Virtual power plant user side load aggregation control method

Abstract

The invention discloses a virtual power plant user side load aggregation control method, which comprises the steps of processing a corrected load prediction demand quantity and a power grid dispatching parameter by adopting a distributed optimization algorithm, generating a multi-period load dispatching scheme and calculating a regulation and control time sequence parameter of each unit; the monitoring instruction executes recorded data and calculates a deviation value between actual response and an expected regulation and control target, and if the deviation value exceeds a preset range, a load response correction parameter is generated; processing the load response correction parameter and the system operation parameter by adopting a progressive control algorithm, generating a dynamic regulation curve and updating a regulation amplitude parameter; establishing a visual interface according to the dynamic adjustment curve, and generating a load distribution diagram and a real-time load curve data set; and analyzing the load distribution diagram and the real-time load curve data set by adopting a model updating mechanism, and if the deviation rate exceeds a preset target, iteratively updating a parameter set of the load prediction model to realize accurate load scheduling and control. According to the invention, the flexibility and reliability of power grid dispatching are improved.

Description

Technical Field

[0001] This invention belongs to the field of virtual power plant technology, and particularly relates to a method for user-side load aggregation control in a virtual power plant. Background Technology

[0002] As the power system transforms towards cleaner and smarter systems, virtual power plants, as a key technology for integrating distributed energy resources, play a crucial role in improving grid flexibility and economy. User-side load aggregation control, a core component of virtual power plants, enables the unified scheduling of massive, dispersed user load resources, achieving power supply and demand balance and optimized system operation.

[0003] Current load aggregation methods suffer from several significant drawbacks. Traditional centralized control methods struggle to adapt to the complexity and diversity of large-scale distributed loads, lacking accurate modeling of the characteristics of different load types. Existing coordination mechanisms are insufficiently responsive to communication delays and data anomalies, failing to guarantee the reliable transmission and execution of control commands. Furthermore, most solutions lack robust security systems and user-friendly interfaces, limiting their practical application effectiveness.

[0004] In the user-side load aggregation control of virtual power plants, the effective integration of distributed heterogeneous load resources faces significant technical challenges. Due to the diverse types and characteristics of user-side loads, ranging from temperature-controlled loads such as air conditioners and water heaters to adjustable loads such as industrial equipment, each load possesses unique dynamic response characteristics and constraints. This heterogeneity makes establishing a unified aggregation model extremely difficult. The complexity of load characteristic modeling directly impacts the design of multi-level coordinated control architectures. Without accurate load models, the system cannot accurately predict the response behavior of various loads, leading to discrepancies between coordinated commands and actual execution results. The uncertainty of coordinated control further exacerbates the difficulty of real-time data processing and analysis. The acquisition, storage, and real-time analysis of massive amounts of multi-source heterogeneous data place extremely high demands on the processing capabilities of data centers, while insufficient data processing capabilities affect the formulation and execution effectiveness of load contingency plans.

[0005] How to construct a virtual power plant user-side load aggregation control method that can effectively integrate heterogeneous load resources, achieve multi-level coordinated control, support real-time processing of massive data, and have sound security and user-friendly interaction has become a key issue that urgently needs to be solved. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a virtual power plant user-side load aggregation control method, comprising: Real-time power data and historical operation records of distributed load units are acquired, and time series analysis algorithms are used to extract features and recognize patterns from the real-time power data to obtain load change parameters and user behavior characteristic parameters. A load forecasting model is established based on the load change parameters. If the deviation of the predicted load demand exceeds a preset threshold, a corrected load forecast demand is generated through a parameter adjustment mechanism. The corrected load forecast demand and power grid dispatch parameters are processed using a distributed optimization algorithm to generate a multi-period load dispatch scheme and calculate the control timing parameters of each distributed load unit. Control commands are sent to the distributed load units according to the control timing parameters. If the communication delay exceeds the preset time window, the locally stored backup command set is triggered to generate command execution record data. The system monitors the execution of the commands, records data, and calculates the deviation between the actual response and the expected control target. If the deviation exceeds a preset range, load response correction parameters are generated. An incremental control algorithm is used to process the load response correction parameters and system operating parameters, generate a dynamic adjustment curve, and update the control amplitude parameters. A visualization interface is established based on the dynamic adjustment curve to generate a load distribution map and a real-time load curve dataset. The load distribution map and real-time load curve dataset are analyzed using a model update mechanism. If the deviation rate exceeds the preset target, the parameter set of the load prediction model is iteratively updated to generate optimized load aggregation control parameters, which are then applied to the next round of load scheduling plan.

[0007] Preferably, the process of obtaining load change parameters and user behavior characteristic parameters includes: The real-time power data and historical operation records of the distributed load units are obtained, and outliers and noise are removed through data cleaning to obtain a preprocessed power dataset. Time series analysis algorithms are used to extract features from the preprocessed power dataset, calculate statistical and periodic features, and obtain a load change feature set. Clustering algorithms are used to perform pattern recognition on the load change feature set to determine the load change trend and the distribution of user behavior patterns; If the fluctuation range of the load change trend exceeds the preset threshold, the abnormal data points are marked to obtain an abnormal load pattern set. Based on the abnormal load pattern set and the distribution of user behavior patterns, a decision tree algorithm is used for classification to obtain user behavior feature parameters; Based on load change trends and user behavior characteristics, a load change parameter set containing peak load and periodic change patterns is generated. For the load change parameter set, a time series forecasting model is used to predict the load change trend in future periods, and a predicted load dataset is obtained.

[0008] Preferably, if the deviation of the predicted load demand exceeds a preset threshold, the process of generating a corrected predicted load demand through a parameter adjustment mechanism includes: Based on historical load data and real-time load data, load change parameters are obtained to construct an initial load prediction model; The predicted load demand is obtained by processing the model training data through the initial load prediction model. If the deviation between the predicted load demand and the actual load data exceeds a preset deviation threshold, then prediction deviation analysis is initiated to determine the source of the deviation. Based on the prediction deviation analysis results, optimize parameters are obtained from the set of adjustment parameters, and the parameter adjustment mechanism is updated. The initial load forecasting model is optimized through a parameter adjustment mechanism to generate a revised load demand. The revised load demand is evaluated by verifying the prediction results, and the model training data is updated by updating the revised load demand to optimize and obtain the target load prediction model.

[0009] Preferably, the process of using a distributed optimization algorithm to process the corrected load forecast demand and power grid dispatch parameters, generating a multi-period load dispatch scheme, and calculating the control timing parameters of each distributed load unit includes: Obtain the corrected load forecast demand and power grid dispatch parameters, and generate a standardized input dataset through data preprocessing; The standardized input dataset is processed using a distributed optimization algorithm to obtain multi-period load allocation results. Based on the multi-period load allocation results, if the allocation results meet the preset power grid stability threshold, a preliminary load dispatching scheme is generated. The preliminary load scheduling scheme is analyzed, and the control timing of each unit is optimized using a linear programming algorithm to obtain the control timing parameters. If there is a deviation between the control timing parameters and the power grid dispatch parameters, the updated load dispatch scheme is obtained by iteratively adjusting the parameters of the distributed optimization algorithm. Based on the updated load scheduling scheme, a time series analysis algorithm is used to verify the stability of multi-period load allocation and determine the final load scheduling scheme. The target control timing parameters for each unit are generated based on the final load scheduling scheme.

[0010] Preferably, the process of generating instruction execution record data includes: If the timing parameters are adjusted, the communication delay data is obtained to determine whether the communication delay exceeds the preset time window. If the communication delay exceeds the preset time window, a backup instruction set is retrieved from local storage to determine the activation status of the triggering mechanism. Based on the activation state of the triggering mechanism, control commands from the standby command set are sent to the load unit to obtain preliminary results of command execution. Based on the preliminary results of the command execution, obtain the response data of the load unit and determine whether the response data matches the control timing parameters; If the response data matches the control timing parameters, the instruction execution data is structured to obtain formatted execution data. Detailed instruction records are generated based on the formatted execution data, and the support vector machine algorithm is used to classify and analyze the instruction records to identify abnormal situations in the execution data and obtain the classification results.

[0011] Preferably, the process of monitoring the execution record data of the instruction and calculating the deviation between the actual response and the expected control target, and generating load response correction parameters if the deviation exceeds a preset range, includes: Obtain instruction execution record data and use time series analysis methods to obtain the feature values ​​of the actual response data; The deviation value is calculated by comparing the characteristic value with the preset control target; If the deviation value exceeds the preset threshold range, a linear regression algorithm is used to generate load response correction parameters; Based on the load response correction parameters, the input of the instruction execution record data is adjusted to obtain the updated instruction sequence; By using the updated instruction sequence, the deviation between the actual response data and the preset control target is recalculated to determine whether the deviation is still within the preset threshold range. If the deviation value still exceeds the preset threshold range, the gradient descent algorithm is used to optimize the load response correction parameters to obtain the optimized correction parameters; Based on the optimized correction parameters, the instruction execution record data is adjusted to generate the final load response control sequence.

[0012] Preferably, the process of using an incremental control algorithm to process the load response correction parameters and system operating parameters, generating a dynamic adjustment curve, and updating the control amplitude parameters includes: Obtain load response correction parameters and system operating parameters, and extract real-time data from the load management system and operating status database to obtain the initial parameter set; The initial parameter set is processed using an incremental control algorithm. If the load response correction parameter exceeds a preset threshold, the parameter is normalized to obtain a standardized parameter set. Based on the standardized parameter set and combined with the system operating parameters, the key points of the adjustment curve are calculated through the dynamic curve generation model to obtain the dynamic adjustment curve. Analyze the changing trend of the dynamic adjustment curve. If the changing trend deviates from the preset system stability range, adjust the curve parameters through a secondary optimization algorithm to obtain an optimized adjustment curve. The control amplitude parameter is extracted from the optimized control curve, and the system control amplitude is updated using a parameter mapping method to obtain the updated control amplitude parameter. The updated system operation status data is obtained and compared with the load response data to obtain a dataset of control effects. Based on the aforementioned control effect dataset, if the system's operating status does not meet the preset performance indicators, the control amplitude parameters are iteratively optimized using the gradient descent algorithm to obtain the final control parameter set.

[0013] Compared with the prior art, the present invention has the following advantages and technical effects: This invention discloses an intelligent control method for distributed load units. By acquiring real-time power data and historical operating records, a time series analysis algorithm is used to extract features and identify patterns to establish a load prediction model. When the prediction deviation exceeds a threshold, a parameter adjustment mechanism corrects the prediction result. A distributed optimization algorithm is used to generate multi-period load dispatching schemes and send control commands to the load units. If there is a communication delay, a local backup command is triggered. This invention also monitors the execution of commands, calculates the deviation between the actual response and the expected target, and uses an incremental control algorithm to generate a dynamic adjustment curve. Finally, the load distribution and real-time curves are displayed through a visual interface, and the prediction model is iteratively updated based on the deviation rate. This method achieves accurate prediction and intelligent control of distributed load units, improving the flexibility and reliability of power grid dispatching. Attached Figure Description

[0014] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation

[0015] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0016] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0017] Example 1 like Figure 1 As shown, this embodiment provides a virtual power plant user-side load aggregation control method, including: Real-time power data and historical operation records of distributed load units are acquired, and time series analysis algorithms are used to extract features and recognize patterns from the real-time power data to obtain load change parameters and user behavior characteristic parameters. A load forecasting model is established based on load change parameters. If the deviation of the predicted load demand exceeds a preset threshold, a corrected load forecast demand is generated through a parameter adjustment mechanism. A distributed optimization algorithm is used to process the corrected load forecast demand and power grid dispatch parameters, generate multi-period load dispatch schemes, and calculate the control timing parameters of each distributed load unit. Control commands are sent to the distributed load units according to the control timing parameters. If the communication delay exceeds the preset time window, the locally stored backup command set is triggered to generate command execution record data. The monitoring command executes and records data and calculates the deviation between the actual response and the expected control target. If the deviation exceeds the preset range, load response correction parameters are generated. An incremental control algorithm is used to process load response correction parameters and system operating parameters, generate dynamic adjustment curves, and update the control amplitude parameters. A visualization interface is created based on the dynamic adjustment curve to generate load distribution maps and real-time load curve datasets. A model update mechanism is used to analyze the load distribution map and real-time load curve dataset. If the deviation rate exceeds the preset target, the parameter set of the load prediction model is iteratively updated to generate optimized load aggregation control parameters, which are then applied to the next round of load scheduling plan.

[0018] Furthermore, the process of obtaining load change parameters and user behavior characteristic parameters includes: The real-time power data and historical operation records of the distributed load units are obtained, and outliers and noise are removed through data cleaning to obtain a preprocessed power dataset. Time series analysis algorithms are used to extract features from the preprocessed power dataset, calculate statistical and periodic features, and obtain a load change feature set. Clustering algorithms are used to perform pattern recognition on the load change feature set to determine the load change trend and the distribution of user behavior patterns; If the fluctuation range of the load change trend exceeds the preset threshold, the abnormal data points are marked to obtain an abnormal load pattern set. Based on the abnormal load pattern set and the distribution of user behavior patterns, a decision tree algorithm is used for classification to obtain user behavior feature parameters; Based on load change trends and user behavior characteristics, a load change parameter set containing peak load and periodic change patterns is generated. For the load change parameter set, a time series forecasting model is used to predict the load change trend in future periods, and a predicted load dataset is obtained.

[0019] For example, during the real-time power data acquisition process of distributed load units, the system collects power data every 15 minutes through smart meters, including parameters such as active power and reactive power.

[0020] For example, the daily power consumption data of 100 households in a residential community shows a normal range of 0.5 kilowatts to 8 kilowatts, but the data contains obvious errors such as sudden anomalies of 50 kilowatts and negative power values. Statistical methods are used in the data cleaning process to identify outliers.

[0021] Specifically, outliers are eliminated by setting a 3x standard deviation rule, while a moving average filter is used to eliminate high-frequency noise.

[0022] In one embodiment, the original dataset contained 14,400 data points, and after cleaning, 13,850 valid data points were retained, resulting in a significant improvement in data quality. The statistical features extracted by the time series analysis algorithm include parameters such as mean, variance, kurtosis, and skewness.

[0023] For example, the average electricity consumption on weekdays is 3.2 kW, while on weekends it is 2.8 kW, showing a clear cyclical difference. Cyclical feature extraction identifies daily, weekly, and seasonal cycles, with the daily cycle showing a pattern of peak electricity consumption in the morning and evening, and a low consumption period at midday. The clustering algorithm uses the K-means method to divide users into three typical patterns. The first category is regular users, accounting for 45%, with stable electricity consumption patterns and significant peak-valley differences; the second category is random users, accounting for 35%, with irregular electricity consumption times; and the third category is high-energy-consuming users, accounting for 20%, with high base loads and frequent fluctuations. Abnormal load pattern identification sets the fluctuation threshold to 200% of the normal load.

[0024] For example, when a user's instantaneous power consumption suddenly increases from 2 kW to over 6 kW, the system automatically marks it as an abnormal mode. This type of abnormality typically corresponds to the use of high-power equipment such as air conditioners starting up or electric water heaters operating. The decision tree algorithm classifies user behavior based on the abnormal load pattern set, extracting key feature parameters such as equipment usage habits, response sensitivity, and load adjustment capabilities.

[0025] For example, highly responsive users will proactively adjust their electricity consumption behavior when electricity prices change, while less responsive users will maintain a fixed electricity consumption pattern. The load change parameter set integrates the peak load time distribution and periodic variation patterns. Specific data shows that peak load occurs between 7 PM and 9 PM in summer and between 6 PM and 8 PM in winter, with a peak-to-valley ratio of 3.5. The time series forecasting model uses an ARIMA model combined with a neural network method, achieving a prediction accuracy of over 92%.

[0026] In one possible implementation, the model successfully predicted that the peak electricity demand for the following day would occur 30 minutes earlier, providing valuable lead time for grid dispatching. This comprehensive analysis method can effectively identify user electricity consumption patterns, improve the accuracy of load forecasting, provide a scientific basis for grid optimization dispatching and demand response management, while reducing grid operating costs and improving power supply reliability.

[0027] Furthermore, if the deviation of the predicted load demand exceeds a preset threshold, the process of generating a corrected predicted load demand through a parameter adjustment mechanism includes: Based on historical load data and real-time load data, load change parameters are obtained to construct an initial load prediction model; The predicted load demand is obtained by processing the model training data through the initial load prediction model. If the deviation between the predicted load demand and the actual load data exceeds the preset deviation threshold, then prediction deviation analysis will be initiated to determine the source of the deviation. Based on the prediction deviation analysis results, optimize parameters are obtained from the set of adjustment parameters, and the parameter adjustment mechanism is updated. The initial load forecasting model is optimized through a parameter adjustment mechanism to generate a revised load demand. The revised load demand is evaluated by verifying the prediction results, and the model training data is updated by updating the revised load demand to optimize and obtain the target load prediction model.

[0028] In one embodiment, obtaining load variation parameters requires comprehensive analysis from multi-dimensional data sources. Historical load data typically includes power sampling points every 15 minutes over the past year, while real-time load data is obtained through instantaneous power values ​​uploaded every second by smart meters. By statistically analyzing this data, key parameters such as daily load peak-to-valley difference, monthly load growth rate, and seasonal fluctuation coefficient can be extracted. These parameters constitute the basic input variables of the initial load forecasting model.

[0029] Specifically, the initial load forecasting model employs a multi-layer neural network architecture, using historical load curves as training samples and external factors such as meteorological data and holiday information as auxiliary inputs. During model training, the system automatically learns the inherent patterns of load changes and establishes a non-linear mapping relationship between input features and load demand. When the model forecasts the next 24 hours, it outputs the expected load demand for each hour.

[0030] In one possible implementation, the prediction deviation analysis mechanism is triggered by real-time monitoring of the difference between the predicted and actual values. When the predicted load for a certain period is 850 kW and the actual load reaches 920 kW, the deviation rate exceeds a preset threshold of 8%, and the system immediately initiates the deviation analysis program. The analysis process examines possible sources of deviation, such as changes in weather conditions, abnormal user behavior, and equipment malfunctions, and quantifies the contribution of each factor to the prediction error.

[0031] For example, the parameter adjustment mechanism includes multiple dimensions such as learning rate adjustment, weight update, and feature importance reassessment. When it is found that the impact of temperature on the load is underestimated, the system selects the temperature sensitivity coefficient from the adjustment parameter set and increases it from 0.15 to 0.23, while reducing the weights of other minor factors. This dynamic adjustment enables the model to better adapt to the new characteristics of load changes.

[0032] It should be noted that generating the revised load demand is an iterative optimization process. The system integrates the original forecast value with the deviation correction amount to generate a more accurate load forecast result. The forecast result verification stage evaluates the reliability of the revised forecast through methods such as cross-validation and residual analysis to ensure that the forecast accuracy meets the actual needs of power grid dispatching.

[0033] In one embodiment, the model training data is updated using a sliding window mechanism, where new actual load data replaces the oldest historical data to maintain the timeliness of the training set. Simultaneously, the system dynamically adjusts the weight allocation of the training data based on changes in prediction accuracy, enabling the model to continuously learn the latest patterns of load changes. This adaptive optimization mechanism significantly improves the long-term stability and accuracy of load forecasting, providing reliable decision support for the safe and economical operation of the power grid.

[0034] Furthermore, the process of using a distributed optimization algorithm to process the corrected load forecast demand and grid dispatch parameters, generating multi-period load dispatch schemes, and calculating the control timing parameters of each distributed load unit includes: Obtain the corrected load forecast demand and power grid dispatch parameters, and generate a standardized input dataset through data preprocessing; A distributed optimization algorithm is used to process the standardized input dataset to obtain load allocation results for multiple time periods; The load allocation results for multiple time periods are used to make a judgment. If the allocation results meet the preset power grid stability threshold, a preliminary load dispatching plan is generated. The preliminary load scheduling scheme is analyzed, and the control timing of each unit is optimized using a linear programming algorithm to obtain the control timing parameters. If there is a deviation between the control timing parameters and the power grid dispatch parameters, the updated load dispatch scheme can be obtained by iteratively adjusting the parameters of the distributed optimization algorithm. Based on the updated load scheduling scheme, time series analysis algorithm is used to verify the stability of multi-period load allocation and determine the final load scheduling scheme. The target control timing parameters for each unit are generated based on the final load scheduling scheme.

[0035] For example, when obtaining the corrected load forecast demand and grid dispatch parameters, the data preprocessing stage needs to standardize the multi-source heterogeneous data.

[0036] For example, the revised load forecast is 512 MW, and the grid dispatch parameters include an upper limit of 800 MW and a lower limit of 200 MW for generator output, and a line transmission capacity limit of 600 MW. Data preprocessing uses the Z-score normalization method to convert the load data into a standard distribution with a mean of 0 and a standard deviation of 1. At the same time, the dispatch parameters are normalized to ensure the comparability of data with different dimensions.

[0037] In one embodiment, the distributed optimization algorithm employs a hybrid strategy combining particle swarm optimization and genetic algorithms to process the standardized dataset. The algorithm divides 24 hours into 6 time periods, each lasting 4 hours, and calculates the optimal load allocation scheme for each time period.

[0038] For example, during the morning peak period from 6:00 to 10:00, the load demand is 2400 MW. The algorithm allocates this demand to three generator sets, which will handle 800 MW, 900 MW, and 700 MW of output respectively. Distributed computing nodes process the optimization problem in parallel for each time period, and the convergence of the global optimal solution is ensured through an information exchange mechanism.

[0039] For example, the grid stability thresholds are set as follows: frequency deviation not exceeding 0.2 Hz and voltage deviation not exceeding 5%. When the load allocation results across multiple time periods meet these constraints, the system generates a preliminary load dispatching plan. This plan includes key information such as the start-up and shutdown times, output curves, and reserve capacity configurations of each generator unit.

[0040] Specifically, when optimizing the timing of control using linear programming algorithms, the objective function is to minimize the system operating cost, and the constraints include power balance, unit ramp-up rate limits, and minimum operating time.

[0041] For example, a coal-fired power unit has a ramp rate of 5 megawatts per minute, and it requires 40 minutes of adjustment time to increase from 400 megawatts to 600 megawatts. Based on this, the algorithm determines the control timing parameters for each unit, including the adjustment start time, adjustment amplitude, and adjustment duration.

[0042] It should be noted that when there is a deviation between the control timing parameters and the power grid dispatch parameters, the system will initiate an iterative adjustment mechanism.

[0043] For example, the calculated unit start-up time is 5:30, but the grid dispatch requirement is 5:45, a deviation of 15 minutes. In this case, by adjusting the weight coefficients and constraint relaxation factors in the distributed optimization algorithm, the load allocation scheme is recalculated, so that the control timing better matches the grid dispatch requirements.

[0044] In one possible implementation, the time series analysis algorithm uses an autoregressive moving average model to verify the stability of load distribution across multiple time periods. The algorithm analyzes the smoothness of load changes between consecutive time periods, detecting any drastic fluctuations or unreasonable jumps.

[0045] For example, if the load variation exceeds 100 MWh between adjacent time periods, the system is deemed to be in an unstable state, requiring re-optimization of the load allocation scheme. Once the final load dispatch scheme is determined, the system generates detailed control timing parameters for each unit. These parameters include adjustment commands accurate to the minute, ensuring the safety and economy of grid operation while improving the accuracy of load forecasting and the reliability of dispatch decisions.

[0046] Furthermore, the process of generating instruction execution record data includes: If the timing parameters are adjusted, the communication delay data is obtained to determine whether the communication delay exceeds the preset time window. If the communication delay exceeds the preset time window, a backup instruction set is retrieved from local storage to determine the activation status of the triggering mechanism. Based on the activation status of the triggering mechanism, control commands from the standby instruction set are sent to the load unit to obtain preliminary results of command execution. Based on the preliminary results of command execution, obtain the response data of the load unit and determine whether the response data matches the control timing parameters; If the response data matches the control timing parameters, the instruction execution data is structured to obtain formatted execution data. Detailed instruction records are generated based on the formatted execution data, and the support vector machine algorithm is used to classify and analyze the instruction records to identify abnormal situations in the execution data and obtain the classification results.

[0047] In one embodiment, this embodiment uses network detection technology to monitor the communication delay between the power grid dispatching system and each load unit in real time.

[0048] Specifically, the system sets a preset time window of 200 milliseconds. When the communication delay of a load unit in a substation reaches 350 milliseconds, exceeding the preset threshold, the backup mechanism is activated. At this time, the system retrieves pre-configured emergency control strategies from the locally stored backup instruction set, including multiple alternatives such as load reduction instructions, power regulation instructions, and equipment protection instructions.

[0049] For example, when the main dispatch center sends a power adjustment command to a load unit in an industrial park, if the communication delay exceeds the limit, the triggering mechanism will automatically activate the local backup command set. This command set includes a three-level response strategy: Level 1 maintains the current output status, Level 2 reduces the load to 80% of the rated power according to a preset ratio, and Level 3 urgently cuts off non-critical loads. The system automatically selects and executes the appropriate backup command based on the current power grid operating status and the importance level of the load.

[0050] In one possible implementation, after receiving the standby control command, the load unit parses the command and operates the equipment through its built-in execution module. Taking a load unit in a commercial complex as an example, upon receiving a secondary response command, the system automatically adjusts the air conditioning load to a preset power level and simultaneously shuts off some lighting equipment. The actual power output drops from 1200 kW to 960 kW, with a response time of 15 seconds. After execution, the load unit sends response data, including actual power data, equipment status information, and execution timestamp, back to the dispatch center.

[0051] It should be noted that the conformity assessment between the response data and the control timing parameters is a crucial step in ensuring the accuracy of command execution. The system compares the deviation between the actual executed power and the target power, and determines that the execution is successful when the deviation is less than 5%. The data generation module then standardizes the raw execution data, encapsulating information such as power values, timestamps, and device numbers in a unified format to generate structured data containing fields such as execution status code, response time, and device health status.

[0052] For example, instruction logging includes generating detailed execution logs that record complete information such as instruction type, issuance time, execution duration, response accuracy, and equipment status changes. These records are not only used for system operation monitoring but also provide data support for subsequent scheduling optimization. By establishing a complete instruction execution archive, the system can trace the entire process of each control operation, ensuring the traceability and reliability of power grid dispatching.

[0053] In one embodiment, the support vector machine algorithm establishes a classification boundary between normal and abnormal execution modes by training on historical execution data. The algorithm extracts feature parameters such as execution duration, power deviation, and communication quality, classifying the execution data into three categories: normal, slightly abnormal, and severely abnormal. When an anomaly is detected in the execution data, the system automatically marks the anomaly type and triggers the corresponding processing mechanism, providing crucial protection for the safe and stable operation of the power grid.

[0054] Furthermore, the process of monitoring command execution, recording data, and calculating the deviation between the actual response and the expected control target, and generating load response correction parameters if the deviation exceeds the preset range, includes: Obtain instruction execution record data and use time series analysis methods to obtain the feature values ​​of the actual response data; The deviation value is calculated by comparing the characteristic value with the preset control target; If the deviation exceeds the preset threshold range, a linear regression algorithm is used to generate load response correction parameters; Based on the load response correction parameters, the input of the instruction execution record data is adjusted to obtain the updated instruction sequence; By using the updated instruction sequence, the deviation between the actual response data and the preset control target is recalculated to determine whether the deviation is still within the preset threshold range. If the deviation value still exceeds the preset threshold range, the gradient descent algorithm is used to optimize the load response correction parameters to obtain the optimized correction parameters; Based on the optimized correction parameters, the instruction execution record data is adjusted to generate the final load response control sequence.

[0055] Furthermore, when acquiring instruction execution record data, the system collects actual power change curves from load units and extracts key feature values ​​through time series analysis.

[0056] For example, when the actual power of a load unit drops from 1200 kW to 900 kW during the regulation process, the system uses an autoregressive moving average model to analyze the power change trend and extract characteristic parameters such as response delay time, steady-state error, and overshoot.

[0057] Specifically, the response delay is 15 seconds, the steady-state error is 20 kilowatts, and the overshoot is 5%.

[0058] In one embodiment, the system compares and calculates the extracted feature values ​​with a preset control target.

[0059] The preset control targets require a response delay of no more than 10 seconds, a steady-state error of no more than 10 kilowatts, and an overshoot of no more than 3%.

[0060] The deviation values ​​obtained by the difference calculation were 5 seconds, 10 kilowatts, and 2%.

[0061] When the deviation exceeds the preset threshold range, the system determines that the current control parameter needs to be corrected.

[0062] For example, the system uses a linear regression algorithm to analyze the relationship between historical control data and actual response effects, and establishes a load response correction parameter model.

[0063] Analysis of 100 historical data samples revealed a linear correlation between power adjustment rate and response delay, with a correlation coefficient of 0.85.

[0064] Based on this relationship, the system generates correction parameters, adjusting the original power adjustment rate from 50 kilowatts per second to 65 kilowatts per second, while shortening the control time window from 20 seconds to 16 seconds.

[0065] In one possible implementation, the system updates the instruction sequence based on the generated correction parameters and recalculates the deviation between the actual response data and the preset target.

[0066] The updated instruction sequence includes a new power setpoint sequence and timing schedule.

[0067] The system simulation verification showed that the corrected response delay time was reduced to 12 seconds and the steady-state error was reduced to 15 kilowatts, but it still exceeded the preset threshold range.

[0068] It should be noted that when the linear regression correction effect is not ideal, the system uses the gradient descent algorithm for deep optimization.

[0069] The algorithm finds the optimal combination of correction parameters through iterative calculation, with the learning rate set to 0.01 and the number of iterations limited to 200.

[0070] During the optimization process, the system simultaneously adjusts three key parameters: power adjustment rate, time window, and control gain.

[0071] After 85 iterations, the algorithm converged, yielding an optimized combination of corrected parameters.

[0072] Specifically, the optimized correction parameters adjust the power adjustment rate to 72 kilowatts per second, the time window to 14 seconds, and the control gain coefficient to 1.2.

[0073] The system regenerates the load response control sequence based on these optimized parameters, including precise power setpoints and execution timing arrangements.

[0074] The final generated control sequence can control the response delay time to within 9 seconds, reduce the steady-state error to 8 kilowatts, and control the overshoot to within 2.5%, all of which meet the preset control target requirements.

[0075] Furthermore, the process of using an incremental control algorithm to process load response correction parameters and system operating parameters, generating dynamic adjustment curves, and updating control amplitude parameters includes: Obtain load response correction parameters and system operating parameters, and extract real-time data from the load management system and operating status database to obtain the initial parameter set; An incremental control algorithm is used to process the initial parameter set. If the load response correction parameter exceeds the preset threshold, the parameter is normalized to obtain a standardized parameter set. Based on the standardized parameter set and combined with the system operating parameters, the key points of the adjustment curve are calculated through the dynamic curve generation model to obtain the dynamic adjustment curve. Analyze the changing trend of the dynamic adjustment curve. If the changing trend deviates from the preset system stability range, adjust the curve parameters through a secondary optimization algorithm to obtain an optimized adjustment curve. The control amplitude parameter is extracted from the optimized control curve, and the system control amplitude is updated using the parameter mapping method to obtain the updated control amplitude parameter. The updated system operation status data is obtained and compared with the load response data to obtain a dataset of control effects. Based on the control effect dataset, if the system operation status does not reach the preset performance index, the control amplitude parameters are iteratively optimized using the gradient descent algorithm to obtain the final control parameter set.

[0076] In one embodiment, when obtaining real-time parameters from the load management system, the system first establishes a connection channel with the distributed data source.

[0077] Specifically, when the system detects that the load fluctuation in a certain area reaches 50 megawatts, it simultaneously extracts operating parameters from the power grid dispatch center, substation monitoring system and user-side smart terminal, including key indicators such as voltage amplitude, frequency deviation and power factor, to form an initial parameter set containing 15 dimensions.

[0078] For example, the incremental control algorithm employs a hierarchical, progressive strategy when processing the initial parameter set. When the load response correction parameter is 25 MW and exceeds the preset threshold of 20 MW, the system initiates a normalization process. This normalization process maps each parameter to a standard range of 0 to 1; voltage parameters are normalized from 220 kV to 0.85, and frequency deviations are normalized from 0.2 Hz to 0.4, ensuring comparability of parameters with different dimensions.

[0079] In one possible implementation, the dynamic curve generation model calculates key points of the adjustment curve based on a standardized parameter set. The system employs cubic spline interpolation, setting 48 control nodes on a 24-hour time axis, each node corresponding to a 30-minute adjustment cycle. Based on load forecast data and historical operating patterns, the model generates a dynamic adjustment curve that includes peak adjustment points, stable operating points, and valley recovery points.

[0080] Specifically, the system stability range is set at a standard where load fluctuations do not exceed ±5%. When the dynamic adjustment curve shows that the load fluctuation reaches 8% during a certain period, the secondary optimization algorithm automatically intervenes to adjust. The optimization process controls the fluctuation amplitude within 4% by adjusting the curve slope and inflection point position, generating a smoother optimized adjustment curve.

[0081] For example, the parameter mapping method converts the optimized regulation curve into specific regulation amplitude parameters. The system extracts key parameters from the curve, such as the peak regulation amplitude of 35 MW, the duration of 45 minutes, and the regulation rate of 0.8 MW per minute, and updates the system regulation setpoint through a linear mapping function.

[0082] In one embodiment, monitoring data showed that the load deviation decreased from 15 MW to 3 MW after regulation, but the system frequency stability index still did not meet the preset 99.5% reliability requirement. At this point, the gradient descent algorithm started iterative optimization, gradually adjusting the regulation amplitude parameters by calculating the gradient direction of the objective function.

[0083] It should be noted that the iterative optimization process employs an adaptive step-size strategy. The initial step size is set to 0.1. When the system performance improvement is less than 1% after three consecutive iterations, the algorithm automatically adjusts the step size to 0.05 to ensure convergence accuracy. After eight iterations, the system obtains the final set of control parameters, achieving load deviation control within 2 MW and frequency stability of 99.8%. This multi-level parameter optimization method can significantly improve the control accuracy of power systems, reduce the need for manual intervention, and ensure that the system maintains stable and reliable regulation capabilities under complex operating conditions.

[0084] Furthermore, the process of establishing a visualization interface based on the dynamic adjustment curve and generating a load distribution map and a real-time load curve dataset includes: Based on the dynamic adjustment curve data, the load distribution characteristics are calculated using an interpolation algorithm to obtain the load distribution dataset; Based on the load distribution dataset, distribution features are extracted, and a real-time load curve is generated using a line graph rendering method to obtain curve graphic data. If the curve graph data meets the preset display threshold, the load distribution map is rendered through the visualization interface framework to obtain the interface display data. Based on the data displayed on the interface, a dynamic update algorithm is used to adjust the visualization interface to obtain a real-time updated load distribution map; By acquiring user interaction data through real-time updated load distribution maps, supplementary real-time load curve datasets are generated.

[0085] In one embodiment, the data source needs to collect raw data in real time from the power grid dispatching system, smart meter network and load monitoring equipment during the process of acquiring load data and curve data.

[0086] For example, a power grid in a certain area collects power data from 1,000 load monitoring points per second, including parameters such as active power, reactive power, and voltage amplitude, with the raw data volume reaching 3.6 GB per hour.

[0087] When cleaning these data, the preprocessing method first identifies outliers and missing values, uses the 3σ criterion to remove data points that deviate from the mean by more than three times the standard deviation, and then uses linear interpolation to fill in the missing data.

[0088] In one embodiment, the integrity threshold of the standardized workload dataset is set to 95%, meaning that the subsequent processing will only begin when the percentage of valid data exceeds 95%.

[0089] The time series segmentation uses a sliding window technique to divide the continuous load data into segments with 15-minute intervals. Each time window contains 900 sampling points, forming the basic data structure for the dynamic adjustment curve.

[0090] This segmentation method can capture the periodic characteristics and sudden fluctuations of load changes.

[0091] Specifically, the interpolation algorithm uses cubic spline interpolation to calculate the load distribution characteristics, and constructs a piecewise cubic polynomial function to smoothly connect the various data points.

[0092] For example, when load data changes drastically within a certain period, the interpolation algorithm can generate a smooth transition curve, avoiding analysis errors caused by sudden data changes.

[0093] The load distribution dataset contains key characteristic parameters such as peak load, valley load, average load, and load factor.

[0094] For example, when generating real-time load curves using line chart rendering methods, adaptive axis scaling technology is employed to automatically adjust the vertical axis scale based on the dynamic range of the load data.

[0095] When the peak load is 85MW and the valley load is 32MW, the system automatically sets the vertical axis range to 0-100MW to ensure the clarity and readability of the curve display.

[0096] The display threshold for curve graph data is set based on data density and refresh rate. Data sampling display is triggered when the data point density exceeds 2 points per pixel.

[0097] In one possible implementation, the visual interface framework adopts a layered rendering architecture, with the bottom layer being a data processing layer responsible for real-time data updates, the middle layer being a graphics rendering layer that handles curve drawing, and the top layer being an interactive control layer that responds to user operations.

[0098] The interface displays data including curve coordinates, color maps, and interactive hotspot definitions, allowing users to view specific values ​​by hovering the mouse over the data.

[0099] For example, the dynamic update algorithm uses an incremental rendering strategy, redrawing only the data areas that have changed, thus avoiding the performance loss caused by full-screen refresh.

[0100] When new load data arrives, the algorithm calculates the magnitude of the data change. If the change exceeds a preset threshold of 0.5MW, a local update is triggered to maintain the real-time responsiveness of the interface.

[0101] User interaction data includes records of actions such as zooming, time range selection, and data point queries. This interaction information is fed back to the data processing module to optimize subsequent data display strategies and improve user experience.

[0102] Furthermore, the process of iteratively updating the parameter set of the load prediction model by analyzing the load distribution map and real-time load curve dataset using a model update mechanism, if the deviation rate exceeds a preset target, includes: Obtain load distribution maps and real-time load curve datasets, and obtain standardized datasets through data preprocessing.

[0103] For the standardized dataset, the deviation rate between the predicted output and the real-time load curve is calculated using analytical processing.

[0104] If the deviation rate exceeds the preset threshold, the gradient descent algorithm is used to update the parameter set of the load prediction model.

[0105] Based on the updated parameter set, the prediction output is regenerated.

[0106] The convergence of the model is determined by comparing the deviation rate between the new forecast output and the real-time load curve.

[0107] If the model does not converge, iteratively update the parameter set and generate the predicted output.

[0108] Based on the final forecast output, determine the set of optimized parameters for the load forecasting model.

[0109] In one embodiment, the system first collects load distribution data of an industrial park over the past 30 days, including hourly power density distribution matrices. Peak load areas are concentrated between 9:00 AM and 11:00 AM and between 2:00 PM and 4:00 PM, with power densities of 85.6 MW / km². 2 and 78.3MW / km 2 .

[0110] Simultaneously, a real-time load curve dataset was acquired, recording load changes at 15-minute intervals. It was found that the average load on weekdays was 156.8MW, and the average load on weekends was 98.4MW.

[0111] Based on this historical data, the system constructs an LSTM neural network load prediction model, setting the number of neurons in the input layer to 48, corresponding to the load data of the past 12 hours, the hidden layer to contain 128 LSTM units, and the output layer to predict the load value of the next 4 hours.

[0112] The initial parameters of the model include a learning rate of 0.001, a weight matrix initialized using the Xavier method, and a bias term of 0.1.

[0113] After training, the system used the model to predict the load for the next day. The predicted value at 10:00 AM was 162.4 MW, while the actual monitored value was 178.9 MW. The calculated deviation rate was 10.2%, which exceeded the preset target threshold of 8%.

[0114] After the model update mechanism is triggered, the system automatically adjusts the learning rate to 0.0008, increases the regularization coefficient to 0.05, introduces attention mechanism weight parameters, and retrains the model parameters through the gradient descent algorithm.

[0115] After 50 iterations, the new model's prediction bias for the same time period dropped to 6.8%, meeting the accuracy requirements. The system then saved the updated parameter set and deployed it to the production environment to continue running.

[0116] Example 2 This embodiment also discloses a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in Embodiment 1.

[0117] Example 3 This embodiment also discloses a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0118] Example 4 This embodiment also discloses a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0119] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A virtual power plant user side load aggregation control method, characterized in that, include: Real-time power data and historical operation records of distributed load units are acquired, and time series analysis algorithms are used to extract features and recognize patterns from the real-time power data to obtain load change parameters and user behavior characteristic parameters. A load forecasting model is established based on the load change parameters. If the deviation of the predicted load demand exceeds a preset threshold, a corrected load forecast demand is generated through a parameter adjustment mechanism. The corrected load forecast demand and power grid dispatch parameters are processed using a distributed optimization algorithm to generate a multi-period load dispatch scheme and calculate the control timing parameters of each distributed load unit. Control commands are sent to the distributed load units according to the control timing parameters. If the communication delay exceeds the preset time window, the locally stored backup command set is triggered to generate command execution record data. The system monitors the execution data of the commands and calculates the deviation between the actual response and the expected control target. If the deviation exceeds the preset range, load response correction parameters are generated. An incremental control algorithm is used to process the load response correction parameters and system operating parameters, generate a dynamic adjustment curve, and update the control amplitude parameters. A visualization interface is established based on the dynamic adjustment curve to generate a load distribution map and a real-time load curve dataset. The load distribution map and real-time load curve dataset are analyzed using a model update mechanism. If the deviation rate exceeds the preset target, the parameter set of the load prediction model is iteratively updated to generate optimized load aggregation control parameters, which are then applied to the next round of load scheduling plan.

2. The method according to claim 1, characterized in that, The process of obtaining load change parameters and user behavior characteristic parameters includes: The real-time power data and historical operation records of the distributed load units are obtained, and outliers and noise are removed through data cleaning to obtain a preprocessed power dataset. Time series analysis algorithms are used to extract features from the preprocessed power dataset, calculate statistical and periodic features, and obtain a load change feature set. Clustering algorithms are used to perform pattern recognition on the load change feature set to determine the load change trend and the distribution of user behavior patterns; If the fluctuation range of the load change trend exceeds the preset threshold, the abnormal data points are marked to obtain an abnormal load pattern set. Based on the abnormal load pattern set and the distribution of user behavior patterns, a decision tree algorithm is used for classification to obtain user behavior feature parameters; Based on load change trends and user behavior characteristics, a load change parameter set containing peak load and periodic change patterns is generated. For the load change parameter set, a time series forecasting model is used to predict the load change trend in future periods, and a predicted load dataset is obtained.

3. The method according to claim 1, characterized in that, If the deviation of the predicted load demand exceeds a preset threshold, the process of generating a corrected predicted load demand through a parameter adjustment mechanism includes: Based on historical and real-time load data, load change parameters are obtained to construct an initial load prediction model. The predicted load demand is obtained by processing the model training data through the initial load prediction model. If the deviation between the predicted load demand and the actual load data exceeds a preset deviation threshold, then prediction deviation analysis is initiated to determine the source of the deviation. Based on the prediction deviation analysis results, optimize parameters are obtained from the set of adjustment parameters, and the parameter adjustment mechanism is updated. The initial load forecasting model is optimized through a parameter adjustment mechanism to generate a revised load demand. The revised load demand is evaluated by verifying the prediction results, and the model training data is updated by updating the revised load demand to optimize and obtain the target load prediction model.

4. The method according to claim 1, characterized in that, The process of using a distributed optimization algorithm to process the corrected load forecast demand and grid dispatch parameters, generating multi-period load dispatch schemes, and calculating the control timing parameters of each distributed load unit includes: Obtain the corrected load forecast demand and power grid dispatch parameters, and generate a standardized input dataset through data preprocessing; The standardized input dataset is processed using a distributed optimization algorithm to obtain multi-period load allocation results. Based on the multi-period load allocation results, if the allocation results meet the preset power grid stability threshold, a preliminary load dispatching scheme is generated. The preliminary load scheduling scheme is analyzed, and the control timing of each unit is optimized using a linear programming algorithm to obtain the control timing parameters. If there is a deviation between the control timing parameters and the power grid dispatch parameters, the updated load dispatch scheme is obtained by iteratively adjusting the parameters of the distributed optimization algorithm. Based on the updated load scheduling scheme, a time series analysis algorithm is used to verify the stability of multi-period load allocation and determine the final load scheduling scheme. The target control timing parameters for each unit are generated based on the final load scheduling scheme.

5. The method according to claim 1, characterized in that, The process of generating instruction execution record data includes: If the timing parameters are adjusted, the communication delay data is obtained to determine whether the communication delay exceeds the preset time window. If the communication delay exceeds the preset time window, a backup instruction set is retrieved from local storage to determine the activation status of the triggering mechanism. Based on the activation state of the triggering mechanism, control commands from the standby command set are sent to the load unit to obtain preliminary results of command execution. Based on the preliminary results of the command execution, obtain the response data of the load unit and determine whether the response data matches the control timing parameters; If the response data matches the control timing parameters, the instruction execution data is structured to obtain formatted execution data. Detailed instruction records are generated based on the formatted execution data, and the support vector machine algorithm is used to classify and analyze the instruction records to identify abnormal situations in the execution data and obtain the classification results.

6. The method according to claim 1, characterized in that, The process of monitoring the execution data of the command and calculating the deviation between the actual response and the expected control target, and generating load response correction parameters if the deviation exceeds a preset range, includes: Obtain instruction execution record data and use time series analysis methods to obtain the feature values ​​of the actual response data; The deviation value is calculated by comparing the characteristic value with the preset control target; If the deviation value exceeds the preset threshold range, a linear regression algorithm is used to generate load response correction parameters; Based on the load response correction parameters, the input of the instruction execution record data is adjusted to obtain the updated instruction sequence; By using the updated instruction sequence, the deviation between the actual response data and the preset control target is recalculated to determine whether the deviation is still within the preset threshold range. If the deviation value still exceeds the preset threshold range, the gradient descent algorithm is used to optimize the load response correction parameters to obtain the optimized correction parameters; Based on the optimized correction parameters, the instruction execution record data is adjusted to generate the final load response control sequence.

7. The method according to claim 1, characterized in that, The process of using an incremental control algorithm to process the load response correction parameters and system operating parameters, generating a dynamic adjustment curve, and updating the control amplitude parameters includes: Obtain load response correction parameters and system operating parameters, and extract real-time data from the load management system and operating status database to obtain the initial parameter set; The initial parameter set is processed using an incremental control algorithm. If the load response correction parameter exceeds a preset threshold, the parameter is normalized to obtain a standardized parameter set. Based on the standardized parameter set and combined with the system operating parameters, the key points of the adjustment curve are calculated through the dynamic curve generation model to obtain the dynamic adjustment curve. Analyze the changing trend of the dynamic adjustment curve. If the changing trend deviates from the preset system stability range, adjust the curve parameters through a secondary optimization algorithm to obtain an optimized adjustment curve. The control amplitude parameter is extracted from the optimized control curve, and the system control amplitude is updated using a parameter mapping method to obtain the updated control amplitude parameter. The updated system operation status data is obtained and compared with the load response data to obtain a dataset of control effects. Based on the aforementioned control effect dataset, if the system's operating status does not meet the preset performance indicators, the control amplitude parameters are iteratively optimized using the gradient descent algorithm to obtain the final control parameter set.

8. A computer apparatus comprising a memory, a processor, and a computer program stored on the memory, wherein the computer program, when executed by the processor, causes the processor to perform the method of any one of claims 1 to 7. The processor executes the computer program to implement the steps of the method according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1-7.

Images