A service area transformer station area flexible interconnection device operation control method

Abstract

The application discloses a service area transformer station area flexible interconnection device operation control method, and particularly relates to the field of power distribution control, and is used for solving the problems of multi-station area load power flow scheduling and dynamic correction of the service area transformer station area flexible interconnection device, and comprises the following steps: collecting the load power of each transformer station area, and constructing a load cluster dynamic migration atlas; planning the elastic power window of the transformer station area according to the atlas; generating a time sequence bias distribution of a direct current bus voltage, and coordinating the power flow direction; synchronously sending a voltage control signal sequence to the flexible interconnection device, and guiding the power flow direction through the direct current bus voltage amplitude bias; collecting voltage bias mutation and transient current dynamic data, establishing a reinforcement learning model, and dynamically adjusting the virtual impedance; performing difference analysis on the actual migration track of the load cluster and the atlas, correcting the load cluster dynamic migration atlas, optimizing the power flow prediction, and improving the power distribution operation stability and engineering implementability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power distribution control, more particularly, to a service area transformer substation flexible interconnection device operation control method. BACKGROUND

[0002] As a comprehensive energy supply place of traffic nodes, the transformer substation flexible interconnection device of the service area on the highway bears the dynamic power supply task of multi-source (such as new energy, energy storage, fast charging pile) and multi-load (such as vehicle flow energy supplement, lighting, catering). However, the service area power load presents obvious "tidal characteristics", such as frequent alternation of peak load scenes such as daytime small car concentrated fast charging, evening heavy truck energy supplement, and large bus group charging during holidays, and the load tide migrates across substation with time and space, forming a short-time multi-load cluster superposition phenomenon. At present, the operation control of the flexible interconnection device generally adopts static topology or local power balance strategy, mainly for passive adjustment after the load has fluctuated, and cannot be adjusted in advance according to the dynamic migration characteristics of the load cluster. When the load cluster migrates rapidly among multiple substation, the power flow distribution between the substation is unbalanced, resulting in problems such as bus voltage return, frequent action of tapping switch, and low-voltage side voltage flicker. The common single-point load prediction or fixed limiting strategy is limited by local information and limited time resolution, and it is difficult to cope with the dynamic nature of the load cluster in multiple substation. SUMMARY

[0003] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a service area transformer substation flexible interconnection device operation control method to solve the problems raised in the background art.

[0004] To achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0005] A service area transformer substation flexible interconnection device operation control method, comprising the following steps:

[0006] S1: Collecting the load power of each transformer substation in the service area, and establishing a load cluster dynamic migration map;

[0007] S2: Based on the load cluster dynamic migration map, planning the elastic power window of the transformer substation according to the space-time power constraint in the map;

[0008] S3: Based on the planning result of the elastic power window, generating a time sequence bias distribution of the DC bus voltage of each transformer substation;

[0009] S4: Synchronously sending the voltage control signal sequence corresponding to the time sequence bias distribution to the flexible interconnection device, and guiding the power flow direction in the flexible interconnection device through the DC bus voltage amplitude bias;

[0010] S5: Collecting voltage bias mutation and transient current dynamic data set, establishing reinforcement learning model for virtual impedance dynamic adjustment;

[0011] S6: Difference analysis is performed on the actual migration trajectory of the load cluster and the dynamic migration atlas of the load cluster, and the dynamic migration atlas of the load cluster is corrected.

[0012] In a preferred embodiment, in S1, the load power of each transformer area in the service area is collected, and the dynamic migration atlas of the load cluster is established, which specifically includes:

[0013] The historical load power data of each transformer area in the service area is collected, and the load power time series data corresponding to the area number and time stamp is constructed;

[0014] The load power time series data is subjected to periodic analysis of a predetermined scale, the load power main period and fluctuation amplitude interval of each transformer area are identified, and a periodic load power data structure is generated;

[0015] In the periodic load power data structure, the periodic load power amplitude and its fluctuation characteristics are clustered and analyzed according to the area number, forming the initial distribution and periodicity of the multi-area load cluster;

[0016] Based on the initial distribution and periodicity of the multi-area load cluster, combined with the dynamic prediction algorithm of the machine learning model, the dynamic change trend of the load cluster is predicted and analyzed, and a dynamic migration data set of the load cluster is generated;

[0017] According to the time stamp sequence of the dynamic migration data set of the load cluster, a dynamic migration atlas of the load cluster is constructed.

[0018] In a preferred embodiment, the dynamic change trend of the load cluster is predicted and analyzed based on the initial distribution and periodicity of the multi-area load cluster, combined with the dynamic prediction algorithm of the machine learning model, which specifically includes:

[0019] The area number, periodic amplitude and covariance in the initial distribution and periodicity structure of the load cluster of a plurality of predetermined scale periods are extracted, and a training data set of the load cluster dynamic change prediction model is established;

[0020] The training data set is input into the structured training process of the machine learning model, and a trained load cluster dynamic change prediction model is output;

[0021] The initial distribution data of the load cluster carrying the area number are input into the load cluster dynamic change prediction model according to the time stamp sequence, and the prediction result of the dynamic change trend of the load cluster is output.

[0022] In a preferred embodiment, in S2, based on the dynamic migration atlas of the load cluster, the elastic power window of the transformer area is planned according to the space-time power constraint in the atlas.

[0023] Analyzing the dynamic migration map of the load cluster, reading the transformer area number, load power expected amplitude and timestamp in the map;

[0024] Segmenting the read area load power distribution data structure according to the timestamp, recording the segmented power expected amplitude change according to the area number;

[0025] The preset power tolerance range, combined with the segmented power expected amplitude change, calculates the dynamic power capacity upper and lower limits of each area and its time sequence span, and divides the time and space boundary according to the area number, as the flexible power window planning result.

[0026] In a preferred embodiment, in S3, based on the flexible power window planning result, the time sequence bias allocation of the DC bus voltage of each transformer area includes:

[0027] Extracting the power planning result of each transformer area in the flexible power window, generating the time sequence demand change sequence of the area power;

[0028] According to the amplitude change of the power time sequence demand change sequence, the preset amplitude grading rule is used to map the power demand amplitude to the DC bus voltage amplitude bias initial value, and the time sequence bias allocation of the DC bus voltage of each area is generated.

[0029] In a preferred embodiment, in S4, the time sequence bias allocation corresponding voltage control signal sequence is sent to the flexible interconnection device, and the power flow is guided by the DC bus voltage amplitude bias in the flexible interconnection device, which includes:

[0030] Convert the time sequence bias allocation of the DC bus voltage of each transformer area into a voltage amplitude bias allocation control signal sequence;

[0031] Synchronize the control signal sequence to the DC side control unit of the flexible interconnection device;

[0032] After receiving the voltage amplitude bias allocation control signal sequence in the DC side control unit, generate the DC bus voltage amplitude bias execution data to adjust the voltage bias of each transformer area.

[0033] In a preferred embodiment, in S5, the voltage bias mutation and transient current dynamic data set is collected, and a reinforcement learning model is established to dynamically adjust the virtual impedance, which includes:

[0034] Extract the control signal timestamp of the historical voltage amplitude bias allocation control signal sequence, collect the historical voltage and current measurement data of each transformer area according to the corresponding timestamp, and construct the voltage bias mutation and transient current dynamic data set;

[0035] In the voltage amplitude bias and current dynamic data set, the actual voltage change amplitude, the instantaneous current change amplitude and the phase characteristics are extracted to form a dynamic response characteristic data set;

[0036] The dynamic response characteristic data set is trained by a reinforcement learning method to establish a virtual impedance adjustment model, and the virtual impedance value is adjusted according to the model output.

[0037] In a preferred embodiment, the reinforcement learning method for training the dynamic response characteristic data set is specifically:

[0038] Based on the historical voltage bias mutation data and the corresponding instantaneous current dynamic data in the dynamic response characteristic data, a training data set is constructed, and the minimum instantaneous current amplitude peak is taken as the reward function evaluation index.

[0039] In a preferred embodiment, in S6, the actual migration trajectory of the load cluster and the load cluster dynamic migration atlas are analyzed, and the load cluster dynamic migration atlas is corrected, specifically including:

[0040] During the operation of the flexible interconnection device of all transformers in the service area, the real-time power execution data is extracted to construct the actual migration trajectory of the load cluster;

[0041] The actual migration trajectory is paired with the load cluster dynamic migration atlas according to the time stamp to generate a load cluster power deviation data set;

[0042] Based on the load cluster power deviation data set, the load cluster dynamic migration atlas is reconstructed.

[0043] The technical effects and advantages of the service area transformer station flexible interconnection device operation control method of the application are:

[0044] By collecting the load power of each area, a load cluster dynamic migration atlas is established, the dynamic characteristics of the multi-area load are comprehensively captured and the trend is deduced, the optimal scheduling of the dynamic distribution of the transformer station load is realized through the elastic power window planning based on the space-time power constraint, the unified coordination of the power flow of the multi-area load is realized through the time sequence bias distribution of the DC bus voltage of each area, the power impact caused by the sudden change of the power flow is avoided through the dynamic guidance of the DC bus voltage amplitude bias, the reinforcement learning model is introduced to adjust the virtual impedance, the transient current impact is effectively suppressed and the dynamic stability is improved, the adaptive correction of the load cluster dynamic migration atlas is realized through the difference analysis of the actual migration trajectory and the dynamic migration atlas of the load cluster, the prediction accuracy and control ability of the power flow distribution are enhanced, and the operation intelligent level and the power flow scheduling ability of the flexible interconnection device of the service area are improved. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 A service area transformer station area flexible interconnection device operation control method schematic diagram of the present application. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0047] Embodiment 1, Figure 1 The present application provides a service area transformer station area flexible interconnection device operation control method, which comprises the following steps:

[0048] S1: collecting the load power of each transformer station area in the service area, and establishing a load cluster dynamic migration atlas;

[0049] S2: based on the load cluster dynamic migration atlas, planning the flexible power window of the transformer station area according to the time-space power constraint in the atlas;

[0050] S3: based on the planning result of the flexible power window, generating the time sequence bias distribution of the DC bus voltage of each transformer station area;

[0051] S4: synchronously sending the voltage control signal sequence corresponding to the time sequence bias distribution to the flexible interconnection device, and guiding the power flow direction in the flexible interconnection device through the DC bus voltage amplitude bias;

[0052] S5: collecting the voltage bias mutation and instantaneous current dynamic data set, and establishing a reinforcement learning model for virtual impedance dynamic adjustment;

[0053] S6: performing difference analysis on the actual migration trajectory of the load cluster and the load cluster dynamic migration atlas, and correcting the load cluster dynamic migration atlas.

[0054] In S1, the load power of each transformer station area in the service area is collected, and a load cluster dynamic migration atlas is established.

[0055] The historical load power data of each transformer area in the service area is collected. First, the numbers and geographical positions of each area are determined, the load measurement interface of each area is clarified, the standardized collection equipment or data interface is adopted, and the load power data of each area in the selected period (such as 1 hour, 24 hours, 7 days or 30 days) is obtained. The collected data format is fixed as a three-tuple of area number, time stamp and corresponding load power value, ensuring the structure and retrievability of the data. During the collection of time series data, fixed time intervals are recorded to ensure that data points are evenly distributed on the time axis, avoid data omission or time misplacement, and ensure data integrity and continuity. The collected raw data is subjected to missing value filling and outlier rejection operation. The missing data is repaired by interpolation method, and the outliers are identified and excluded by statistical method. After data cleaning, the load power time series data set with area number as index and time stamp as vertical axis is generated, and the data format is unified as CSV or database table for structured storage, which is convenient for subsequent processing and calling.

[0056] The generated load power time series data set is read and analyzed with a preset scale period. The setting of the period scale directly affects the granularity of the analysis result. Here, the period scale is set to 24 hours to ensure coverage of daily load variation rules. The data under each area number is segmented by 24 hours as a period, the maximum value, minimum value and mean value of the load power in each period are calculated, and the main period and fluctuation amplitude interval in the period are further extracted. The main period is extracted by the peak period extraction algorithm, and the main period position is marked according to the peak value frequency of the load power; the fluctuation amplitude interval is quantified by the difference between the maximum value and the minimum value in the period. The main period value and fluctuation amplitude interval (combined with the standard deviation of the fluctuation amplitude to integrate into the fluctuation characteristics) of each area number are calculated and output, and the period load power data structure is formed by summarizing, which includes area number, period number, main period value and fluctuation amplitude interval. The data structure is stored by area number index to provide direct input for subsequent load clustering and dynamic prediction.

[0057] The period amplitude and fluctuation amplitude interval of each area are extracted to construct a feature vector for clustering analysis. The K-means clustering algorithm is selected, the number of clustering categories is preset to 3, and the period data of each area is taken as a sample containing the period amplitude and fluctuation amplitude interval characteristics, which is input into the K-means clustering algorithm for processing, and the clustering label of each area in different periods is output. The clustering label is directly mapped to the initial distribution of multi-area load cluster to form a load cluster category index table. Combined with the period number, a periodicity index is formed to describe the dynamic change of the load cluster in different periods. The initial distribution of the load cluster and the periodicity are stored by area number, and the data structure contains area number, clustering category, period number and periodicity characteristics, which are used as input data for the next step of dynamic prediction.

[0058] The initial distribution of the extracted output load cluster and the periodicity law of the substation number, the periodic amplitude, and the periodic fluctuation covariance characteristics are extracted to establish a training data set of the load cluster dynamic change prediction model. Each sample data includes a substation number, a periodic number, a periodic amplitude, a fluctuation amplitude interval, and a periodic fluctuation covariance, ensuring the completeness of the multi-dimensional feature information input to the model. The training data set is preprocessed using a standardization method, and the periodic amplitude and the fluctuation amplitude interval are normalized to ensure uniform weight distribution of different feature dimensions. The structured training process of the machine learning model is input into the training data set, the long short-term memory network model architecture is selected to capture the time sequence dependency, the number of model layers is set to two, the number of nodes in each layer is set to 64, the number of training rounds is set to 100, the learning rate is set to 0.001, and the loss function is mean square error. After the model training is completed, the trained model is used as the load cluster dynamic change prediction model. The load cluster initial distribution data carrying the substation number is input into the prediction model in chronological order, and the load cluster dynamic change trend prediction results at each timestamp are output in turn, and are integrated in chronological order to form a load cluster dynamic migration data set, which includes timestamp, substation number, load cluster category, and predicted load amplitude.

[0059] From the generated load cluster dynamic migration data set, the substation number, the load cluster category, and the predicted load amplitude at each timestamp are extracted, arranged in chronological order, and formed into a multi-substation load cluster migration sequence data. The multi-substation load cluster migration sequence data is indexed according to the substation number to form a dynamic migration trajectory data set for each substation. Each dynamic migration trajectory data set is concatenated in chronological order to form a load cluster dynamic migration trajectory. The trajectory node data includes a substation number, a timestamp, a load cluster category, and a predicted load amplitude. Through the connection relationship between nodes, a directed graph structure is formed, and the weight of each edge in the graph represents the dynamic migration amplitude of the load cluster, and the edge direction reflects the migration trend direction of the load cluster. The dynamic migration trajectory data of all substations and the corresponding directed graph structure are integrated to output a load cluster dynamic migration map, and the data structure includes a timestamp, a substation number, a load cluster category, and a predicted load amplitude.

[0060] In S2, based on the load cluster dynamic migration map, the flexible power window of the transformer substation is planned according to the space-time power constraint in the map.

[0061] The load cluster dynamic migration atlas data structure is obtained, and the data structure includes the unique number of each substation, the expected amplitude of load power at each time, and timestamp information. The substation number is identified by engineering coding rules, for example, using a four-digit number: the first two digits represent the substation or regional code, and the last two digits represent the specific substation number, ensuring global uniqueness. The expected amplitude of load power is in kilowatts (kW), recording the target load power distribution of each substation in the future period by the dispatching layer. The expected amplitude of load power of each substation is sorted in timestamp order to construct a three-tuple data structure of substation number, timestamp, and power expected amplitude, facilitating subsequent time series segmentation analysis. When segmenting the data, set the time segmentation interval, which can be set to 15 minutes, 30 minutes, or 1 hour according to engineering requirements. This embodiment selects 30 minutes (or other time intervals that meet the charging time of the service area) as the segmentation interval for detailed management of daily load dynamics. According to the 30-minute segmentation interval, the time series data under each substation number is divided into several time periods, and each time period contains the expected amplitude of load power with the corresponding timestamp. For each segment, record the change in the expected amplitude of power in that time period, including the starting power amplitude, ending power amplitude, and change amplitude. All data are indexed by substation number and stored as structured data of substation number-time period number-power expected amplitude start and end values.

[0062] A power tolerance range is set, which is used to determine the upper and lower boundary conditions when power dynamic distribution is determined. The setting of the power tolerance range is based on engineering requirements combined with the safe operation range of the equipment. The example setting is between ±10% to ±20%, and the specific value is determined by statistical data of load fluctuation history. The default selection is ±15% as the power tolerance range, in order to balance the safety and flexible adjustment ability. The unit of the power tolerance range is consistent with the unit of the power amplitude, which is kilowatt (kW). Combined with the above segmented power expected amplitude change, the time period number, the starting power amplitude and the ending power amplitude of each transformer area are read in turn, the power change rate in the time period is calculated, and the dynamic capacity upper and lower limits are calculated based on the tolerance range. The dynamic capacity upper limit is the power change rate multiplied by (1+power tolerance), and the dynamic capacity lower limit is the power change rate multiplied by (1-power tolerance). For example, when the power change rate of transformer area A in a certain time period is 100 kW, and the tolerance is set to ±15%, the dynamic capacity upper limit is 115 kW, and the lower limit is 85 kW. The dynamic capacity upper and lower limit data structure is indexed by transformer area number and time period number, and records the upper and lower limit values and their corresponding time span (start time stamp and end time stamp of the time period), forming a dynamic power capacity data structure. The dynamic power capacity data structure is aggregated according to the transformer area number, and the time-space power capacity distribution boundary is output as the flexible power window planning result. The planning result of the flexible power window refers to the target value or target range of the load power of each transformer area in a given time period, which is used to guide the power flow dispatching and load distribution. In engineering, the power planning result of the flexible power window is stored in the form of a structured data table of transformer area number, time stamp and target power value. Each transformer area number corresponds to a group of power target values under the time stamp.

[0063] In S3, based on the planning result of the flexible power window, the time sequence bias distribution of the DC bus voltage of each transformer area is generated.

[0064] First, the power planning data of each transformer area is extracted from the planning result of the flexible power window. After extraction, the data of each transformer area is arranged in time stamp order to form a time sequence demand change sequence of the transformer area power. The data structure includes transformer area number, time stamp and corresponding power target value. The time sequence demand change sequence of each transformer area describes the power demand change trend of the transformer area in the entire dispatching period.

[0065] On the basis of the step output of the substation power time sequence demand change sequence, the power demand data is classified and mapped according to the amplitude change. For the power time sequence demand change sequence of each substation, the target power value at each timestamp is read in turn, and the amplitude difference value between adjacent timestamps is calculated, which is used to quantify the dynamic characteristics of power change. For the dynamic classification of amplitude change, a preset amplitude classification rule is used for processing. The setting method of the amplitude classification rule is: according to engineering experience and equipment carrying capacity, the classification threshold of amplitude difference value is divided, for example, the interval of 0-10kW is set as the first level amplitude, the interval of 10kW-20kW is set as the second level amplitude, the interval of 20kW-30kW is set as the third level amplitude, and the interval of 30kW and above is set as the fourth level amplitude. Through the amplitude classification rule, each power demand amplitude difference value is mapped to the corresponding amplitude level according to the classification threshold, forming a power demand classification index. The power demand classification index and the direct current bus voltage amplitude offset initial value establish a corresponding mapping relationship. The setting method of the default mapping relationship is: the first level amplitude corresponds to the direct current bus voltage offset initial value of 0.5V, the second level amplitude corresponds to 1.0V, the third level amplitude corresponds to 1.5V, and the fourth level amplitude corresponds to 2.0V, which ensures that power demands of different amplitude levels can realize power flow guidance and power distribution through voltage offset (the specific setting of the initial offset value is carried out after power distribution test of equipment of different specifications). Apply this mapping relationship to the substation power time sequence demand change sequence, and classify and calculate each power demand amplitude difference value to output the corresponding direct current bus voltage amplitude offset initial value. Finally, the voltage amplitude offset initial value of each substation is output in timestamp order to form the time sequence offset distribution result of the direct current bus voltage of each substation.

[0066] In S4, the time sequence offset distribution corresponding voltage control signal sequence is sent to the flexible interconnection device, and the power flow is guided by the direct current bus voltage amplitude offset in the flexible interconnection device.

[0067] The time sequence bias allocation results of the DC bus voltage of each transformer area are obtained, and the time sequence bias allocation result data format of each area is unified. For each area number, the voltage amplitude bias initial value is extracted in time stamp order, the voltage amplitude bias allocation control signal sequence is constructed, the signal sequence adopts the form of digital control instruction, each instruction includes three tuples of area number, time stamp and voltage bias initial value, the data precision is set to three decimal places according to the system requirement, and the unit is volt. For the voltage bias allocation control signal sequence of multiple areas, the signal sequences of each area are arranged synchronously according to the time stamp, so that the control signals of different areas can be issued at the same time step, and the unbalanced power flow or system oscillation caused by scheduling delay is avoided. The signal issuing strategy adopts the unified scheduling mode of the main scheduling center, the control signal sequences of all areas are packaged according to the unified time stamp to form a multi-area voltage amplitude bias allocation control signal package. The signal package structure adopts binary data stream or standard communication protocol frame for synchronous broadcast issuing, so as to determine the execution of data in the flexible interconnection device on time.

[0068] For the control signal sequence of each area number, the DC side control unit parses the data according to the time stamp order, and generates voltage amplitude bias execution data in turn. The voltage amplitude bias execution data is generated by the controller according to the preset voltage control step value, and the setting of the step value is determined according to the equipment response ability and system dynamic requirement. The default step value is set to 0.1V / step, which ensures that the equipment can gradually adjust the voltage amplitude within the speed range allowed by the project, and tries to avoid the frequent occurrence of sudden large voltage changes to cause current impact or equipment protection action. The DC side control unit adjusts the voltage bias of the target area according to the voltage amplitude bias execution data, and the adjustment mode is to directly modify the bus voltage reference value of the target area in the controller, and to execute the output in real time through the internal converter control logic of the equipment.

[0069] In S5, the voltage bias mutation and transient current dynamic data set are collected, and the reinforcement learning model is established to perform virtual impedance dynamic adjustment.

[0070] The timestamp of each historical control signal is extracted, and according to the correspondence between the substation number and the timestamp, the historical voltage measurement database and the historical current measurement database are called to obtain the voltage value and the current value of each substation at the timestamp. The voltage measurement data is in volts, accurate to three decimal places, and the current measurement data is in amperes, also accurate to three decimal places. After the collection is completed, the voltage offset data in the voltage offset bias control signal sequence is integrated with the voltage measurement data and the instantaneous current measurement data corresponding to the timestamp, and the voltage offset mutation and the instantaneous current dynamic data set are formed according to the substation number and the timestamp index. The data structure of the data set contains the substation number, the timestamp, the voltage offset data, the measured voltage data and the instantaneous current data. In the voltage offset mutation and the instantaneous current dynamic data set, the actual voltage change amplitude is extracted according to the substation number and the timestamp order. The voltage change amplitude is calculated by comparing the measured voltage values of adjacent timestamps. The instantaneous current change amplitude is also calculated by comparing the measured current values of adjacent timestamps. The phase feature is extracted by comparing the time sequence of the voltage and current measurement data, and the phase difference is calculated by using a digital signal processing method (such as the time sequence intersection point method), with the unit being degrees. The extracted feature data is indexed according to the substation number and the timestamp to form the dynamic response feature data set.

[0071] For the dynamic response feature data set, a reinforcement learning method is used for policy iteration training, and a deep reinforcement learning framework (such as DQN or DDPG) is selected for the algorithm architecture, with historical data as the training input. The training data set is constructed in the following way: the historical voltage offset mutation data and the corresponding instantaneous current dynamic data are aligned according to the timestamp order, grouped according to the substation number, and formed into a sample set. The reward function is set in the following way: the minimum instantaneous current amplitude peak value is taken as the evaluation index, and when the model output strategy reduces the instantaneous current amplitude peak value, the reward value increases; otherwise, the reward value decreases, and the reward value interval is set to 0 to 1. During the model training process, the stochastic gradient descent method is used to optimize the policy network, the learning rate is set to 0.001, and the training number of rounds is set to 100 rounds to ensure the stability of the model convergence. After the training is completed, the virtual impedance adjustment model is output, which can calculate the target virtual impedance value in real time according to the input dynamic response feature data. The model output is directly called by the flexible interconnection device controller to perform virtual impedance adjustment on each substation in the project, realize instantaneous current dynamic control, and avoid the risk of power flow impact and equipment overload.

[0072] In S6, the actual migration trajectory of the load cluster and the load cluster dynamic migration map are analyzed, and the load cluster dynamic migration map is corrected.

[0073] The real-time power execution data in the running process of the flexible interconnection device is obtained for all transformer stations in the service area. The real-time power execution data is collected through the digital controller interface of the equipment, and the data structure includes the station number, timestamp, and real-time power value. The collected real-time power execution data is arranged in order of station number and timestamp to form a station power execution time series data set. The timestamp and predicted power data under the same station number are read from the load cluster dynamic migration map, and the real-time power execution data and the predicted power data are correspondingly paired using the timestamp as an index to form a paired data structure. If there is a slight deviation between the real-time power execution data and the predicted data in the timestamp, the paired data structure includes the station number, timestamp, real-time power execution value, and predicted power value, and a load cluster power deviation data set is constructed.

[0074] On the basis of the formed load cluster power deviation data set, the deviation data is summarized and analyzed according to the station number, and the power deviation value under each timestamp is extracted. The deviation data and the training data of the original machine learning model are combined to form a new model incremental learning data set. The training data set of the original machine learning model includes station number, cycle amplitude, fluctuation characteristics, and periodicity rules and other feature data, while the data in the load cluster power deviation data set includes station number, timestamp, real-time power value, predicted power value, and power deviation value. When the data is combined, the original training data and the power deviation data are structured and spliced according to the station number and the timestamp order, and each sample data includes the station number, the timestamp, the cycle amplitude, the fluctuation characteristics, the power deviation value, and the corresponding predicted label. In the model incremental training process, the original weight of the model is used as the initial value, and the new data is combined for dynamic weight update, and the updated load cluster dynamic migration prediction model is output. Finally, the station number, timestamp, and periodicity rule characteristics are input into the corrected model in timestamp order, and the corrected load cluster dynamic migration prediction result is output. The prediction result data structure includes the station number, the timestamp, and the corrected predicted power value. All timestamps of the prediction result are spliced according to the station number and the timestamp order to reconstruct the load cluster dynamic migration map.

[0075] The above formulas are dimensionless values calculated, and the formulas are obtained by software simulation of a large amount of data to obtain a formula of the latest real situation. The preset parameters and threshold values in the formula are set by a person skilled in the art according to the actual situation.

[0076] The above-described embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented by software, the above-described embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center through a wired (for example, infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state disk.

[0077] Those of ordinary skill in the art can realize that the modules and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0078] Those of ordinary skill in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, device, and module can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0079] In several embodiments provided in the present application, it should be understood that the disclosed system, device, and method can be implemented in other ways. For example, the above-described device embodiments are only schematic, for example, the division of the modules is only a logical function division, and actual implementation can have another division manner, for example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the shown or discussed ones can be indirect coupling or communication connection through some interfaces, devices, or modules, which can be electrical, mechanical, or other forms.

[0080] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, and may be located in one place or distributed on multiple network modules. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment.

[0081] In addition, the functional modules in each embodiment of the present application can be integrated into one processing module, or each module can exist physically alone, or two or more modules can be integrated into one module.

[0082] If the functions are realized in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application can be embodied in the form of software products, and the computer software products are stored in a storage medium, including a plurality of instructions for making a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, and various program code storage media.

[0083] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0084] Finally: the above is only the preferred embodiment of the present application, and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for operating and controlling a flexible interconnection device for transformer substations in a service area, characterized in that, Includes the following steps: S1: Collect the load power of each transformer substation in the service area and establish a dynamic migration map of load clusters; S2: Based on the dynamic migration map of load clusters, plan the elastic power window of the transformer substation according to the spatiotemporal power constraints in the map; S3: Based on the results of the flexible power window planning, generate the timing bias allocation of the DC bus voltage for each transformer substation. S4: Synchronously send the voltage control signal sequence corresponding to the timing bias allocation to the flexible interconnect device, and guide the power flow in the flexible interconnect device through the DC bus voltage amplitude bias; S5: Collect dynamic datasets of voltage bias changes and instantaneous current, and establish a reinforcement learning model for dynamic adjustment of virtual impedance; S6: Perform a difference analysis between the actual migration trajectory of the load cluster and the dynamic migration map of the load cluster, and correct the dynamic migration map of the load cluster. In S1, the load power of each transformer substation in the service area is collected, and a dynamic migration map of load clusters is established, specifically including: Collect historical load power data for each transformer substation in the service area, and construct load power time-series data corresponding to substation number and timestamp; Perform a periodic analysis on the load power time series data with a preset scale to identify the main period and fluctuation amplitude range of the load power in each transformer area and generate a periodic load power data structure; In the periodic load power data structure, cluster analysis is performed on the periodic load power amplitude and its fluctuation characteristics according to the transformer area number to form the initial distribution and periodic pattern of multi-transformer area load clusters; Based on the initial distribution and periodicity of load clusters in multiple areas, and combined with the dynamic prediction algorithm of the machine learning model, the dynamic change trend of load clusters is predicted and analyzed, and a dynamic migration data set of load clusters is generated. Construct a dynamic migration map of load clusters according to the timestamp order of the dynamic migration data set of load clusters; In S2, based on the dynamic migration map of load clusters, the flexible power window of the transformer substation is planned according to the spatiotemporal power constraints in the map, specifically including: Analyze the dynamic migration map of load clusters and read the transformer substation number, expected load power amplitude and timestamp of each transformer substation in the map; The read load power distribution data of the transformer area is segmented according to the timestamp, and the expected amplitude change of the segmented power is recorded according to the transformer area number. The preset power tolerance range, combined with the expected amplitude change of segmented power, calculates the upper and lower limits of dynamic power capacity and its time span for each transformer area, and divides the time and space boundaries according to the transformer area number, which serves as the result of flexible power window planning. In S3, based on the results of the flexible power window planning, the timing bias allocation of the DC bus voltage for each transformer substation is generated, specifically including: Extract the power planning results of each transformer substation in the elastic power window and generate a time-series power demand change sequence for the substation. Based on the amplitude changes of the power demand time-series change sequence, a preset amplitude grading rule is used to map the power demand amplitude grading to the initial value of the DC bus voltage amplitude offset, thereby generating the time-series offset allocation of the DC bus voltage for each transformer area.

2. The operation control method for a flexible interconnection device in a service area transformer substation according to claim 1, characterized in that, The method of predicting and analyzing the dynamic change trend of load clusters based on the initial distribution and periodicity of load clusters in multiple areas, combined with a machine learning model and dynamic prediction algorithm, specifically includes: Extract the transformer area number, periodic amplitude, and covariance from the initial distribution and periodic structure of the load clusters in the first few preset scale cycles to establish a training dataset for the load cluster dynamic change prediction model. A structured training process that inputs the training dataset into the machine learning model outputs a trained load cluster dynamic change prediction model. The initial distribution data of load clusters carrying the substation number are input into the load cluster dynamic change prediction model in the order of timestamps, and the prediction results of the dynamic change trend of load clusters are output.

3. The operation control method for a flexible interconnection device in a service area transformer substation according to claim 1, characterized in that, In S4, the voltage control signal sequence corresponding to the timing bias allocation is synchronously sent to the flexible interconnect device. The power flow is guided in the flexible interconnect device through DC bus voltage amplitude bias, specifically including: The timing bias distribution of DC bus voltage in each transformer substation is converted into a voltage amplitude bias distribution control signal sequence. The control signal sequence is synchronously sent to the DC-side control unit of the flexible interconnect device; After receiving the voltage amplitude bias allocation control signal sequence, the DC side control unit generates DC bus voltage amplitude bias execution data to adjust the voltage bias of each transformer substation.

4. The operation control method for a flexible interconnection device in a service area transformer substation according to claim 1, characterized in that, In S5, collecting dynamic datasets of voltage bias changes and instantaneous current, and establishing a reinforcement learning model for dynamic virtual impedance adjustment specifically includes: Extract the control signal timestamps of the historical voltage amplitude bias allocation control signal sequence, collect historical voltage and current measurement data of each transformer area according to the corresponding timestamps, and construct a dynamic data set of voltage bias change and instantaneous current. In the voltage amplitude bias and current dynamic data set, the actual voltage change amplitude, instantaneous current change amplitude and phase characteristics are extracted to form a dynamic response characteristic data set; A virtual impedance adjustment model is established by iteratively training a set of dynamic response feature data using reinforcement learning methods, and the virtual impedance value is adjusted according to the model output.

5. The operation control method for a flexible interconnection device in a service area transformer substation according to claim 4, characterized in that, The specific steps of iteratively training the policy on the dynamic response feature data set based on reinforcement learning are as follows: Based on historical voltage bias abrupt change data and corresponding instantaneous current dynamic data in the dynamic response feature data, a training data set is constructed, and minimizing the peak value of the instantaneous current amplitude is used as the evaluation index of the reward function.

6. The operation control method for a flexible interconnection device in a service area transformer substation according to claim 1, characterized in that, In S6, a difference analysis is performed between the actual migration trajectory of the load clusters and the dynamic migration map of the load clusters. The correction of the dynamic migration map of the load clusters specifically includes: During the operation of the flexible interconnection device in all transformer substations in the service area, real-time power execution data is extracted to construct the actual migration trajectory of the load cluster; The actual migration trajectories are paired with the load cluster dynamic migration map according to the timestamp to generate a load cluster power deviation dataset; Based on the load cluster power deviation dataset, the dynamic migration map of the load cluster is reconstructed.

Images