Flexible load controllable characteristic modeling method and system

Abstract

The invention discloses a flexible load controllable characteristic modeling method and system, and the method comprises the steps: collecting industrial and resident flexible load operation data, building a database, and extracting multi-dimensional controllable characteristic parameters; a load classification module is constructed to realize accurate load classification, a regulation and control strategy model is constructed to determine a load regulation and control priority, and a differentiation strategy is designed; and deploying a communication interaction system to realize real-time data transmission between the virtual power plant and the load terminal, and dynamically correcting regulation and control strategy parameters based on real-time data to complete modeling. The system comprises a load information acquisition and storage unit, a classification processing unit, a regulation and control strategy generation unit, a communication interaction control unit, a strategy dynamic correction unit and a power grid state monitoring unit which work cooperatively. According to the method and the system, the problems of load classification framework deficiency and incomplete communication interaction in the prior art are solved, the regulation and control precision and efficiency of the virtual power plant on the flexible load are improved, the power grid absorption capability is enhanced, and stable operation of the power grid is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of flexible load control technology in virtual power plants, and in particular to a method and system for modeling the controllable characteristics of flexible loads. Background Technology

[0002] As virtual power plants play an increasingly prominent role in power grid dispatch, the demand for regulating flexible loads, such as production equipment loads in the industrial sector with continuous operation requirements and electrical equipment loads in the residential sector with time-shifting capabilities, is constantly increasing. Currently, flexible loads in power grid operation are diverse in type and exhibit significant differences in operating characteristics. Therefore, it is necessary to achieve efficient collaboration between virtual power plants and flexible loads through precise classification, scientific formulation of control strategies, and real-time data interaction. This is crucial to adapt to changes in the real-time operating status of the power grid, improve the accuracy and effectiveness of load regulation, and ensure stable power grid operation. Against this backdrop, the development of modeling methods and systems for the controllable characteristics of flexible loads has become critical.

[0003] Existing technologies have significant shortcomings in flexible load regulation. On the one hand, there is a lack of a unified load classification architecture. The classification criteria for continuous industrial production loads, intermittent industrial production loads, rigid residential electricity loads, and flexible residential electricity loads are not clear enough. A precise classification method based on multi-dimensional characteristic parameters has not been formed, resulting in insufficient accuracy of load classification results and making it difficult to support the formulation of subsequent differentiated regulation strategies. On the other hand, the communication and interaction mechanism is imperfect. The data transmission protocol between the virtual power plant control center and the flexible load control terminal lacks standardized design, data transmission delay control is poor, and an effective data verification and optimization adjustment mechanism has not been established. This makes it difficult to guarantee the real-time performance and stability of control command issuance and load operation data upload, affecting the timely execution and dynamic correction of regulation strategies, and thus leading to poor overall regulation effect. Summary of the Invention

[0004] In order to overcome the shortcomings and deficiencies of the existing technology, the present invention provides a method and system for modeling the controllable characteristics of flexible loads.

[0005] The technical solution adopted in this invention is:

[0006] A method for modeling the controllable characteristics of flexible loads includes the following steps:

[0007] S1 classifies and identifies the loads of production equipment with continuous operation requirements in the industrial sector and the loads of electrical equipment with time-shifting capabilities in the residential sector. By collecting voltage fluctuation data, current change data and power consumption data of different types of loads under typical operating conditions, a load basic information database is established.

[0008] S2, Based on the load basic information database, controllable characteristic parameters are extracted using a multi-dimensional feature extraction method. The controllable characteristic parameters are input into a preset classification algorithm to calculate a similarity threshold and perform accurate load classification. The controllable characteristic parameters include voltage fluctuation frequency, current change amplitude, and peak power consumption duration.

[0009] S3, construct a load classification module, input the controllable characteristic feature parameters into a preset classification algorithm, and accurately classify the controllable characteristics of industrial continuous production load, industrial intermittent production load, residential rigid electricity load, and residential flexible electricity load by calculating the similarity threshold between different feature parameters;

[0010] S4. Build a control strategy model, determine the control priority of each type of load based on the controllability characteristics of different types of loads and the real-time operating status parameters of the power grid, formulate a power rapid adjustment strategy for high-priority loads, and formulate a time-based reasonable transfer strategy for low-priority loads.

[0011] S5 deploys a communication and interaction system, adopts a two-way data transmission protocol, establishes a communication connection between the virtual power plant control center and various flexible load control terminals, and performs real-time issuance of control commands and real-time uploading of load operation data.

[0012] S6. Based on the real-time load operation data uploaded by the communication interaction system, calculate the deviation between the actual control effect and the preset control target, adjust the parameter weights in the control strategy model, and complete the modeling of the controllable characteristics of flexible load.

[0013] Furthermore, in S3, the following formula is used to calculate the similarity threshold between different feature parameters:

[0014] ,

[0015] in, This represents the similarity value between the i-th feature parameter and the j-th feature parameter. This represents the value of the i-th feature parameter in the k-th sample. This represents the value of the j-th feature parameter in the k-th sample, where n represents the number of samples. This represents the cosine similarity weight coefficient. This represents the index similarity weighting coefficient. This represents the exponential decay coefficient.

[0016] Furthermore, in step S4, when determining the control priority of various load types, the control priority value of the m-th load type is adopted. This represents the maximum allowable power value for the m-th type of load. This represents the current operating power value of the m-th type of load. This indicates the adjustable time length of the m-th type of load. This represents the total running time of the m-th type of load. This represents the power supply reliability requirement factor for the m-th type of load. These represent the weighting parameters for the power difference percentage, the adjustable time percentage, and the reliability requirement coefficient, respectively.

[0017] Furthermore, in step S6, when adjusting the weights of the control strategy model parameters, the following formula is used:

[0018] ,

[0019] in, Indicates the first The weight value of the k-th parameter in the next iteration. This represents the weight value of the k-th parameter in the t-th iteration. Indicates the first The weight value of the k-th parameter in the next iteration. This represents the learning rate parameter. This represents the deviation function between the actual control effect and the preset control target. This represents the partial derivative of the bias function with respect to the weight of the k-th parameter at the t-th iteration. This represents the momentum coefficient.

[0020] Furthermore, in S2, the following formula is used to calculate the duration of the peak power consumption:

[0021] ,

[0022] in, Indicates the duration of peak power consumption. This indicates the moment when the power first reaches its peak. This indicates the moment when power first falls below its peak value. This represents the power consumption value at time t. Indicates peak power consumption. This indicates an indicator function. The function value is 1 when the value inside the parentheses is greater than or equal to 0, and 0 otherwise.

[0023] Furthermore, in step S5, if a bidirectional data transmission delay occurs, the following formula is used:

[0024] ,

[0025] in, Indicates the delay in bidirectional data transmission. Indicates the length of the transmitted data. Indicates the data transmission rate. Indicates signal propagation delay, This indicates a data processing delay.

[0026] Furthermore, step S3 includes the following sub-steps:

[0027] S31, retrieve the controllable characteristic parameters of all load samples from the load basic information database, make a preliminary division according to the two major categories of industrial load and residential load, screen out samples with complete characteristic parameters in each type of load, and remove samples with missing data or abnormal fluctuations exceeding the preset range to ensure the validity of the data used in subsequent classification calculations.

[0028] S32, after the initial division of industrial load samples and residential load samples, the characteristic parameters are standardized. The voltage fluctuation frequency, current change amplitude and power consumption peak duration parameters with different dimensions are converted into standardized parameters with the same numerical range to eliminate the influence of dimension differences on the classification results.

[0029] S33, input the standardized feature parameters into the preset classification algorithm, set the initial similarity threshold, and continuously adjust the threshold size through iterative calculation so that the feature parameter similarity of samples of the same type of load in the classification result is higher than the preset threshold, and the feature parameter similarity of samples of different types of load is lower than the preset threshold, and determine the optimal similarity threshold;

[0030] S34. Based on the optimal similarity threshold, all load samples are finally classified to generate a classification list of industrial continuous production load, industrial intermittent production load, residential rigid electricity load, and residential flexible electricity load. The classification results are stored in the load basic information database to provide a classification basis for the subsequent construction of the control strategy model.

[0031] Furthermore, step S4 includes the following sub-steps:

[0032] S41: Collect real-time operating status parameters of the power grid, including the current active power load, reactive power load, bus voltage level and frequency fluctuation of the power grid. Remove noise interference from the parameters through data preprocessing, extract the calibration index of the power grid operating status, and establish a real-time power grid status assessment dataset.

[0033] S42, Combining the load classification results obtained in step S3, analyze the degree of influence of different types of loads on the power grid operation status, and determine the control sensitivity coefficient of each type of load by calculating the correlation between the power changes of various types of loads and the frequency fluctuations of the power grid.

[0034] S43. Based on the control sensitivity coefficient and the real-time power grid status assessment dataset, formulate a control priority evaluation standard, set loads with high control sensitivity coefficients and significant impact on the stable operation of the power grid as high priority, and set loads with low control sensitivity coefficients and minimal impact on the stable operation of the power grid as low priority.

[0035] S44. Design control strategies for loads with different priorities. High-priority loads adopt a power fast adjustment strategy, which determines the power adjustment step size, frequency and maximum adjustment range. Low-priority loads adopt a time-switching strategy, which plans the switching range, switching duration and switching power after the switching of the load during the running period, forming a complete set of control strategies.

[0036] Furthermore, step S5 includes the following sub-steps:

[0037] S51, determine the hardware architecture of the communication interaction system, select a communication module with high bandwidth and low latency as the communication interface between the virtual power plant control center and the load control terminal, configure a data storage server to store load operation data and control commands during transmission, and deploy a data processing unit to perform real-time parsing of the transmitted data.

[0038] S52 designs a bidirectional data transmission protocol, specifying the data packet format, transmission rate, and verification method for control commands sent from the virtual power plant control center to the load control terminal. It also determines the data packet structure, sampling frequency, and data compression algorithm for load operation data uploaded from the control terminal to the control center to ensure the integrity and accuracy of data transmission.

[0039] S53, build a communication connection test environment to simulate communication scenarios under different network conditions, including network bandwidth fluctuations, signal interference and network outage recovery, test the transmission delay, data packet loss rate and command response speed of the communication interaction system under different scenarios, and record the calibration performance indicators during the test process;

[0040] S54, based on the communication connection test results, optimize and adjust the hardware parameters and transmission protocol of the communication interaction system, replace communication modules whose performance does not meet the requirements, and correct the data verification algorithm in the transmission protocol to ensure that the communication interaction system can still carry out stable data transmission and command interaction in complex network environments.

[0041] A flexible load controllable characteristic modeling system, the system being applied to the aforementioned flexible load controllable characteristic modeling method, comprising:

[0042] The load information acquisition and storage unit is used to classify and identify the loads of production equipment with continuous operation requirements in the industrial field and the loads of electrical equipment with time-shifting capabilities in the residential field. By collecting voltage fluctuation data, current change data and power consumption data of different types of loads under typical operating conditions, a load basic information database is established.

[0043] The load classification processing unit is used to receive load data transmitted by the load information acquisition and storage unit, extract controllable characteristic parameters using a multi-dimensional feature extraction method, input the controllable characteristic parameters into a preset classification algorithm to calculate a similarity threshold, and perform accurate load classification. The controllable characteristic parameters include voltage fluctuation frequency, current change amplitude, and peak power consumption duration. The load classification processing unit is connected to the load information acquisition and storage unit and the control strategy generation unit, respectively, and synchronizes the classification results to the load information acquisition and storage unit and transmits them to the control strategy generation unit.

[0044] The control strategy generation unit is used to build a control strategy model, receive the load classification results transmitted by the load classification processing unit, determine the control priority of various loads in combination with the real-time operating status parameters of the power grid, formulate a power rapid adjustment strategy for high priority loads, formulate a time period reasonable transfer strategy for low priority loads, and convert the generated control strategy into control commands and transmit them to the communication interaction control unit.

[0045] The communication and interaction control unit is used to receive control commands transmitted by the regulation strategy generation unit, establish a communication connection with the flexible load control terminal using a two-way data transmission protocol, and send control commands and upload load operation data. This unit is connected to the regulation strategy generation unit and the strategy dynamic correction unit respectively, and transmits the uploaded load operation data to the strategy dynamic correction unit.

[0046] The strategy dynamic correction unit is used to receive real-time load operation data transmitted by the communication interaction control unit, calculate the deviation between the actual control effect and the preset control target, and adjust the parameter weights in the control strategy model. This unit is connected to the control strategy generation unit and feeds back the corrected parameter weights to the control strategy generation unit.

[0047] The power grid status monitoring unit is used to collect real-time active power load, reactive power load, bus voltage level and frequency fluctuation of the power grid. After preprocessing the data, it is transmitted to the control strategy generation unit to provide power grid status data support for determining control priorities. This unit is connected to the control strategy generation unit through a wireless communication module to ensure real-time transmission of power grid status data.

[0048] Beneficial effects:

[0049] This invention proposes a method and system for modeling the controllable characteristics of flexible loads. Through multi-dimensional feature extraction, it achieves accurate classification of industrial and residential flexible loads. Combined with real-time grid status, it formulates differentiated control strategies. Coupled with a two-way communication interaction system and a dynamic correction mechanism, it significantly improves the accuracy and efficiency of virtual power plant control over flexible loads, enhances the grid's capacity to absorb various loads, and ensures stable grid operation. By constructing a load classification module, using voltage fluctuation frequency, current variation amplitude, and peak power consumption duration as feature parameters, and combining classification algorithms to determine similarity thresholds, it achieves accurate classification of continuous industrial production loads, intermittent industrial production loads, rigid residential electricity loads, and flexible residential electricity loads, providing a reliable basis for formulating differentiated control strategies. To address the issue of imperfect communication interaction mechanisms, it deploys a standardized two-way data transmission protocol communication interaction system, configures a high-bandwidth, low-latency communication module and a data verification mechanism, and ensures the real-time performance and stability of control command issuance and load data upload through testing and optimization. Simultaneously, relying on a dynamic correction mechanism to adjust control parameters ensures timely execution and optimization of control strategies, effectively improving the overall control effect. Attached Figure Description

[0050] Figure 1 This is a flowchart illustrating the steps of a flexible load controllable characteristic modeling method according to the present invention.

[0051] Figure 2 This is a unit composition diagram of a flexible load controllable characteristic modeling system according to the present invention. Detailed Implementation

[0052] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0053] like Figure 1 As shown, a method for modeling the controllable characteristics of flexible loads includes the following steps:

[0054] S1 classifies and identifies the loads of production equipment with continuous operation requirements in the industrial sector and the loads of electrical equipment with time-shifting capabilities in the residential sector. By collecting voltage fluctuation data, current change data and power consumption data of different types of loads under typical operating conditions, a load basic information database is established.

[0055] Specifically, step S1 completes the basic data collection and database construction for flexible loads, providing data support for subsequent modeling. During implementation, the collection scope for two types of flexible loads—industrial and residential—is first defined. In the industrial sector, the focus is on production equipment with continuous operation requirements, such as blast furnace cooling systems in steel enterprises and reaction vessel equipment in chemical enterprises. In the residential sector, the focus is on electrical equipment with time-shifting capabilities, such as household energy storage devices, electric water heaters with adjustable operating times, and smart air conditioners. During the data collection process, specific data collection parameters need to be set: voltage fluctuation data collection accuracy is controlled within ±0.01V, with a sampling frequency of 50 times per second; current change data collection accuracy is ±0.001A, also with a sampling frequency of 50 times per second; power consumption data collection accuracy is ±0.1W, with a sampling frequency of once per minute. The collection period lasts 72 hours, covering the complete production shift of industrial equipment and peak and off-peak periods for residential electricity consumption, ensuring that the data reflects typical load operating conditions. After data collection, the data is filtered to remove abnormal data such as voltage fluctuations exceeding ±10% of rated voltage, current changes exceeding ±15% of rated current, and power consumption fluctuations exceeding ±20% of rated power. The valid data is then organized in the format of "load type-equipment number-collection time-voltage value-current value-power value" and imported into a MySQL database to build a load basic information database. This database supports read and write operations of 1000 data entries per second, providing stable data access support for subsequent feature extraction and classification.

[0056] S2, Based on the load basic information database, a multi-dimensional feature extraction method is adopted to extract controllable characteristic parameters of various loads from three dimensions: voltage fluctuation frequency, current change amplitude, and power consumption peak duration.

[0057] Specifically, step S2 extracts key parameters from the collected basic data through multi-dimensional feature extraction, which reflect the controllable characteristics of flexible loads, providing a basis for load classification. During implementation, the load basic information database constructed in step S1 is used as the data source. A sliding window method is employed to segment the data, with a window duration of 10 minutes and a sliding step size of 5 minutes, ensuring that dynamic changes in load operation can be captured. In terms of voltage fluctuation frequency extraction, the number of times the voltage value exceeds the rated voltage ±5% within each window is counted, and this number is defined as the voltage fluctuation frequency. For example, if the voltage of a cooling system for an industrial blast furnace exceeds the range 8 times within a window, then its voltage fluctuation frequency is 8 times / 10 minutes. In terms of current change amplitude extraction, the difference between the maximum and minimum current values ​​within each window is calculated, and this difference is the current change amplitude. For example, if the maximum current of a residential smart air conditioner is 5A and the minimum current is 2A within a window, its current change amplitude is 3A. In terms of power consumption peak duration extraction, the peak power consumption within each window (i.e., the maximum power within the window) is first determined, and then the duration for which the power value is maintained at 90% or above the peak value is counted. This time is the power consumption peak duration. For example, if the peak power of an industrial reactor is 100kW within a window, and the power is maintained at 90kW or above for 3 minutes, then its power consumption peak duration is 3 minutes. After extraction, the three types of feature parameters are associated with and stored with the corresponding load identification information to form a feature parameter dataset. The completeness of this dataset directly determines the accuracy of subsequent load classification and lays the foundation for load classification in step S3.

[0058] S3, construct a load classification module, input the controllable characteristic parameters into a preset classification algorithm, and accurately classify industrial continuous production load, industrial intermittent production load, residential rigid electricity load, and residential flexible electricity load by calculating the similarity threshold between different characteristic parameters;

[0059] Specifically, step S3 constructs a load classification module to accurately classify flexible loads, providing a classification basis for formulating differentiated control strategies. In implementation, the feature parameter dataset obtained in step S2 is first divided into a training set and a test set in a 7:3 ratio. The training set is used to optimize the classification algorithm parameters, and the test set is used to verify the classification effect. An improved K-nearest neighbor algorithm is selected for classification. First, the initial number of cluster centers is determined to be four (corresponding to four target categories: continuous industrial production load, intermittent industrial production load, rigid residential electricity load, and flexible residential electricity load). Then, the similarity between each sample's feature parameters (voltage fluctuation frequency, current change amplitude, and peak power consumption duration) and each cluster center is calculated. The similarity calculation is based on Euclidean distance, while introducing weighting coefficients. The weight for voltage fluctuation frequency is set to 0.3, the weight for current change amplitude is set to 0.4, and the weight for peak power consumption duration is set to 0.3. The comprehensive similarity between the sample and the cluster center is obtained through weighted calculation. An initial similarity threshold of 0.6 is set, and samples are assigned to the cluster category with the highest similarity exceeding the threshold. If the similarity between a sample and all cluster centers is below the threshold, the cluster center positions are readjusted and iterative calculations are performed, with a maximum of 50 iterations, until the clustering results stabilize (the change in cluster centers is less than 0.01 after three consecutive iterations). The classification effect is verified using a test set, requiring a classification accuracy of no less than 92%. If this is not achieved, the weight coefficients and similarity threshold are adjusted, and retraining is performed until the requirement is met. Finally, all load samples are classified according to the optimal classification model, generating a classification list and storing it in the load basic information database. The classification results can clearly distinguish the differences in the controllable characteristics of different loads. For example, industrial continuous production loads typically have low voltage fluctuation frequency (less than 3 times / 10 minutes), small current change amplitude (less than 2A), and long peak power consumption duration (greater than 15 minutes / 10-minute window), providing a clear load classification basis for the construction of the control strategy model in step S4.

[0060] S4. Build a control strategy model. Based on the controllability characteristics of different types of loads and combined with the real-time operating status parameters of the power grid, determine the control priority of each type of load. Formulate a power rapid adjustment strategy for high-priority loads and a time-based reasonable transfer strategy for low-priority loads.

[0061] Specifically, step S4 involves building a control strategy model and, by combining load classification results with grid status, formulating differentiated control strategies to achieve scientific control of flexible loads. During implementation, real-time grid operating status parameters are first collected, including total active power load, total reactive power load, bus voltage, and grid frequency. The data is collected once per second from the SCADA system of the grid dispatch center to ensure real-time data accuracy. The grid status parameters are then correlated with the load classification results from step S3 to analyze the impact of different load categories on the grid: continuous industrial production loads, due to their high power and continuous operation, significantly affect grid frequency stability and are defined as highly sensitive loads; intermittent industrial production loads, which can be suspended during specific periods, have a moderate impact on the grid and are defined as moderately sensitive loads; rigid residential electricity loads (such as lighting and refrigerators) cannot be interrupted and have a fixed impact on the grid and are defined as moderately sensitive loads; flexible residential electricity loads (such as air conditioners and electric water heaters), whose operating times and power can be flexibly adjusted, have a small impact on the grid and are defined as low-sensitive loads. Based on sensitivity levels, control priorities are determined, with high-sensitivity loads having the highest priority (priority coefficient set to 0.9), followed by medium-sensitivity loads (priority coefficient set to 0.6), and low-sensitivity loads having the lowest priority (priority coefficient set to 0.3). For high-priority loads, a rapid power adjustment strategy is implemented, setting the power adjustment step size to 5% of rated power, the adjustment frequency to once per minute, and the maximum single adjustment amplitude not exceeding 10% of rated power to ensure rapid response to grid frequency fluctuations. For low-priority loads, a time-shifting strategy is implemented, determining the range of time periods for load shifting (e.g., residential air conditioning can be shifted from peak electricity consumption of 18:00-22:00 to off-peak electricity consumption of 00:00-06:00), with a shift duration of no less than 2 hours. After shifting, the operating power is maintained at 70%-90% of rated power to avoid significant impact on user experience. After the control strategies are formulated, a strategy library is created and stored on the virtual power plant control center server. The server adopts a dual-machine hot standby architecture to ensure the security of strategy data and stable retrieval, providing strategy support for communication interaction and command issuance in step S5.

[0062] S5 deploys a communication and interaction system, adopts a two-way data transmission protocol, establishes a communication connection between the virtual power plant control center and various flexible load control terminals, and enables real-time issuance of control commands and real-time uploading of load operation data.

[0063] Specifically, step S5 establishes a real-time communication link between the virtual power plant and the flexible load terminal by deploying a communication interaction system, enabling bidirectional transmission of control commands and operational data. During implementation, the communication interaction system adopts a three-tier architecture of "control center - regional gateway - load terminal." The control center deploys a core communication server using an industrial Ethernet switch with a port rate of 1000Mbps, supporting both TCP / IP and MQTT transmission protocols. Regional gateways are deployed according to power grid zones, with each gateway covering a radius of no more than 5 kilometers, supporting dual-link communication of 4G / 5G and fiber optics. When the fiber optic link is interrupted, it automatically switches to the 4G / 5G link within 10 seconds to ensure communication continuity. Load terminals are equipped with communication modules; industrial equipment terminals communicate with the gateway using an RS485 interface at a communication rate of 9600bps, while residential equipment terminals communicate with the gateway using WiFi or Bluetooth at a rate of 150Mbps for WiFi and 2Mbps for Bluetooth. In terms of communication protocol design, control commands are issued using the MQTT protocol. Command data packets include load identifier, control type (power adjustment / time shift), control parameters, and execution time limit, with the data packet size controlled within 128 bytes to ensure fast transmission. Load operation data is uploaded using the TCP / IP protocol, including real-time power, voltage, current, and operating status. The upload frequency is once every 30 seconds, with the data packet size controlled within 64 bytes to reduce network bandwidth consumption. After deployment, performance testing of the communication system is conducted, including transmission latency, packet loss rate, and link switching time. The requirements are that the transmission latency not exceed 500ms, the packet loss rate be less than 0.1%, and the link switching time not exceed 10 seconds. If the test fails to meet the standards, the gateway deployment location is optimized or the communication parameters are adjusted until the requirements are met. The stable operation of the communication interaction system is crucial for realizing real-time control of flexible loads by the virtual power plant, providing real-time data support for the dynamic correction of the strategy in step S6.

[0064] S6. Based on the real-time load operation data uploaded by the communication interaction system, the output result of the control strategy model is dynamically corrected. By comparing the deviation between the actual load control effect and the preset control target, the parameter weights in the control strategy model are adjusted to complete the modeling of the controllable characteristics of flexible load.

[0065] Specifically, step S6 involves dynamically modifying the control strategy model, optimizing control parameters based on real-time load operation data, improving control effectiveness, and completing the modeling of the controllable characteristics of flexible loads. During implementation, the real-time load operation data uploaded by the communication interaction system in step S5 is used as the basis. This data includes the actual power, voltage, and current changes after the load executes the control command, as well as feedback data on the grid operation status (such as changes in grid frequency and bus voltage after control). First, the deviation between the actual control effect and the preset control target is calculated. The preset control target is set according to the power grid demand. For example, for power adjustment of continuous industrial production load, the preset target is that the deviation between the load power after control and the commanded power does not exceed ±3%, and the power grid frequency fluctuation is controlled within 50Hz±0.1Hz. If the actual data shows that the load power deviation is 5% and the power grid frequency fluctuation is 50Hz±0.2Hz, then the deviation value is calculated as the weighted average of the power deviation ratio (5% / 3%≈1.67) and the frequency deviation ratio (0.2Hz / 0.1Hz=2), with weights set to 0.6 and 0.4 respectively. The comprehensive deviation value is 1.67×0.6+2×0.4=1.802. When the comprehensive deviation value is greater than 1.2, the control strategy model parameter correction is initiated. The correction objects include the control priority coefficient, power adjustment step size, time period transfer range, and other parameters in step S4. For example, if the deviation is caused by insufficient response of high-priority load regulation, the priority coefficient of high-priority load can be increased from 0.9 to 0.95, and the power adjustment step size can be increased from 5% of rated power to 6% of rated power. If the transfer effect of low-priority load periods is not good, the transferable period range can be expanded by 1 hour, and the lower limit of power after transfer can be reduced from 70% of rated power to 65% of rated power. After correction, the new parameters are imported into the regulation strategy model, new control commands are issued through the communication interaction system, the regulation effect is observed, and the deviation calculation and parameter correction process is repeated until the comprehensive deviation value is less than 1.2 and remains stable for 30 consecutive minutes. Finally, an optimized flexible load controllable characteristic model is formed. The model is stored in the model library of the virtual power plant control center, which supports calling and updating at any time according to changes in the power grid operating status, realizing long-term stable and precise regulation of flexible loads, and completing the entire modeling process.

[0066] Preferably, in step S3, the following formula is used to calculate the similarity threshold between different feature parameters: ,in, This represents the similarity value between the i-th feature parameter and the j-th feature parameter. This represents the value of the i-th feature parameter in the k-th sample. This represents the value of the j-th feature parameter in the k-th sample, where n represents the number of samples. This represents the cosine similarity weight coefficient. This represents the index similarity weighting coefficient. This represents the exponential decay coefficient.

[0067] Specifically, the implementation of the feature parameter similarity threshold calculation in step S3 achieves more accurate similarity assessment through multi-factor weighting, providing a reliable basis for load classification. During implementation, the required basic data for calculation is first defined, namely the values ​​of different feature parameters extracted in step S2 in each sample. The number of samples is determined according to the load type, with no less than 500 samples for industrial loads and no less than 1000 samples for residential loads, ensuring that the data volume meets the reliability requirements of the calculation. Regarding parameter settings, the cosine similarity weight coefficient is set to 0.6, which is verified based on historical classification data and effectively reflects the degree of linear correlation between feature parameters; the exponential similarity weight coefficient is set to 0.4, complementing the cosine similarity weight coefficient and highlighting the impact of absolute differences in feature parameters on similarity; the exponential decay coefficient is set to 0.8, determined through multiple iterative tests, to avoid extreme interference in the similarity calculation results when absolute differences are too large. In the calculation process, cosine similarity and exponential similarity are calculated separately. Cosine similarity is obtained by the ratio of the dot product of the feature parameter vectors to the product of their moduli. Exponential similarity is obtained by attenuating the absolute differences in the feature parameters using an exponential function. Then, the two types of similarities are weighted and summed according to a set weighting coefficient to obtain the final feature parameter similarity value. This calculation method can simultaneously consider the linear correlation and absolute differences of the feature parameters. Compared with cosine similarity calculation alone, the classification accuracy can be improved by 8%-12%, effectively solving the classification bias problem caused by different types of parameter differences in traditional classification. This provides key computational support for determining the optimal similarity threshold in step S3 and achieving accurate load classification.

[0068] Preferably, in step S4, when determining the control priority of various load types, the control priority value of the m-th load type is adopted. This represents the maximum allowable power value for the m-th type of load. This represents the current operating power value of the m-th type of load. This indicates the adjustable time length of the m-th type of load. This represents the total running time of the m-th type of load. This represents the power supply reliability requirement factor for the m-th type of load. These represent the weighting parameters for the power difference percentage, the adjustable time percentage, and the reliability requirement coefficient, respectively.

[0069] Specifically, in step S4, the load control priority calculation uses multi-dimensional parameter weighting to quantify priority values, providing an objective basis for formulating differentiated control strategies. During implementation, the acquisition method and value range of each parameter are first clarified: the maximum allowable power value of the m-th type of load is determined based on the rated power of the equipment and the grid's carrying capacity; the maximum allowable power value for continuous industrial production loads does not exceed 110% of the rated power of the equipment, and for flexible residential loads, it does not exceed 100% of the rated power; the current operating power value is collected in real time through the communication interaction system in step S5, with a collection frequency of once per second; the adjustable time length is determined based on the load's operating characteristics; the adjustable time length for intermittent industrial production loads is not less than 2 hours, and for flexible residential loads, it is not less than 4 hours; the total operating time length is the actual daily operating time of the load, typically 8-16 hours for industrial loads and 10-18 hours for residential loads; the power supply reliability demand coefficient is set according to the load's importance, with a coefficient of 0.9-1.0 for continuous industrial production loads, 0.7-0.8 for rigid residential loads, and 0.4-0.6 for flexible residential loads. In terms of weighting parameter settings, the power difference has a weight of 0.4, focusing on the remaining space between the current load power and the maximum allowable power; the adjustable time has a weight of 0.3, reflecting the time flexibility of load regulation; and the reliability requirement coefficient has a weight of 0.3, reflecting the load's dependence on power supply stability. During calculation, the weighting or value of each parameter is calculated separately, and then the weighted sum is obtained to obtain the priority value. The value range is controlled between 0 and 1, with higher values ​​indicating higher regulation priority. This calculation method can control the priority determination error within 5%, avoiding the regulation chaos caused by traditional subjective priority determination, ensuring that high-impact loads are regulated first, and improving the stability of power grid operation.

[0070] Preferably, in step S6, when adjusting the weights of the control strategy model parameters, the following formula is used: ,in, Indicates the first The weight value of the k-th parameter in the next iteration. This represents the weight value of the k-th parameter in the t-th iteration. Indicates the first The weight value of the k-th parameter in the next iteration. This represents the learning rate parameter. This represents the deviation function between the actual control effect and the preset control target. This represents the partial derivative of the bias function with respect to the weight of the k-th parameter at the t-th iteration. This represents the momentum coefficient.

[0071] Specifically, in step S6, the parameter weights of the control strategy model are adjusted using gradient descent combined with momentum term calculation to achieve dynamic optimization of parameter weights, ensuring that the control strategy continuously adapts to actual needs. During implementation, the iteration-related parameters are first determined: the learning rate is set to 0.01, a value determined through multiple adjustments to ensure both the speed of parameter adjustment and to avoid parameter oscillations caused by an excessively large learning rate; the momentum coefficient is set to 0.9 to retain the parameter adjustment trend of the previous iteration and reduce the impact of local optima on the adjustment results; the number of iterations is determined based on the deviation changes, with each adjustment iteration involving no less than 30 iterations to ensure sufficient parameter adjustment. The deviation function is constructed based on the difference between the actual control effect and the preset target. The difference indicators include load power adjustment deviation, grid frequency fluctuation deviation, and control response time deviation. The weights of each indicator are set according to the grid operation requirements: power adjustment deviation weight 0.5, frequency fluctuation deviation weight 0.3, and response time deviation weight 0.2. During the calculation, the partial derivative of the deviation function with respect to the current parameter weights is first calculated to reflect the degree of influence of parameter weight changes on the deviation. Then, combined with the previous parameter weight adjustment and the learning rate, the current parameter weight adjustment value is calculated. The parameter weights are continuously updated iteratively until the deviation function value is less than a preset threshold (usually set to 0.05). Compared with traditional gradient descent, this adjustment method can improve the parameter convergence speed by 20%-30% and effectively avoid parameters getting trapped in local optima, ensuring that the parameters of the control strategy model are always in an optimal state. This keeps the deviation of load control effect continuously controlled within an acceptable range, ensuring the long-term stable operation of the power grid.

[0072] Preferably, in step S2, the following formula is used to calculate the duration of the peak power consumption: ,in, Indicates the duration of peak power consumption. This indicates the moment when the power first reaches its peak. This indicates the moment when power first falls below its peak value. This represents the power consumption value at time t. Indicates peak power consumption. This indicates an indicator function. The function value is 1 when the value inside the parentheses is greater than or equal to 0, and 0 otherwise.

[0073] Specifically, in step S2, the peak power consumption duration is calculated by accurately determining the time the power remains within the peak range using an integral method, providing support for extracting key parameters of load controllability characteristics. In implementation, the criteria for determining the peak power are first established. A sliding window method (10-minute window duration) is used to extract the maximum power value within each window as the peak power value for that window, ensuring that the peak value reflects the highest power demand of the load during a specific period. The peak power range is set to 90% or higher of the peak value. This threshold is determined by analyzing a large amount of load operation data, accurately capturing the peak duration while eliminating the interference of occasional power fluctuations on the statistical results. For time point determination, the moment when the power first reaches its peak is determined by real-time monitoring of power changes. When the power value rises from below the peak to the peak, it is recorded as t1; the moment when the power first falls below the peak is recorded as t2, when the power value falls from the peak range to below the peak. The time recording accuracy is controlled within 1 second to ensure the accuracy of the duration calculation. During the calculation, an indicator function is used to determine the power value within the time period t1 to t2. When the power value is in the peak range, the indicator function is set to 1; otherwise, it is 0. The indicator function is then integrated over this time period to obtain the duration of the peak power consumption. This calculation method can accurately count the duration of the load under high power conditions with an error controlled within 5 seconds. Compared with traditional manual statistics methods, the efficiency is improved by more than 90%. It provides key parameters for distinguishing between continuous industrial production loads (long peak duration) and intermittent loads (short peak duration), supporting subsequent accurate load classification.

[0074] Preferably, in step S5, if a bidirectional data transmission delay occurs, the following formula is used: ,in, Indicates the delay in bidirectional data transmission. Indicates the length of the transmitted data. Indicates the data transmission rate. Indicates signal propagation delay, This indicates a data processing delay.

[0075] Specifically, in step S5, the bidirectional data transmission delay calculation achieves precise quantification of delay by decomposing the delay of each stage of the transmission process, providing a basis for optimizing the communication interaction system. During implementation, the calculation basis for each delay component is first clarified: the transmission data length is determined according to the data type; the control command data packet length is fixed at 128 bytes (including necessary information such as load identifier and control type), and the load operation data packet length is 64 bytes (including core data such as real-time power and voltage); the data transmission rate is set according to the communication link type: 1000Mbps for fiber optic links, 100Mbps for 4G / 5G links, 150Mbps for WiFi links, and 2Mbps for Bluetooth links, ensuring that the rate values ​​match the actual communication link performance; the signal propagation delay is calculated based on the transmission distance: the propagation delay from the control center to the regional gateway is calculated at 0.005 milliseconds per kilometer, and the propagation delay from the regional gateway to the load terminal is calculated at 0.001 milliseconds per 100 meters; the data processing delay is determined according to equipment performance: the processing delay of the control center server does not exceed 10 milliseconds, the processing delay of the regional gateway does not exceed 5 milliseconds, and the processing delay of the load terminal does not exceed 3 milliseconds. The calculation process first calculates the data transmission time (the ratio of data length to transmission rate), signal propagation delay, and data processing delay separately, and then sums these three to obtain the bidirectional data transmission delay. This calculation method can comprehensively reflect the impact of each link in the communication on the delay, with the calculation error controlled within 2 milliseconds. By periodically calculating the transmission delay, communication bottlenecks (such as increased transmission time due to a decrease in transmission rate) can be identified in a timely manner, providing data support for optimizing communication module configuration and adjusting transmission protocols, ensuring that the communication interaction system meets real-time control requirements (delay not exceeding 500ms).

[0076] Preferably, step S3 includes the following sub-steps:

[0077] S31, retrieve the controllable characteristic parameters of all load samples from the load basic information database, make a preliminary division according to the two major categories of industrial load and residential load, screen out samples with complete characteristic parameters in each type of load, and remove samples with missing data or abnormal fluctuations exceeding the preset range to ensure the validity of the data used in subsequent classification calculations.

[0078] S32, after the initial division of industrial load samples and residential load samples, the characteristic parameters are standardized. The voltage fluctuation frequency, current change amplitude and power consumption peak duration parameters with different dimensions are converted into standardized parameters with the same numerical range to eliminate the influence of dimension differences on the classification results.

[0079] S33, input the standardized feature parameters into the preset classification algorithm, set the initial similarity threshold, and continuously adjust the threshold size through iterative calculation so that the feature parameter similarity of samples of the same type of load in the classification result is higher than the preset threshold, and the feature parameter similarity of samples of different types of load is lower than the preset threshold, and determine the optimal similarity threshold;

[0080] S34. Based on the optimal similarity threshold, all load samples are finally classified to generate a classification list of industrial continuous production load, industrial intermittent production load, residential rigid electricity load, and residential flexible electricity load. The classification results are stored in the load basic information database to provide a classification basis for the subsequent construction of the control strategy model.

[0081] Specifically, step S3 involves detailed implementation of the entire load classification process to ensure classification accuracy. During S31, controllable characteristic parameters of all load samples are retrieved from the load basic information database. After initial classification into industrial and residential categories, data screening criteria are set: voltage fluctuation data missing rate not exceeding 3%, current change data anomalies (exceeding rated value ±15%) not exceeding 2%, and power consumption data remaining blank for no more than 5 minutes. Samples that do not meet these criteria are removed. The effectiveness rate of industrial load samples must reach over 95%, and the effectiveness rate of residential load samples must reach over 98% to ensure the validity of subsequent calculation data. In S32, a linear transformation method is used to standardize the feature parameters, unifying the voltage fluctuation frequency (original range 0-20 times / 10 minutes), current change amplitude (original range 0-10A), and peak power consumption duration (original range 0-10 minutes) to the 0-1 range. The transformation formula is calculated based on the maximum and minimum values ​​of the samples to ensure the elimination of dimensional differences. For example, the standardized voltage fluctuation frequency value = (original value - minimum sample value) / (maximum sample value - minimum sample value). The deviation of the standardized parameters is controlled within ±0.02. In S33, the standardized parameters are input into the improved K-nearest neighbor algorithm. The initial similarity threshold is set to 0.6, and the adjustment increment is 0.05 in each iteration. During the iteration, the mean similarity of load samples of the same and different classes is recorded. The iteration stops when the difference between the mean of the same class and the mean of the different classes is the largest. Usually, 10-15 iterations are needed to determine the optimal threshold. The optimal threshold must satisfy the following conditions: similarity of samples of the same class ≥ 0.75 and similarity of samples of different classes ≤ 0.45. In step S34, after classifying according to the optimal threshold, a classification list including load type, equipment number, classification result, and characteristic parameters is generated and stored in the database in CSV format. The storage path is divided by "year / month / load category". At the same time, a classification report is generated, which must include the total number of samples, the number of valid samples, and the classification accuracy (≥92%), providing a clear classification basis for the construction of the control strategy model in step S4.

[0082] Preferably, step S4 includes the following sub-steps:

[0083] S41: Collect real-time operating status parameters of the power grid, including the current active power load, reactive power load, bus voltage level and frequency fluctuation of the power grid. Remove noise interference from the parameters through data preprocessing, extract the calibration index of the power grid operating status, and establish a real-time power grid status assessment dataset.

[0084] S42, Combining the load classification results obtained in step S3, analyze the degree of influence of different types of loads on the power grid operation status, and determine the control sensitivity coefficient of each type of load by calculating the correlation between the power changes of various types of loads and the frequency fluctuations of the power grid.

[0085] S43. Based on the control sensitivity coefficient and the real-time power grid status assessment dataset, formulate a control priority evaluation standard, set loads with high control sensitivity coefficients and significant impact on the stable operation of the power grid as high priority, and set loads with low control sensitivity coefficients and minimal impact on the stable operation of the power grid as low priority.

[0086] S44. Design control strategies for loads with different priorities. High-priority loads adopt a power fast adjustment strategy, which determines the power adjustment step size, frequency and maximum adjustment range. Low-priority loads adopt a time-switching strategy, which plans the switching range, switching duration and switching power after the switching of the load during the running period, forming a complete set of control strategies.

[0087] Specifically, step S4 involves detailed procedures based on control strategies to ensure the strategies are scientifically sound and feasible. During S41 implementation, real-time grid status parameters are collected via the SCADA system of the power grid dispatch center at a frequency of once per second. These parameters include active power load (measurement range 0-5000MW, accuracy ±0.5MW), reactive power load (measurement range 0-2000Mvar, accuracy ±0.2Mvar), bus voltage (measurement range 0-380kV, accuracy ±0.1kV), and frequency (measurement range 49.5-50.5Hz, accuracy ±0.01Hz). A Kalman filter algorithm is used to remove data noise, with a filter window size of 5 sampling points. The processed data is then stored in the power grid status database, with a data update delay of no more than 1 second, thus constructing a real-time status assessment dataset. In S42, combining the classification results of S3, the Pearson correlation coefficient is used to calculate the correlation between various load power changes and grid frequency fluctuations. The calculation window duration is set to 1 minute, and the correlation coefficient ranges from -1 to 1. The larger the absolute value, the greater the influence. The correlation coefficient for continuous industrial production loads is typically 0.7-0.9, and for residential flexible electricity loads it is 0.3-0.5. Based on this, loads with a correlation coefficient ≥0.6 are defined as highly sensitive loads, 0.3-0.6 as medium sensitive loads, and <0.3 as low sensitive loads. The sensitivity coefficients for regulating various loads are determined. In S43, priority evaluation criteria are established: the priority coefficient for highly sensitive loads is 0.8-1.0, for medium sensitive loads it is 0.5-0.7, and for low sensitive loads it is 0.2-0.4. At the same time, combined with the real-time active power gap of the grid (when the gap is >100MW, high sensitive loads are prioritized for regulation), the final regulation priority is comprehensively determined. The priority determination cycle is 5 minutes to ensure timely adaptation to changes in grid status. In S44, the power adjustment step size for high-priority loads is set to 5% of the rated power, the adjustment frequency is once per minute, and the maximum single adjustment amplitude does not exceed 10% of the rated power; the transfer range for low-priority loads is from 2 hours before the peak electricity consumption period to 3 hours after the peak, the transfer duration is not less than 2 hours, and the operating power is maintained at 70%-90% of the rated power after the transfer, forming a set of control strategies including control objects, strategy types, and parameter ranges, which are stored in the strategy library for later use.

[0088] Preferably, step S5 includes the following sub-steps:

[0089] S51, determine the hardware architecture of the communication interaction system, select a communication module with high bandwidth and low latency as the communication interface between the virtual power plant control center and the load control terminal, configure a data storage server to store load operation data and control commands during transmission, and deploy a data processing unit to perform real-time parsing of the transmitted data.

[0090] S52 designs a bidirectional data transmission protocol, specifying the data packet format, transmission rate, and verification method for control commands sent from the virtual power plant control center to the load control terminal. It also determines the data packet structure, sampling frequency, and data compression algorithm for load operation data uploaded from the control terminal to the control center to ensure the integrity and accuracy of data transmission.

[0091] S53, build a communication connection test environment to simulate communication scenarios under different network conditions, including network bandwidth fluctuations, signal interference and network outage recovery, test the transmission delay, data packet loss rate and command response speed of the communication interaction system under different scenarios, and record the calibration performance indicators during the test process;

[0092] S54, based on the communication connection test results, optimize and adjust the hardware parameters and transmission protocol of the communication interaction system, replace communication modules whose performance does not meet the requirements, and correct the data verification algorithm in the transmission protocol to ensure that the communication interaction system can still carry out stable data transmission and command interaction in complex network environments.

[0093] Specifically, step S5 involves building and optimizing a communication interaction system to ensure stable data transmission. During the implementation of S51, the hardware architecture is determined as follows: the virtual power plant control center is configured with two redundant communication servers (CPU: Intel Xeon Gold 6330, memory: 64GB, hard drive: 1TB SSD), using an industrial Ethernet switch (24 ports, 1000Mbps speed); the area gateway is an industrial-grade 4G / 5G dual-mode gateway (supporting LTE Cat.12, downlink speed: 600Mbps, uplink speed: 150Mbps), deployed one per 5 square kilometers, with each gateway equipped with dual SIM cards for link redundancy; for load terminal communication modules, industrial equipment uses RS485 modules (transmission distance ≤1200 meters, speed: 9600bps), while residential equipment uses WiFi 6 modules (speed: 1.2Gbps, coverage range ≤100 meters) or Bluetooth 5.2 modules (speed: 2Mbps, coverage range ≤30 meters), and an 8TB data storage server (RAID 5 array) and a quad-core data processing unit (2.4GHz clock speed) are configured to ensure hardware support for communication requirements. In S52, a bidirectional transmission protocol is designed: control commands are sent using the MQTT protocol, with a data packet format of "frame header (2 bytes) + payload ID (8 bytes) + control type (1 byte) + control parameters (4 bytes) + checksum (1 byte) + frame trailer (2 bytes)", the transmission rate is set to 1Mbps, and the checksum method is CRC16; payload data is uploaded using the TCP / IP protocol, with a data packet structure of "frame header (2 bytes) + payload ID (8 bytes) + running data (8 bytes) + timestamp (4 bytes) + checksum (1 byte) + frame trailer (2 bytes)", the sampling frequency is 30 seconds / time, and the LZ77 compression algorithm (compression rate ≥50%) is used to ensure complete and accurate data transmission. In S53, a test environment was set up to simulate network bandwidth fluctuations (50Mbps-1000Mbps), signal interference (signal-to-noise ratio 20dB-40dB), and network outage recovery (outage duration 10 seconds-5 minutes). Each scenario was tested for 2 hours, recording transmission latency (≤500ms), packet loss rate (≤0.1%), and command response speed (≤1 second). The test sample size was no less than 1000 sets, and a performance test report was generated. In S54, optimizations were made based on the test results: for gateways with excessive transmission latency, the antenna gain was adjusted (from 5dBi to 8dBi); for gateways with excessive packet loss rate, the verification algorithm was changed to CRC32; for gateways with slow network outage recovery, the link detection cycle was shortened (from 30 seconds to 10 seconds). After optimization, the tests were repeated until all performance indicators met the standards, ensuring the communication system operated stably under complex conditions and supporting real-time interaction between the virtual power plant and the load terminal.

[0094] like Figure 2As shown, a flexible load controllable characteristic modeling system is applied to a flexible load controllable characteristic modeling method, comprising:

[0095] The load information acquisition and storage unit is used to collect voltage fluctuation data, current change data, and power consumption data of production equipment loads with continuous operation requirements in the industrial field and electrical equipment loads with time-shifting capabilities in the residential field, and to establish a load basic information database. This unit is connected to the load classification and processing unit through a data transmission line to transmit the collected and stored load data to the load classification and processing unit.

[0096] The load classification processing unit receives load data transmitted from the load information acquisition and storage unit, extracts controllable characteristic parameters using a multi-dimensional feature extraction method, inputs the controllable characteristic parameters into a preset classification algorithm to calculate a similarity threshold, and performs accurate load classification. This unit is connected to the load information acquisition and storage unit and the control strategy generation unit, and synchronizes the classification results to the load information acquisition and storage unit and transmits them to the control strategy generation unit. The controllable characteristic parameters include voltage fluctuation frequency, current change amplitude, and peak power consumption duration.

[0097] The regulation strategy generation unit is used to build a regulation strategy model, receive the load classification results transmitted by the load classification processing unit, determine the load regulation priority by combining the real-time operating status parameters of the power grid, and design regulation strategies for loads of different priorities (a power rapid adjustment strategy is formulated for high-priority loads, and a time period reasonable transfer strategy is formulated for low-priority loads). This unit is connected to the communication interaction control unit and converts the generated regulation strategy into control commands and transmits them to the communication interaction control unit.

[0098] The communication and interaction control unit is used to receive control commands transmitted by the regulation strategy generation unit, establish a communication connection with the flexible load control terminal using a two-way data transmission protocol, and send control commands and upload load operation data. This unit is connected to the regulation strategy generation unit and the strategy dynamic correction unit respectively, and transmits the uploaded load operation data to the strategy dynamic correction unit.

[0099] The strategy dynamic correction unit is used to receive real-time load operation data transmitted by the communication interaction control unit, calculate the deviation between the actual control effect and the preset control target, and adjust the parameter weights in the control strategy model. This unit is connected to the control strategy generation unit and feeds back the corrected parameter weights to the control strategy generation unit.

[0100] The power grid status monitoring unit is used to collect real-time active power load, reactive power load, bus voltage level and frequency fluctuation of the power grid. After preprocessing the data, it is transmitted to the control strategy generation unit to provide power grid status data support for determining control priorities. This unit is connected to the control strategy generation unit through a wireless communication module to ensure real-time transmission of power grid status data.

[0101] In terms of load classification, this invention collects multi-dimensional operational data and extracts feature parameters, then combines them with a classification algorithm to accurately classify different types of flexible loads. The classification results provide a precise basis for subsequent regulation. In terms of regulation strategy formulation, the invention determines the load regulation priority based on the real-time operating status of the power grid and designs differentiated strategies for loads with different priorities, thereby improving the accuracy and rationality of regulation. In terms of data interaction and strategy optimization, the established communication interaction system enables real-time transmission of control commands and operational data. At the same time, a dynamic correction mechanism adjusts model parameters to ensure continuous optimization of regulation effects. Overall, this invention can improve the load regulation efficiency of the virtual power plant and enhance the power grid's absorption capacity.

[0102] To address the lack of a unified load classification architecture, this invention constructs a dedicated load classification module. Based on key data from load operation, it extracts feature parameters and uses classification algorithms to calculate similarity thresholds, achieving clear and accurate classification of various flexible loads. This changes the previous situation of chaotic classification and insufficient data. Regarding the imperfect communication interaction mechanism, it deploys a standardized two-way communication interaction system, selects suitable communication modules, and designs standardized transmission protocols. Furthermore, it optimizes and tests complex network conditions to ensure real-time and stable data transmission. Simultaneously, it incorporates a dynamic correction mechanism to ensure the effective execution of control strategies, solving the problems of communication delays and unreliable data transmission.

[0103] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0104] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for modeling the controllable characteristics of flexible loads, characterized in that, Includes the following steps: S1 classifies and identifies the loads of production equipment with continuous operation requirements in the industrial sector and the loads of electrical equipment with time-shifting capabilities in the residential sector. By collecting voltage fluctuation data, current change data and power consumption data of different types of loads under typical operating conditions, a load basic information database is established. S2, Based on the load basic information database, controllable characteristic parameters are extracted using a multi-dimensional feature extraction method. The controllable characteristic parameters are input into a preset classification algorithm to calculate a similarity threshold and perform accurate load classification. The controllable characteristic parameters include voltage fluctuation frequency, current change amplitude, and peak power consumption duration. S3, construct a load classification module, input the controllable characteristic feature parameters into a preset classification algorithm, and accurately classify the controllable characteristics of industrial continuous production load, industrial intermittent production load, residential rigid electricity load, and residential flexible electricity load by calculating the similarity threshold between different feature parameters; S4. Build a control strategy model, determine the control priority of each type of load based on the controllability characteristics of different types of loads and the real-time operating status parameters of the power grid, formulate a power rapid adjustment strategy for high-priority loads, and formulate a time-based reasonable transfer strategy for low-priority loads. S5 deploys a communication and interaction system, adopts a two-way data transmission protocol, establishes a communication connection between the virtual power plant control center and various flexible load control terminals, and performs real-time issuance of control commands and real-time uploading of load operation data. S6. Based on the real-time load operation data uploaded by the communication interaction system, calculate the deviation between the actual control effect and the preset control target, adjust the parameter weights in the control strategy model, and complete the modeling of the controllable characteristics of flexible load.

2. The method for modeling the controllable characteristics of flexible loads according to claim 1, characterized in that, In step S3, the following formula is used to calculate the similarity threshold between different feature parameters: ， in, This represents the similarity value between the i-th feature parameter and the j-th feature parameter. This represents the value of the i-th feature parameter in the k-th sample. This represents the value of the j-th feature parameter in the k-th sample, where n represents the number of samples. This represents the cosine similarity weight coefficient. This represents the index similarity weighting coefficient. This represents the exponential decay coefficient.

3. The method for modeling the controllable characteristics of flexible loads according to claim 1, characterized in that, In step S4, when determining the control priority of various load types, the control priority value of the m-th load type is adopted. This represents the maximum allowable power value for the m-th type of load. This represents the current operating power value of the m-th type of load. This indicates the adjustable time length of the m-th type of load. This represents the total running time of the m-th type of load. This represents the power supply reliability requirement factor for the m-th type of load. These represent the weighting parameters for the power difference percentage, the adjustable time percentage, and the reliability requirement coefficient, respectively.

4. The method for modeling the controllable characteristics of flexible loads according to claim 1, characterized in that, In step S6, the following formula is used when adjusting the weights of the control strategy model parameters: ， in, Indicates the first The weight value of the k-th parameter in the next iteration. This represents the weight value of the k-th parameter in the t-th iteration. Indicates the first The weight value of the k-th parameter in the next iteration. This represents the learning rate parameter. This represents the deviation function between the actual control effect and the preset control target. This represents the partial derivative of the bias function with respect to the weight of the k-th parameter at the t-th iteration. This represents the momentum coefficient.

5. The method for modeling the controllable characteristics of flexible loads according to claim 1, characterized in that, In step S2, the following formula is used to calculate the duration of the peak power consumption: ， in, Indicates the duration of peak power consumption. This indicates the moment when the power first reaches its peak. This indicates the moment when power first falls below its peak value. This represents the power consumption value at time t. Indicates peak power consumption. This indicates an indicator function. The function value is 1 when the value inside the parentheses is greater than or equal to 0, and 0 otherwise.

6. The method for modeling the controllable characteristics of flexible loads according to claim 1, characterized in that, In step S5, if a bidirectional data transmission delay occurs, the following formula is used: ， in, Indicates the bidirectional data transmission delay. Indicates the length of the transmitted data. Indicates the data transmission rate. Indicates signal propagation delay, This indicates a data processing delay.

7. The method for modeling the controllable characteristics of flexible loads according to claim 1, characterized in that, S3 includes the following steps: S31, retrieve the controllable characteristic parameters of all load samples from the load basic information database, make a preliminary division according to the two major categories of industrial load and residential load, screen out samples with complete characteristic parameters in each type of load, and remove samples with missing data or abnormal fluctuations exceeding the preset range to ensure the validity of the data used in subsequent classification calculations. S32, after the initial division of industrial load samples and residential load samples, the characteristic parameters are standardized. The voltage fluctuation frequency, current change amplitude and power consumption peak duration parameters with different dimensions are converted into standardized parameters with the same numerical range to eliminate the influence of dimension differences on the classification results. S33, input the standardized feature parameters into the preset classification algorithm, set the initial similarity threshold, and continuously adjust the threshold size through iterative calculation so that the feature parameter similarity of samples of the same type of load in the classification result is higher than the preset threshold, and the feature parameter similarity of samples of different types of load is lower than the preset threshold, and determine the optimal similarity threshold; S34. Based on the optimal similarity threshold, all load samples are finally classified to generate a classification list of industrial continuous production load, industrial intermittent production load, residential rigid electricity load, and residential flexible electricity load. The classification results are stored in the load basic information database to provide a classification basis for the subsequent construction of the control strategy model.

8. The method for modeling the controllable characteristics of flexible loads according to claim 1, characterized in that, S4 includes the following steps: S41: Collect real-time operating status parameters of the power grid, including the current active power load, reactive power load, bus voltage level and frequency fluctuation of the power grid. Remove noise interference from the parameters through data preprocessing, extract the calibration index of the power grid operating status, and establish a real-time power grid status assessment dataset. S42, Combining the load classification results obtained in step S3, analyze the degree of influence of different types of loads on the power grid operation status, and determine the control sensitivity coefficient of each type of load by calculating the correlation between the power changes of various types of loads and the frequency fluctuations of the power grid. S43. Based on the control sensitivity coefficient and the real-time power grid status assessment dataset, formulate a control priority evaluation standard, set loads with high control sensitivity coefficients and significant impact on the stable operation of the power grid as high priority, and set loads with low control sensitivity coefficients and minimal impact on the stable operation of the power grid as low priority. S44. Design control strategies for loads with different priorities. High-priority loads adopt a power fast adjustment strategy, which determines the power adjustment step size, frequency and maximum adjustment range. Low-priority loads adopt a time-switching strategy, which plans the switching range, switching duration and switching power after the switching of the load during the running period, forming a complete set of control strategies.

9. The method for modeling the controllable characteristics of flexible loads according to claim 1, characterized in that, S5 includes the following steps: S51, determine the hardware architecture of the communication interaction system, select a communication module with high bandwidth and low latency as the communication interface between the virtual power plant control center and the load control terminal, configure a data storage server to store load operation data and control commands during transmission, and deploy a data processing unit to perform real-time parsing of the transmitted data. S52 designs a bidirectional data transmission protocol, specifying the data packet format, transmission rate, and verification method for control commands sent from the virtual power plant control center to the load control terminal. It also determines the data packet structure, sampling frequency, and data compression algorithm for load operation data uploaded from the control terminal to the control center to ensure the integrity and accuracy of data transmission. S53, build a communication connection test environment to simulate communication scenarios under different network conditions, including network bandwidth fluctuations, signal interference and network outage recovery, test the transmission delay, data packet loss rate and command response speed of the communication interaction system under different scenarios, and record the calibration performance indicators during the test process; S54, based on the communication connection test results, optimize and adjust the hardware parameters and transmission protocol of the communication interaction system, replace communication modules whose performance does not meet the requirements, and correct the data verification algorithm in the transmission protocol to ensure that the communication interaction system can still carry out stable data transmission and command interaction in complex network environments.

10. A flexible load controllable characteristic modeling system, characterized in that, The system is applied to the flexible load controllable characteristic modeling method according to any one of claims 1-9, comprising: The load information acquisition and storage unit is used to classify and identify the loads of production equipment with continuous operation requirements in the industrial field and the loads of electrical equipment with time-shifting capabilities in the residential field. By collecting voltage fluctuation data, current change data and power consumption data of different types of loads under typical operating conditions, a load basic information database is established. The load classification processing unit is used to receive load data transmitted by the load information acquisition and storage unit, extract controllable characteristic parameters using a multi-dimensional feature extraction method, input the controllable characteristic parameters into a preset classification algorithm to calculate a similarity threshold, and perform accurate load classification. The controllable characteristic parameters include voltage fluctuation frequency, current change amplitude, and peak power consumption duration. The load classification processing unit is connected to the load information acquisition and storage unit and the control strategy generation unit, respectively, and synchronizes the classification results to the load information acquisition and storage unit and transmits them to the control strategy generation unit. The control strategy generation unit is used to build a control strategy model, receive the load classification results transmitted by the load classification processing unit, determine the control priority of various loads in combination with the real-time operating status parameters of the power grid, formulate a power rapid adjustment strategy for high priority loads, formulate a time period reasonable transfer strategy for low priority loads, and convert the generated control strategy into control commands and transmit them to the communication interaction control unit. The communication and interaction control unit is used to receive control commands transmitted by the regulation strategy generation unit, establish a communication connection with the flexible load control terminal using a two-way data transmission protocol, and send control commands and upload load operation data. This unit is connected to the regulation strategy generation unit and the strategy dynamic correction unit respectively, and transmits the uploaded load operation data to the strategy dynamic correction unit. The strategy dynamic correction unit is used to receive real-time load operation data transmitted by the communication interaction control unit, calculate the deviation between the actual control effect and the preset control target, and adjust the parameter weights in the control strategy model. This unit is connected to the control strategy generation unit and feeds back the corrected parameter weights to the control strategy generation unit. The power grid status monitoring unit is used to collect real-time active power load, reactive power load, bus voltage level and frequency fluctuation of the power grid. After preprocessing the data, it is transmitted to the control strategy generation unit to provide power grid status data support for determining control priorities. This unit is connected to the control strategy generation unit through a wireless communication module to ensure real-time transmission of power grid status data.

Images