolution process is recorded synchronously.

[0063] Preferably, in one scenario, step 233 is specifically implemented as follows: the driving iterative evolution module takes the new generation of population as input and repeatedly executes the fitness evaluation and selection, crossover, and mutation operations in step 232. During each iteration, the optimal fitness value and corresponding solution vector of the population are recorded, generating an iterative evolution log that clearly presents the fitness change trend and the evolution process of the optimal solution vector in each generation. When the iteration process meets the preset iteration termination criterion (fitness value convergence and optimization objective achieved, or the maximum number of iterations is reached), the iteration is terminated, and all non-dominated solution vectors that satisfy the optimization objective (i.e., no other solution vector is better than the optimal solution vector in all single objective values) are extracted during the iteration process to form a Pareto optimal solution set. The Pareto optimal solution set is further screened based on whether the corresponding collaborative scheme of the solution vector best suits the current distribution network conditions and equipment operating status. Combining real-time distribution network condition indicators (e.g., heavy load, severe voltage fluctuations), solution vectors with higher weight coefficients and better target values ​​under the corresponding conditions are prioritized. Combining real-time equipment operating parameters (e.g., remaining capacity, recent action count), solution vectors with fewer equipment actions and higher remaining capacity utilization are prioritized, ultimately selecting a unique Pareto optimal solution. This Pareto optimal solution vector is then analyzed to extract the compensation capacity allocation ratio for each type of reactive power compensation equipment, converting it into a specific compensation capacity allocation coefficient (allocation coefficient = allocation ratio × equipment rated capacity), ensuring the coefficient value is within the equipment rated capacity threshold range. The action timing nodes of each device are extracted and organized into an action timing topology table in timestamp order. This table includes information such as equipment type, action trigger time node, action duration, compensation capacity allocation coefficient, and triggering conditions (e.g., triggering when voltage fluctuation reaches a certain value). Based on the compensation capacity allocation coefficient and the action timing topology table, a coordinated control strategy sequence is generated. The sequence specifies the action order, execution parameters and coordination logic of each device. The strategy sequence is then subjected to a final constraint verification (verifying whether the device action timing meets the response rate, whether the total compensation capacity exceeds the line carrying capacity, etc.). After the verification is passed, the coordinated control strategy sequence and strategy description document are output. This sequence can be directly used to drive multiple types of reactive power compensation devices to perform coordinated control actions.

[0064] Optionally, step 3 includes: Step 31: Analyze the compensation capacity allocation coefficients and action timing topology table in the collaborative control strategy sequence, and integrate them with the control interface protocol specifications of various types of reactive power compensation equipment to generate a prototype of equipment-specific control instructions; Step 32: Based on the real-time update data of the topology of the distribution network-compensation equipment collaborative state characteristics, perform feasibility verification and parameter iterative correction on the prototype of the equipment-specific control command to generate heterogeneous adaptation execution commands; Step 33: Send the heterogeneous adaptation execution command to the corresponding type of reactive power compensation equipment, and simultaneously collect the action feedback data stream of each equipment to form a feedback control link, so as to coordinate the action of each type of reactive power compensation equipment and complementarily adjust the performance gain.

[0065] Optionally, step 31 includes: Step 311: Analyze the compensation capacity allocation coefficient and action timing topology table in the collaborative control strategy sequence, determine the action parameter thresholds and execution time nodes of each type of reactive power compensation equipment, and form a list of equipment action instructions; Step 312: Retrieve the control interface protocol specifications of various types of reactive power compensation equipment. The specifications include instruction encoding format, communication baud rate, and data verification mechanism. Based on the control interface specifications, convert the parameters in the equipment action instruction list into instruction formats that can be recognized by the corresponding equipment. Step 313: Perform syntax compliance verification on the converted instruction format, correct instruction fields with abnormal formats, and generate a prototype of device-specific control instructions.

[0066] Steps 311-313 construct a progressive device control command generation logic of "strategy parsing - interface adaptation - compliance verification" adapted to the collaborative management and control scenario of multiple types of reactive power compensation equipment in the distribution network. This solves the problems of generalized strategy parsing, failure to adapt to differences in interface protocols of different devices, and insufficient verification of command format compliance, which lead to commands not being recognized by devices or executing abnormally. This implementation closely revolves around the core requirement of collaborative operation of multiple types of reactive power compensation equipment (such as static var generators, parallel capacitor banks, etc.). It extracts the core parameters of device actions by accurately parsing the collaborative control strategy sequence, specifically adapts to the control interface protocol specifications of different devices, and then performs multi-dimensional syntax compliance verification and correction to generate accurate, adapted, and compliant device-specific control command prototypes. This ensures that the commands can be accurately recognized and executed by various types of devices, guaranteeing the reliability and effectiveness of multi-device collaborative control.

[0067] Preferably, the specific implementation process of step 311 is as follows: The drive strategy parsing module loads the cooperative control strategy sequence, first parses the structure of the sequence, clarifies the type identifiers of multiple types of reactive power compensation equipment (such as Static Var Generator, abbreviated as SVG; parallel capacitor bank), the compensation capacity allocation coefficients corresponding to each device, the action timing topology table and the cooperative logic description, and generates a strategy parsing list. Based on the strategy parsing list, the core information of the compensation capacity allocation coefficient and action timing topology table is extracted: from the compensation capacity allocation coefficient, the real-time compensation capacity target value, capacity adjustment step size (taken as a certain proportion of the equipment's rated capacity, for example: 5%-10%), and capacity adjustment direction (increase or decrease capacity) of each device are extracted, and the rated capacity parameters of the corresponding devices are retrieved simultaneously for verification to ensure that the extracted compensation capacity target value does not exceed the rated capacity threshold of the device; from the action timing topology table, the action trigger time node (based on the unified timestamp of the distribution network, for example: with 1ms as the time granularity, the trigger time node is 10ms, 25ms, etc.), action duration, action trigger preconditions (such as the distribution network voltage fluctuation coefficient reaching a certain value), and action termination conditions are extracted to generate a core parameter table of device actions. The core parameters of each device's actions are categorized and organized according to device type. The correspondence between the action parameter thresholds (such as the upper and lower limits of compensation capacity and the allowable range of action timing deviation) and the execution time nodes of each type of device is clarified, forming a structured list of device action instructions. The list includes fields such as unique device identifier, action type, core parameters, execution timing, and trigger / termination conditions, ensuring that the list information is complete and accurately corresponds to the action requirements of each device.

[0068] Preferably, in the specific technical implementation of step 312: the drive interface adaptation module retrieves a preset multi-type reactive power compensation equipment control interface protocol specification library. This specification library stores the corresponding interface protocol specifications according to the equipment type. Each specification contains three core elements: instruction encoding format (such as binary encoding, ASCII encoding, different formats are adapted to different devices, for example: SVG is adapted to binary encoding, and parallel capacitor banks are adapted to ASCII encoding), communication baud rate (the value is within a certain range, for example: 9600bps-115200bps, matched according to the device communication performance), and data verification mechanism (such as parity check, cyclic redundancy check, abbreviated as CRC, different verification methods are supported by different devices), and generates a device interface protocol parameter table. Based on the device interface protocol parameter table, the core parameters in the device action instruction list are converted one by one into the corresponding device-recognizable instruction format according to device type: for instruction encoding format, the decimal parameters in the list (such as compensation capacity value) are converted into byte streams of the corresponding encoding format; for communication baud rate, it is embedded as an additional configuration parameter for instruction transmission in the instruction header; for data verification mechanism, the checksum of the instruction data is calculated according to the corresponding rules and added to the instruction tail. During the conversion process, the association mapping between device type and interface protocol parameters is established simultaneously to ensure that the instruction conversion of each type of device accurately adapts to its interface specification, generate a preliminary conversion instruction set, record the conversion log synchronously, and clarify the conversion rules and corresponding relationships of each parameter.

[0069] Preferably, in one scenario, step 313 is specifically implemented as follows: The driver compliance verification module constructs a multi-dimensional syntax compliance verification rule base based on the control interface protocol specifications of multiple types of reactive power compensation equipment. The rule base includes five core rules: instruction field length verification, encoding format verification, check code validity verification, timing parameter rationality verification, and parameter threshold compliance verification. The specific verification logic of each rule is as follows: Instruction field length verification: Based on the instruction length range specified in the corresponding device interface specification, verify whether the total length of the fields of each instruction in the preliminary converted instruction set is within the allowable range; Encoding format verification: Verify whether the encoding format of the instruction is consistent with the device requirements through a format parsing tool (e.g., verify whether the byte bits of the binary encoding conform to the specification); Check code validity verification: Recalculate the check code of the instruction and compare it with the check code at the end of the instruction to determine whether they are consistent; Timing parameter rationality verification: Verify whether the action triggering time node in the instruction conforms to the device's response rate constraint (e.g., whether the triggering time interval is not less than the device's minimum response delay time, for example: not less than 5ms); Parameter threshold compliance verification: Verify whether the compensation capacity and other parameters in the instruction are within the device's rated parameter threshold range. The constructed verification rule base is used to verify the initial converted instruction set instruction by instruction and rule by rule. Instruction fields with format abnormalities (such as excessive field length, incorrect checksum, unreasonable timing parameters, etc.) are marked, and an abnormal instruction list is generated. For problematic fields in the abnormal instruction list, corrections are made based on the interface protocol specification and the core parameter table of device actions: if the field length exceeds the specification, it is truncated or padded with zeros according to the specification; if the checksum is incorrect, it is recalculated and replaced; if the timing parameter is unreasonable, the trigger time node is recalibrated in conjunction with the device response rate; if the parameter threshold exceeds the standard, the parameters are adjusted back to the device action instruction list and re-converted. After the correction is completed, compliance verification is performed again until all instructions comply with the verification rules, generating a prototype of device-specific control instructions. Verification and correction reports are output simultaneously, clearly stating the verification results, abnormal issues, and correction measures for each instruction.

[0070] Optionally, step 32 includes: Step 321: Extract the updated data of the cooperative status feature topology of the distribution network and compensation equipment in real time to obtain the current operating status parameters of various types of reactive power compensation equipment and the real-time voltage and power data of the distribution network, and generate a real-time verification data stack accordingly. Step 322: Compare and verify the action parameters in the prototype of the device-specific control command with the current status parameters of the device in the generated real-time verification data cluster to determine whether the action parameters are within the safe operating threshold range of the device. Step 323: Iteratively correct the action parameters that exceed the safe operation threshold range so that the adjusted parameters match the equipment operation constraints and distribution network operating conditions, and generate heterogeneous adaptation execution instructions accordingly.

[0071] Steps 321-323 construct a real-time closed-loop control logic adapted to the collaborative management and control scenario of distribution network and compensation equipment. This logic, consisting of "time-sequential real-time data acquisition, dual-dimensional comparison and verification of equipment and operating conditions, and iterative correction based on operating conditions," addresses the problems of lacking real-time operating condition verification, generalized parameter correction, and high risk of command execution due to failure to adapt to the dynamic operating status of equipment during the traditional control command generation process. This implementation closely revolves around the core requirements of safe operation of various types of reactive power compensation equipment and dynamic adaptation to distribution network operating conditions. It constructs a verification benchmark by extracting real-time updated data of the collaborative state topology, conducts targeted bidirectional comparisons of command parameters with real-time equipment status and distribution network operating conditions, and then precisely adjusts parameters through an iterative correction mechanism based on operating conditions. This ensures that the final generated heterogeneous adaptation execution command not only meets the constraints of safe equipment operation but also accurately matches the real-time control requirements of the distribution network, guaranteeing the safety and effectiveness of collaborative control.

[0072] Preferably, the specific implementation process of step 321 is as follows: The real-time data extraction module loads the distribution network-compensation equipment collaborative status feature topology. First, the updated data identifier of the topology is parsed to determine the latest data's acquisition timestamp, data update range (such as updates to operating parameters of a certain type of compensation equipment, or updates to distribution network voltage data), and operating condition change identifier (such as switching from normal operating condition to heavy load operating condition), generating a data extraction range list. Based on this list, two types of core data are accurately extracted: one type is the current operating status parameters of various types of reactive power compensation equipment, including but not limited to current compensation capacity, remaining compensation capacity, contact temperature, real-time response rate, recent action count, and current fault alarm status, synchronously associated with the equipment's unique identifier and acquisition timestamp; the other type is the distribution network's real-time voltage and power data, including real-time voltage values ​​at each node, voltage fluctuation coefficient, total system active power, reactive power deficit, and power factor, synchronously associated with the distribution network operating condition identifier. The two types of extracted data are sorted in ascending order by collection timestamp and organized in a hierarchical structure of "equipment type-data type-timestamp" to construct a time-series real-time verification data stack. Each stack element contains complete equipment-distribution network correlation data under a single timestamp, ensuring the time-series correlation and integrity of the data, and providing an accurate real-time benchmark for subsequent comparison and verification.

[0073] Preferably, in the specific technical implementation of step 322: the drive parameter verification module extracts the real-time operating status parameters of the corresponding device from the real-time verification data stack according to the device's unique identifier, groups them according to the mapping relationship of "device-specific control instruction prototype - corresponding device real-time parameters", and generates real-time verification data clusters (each data cluster corresponds to the instructions and real-time data of a type of device). Based on the technical parameter manuals of various types of reactive power compensation equipment and the safe operation threshold configuration in the topology of the distribution network-compensation equipment collaborative state characteristics, a device safe operation threshold matrix is ​​constructed. The row index of this matrix is ​​the device type, the column index is the type of action parameter to be verified (compensation capacity, action timing interval, action amplitude, response rate), and the matrix element value is the safe operation threshold range of the corresponding parameter of the corresponding device (for example, the safe threshold of the compensation capacity of a certain type of static var generator, i.e., SVG, is a certain range of kVar, and the safe threshold of the action timing interval is a certain range of ms). The action parameters (target compensation capacity value, action trigger time node, and action amplitude parameters) in the prototype of the equipment-specific control command are bidirectionally compared and verified with the corresponding current equipment status parameters in the real-time verification data cluster and the corresponding thresholds in the safe operation threshold matrix. This comparison involves three aspects: first, verifying the compatibility between the action parameters and the current equipment status parameters (e.g., whether the target compensation capacity value exceeds the equipment's remaining compensation capacity); second, verifying whether the action parameters are within the safe operation threshold range; and third, verifying the compatibility between the action parameters and the real-time operating conditions of the distribution network (e.g., whether the action amplitude meets the distribution network voltage stability requirements under heavy load conditions). Each verification result is recorded, and action parameters exceeding the safe operation threshold range or not matching the real-time operating conditions are marked. A parameter verification result table is generated, clearly defining the abnormal parameter items, the current real-time value, the safe threshold range, and the operating condition compatibility requirements.

[0074] Preferably, in one scenario, step 323 is specifically implemented as follows: the driving parameter correction module loads the parameter verification result table, the equipment safe operation threshold matrix, and the real-time operating condition identifier of the distribution network, and constructs an operating condition linkage iterative correction logic. This logic includes a parameter correction rule base and a convergence criterion: the parameter correction rule base designs correction strategies according to the types of abnormal parameters. For cases where the compensation capacity parameter exceeds the threshold, the minimum value between the remaining compensation capacity of the equipment and the reactive power gap of the distribution network is used as the benchmark, and the compensation capacity target value is corrected downward according to the equipment capacity adjustment step size (a certain proportion of the equipment's rated capacity, for example: 5%); for cases where the action timing interval parameter exceeds the threshold, the result is... The current real-time response rate of the equipment (extracted from the real-time verification data cluster) is used to correct the trigger time node according to the logic of "action timing interval ≥ 1 / real-time response rate". For cases where the action amplitude parameter exceeds the threshold, it is dynamically adjusted in conjunction with the distribution network voltage fluctuation coefficient (extracted from the real-time verification data cluster). The larger the voltage fluctuation coefficient, the smaller the action amplitude correction, to avoid aggravating voltage fluctuations. The convergence criterion is that the corrected action parameter simultaneously meets the requirements of "within the safe operation threshold range", "adapted to the current status parameters of the equipment", and "adapted to the real-time operating conditions of the distribution network", and the parameter change amplitude after two consecutive corrections is less than the preset minimum value (e.g., 0.5%). The abnormal parameters in the parameter verification result table are corrected one by one according to the correction rule library. The corrected parameters are then resubmitted into the comparison and verification process in step 322. If abnormalities still exist, the correction-verification process is repeated until all parameters meet the convergence criteria. Integrate all the corrected action parameters, format them according to the control interface protocol specifications of each type of reactive power compensation equipment (consistent with the interface specifications in step 312), generate heterogeneous adaptation execution instructions ("heterogeneous" reflects the differences in interface and operating status of different types of equipment), and output parameter correction logs synchronously to clarify the initial value, correction process, final value and correction basis of each abnormal parameter.

[0075] Optionally, step 33 includes: Step 331: Send the heterogeneous adaptation execution command to different types of reactive power compensation equipment, including parallel capacitor banks, SVG, and energy storage type reactive power compensation devices, through the corresponding communication interface; Step 332: Collect action feedback data streams of various types of reactive power compensation equipment. The feedback data streams include action completion status, actual output compensation capacity, and equipment operating thermal field data. Step 333: Compare the deviation between the action feedback data stream and the expected parameters in the collaborative control strategy sequence. If there is a deviation, fine-tune the subsequent execution instructions based on the deviation value to form a feedback control link, so as to coordinate the action of various types of reactive power compensation equipment and complementarily adjust the performance gain.

[0076] Steps 331-333 construct a closed-loop collaborative control link—"heterogeneous interface adaptation - multi-dimensional feedback acquisition - scenario-based deviation fine-tuning"—adapted to the collaborative management and control scenarios of various types of reactive power compensation equipment in the distribution network. This solves the problems of interface incompatibility, single feedback data, and generalized deviation adjustment leading to poor multi-device collaborative effects and low control accuracy in traditional command issuance. This implementation closely revolves around the characteristic differences of parallel capacitor banks, static var generators (SVG), and energy storage reactive power compensation devices, as well as the dynamic operating conditions of the distribution network. Precise heterogeneous interface adaptation ensures the reliability of command issuance; multi-dimensional feedback data acquisition builds a comprehensive operational status perception benchmark; and scenario-based deviation fine-tuning achieves dynamic optimization of commands. Ultimately, a collaborative linkage mechanism of "decrease-feedback-control" is formed, ensuring consistent action of various types of equipment and achieving complementary performance gains in distribution network voltage stability and power optimization.

[0077] Preferably, the specific implementation process of step 331 is as follows: The driving communication adaptation module parses the unique device identifier in the heterogeneous adaptation execution instruction, matches the corresponding device type (parallel capacitor bank, SVG, energy storage reactive power compensation device), and retrieves a preset multi-type reactive power compensation device communication interface parameter library. This parameter library contains the communication protocol type (such as RS485, Ethernet, CAN bus), communication baud rate (within a certain range, for example: 9600bps-115200bps), data frame format, address encoding, and verification mechanism (parity check, cyclic redundancy check, abbreviated as CRC) corresponding to each type of device, generating a device-interface parameter mapping table. Based on this mapping table, the heterogeneous adaptation execution instruction is encapsulated in a communication format, and a frame header, address code, instruction type code, check code, and frame tail are added according to the communication protocol requirements of the corresponding device to generate a command frame that the device can directly recognize. A command issuance scheduling queue is constructed, and command frames are issued in order according to the action timing requirements in the coordinated control strategy sequence to avoid command issuance conflicts between different devices (e.g., when SVG and parallel capacitor banks need to coordinate their actions, they are issued sequentially at preset time intervals). A issuance timestamp and acknowledgment identifier are added to each command frame. The encapsulated command frames are then issued to each type of reactive power compensation device via the corresponding physical communication interface. A command issuance acknowledgment mechanism is simultaneously initiated, waiting for the device to return a reception acknowledgment signal. If no acknowledgment signal is received within a preset timeout period (e.g., 50ms), the command is reissued, and the number of retries is recorded. If the number of retries exceeds a preset limit (e.g., 3 times), it is marked as an issuance anomaly, and an alarm log is generated. Finally, a command issuance completion list is generated, clearly indicating the issuance status, timestamp, and acknowledgment information of each device's command.

[0078] Preferably, in the specific technical implementation of step 332: the drive feedback acquisition module constructs a feedback data acquisition channel based on the device's unique identifier, and the channel configuration is consistent with the communication interface parameters in step 331 to ensure data transmission compatibility. Different acquisition frequencies and triggering methods are set according to the feedback characteristics of different types of reactive power compensation devices: trigger-based acquisition is used for the action completion status (the device actively reports after completing the action, or the acquisition is triggered after a preset delay after the command is issued, for example, 100ms); periodic acquisition is used for the actual output compensation capacity (the acquisition period is within a certain range, for example, 20ms-50ms); and dual acquisition of the device's operating thermal field data is used (the regular period is within a certain range, for example, 100ms, and immediate acquisition is triggered when the thermal field data exceeds a preset warning threshold). The acquired feedback data stream includes three core dimensions: action completion status (completed / incomplete / abnormal, completion time), actual output compensation capacity (real-time value, capacity fluctuation coefficient), and device operating thermal field data (key component temperature, temperature change rate, thermal field distribution uniformity index), synchronously associated with the device's unique identifier, acquisition timestamp, and the identifier of the corresponding issued command. The collected feedback data stream undergoes preprocessing: first, format validation to remove invalid data with format errors or missing data; second, time sequence alignment to associate feedback data from different devices with the same time dimension based on the collection timestamp; and third, anomaly filtering to remove transient interference data through comparison with neighboring data, generating a structured action feedback data stream, which is organized into feedback data clusters according to the hierarchy of "device type-data dimension-timestamp" to ensure the integrity, accuracy, and time sequence correlation of the data.

[0079] Preferably, in one scenario, step 333 is specifically implemented as follows: The drive deviation control module extracts the expected parameters of each type of reactive power compensation equipment from the coordinated control strategy sequence, including the expected action completion time, the expected compensation capacity output range, and the expected operating thermal field threshold. It associates these parameters with the corresponding data in the action feedback data stream based on the unique identifier of the equipment, constructing an expected-actual parameter comparison matrix. The row index of this matrix represents the equipment type, and the column index represents the comparison parameter type (action completion time, compensation capacity deviation, thermal field temperature deviation). The matrix element values ​​are the deviation amount and percentage between the expected and actual values. A scenario-based deviation threshold matrix is ​​set. This matrix combines the real-time operating condition identifier of the distribution network (normal / heavy load / severe voltage fluctuation) with the equipment type to set differentiated deviation thresholds (for example, the compensation capacity deviation threshold is relaxed to a certain range under heavy load conditions and tightened to a certain range under normal operating conditions). If the deviation exceeds the corresponding threshold, it is determined that the deviation needs to be controlled. Based on the deviation and real-time operating conditions of the distribution network, a fine-tuning strategy library for equipment coordination is constructed: For compensation capacity deviation, if the actual output of the SVG is lower than expected and the parallel capacitor bank still has remaining capacity, the compensation capacity allocation coefficient of the parallel capacitor bank is fine-tuned to compensate for the deviation, while the action amplitude parameters of the SVG are fine-tuned to avoid superimposed fluctuations; for action timing deviation, if the action completion time of a certain device is too long, the action trigger time node of subsequent related devices is fine-tuned to ensure consistency of coordination timing; for thermal field temperature deviation, if the temperature of a certain device exceeds the expected threshold, its action amplitude and action interval are fine-tuned to reduce the operating load, while other devices are scheduled to share its compensation task. The fine-tuning strategy is converted into parameter adjustment amounts for subsequent execution instructions, updating the corresponding parameters in the heterogeneous adaptation execution instructions to form a feedback control link. Steps 331-333 are repeated until the expected-actual parameter deviation is within the scenario-based deviation threshold range, realizing coordinated action and complementary performance gain adjustment of various types of reactive power compensation equipment, synchronously generating control logs, clarifying the deviation type, adjustment amount, adjustment basis, and control effect.

[0080] like Figure 2 The diagram shown is a structural schematic of a multi-type reactive power compensation coordinated control device according to an embodiment of this application, which includes: The data deconstruction module is used to perform cross-dimensional heterogeneous correlation deconstruction of real-time operation data of the distribution network and the status data of various types of reactive power compensation equipment, so as to generate a collaborative status feature topology of the distribution network and compensation equipment. The optimization modeling module is used to construct a reactive power compensation collaborative optimization model under multi-objective coupling constraints based on the topology of the distribution network-compensation equipment collaborative state characteristics. It generates a collaborative control strategy sequence through dynamic balancing configuration of compensation capacity and collaborative matching of response timing. The instruction execution module is used to generate heterogeneous adaptation execution instructions for various types of reactive power compensation devices according to the coordinated control strategy sequence, so as to coordinate the actions of various types of reactive power compensation devices and to perform complementary adjustment of performance gains.

[0081] like Figure 3 The diagram shown is a structural schematic of an electronic device according to an embodiment of this application, which includes a processor and a memory; The memory is used to store computer programs; When the processor executes the program stored in the memory, it implements the functions of each module of the multi-type reactive power compensation collaborative control device as described above, or implements the steps of the multi-type reactive power compensation collaborative control method.

Claims

1. A multi-type reactive power compensation cooperative control method, characterized in that, include: Step 1: Perform cross-dimensional heterogeneous correlation deconstruction on the real-time operation data of the distribution network and the status data of various types of reactive power compensation equipment to generate a collaborative status feature topology of the distribution network and compensation equipment. Step 2: Based on the topology of the coordinated state characteristics of the distribution network and compensation equipment, construct a reactive power compensation coordinated optimization model under multi-objective coupling constraints, and generate a coordinated control strategy sequence through dynamic balancing configuration of compensation capacity and coordinated matching of response timing. Step 3: Based on the coordinated control strategy sequence, generate heterogeneous adaptation execution instructions for various types of reactive power compensation devices, coordinate the actions of each type of reactive power compensation device, and make complementary adjustments to the performance gains.

2. The multi-type reactive power compensation cooperative control method according to claim 1, characterized in that, Step 1 includes: Step 11: Perform timestamp synchronization and dimension normalization on the real-time operation data of the distribution network and the status data of various types of reactive power compensation equipment to generate a time-series unified multi-source dataset. Step 12: Based on the correlation criteria between equipment type attributes and operating conditions, perform feature clustering, divide-and-conquer, and correlation mapping modeling on the time-series collaborative multi-source dataset to generate equipment-operating condition correlation feature tensors; Step 13: Use the constructed feature enhancement fusion model to perform redundant information pruning and key information aggregation and reconstruction on the equipment-operating condition association feature tensor to generate the distribution network-compensation equipment collaborative state feature topology.

3. The multi-type reactive power compensation cooperative control method according to claim 2, characterized in that, Step 11 includes: Step 111: Collect real-time operation data of the power distribution network and status data of various types of reactive power compensation equipment, and perform format compliance verification and missing value reconstruction and repair on various types of data to generate a complete original data stack. Step 112: Using the timestamp of the distribution network data acquisition terminal as the reference source, perform time axis homogeneous alignment processing on the complete original data stack, and integrate it into a dataset with a unified time granularity as a time-series aligned data cluster through interpolation reconstruction algorithm; Step 113: Perform numerical interval standardization mapping on the time-series aligned data clusters to eliminate the dimensional differences between different types of data, so as to generate a time-series unified multi-source dataset.

4. The multi-type reactive power compensation coordinated control method according to claim 2, characterized in that, Step 12 includes: Step 121: Based on the technical attribute parameters of various types of reactive power compensation equipment, divide the equipment status data in the time-series unified multi-source dataset into static attribute parameters and dynamic response characteristic data to form an equipment feature classification system. Step 122: Construct an association mapping rule engine. The association mapping rule engine contains the mapping relationship between different operating conditions and equipment characteristics. Based on the rule engine, pairwise association matching modeling is performed on the distribution network operation data and equipment characteristic data in the time-series unified multi-source dataset to generate an initial association tensor. Step 123: Based on the equipment feature classification spectrum, the initial association tensor is validated and screened to remove tensor elements without actual physical association and retain effective association feature dimensions to generate equipment-operating condition association feature tensor.

5. The multi-type reactive power compensation coordinated control method according to claim 2, characterized in that, Step 13 includes: Step 131: Calculate the information redundancy of each feature column in the device-operating condition correlation feature tensor using the mutual information entropy quantization layer, and mark the feature columns with redundancy higher than the preset threshold to obtain redundant feature columns with removal labels. Step 132: Use a feature dimension pruning and purification layer to prune and remove the marked redundant feature columns, sort the remaining feature columns by importance weight, and extract the core feature dimensions with the highest ranking to generate the core feature tensor. Step 133: Employ an attention mechanism layer to enhance key information in the aggregated core feature tensor, and assign weight coefficients to the operating conditions corresponding to different core features to generate a topology of the distribution network-compensation equipment collaborative state features.

6. The multi-type reactive power compensation coordinated control method according to claim 1, characterized in that, Step 2 includes: Step 21: Extract voltage deviation gradient features, power factor resonance features, and equipment response delay features from the topology of the distribution network-compensation equipment cooperative state features, and construct a multi-objective optimization objective functional and constraint boundary conditions; Step 22: Based on the multi-objective optimization objective functional and constraint boundary conditions, construct a multi-type reactive power compensation collaborative optimization model, wherein the collaborative optimization model embeds a device response characteristic differential weight adaptation mechanism; Step 23: Solve the collaborative optimization model using the constructed multi-objective particle swarm collaborative optimization model, and output the compensation capacity allocation coefficients and action timing topology tables of various types of reactive power compensation equipment to generate a collaborative control strategy sequence.

7. The multi-type reactive power compensation coordinated control method according to claim 6, characterized in that, Step 21 includes: Step 211: Extract the node voltage deviation gradient value, system power factor resonance coefficient, and voltage fluctuation attenuation coefficient from the topology of the distribution network-compensation equipment cooperative state characteristics as optimization target parameters, and construct a multi-objective optimization objective functional with the objectives of minimizing voltage deviation gradient, optimizing power factor resonance, and maximizing voltage fluctuation attenuation. Step 212: Combine the rated capacity threshold, maximum response rate, and action number threshold constraints of various types of reactive power compensation equipment with the capacity carrying capacity constraints of the distribution network lines to construct collaborative optimization constraint boundary conditions; Step 213: Standardize and normalize the multi-objective optimization objective functional and the collaborative optimization constraint boundary conditions to generate the multi-objective optimization objective functional and constraint boundary conditions.

8. The multi-type reactive power compensation coordinated control method according to claim 6, characterized in that, Step 22 includes: Step 221: Based on the multi-objective optimization objective functional and constraint boundary conditions, construct a multi-type reactive power compensation collaborative optimization model; Step 222: Analyze the differences in response characteristics of various types of reactive power compensation equipment, construct a differentiated weight adaptation mechanism, assign dynamic response weight coefficients to fast response equipment, and assign capacity allocation weight coefficients to large-capacity compensation equipment. Step 223: Embed the differentiated weight adaptation mechanism into the objective functional solution process of the collaborative optimization model, so that the model solution process can adapt to the differences in technical characteristics of different types of equipment, thereby completing the construction of a multi-type reactive power compensation collaborative optimization model.

9. The multi-type reactive power compensation coordinated control method according to claim 6, characterized in that, Step 23 includes: Step 231: Use a multi-objective particle swarm cooperative optimization model to solve the multi-type reactive power compensation cooperative optimization model. Initialize the population as a combination of compensation capacity and action timing solution vector, and set the iteration termination criterion. Step 232: Calculate the fitness value of each combination solution vector in the initial population using the fitness evaluation function, and perform selection, crossover, mutation and evolution operations based on the fitness value to generate a new generation population; Step 233: Repeat the iterative evolution process until the termination criterion is met. Select the Pareto optimal solution from the final population, analyze the optimal solution to obtain the compensation capacity allocation coefficient and action timing topology table of each type of reactive power compensation equipment, and generate a coordinated control strategy sequence.

10. The multi-type reactive power compensation coordinated control method according to claim 1, characterized in that, Step 3 includes: Step 31: Analyze the compensation capacity allocation coefficients and action timing topology table in the collaborative control strategy sequence, and integrate them with the control interface protocol specifications of various types of reactive power compensation equipment to generate a prototype of equipment-specific control instructions. Step 32: Based on the real-time update data of the topology of the distribution network-compensation equipment collaborative state characteristics, perform feasibility verification and parameter iterative correction on the prototype of the equipment-specific control command to generate heterogeneous adaptation execution commands; Step 33: Send the heterogeneous adaptation execution command to the corresponding type of reactive power compensation equipment, and simultaneously collect the action feedback data stream of each equipment to form a feedback control link, so as to coordinate the action of each type of reactive power compensation equipment and complementarily adjust the performance gain.

Images