Virtual power plant distributed energy aggregation and prediction method

Abstract

The invention discloses a virtual power plant distributed energy aggregation and prediction method, which comprises the steps of establishing a dynamic aggregation weight distribution mechanism for a high-priority resource combination, and dynamically adjusting the weight coefficient of each resource in an aggregation process according to a real-time meteorological condition and a load demand prediction result, calculating through a weighted fusion algorithm to obtain an overall output prediction value of the virtual power plant, and generating a corresponding uncertainty interval range at the same time; and recalculating overall operation state evaluation indexes of the virtual power plant through the corrected aggregation mechanism parameters, including key performance parameters such as prediction precision, adjustment capability and economic benefits, establishing an operation state database to record historical optimization results, and forming a continuously improved intelligent distributed energy aggregation and prediction system.

Description

Technical Field

[0001] This invention belongs to the field of power integration technology, and in particular relates to a method for the aggregation and prediction of distributed energy resources in a virtual power plant. Background Technology

[0002] With the deepening of the global energy transition, virtual power plants, as a key technology platform for integrating distributed energy resources, have become a core support for building new power systems. By unifying the scheduling and management of various distributed energy resources such as wind power, photovoltaics, energy storage, and controllable loads, virtual power plants can significantly improve the flexibility and stability of the power grid, and their technological development level directly affects the overall operating efficiency of the energy system.

[0003] Current distributed energy aggregation methods mainly employ simple superposition or averaging, lacking in-depth consideration of the differences in characteristics among various energy resources. These traditional methods often overlook the inherent correlations between different types of energy resources, resulting in aggregation results that fail to fully reflect the actual operating state of the system and hinder the realization of the synergistic advantages of distributed energy.

[0004] In the actual operation of virtual power plants, different types of distributed energy resources have drastically different output characteristics and response patterns. These differences in characteristics lead to complex interactions during the aggregation process. Due to the lack of effective methods for identifying these interaction mechanisms, existing technologies struggle to accurately capture key information such as the complementary output patterns between wind and solar power, and the coordinated operation patterns between energy storage and controllable loads. This ambiguity in the interaction mechanisms further results in a lack of aggregation optimization strategies, preventing the system from intelligently adjusting resource allocation based on the synergistic effects of different energy resources. When the aggregation process lacks targeted optimization strategies, the entire virtual power plant system cannot achieve the optimal combination of various distributed energy resources, severely restricting the system's economic efficiency and reliability. Summary of the Invention

[0005] This invention proposes a method for distributed energy aggregation and prediction in virtual power plants to address the problems existing in the prior art.

[0006] To achieve the above objectives, the present invention provides a method for virtual power plant distributed energy aggregation and prediction, comprising the following steps:

[0007] Historical operational data of distributed energy resources are acquired, and the historical operational data are processed using multidimensional feature extraction algorithms and time series analysis methods to obtain a characteristic difference matrix;

[0008] Based on the characteristic difference matrix, the time-series complementarity coefficient is calculated for the correlation degree between resources, and the intra-group complementarity strength value is obtained by using the Pearson correlation coefficient method. Based on the intra-group complementarity strength value, the high-priority resource combination is determined.

[0009] A dynamic aggregation weight allocation mechanism is established for high-priority resource combinations. The weight coefficients of each resource in the aggregation process are dynamically adjusted according to real-time meteorological conditions and load demand forecast results. The overall output forecast value of the virtual power plant is calculated through a weighted fusion algorithm, and the corresponding uncertainty range is generated.

[0010] After obtaining the overall output prediction value, a multi-objective optimization model is established with the objective functions of maximizing economic efficiency and optimizing operational stability. The optimal resource allocation scheme is obtained by using the particle swarm optimization algorithm.

[0011] Real-time dispatch instructions are formulated based on the optimal resource allocation scheme, and the virtual power plant operation status is always met by the grid dispatching requirements through a closed-loop feedback control mechanism.

[0012] A sliding time window method is used to continuously monitor the deviation between the actual operating data and the predicted values ​​of each resource. When the cumulative deviation exceeds the preset error threshold, the aggregation parameters are automatically corrected, and the corrected aggregation mechanism parameters are used for the next round of prediction calculation.

[0013] The overall operating status evaluation index of the virtual power plant is recalculated using the modified aggregation mechanism parameters. An operating status database is established to record historical optimization results, resulting in a continuously improving intelligent distributed energy aggregation and prediction system.

[0014] Optionally, obtaining the characteristic difference matrix includes:

[0015] Historical operational data of distributed energy resources are acquired, and a multi-dimensional feature extraction algorithm is used to extract power characteristics from the historical operational data to obtain an initial characteristic dataset; the historical operational data includes wind power, photovoltaic, energy storage, and controllable loads;

[0016] The initial characteristic dataset is processed by time series analysis to extract power fluctuation amplitude, response time constant and regulation range, and obtain standardized output characteristic parameters.

[0017] The output characteristic parameters are normalized using a standardization method to eliminate dimensional differences and obtain a standardized characteristic dataset.

[0018] Based on the standardized characteristic dataset, a characteristic difference matrix is ​​constructed, and the output characteristic differences among distributed energy resources are calculated to obtain the characteristic difference matrix.

[0019] Optionally, determining the high-priority resource combination includes:

[0020] Based on the characteristic difference matrix, the key feature vectors of each distributed energy resource are extracted by principal component analysis to obtain the dimensionality-reduced feature dataset.

[0021] Based on the dimensionality-reduced feature dataset, wind power, photovoltaic power, energy storage and controllable load are grouped using the k-means clustering algorithm to obtain resource collaborative clustering results;

[0022] Based on the resource collaborative clustering results, the temporal complementarity coefficient of resources within each group is calculated, and the Pearson correlation coefficient method is used to obtain the complementarity strength value within the group;

[0023] If the complementary strength value within the group exceeds the preset threshold, the coordination strength between energy storage and controllable load within the group is calculated using the grey relational analysis method to obtain the coordination coefficient within the group.

[0024] Based on the coordination coefficient within the group, a weighted average method is used to integrate the complementary strength value and the coordination coefficient within the group to obtain the comprehensive synergy score for each group.

[0025] By comprehensively evaluating collaboration scores, the collaboration priorities of each group are ranked, and high-priority resource combinations are determined.

[0026] Optionally, the calculation to obtain the overall power output prediction value of the virtual power plant and generate the corresponding uncertainty interval range includes:

[0027] Based on real-time meteorological conditions, wind power and photovoltaic power output data are obtained, and time series analysis is used to obtain the output change trend of each resource.

[0028] If the output change trend exceeds the preset fluctuation threshold, short-term feature data is extracted using the sliding window method to obtain the resource output feature vector;

[0029] Based on the resource output feature vector, the support vector regression algorithm is used to predict the future output value of each resource, and the single resource output prediction result is obtained.

[0030] By using the load demand forecast results, the real-time load demand curve of the virtual power plant is obtained, and the range of load demand fluctuations is determined.

[0031] If the deviation between the load demand fluctuation range and the single resource output prediction result exceeds the preset threshold, the linear interpolation method is used to adjust the dynamic weight coefficient of each resource to obtain the optimized weight allocation scheme.

[0032] Based on the optimized weight allocation scheme, a weighted fusion algorithm is used to calculate the overall power output prediction value of the virtual power plant, and obtain the comprehensive power output prediction result.

[0033] Optionally, obtaining the optimal resource allocation scheme includes:

[0034] After obtaining the overall power output prediction value, a multi-objective optimization model is constructed. The objective function is defined as maximizing economic efficiency and optimizing operational stability. The constraints are set as the technical parameter limits of each resource and the requirements for safe operation of the power grid. The particle swarm optimization algorithm is used to calculate the optimal power output allocation ratio and resource configuration scheme of each resource.

[0035] Based on the optimal output allocation ratio of each resource, real-time output data of each resource is obtained. Through time series analysis, the output change trend of each resource is extracted to determine whether the output of each resource meets the technical parameter limits.

[0036] If the resource output exceeds the technical parameter limit, the output allocation ratio is adjusted by linear interpolation to obtain a corrected output allocation scheme.

[0037] Based on the revised power allocation scheme, real-time operating status data of the virtual power plant is obtained. The uncertainty of each resource power allocation scheme is analyzed using the Monte Carlo simulation method to determine the range of uncertainty in the operating status.

[0038] Based on the uncertainty range of the operating status, real-time load demand data of the power grid is obtained to determine whether the load demand fluctuation exceeds the preset threshold.

[0039] If the load demand fluctuation exceeds the preset threshold, the output allocation ratio of each resource is adjusted by weighted average method to obtain an optimized resource scheduling scheme.

[0040] Based on the optimized resource scheduling scheme, real-time power output monitoring data of each resource is obtained. Short-term power output feature vectors are extracted using the sliding window method to determine whether the power output of each resource meets the requirements for safe operation of the power grid.

[0041] If the output of resources does not meet the requirements for safe operation of the power grid, the allocation ratio of each resource output is recalculated using the quadratic programming method to obtain the adjusted scheduling scheme.

[0042] Based on the adjusted scheduling plan, obtain comprehensive operation data of the virtual power plant, generate overall operation status characteristics of the virtual power plant through data fusion methods, and determine the operation stability indicators of the virtual power plant.

[0043] Based on the operational stability indicators of the virtual power plant, economic evaluation data is obtained. Through cost accounting methods, the economic indicators of the virtual power plant are calculated to obtain a comprehensive optimized operation plan.

[0044] Optionally, ensuring that the virtual power plant's operating state always meets the grid dispatch requirements through a closed-loop feedback control mechanism includes:

[0045] Acquire real-time operating data of the virtual power plant and generate a real-time status feature vector of the virtual power plant through data fusion technology;

[0046] Based on the real-time status feature vector, a preset load demand prediction model is used to calculate the predicted load demand value for a future period of time.

[0047] After obtaining the load demand forecast, the increase in load demand is calculated by comparing it with historical data;

[0048] If the increase in load demand exceeds a preset threshold, the order of energy storage resource allocation will be determined by a priority sorting algorithm.

[0049] Based on the order of energy storage resource allocation, dispatch instructions for energy storage resources are generated to quickly respond to changes in load demand;

[0050] Obtain real-time wind power output data and calculate the wind power output prediction deviation rate by comparing it with the wind power output prediction model;

[0051] If the wind power output prediction deviation rate is greater than the preset threshold, the participation level of controllable loads will be determined through the controllable load management platform.

[0052] Based on the participation level of controllable loads, scheduling instructions for controllable loads are generated for balancing and adjustment.

[0053] The system acquires the real-time operating status of the virtual power plant and adjusts resource scheduling instructions through a closed-loop feedback control algorithm.

[0054] Update the virtual power plant's operating status according to the adjusted resource scheduling instructions to ensure that the grid scheduling requirements are met.

[0055] Optionally, the step of adjusting the aggregation parameters includes:

[0056] The sliding time window method is used to continuously collect the operation data of each resource. The collected data is cleaned by the data preprocessing module to obtain a standardized operation dataset.

[0057] By comparing the standardized operational dataset with the prediction model, the deviation between the actual operational data and the predicted values ​​of each resource is calculated, and the deviation dataset is obtained.

[0058] If the cumulative deviation value in the deviation dataset exceeds the preset error threshold, the deviation analysis module will identify the source of the deviation and determine the resource characteristic parameters that need to be corrected.

[0059] Based on the determined resource characteristic parameters, an online learning algorithm is used to update the parameter weights, resulting in a corrected aggregate parameter set.

[0060] Optionally, the continuously improved intelligent distributed energy aggregation and prediction system includes:

[0061] The overall operating status evaluation index of the virtual power plant is recalculated using the revised aggregation mechanism parameters.

[0062] By analyzing the trend of virtual power plant operation status changes through state change feature set analysis, key influencing factors are extracted using principal component analysis algorithm to obtain key factor set;

[0063] If the factor values ​​in the key factor set exceed the preset threshold, the factor screening module determines the operating parameters that need to be optimized and generates an optimization parameter set.

[0064] The resource allocation strategy of the virtual power plant is adjusted according to the optimized parameter set, and the resource scheduling weights are updated using a linear regression algorithm to obtain the updated scheduling weight set.

[0065] Real-time resource scheduling instructions are generated by updating the scheduling weight set and output to the virtual power plant control system to obtain the scheduling execution dataset.

[0066] If the execution deviation in the scheduled execution dataset exceeds the preset threshold, the deviation analysis module identifies the source of the deviation and determines the set of control parameters that need to be corrected.

[0067] The virtual power plant operation state model is updated based on the revised control parameter set, and the model parameters are optimized using the support vector machine algorithm to obtain the optimized operation state model.

[0068] The optimized operating status model generates a sequence of predicted values ​​for the virtual power plant, which is then output to the operating status database to update the historical optimization dataset.

[0069] The present invention also provides 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.

[0070] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method.

[0071] Compared with the prior art, the present invention has the following advantages and technical effects:

[0072] This invention discloses an intelligent aggregation and prediction method for distributed energy resources. It obtains historical operational data from wind power, photovoltaics, energy storage, and controllable loads through multi-dimensional feature extraction, establishes a resource correlation calculation model to analyze complementarity, constructs a dynamic aggregation weight allocation mechanism for resource pairs with significant synergistic effects, dynamically adjusts weight coefficients based on real-time meteorological conditions and load forecasts, and uses a multi-objective optimization model to solve for the optimal resource allocation scheme. Real-time dispatch instructions are formulated, and closed-loop feedback control ensures that grid requirements are met. This invention also employs a sliding time window to monitor deviations, triggering an aggregation parameter correction program. Resource characteristic parameters are updated through online learning, achieving continuous improvement in intelligent aggregation and prediction. This method can improve the synergistic utilization efficiency of distributed energy resources, enhance the overall regulation capability and economic benefits of virtual power plants, and provide strong support for the flexibility and reliability of power systems. Attached Figure Description

[0073] 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:

[0074] Figure 1 This is a flowchart of a method according to an embodiment of the present invention. Detailed Implementation

[0075] 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.

[0076] 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.

[0077] Example 1

[0078] like Figure 1 As shown, this embodiment provides a method for virtual power plant distributed energy aggregation and prediction, including:

[0079] Historical operational data of distributed energy resources are acquired, and the historical operational data are processed using multidimensional feature extraction algorithms and time series analysis methods to obtain a characteristic difference matrix;

[0080] Based on the characteristic difference matrix, the time-series complementarity coefficient is calculated for the correlation degree between resources, and the intra-group complementarity strength value is obtained by using the Pearson correlation coefficient method. Based on the intra-group complementarity strength value, the high-priority resource combination is determined.

[0081] A dynamic aggregation weight allocation mechanism is established for high-priority resource combinations. The weight coefficients of each resource in the aggregation process are dynamically adjusted according to real-time meteorological conditions and load demand forecast results. The overall output forecast value of the virtual power plant is calculated through a weighted fusion algorithm, and the corresponding uncertainty range is generated.

[0082] After obtaining the overall output prediction value, a multi-objective optimization model is established with the objective functions of maximizing economic efficiency and optimizing operational stability. The optimal resource allocation scheme is obtained by using the particle swarm optimization algorithm.

[0083] Real-time dispatch instructions are formulated based on the optimal resource allocation scheme, and the virtual power plant operation status is always met by the grid dispatching requirements through a closed-loop feedback control mechanism.

[0084] A sliding time window method is used to continuously monitor the deviation between the actual operating data and the predicted values ​​of each resource. When the cumulative deviation exceeds the preset error threshold, the aggregation parameters are automatically corrected, and the corrected aggregation mechanism parameters are used for the next round of prediction calculation.

[0085] The overall operating status evaluation index of the virtual power plant is recalculated using the modified aggregation mechanism parameters. An operating status database is established to record historical optimization results, resulting in a continuously improving intelligent distributed energy aggregation and prediction system.

[0086] Specifically, the following steps are included:

[0087] Step S101: Use a multi-dimensional feature extraction algorithm to obtain historical operation data of various distributed energy resources such as wind power, photovoltaic, energy storage, and controllable load. Use time series analysis to extract the output characteristic parameters of each resource, including power fluctuation amplitude, response time constant, and regulation range, to obtain a standardized characteristic difference matrix for subsequent complementarity analysis.

[0088] Specifically, historical operational data of distributed energy resources are acquired, and multi-dimensional feature extraction algorithms are used to extract output characteristics from wind power, photovoltaics, energy storage, and controllable loads to obtain an initial characteristic dataset. Time series analysis is then used to process this initial characteristic dataset to extract power fluctuation amplitude, response time constant, and regulation capacity range, resulting in standardized output characteristic parameters. These parameters are then normalized using standardization methods to eliminate dimensional differences, yielding a standardized characteristic dataset. Based on this standardized dataset, a characteristic difference matrix is ​​constructed, and the output characteristic differences among various distributed energy resources are calculated to obtain the characteristic difference matrix.

[0089] For example, historical operational data for distributed energy resources typically comes from real-time monitoring systems of wind power, solar power, energy storage, and controllable loads. Historical wind power data may include wind speed, wind direction, and power generation; solar power data includes solar intensity, module temperature, and output power; energy storage systems record charging and discharging power and battery status; and controllable loads focus on changes in electricity demand. This data is recorded in real-time by sensors and data acquisition systems, forming time-series datasets that provide the foundation for subsequent analysis. When acquiring this data, it is crucial to ensure its integrity and accuracy, for example, by removing outliers or filling in missing values ​​through data cleaning to improve the reliability of subsequent analysis.

[0090] In this embodiment, a multi-dimensional feature extraction algorithm can be used to extract power characteristics from these distributed energy resources. For wind power, extractable features include average power, power fluctuation frequency, and maximum output; for photovoltaics, the focus is on peak solar power and power response under varying solar irradiance; for energy storage systems, the charging and discharging rates and capacity utilization are extracted; and for controllable loads, the load shedding capacity and response speed are analyzed.

[0091] For example, a wind farm recorded an average power output of 2.5 MW over the past year, with high frequency of power fluctuations, indicating strong randomness. A photovoltaic system might reach a peak power output of 5 MW on sunny days, but its power fluctuates significantly under cloudy conditions. Such initial characteristic datasets lay the foundation for subsequent analysis and help to gain a deeper understanding of the operational patterns of various resources.

[0092] Specifically, time series analysis methods can further process the initial characteristic dataset to extract key parameters such as power fluctuation amplitude, response time constant, and regulation range.

[0093] For example, time series analysis can reveal the magnitude of power fluctuations in wind power. For instance, a wind farm might experience power fluctuations of up to 30% during strong winds, indicating high instability. The response time constant measures how quickly a resource responds to control signals. For example, an energy storage system might complete power adjustments within 2 seconds of receiving a frequency regulation command, demonstrating a short response time constant and high flexibility. The regulation capacity range reflects the range of power that can be mobilized by the resource; for example, controllable loads can reduce electricity demand by 1.2 MW during peak hours. This analytical method helps quantify the dynamic characteristics of resources, providing data support for grid dispatching.

[0094] For example, standardization methods are used to normalize output characteristic parameters and eliminate dimensional differences. Power fluctuation amplitude may be expressed as a percentage, response time constant in seconds, and regulation capacity range in megawatts. These parameters have different dimensions, and direct comparison will be distorted.

[0095] In this embodiment, a minimum-maximum normalization method can be used to map all parameters to the range of 0 to 1.

[0096] For example, a 30% power fluctuation in wind power might be normalized to 0.6, a 2-second response time constant in energy storage might be normalized to 0.4, and a regulation capacity range of 1.2 MW might be normalized to 0.8. Such standardized characteristic datasets eliminate the influence of dimensions, facilitating subsequent comparisons and analyses, and improving the scientific rigor and consistency of data processing.

[0097] In this embodiment, a characteristic difference matrix is ​​constructed based on a standardized characteristic dataset to calculate the output characteristic differences among distributed energy resources.

[0098] For example, the power fluctuation amplitudes of wind power and solar power are 0.6 and 0.5 respectively, with a difference of 0.1; the response time constants are 0.7 and 0.4 respectively, with a difference of 0.3; and the regulation capacity ranges are 0.9 and 0.8 respectively, with a difference of 0.1. These differences are recorded in matrix form, forming a characteristic difference matrix. Such a matrix intuitively reflects the similarities and differences in output characteristics of various resources. For example, wind power and solar power have relatively small differences in volatility but large differences in response speed. This analysis provides a basis for resource combination optimization. For instance, in frequency regulation scenarios, energy storage systems with fast response speeds can be prioritized, while in peak shaving and valley filling, the regulation capacity of wind power and controllable loads can be combined.

[0099] For example, the application value of the characteristic difference matrix can be analyzed from multiple perspectives. In grid dispatching, the characteristic difference matrix can help identify resource combinations with strong complementarity. For instance, wind power is highly volatile but has strong regulation capabilities, while energy storage has a fast response speed but limited capacity; combining the two can improve system stability. In energy trading, the matrix can guide market strategies. For example, when photovoltaic output is stable, its electricity can be sold first, while wind power, which fluctuates greatly, can be paired with energy storage to smooth its output. Such analysis not only optimizes resource allocation but also improves the reliability and economy of the grid, providing technical support for the efficient utilization of distributed energy resources.

[0100] Step S102: Construct a resource correlation calculation model based on the characteristic difference matrix, analyze the temporal complementary law of wind power and photovoltaic output through Pearson correlation coefficient, quantify the coordination mode strength of energy storage and controllable load using grey relational analysis method, and determine whether the correlation coefficient between each resource pair exceeds the preset threshold of 0.6. If it exceeds, it is determined that there is a significant synergistic effect.

[0101] Specifically, key feature vectors for each distributed energy resource are extracted using principal component analysis (PCA) based on the characteristic difference matrix, resulting in a dimensionality-reduced feature dataset. Based on this dataset, k-means clustering is applied to group wind power, solar power, energy storage, and controllable loads, yielding resource synergy clustering results. For each cluster, the temporal complementarity coefficient is calculated, and the Pearson correlation coefficient method is used to obtain the complementarity strength value. If the complementarity strength value exceeds a preset threshold, grey relational analysis is used to calculate the coordination strength between energy storage and controllable loads, resulting in a coordination coefficient. Based on the coordination coefficient, a weighted average method is used to fuse the complementarity strength value and coordination coefficient, yielding a comprehensive synergy score for each group. The comprehensive synergy score is then used to rank the groups by synergy priority, determining high-priority resource combinations.

[0102] For example, after constructing the characteristic difference matrix, when extracting key feature vectors using principal component analysis (PCA), standardized characteristic datasets of wind power, photovoltaics, energy storage, and controllable loads can be used as input. PCA can then be used to reduce dimensionality and retain key information. The core of PCA lies in identifying the direction in the data that contributes the most to variance.

[0103] For example, the power fluctuation range of wind power may vary significantly due to changes in wind speed, while that of photovoltaic power is affected by sunlight intensity, resulting in different fluctuation characteristics. Assuming the standard deviation of wind power power fluctuation is 100 kW, photovoltaic is 50 kW, energy storage response time constant is 0.5 seconds, and controllable load is 2 seconds, principal component analysis can extract the principal components of these characteristics, generating dimensionality-reduced feature vectors, such as the fluctuation characteristic vectors for wind power and photovoltaic, and the response characteristic vectors for energy storage and controllable load.

[0104] In this embodiment, when applying the k-means clustering algorithm to group the dimensionality-reduced feature dataset, the number of clusters k can be set to 3, and classification is performed based on the Euclidean distance of the feature vectors.

[0105] For example, wind power and solar power may be grouped together due to their similar output fluctuation characteristics, while energy storage and controllable loads may be grouped together due to their similar response time characteristics. Clustering results can reflect the synergistic potential of resources; for instance, wind power and solar power, which have large output fluctuations, need to be paired with fast-responding energy storage. After grouping, the temporal complementarity coefficients of resources within each group are calculated using the Pearson correlation coefficient method.

[0106] For example, the time series data of wind power and photovoltaic power output may show a negative correlation, indicating that when wind power output is high, photovoltaic power output is low, and the complementarity is strong. The Pearson coefficient may be -0.8, which exceeds the preset threshold of -0.7.

[0107] Specifically, for groups where the complementary strength value exceeds the threshold, grey relational analysis can further evaluate the coordination strength between energy storage and controllable load.

[0108] For example, in a combination of energy storage and controllable loads, the rapid response of energy storage can compensate for the adjustment delay of controllable loads, and the grey relational degree may reach 0.85, indicating that the two are closely coordinated. Next, when using a weighted average method to integrate the complementary strength value and the coordination coefficient, the complementary strength weight can be set to 0.6, and the coordination coefficient weight to 0.4.

[0109] For example, a group has a complementarity strength value of 0.8, a coordination coefficient of 0.85, and a weighted average comprehensive synergy score of 0.82. Based on this score, the synergy priority of each group is ranked, and high-scoring groups, such as those containing wind power, photovoltaics, and energy storage, can be given priority for grid dispatch.

[0110] In this embodiment, resource combinations with high comprehensive synergy scores can be used to optimize the distributed energy configuration of the power grid.

[0111] For example, the highest priority combination might include wind power, solar power, and energy storage, due to their strong complementarity and rapid response, effectively smoothing grid load fluctuations. In contrast, lower priority combinations might lack complementarity and require additional resources. This approach significantly improves the synergistic efficiency of distributed energy resources through multi-dimensional analysis and group optimization.

[0112] Step S103: Establish a dynamic aggregation weight allocation mechanism for high-priority resource combinations. Adjust the weight coefficients of each resource in the aggregation process dynamically based on real-time meteorological conditions and load demand forecast results. Calculate the overall output forecast value of the virtual power plant through a weighted fusion algorithm, and generate the corresponding uncertainty range.

[0113] Specifically, based on real-time meteorological conditions, wind power and photovoltaic power output data are acquired, and time series analysis is used to obtain the output change trends of each resource. If the output change trend exceeds a preset fluctuation threshold, short-term feature data is extracted using a sliding window method to obtain resource output feature vectors. Based on the resource output feature vectors, a support vector regression algorithm is used to predict the future output value of each resource, obtaining the single-resource output prediction result. Using the load demand prediction results, the real-time load demand curve of the virtual power plant is obtained to determine the load demand fluctuation range. If the deviation between the load demand fluctuation range and the single-resource output prediction result exceeds a preset threshold, a linear interpolation method is used to adjust the dynamic weight coefficients of each resource, obtaining an optimized weight allocation scheme. Based on the optimized weight allocation scheme, a weighted fusion algorithm is used to calculate the overall output prediction value of the virtual power plant, obtaining the comprehensive output prediction result.

[0114] For example, in a virtual power plant operation scenario, real-time weather conditions significantly impact wind and solar power output. Suppose that weather data for a certain region on a certain day indicates wind speeds fluctuating between 8 m / s and 12 m / s, and solar radiation intensity varying between 600 W / m² and 800 W / m². By collecting real-time output data from wind farms and solar power plants, we can obtain data showing wind power output fluctuating between 200 MW and 300 MW, and solar power output varying between 150 MW and 250 MW. Time series analysis can employ an autoregressive moving average model to analyze output data from the past 24 hours, identifying an upward trend in wind power output during the nighttime peak and a peak in solar power output at midday. This trend analysis helps to understand resource output patterns and provides a basis for subsequent forecasting.

[0115] In this embodiment, if the fluctuation range of wind power output exceeds a preset threshold of 20%, or the fluctuation range of photovoltaic output exceeds 15%, short-term feature data is extracted using a sliding window method. Assuming a window length of 1 hour, the peak value, mean, and variance of wind power output are extracted to obtain a feature vector such as [250MW, 20MW, 15%]. Similarly, the photovoltaic feature vector can be [200MW, 10MW, 10%]. These feature vectors capture the dynamic characteristics of resource changes and provide input for the prediction model.

[0116] Specifically, support vector regression (SVR) can be used to predict single-resource output for the next hour. Assuming the model is trained based on historical data, the predicted wind power output is 260MW, and the predicted photovoltaic output is 210MW. The prediction results are calibrated in conjunction with meteorological conditions to ensure the error is controlled within ±5%. This method effectively addresses the uncertainties caused by meteorological fluctuations.

[0117] For example, virtual power plant load demand forecasting can be based on historical electricity consumption data and real-time user behavior. Suppose a daily load demand curve shows peak evening demand of 450MW, fluctuating between 400MW and 500MW. If the total predicted output of a single resource is 470MW, and the deviation from the load demand is less than a preset threshold of 10%, then no significant weight adjustment is needed. If the deviation exceeds the threshold, such as a total predicted output of 380MW and a deviation of 15%, then the weights are adjusted using linear interpolation. Assuming an initial wind power weight of 0.6 and a solar power weight of 0.4, after interpolation adjustment, the wind power weight increases to 0.65, and the solar power weight decreases to 0.35 to balance insufficient output.

[0118] In this embodiment, the optimized weight allocation scheme is input into the weighted fusion algorithm to calculate the predicted comprehensive output of the virtual power plant. Assuming a predicted wind power output of 260MW and a photovoltaic output of 210MW, the weighted comprehensive output is 260×0.65 + 210×0.35 = 242.5MW. This result can be used as a scheduling basis to optimize resource allocation, smooth output fluctuations, and improve the operational stability of the virtual power plant.

[0119] It should be noted that the above methods closely integrate meteorological and load data to ensure forecast accuracy. Sliding window feature extraction and support vector regression forecasting can dynamically adapt to changes in resource output. Weight adjustment and weighted fusion enhance the synergy between resources, ensuring that overall output meets load demand and optimizing the operating efficiency of the virtual power plant.

[0120] Step S104: After obtaining the overall power output prediction value, establish a multi-objective optimization model with the objective functions of maximizing economic efficiency and optimizing operational stability. The constraints include the limitations of various resource technical parameters and the requirements for safe operation of the power grid. The optimal resource allocation scheme, including the power output allocation ratio of each resource, is obtained by using the particle swarm optimization algorithm.

[0121] Specifically, after obtaining the overall power output forecast, a multi-objective optimization model is constructed. The objective functions are defined as maximizing economic efficiency and optimizing operational stability. Constraints are set as limitations on the technical parameters of each resource and the requirements for safe operation of the power grid. A particle swarm optimization algorithm is used to calculate the optimal power output allocation ratio and resource configuration scheme for each resource. Based on the optimal power output allocation ratio, real-time power output data for each resource is obtained. Through time series analysis, the trend of power output changes for each resource is extracted to determine whether the power output of each resource meets the technical parameter limitations. If the power output of a resource exceeds the technical parameter limitations, the power output allocation ratio is adjusted using a linear interpolation method to obtain a corrected power output allocation scheme. Based on the corrected power output allocation scheme, real-time operating status data of the virtual power plant is obtained. The uncertainty of the power output allocation scheme for each resource is analyzed using a Monte Carlo simulation method to determine the range of uncertainty in the operating status. Based on the range of uncertainty in the operating status, real-time load demand data of the power grid is obtained to determine whether load demand fluctuations exceed a preset threshold. If load demand fluctuations exceed the preset threshold, the power output allocation ratio of each resource is adjusted using a weighted average method to obtain an optimized resource scheduling scheme. Based on the optimized resource scheduling scheme, real-time output monitoring data for each resource is acquired. Using a sliding window method, short-term output feature vectors are extracted to determine whether the output of each resource meets the requirements for safe grid operation. If the resource output does not meet the requirements, a quadratic programming method is used to recalculate the output allocation ratio of each resource, resulting in an adjusted scheduling scheme. Based on the adjusted scheduling scheme, comprehensive operation data of the virtual power plant is acquired. Using data fusion methods, the overall operating status characteristics of the virtual power plant are generated, and the operational stability indicators of the virtual power plant are determined. Based on the operational stability indicators of the virtual power plant, economic evaluation data is acquired. Using cost accounting methods, the economic indicators of the virtual power plant are calculated, resulting in a comprehensive optimized operation scheme.

[0122] In this embodiment, when constructing a multi-objective optimization model, the objective function focuses on maximizing economic efficiency and optimizing operational stability.

[0123] For example, a virtual power plant includes wind power, solar power, and energy storage systems. The economic objective can be achieved by minimizing operating costs, considering generation costs, maintenance costs, and electricity price fluctuations. The stability objective focuses on output volatility, ensuring smooth changes in output across all resources. Constraints include a wind power output cap of 1000 MW, solar power output limited by sunlight intensity, energy storage charging and discharging efficiency of 90%, and a stable range for grid voltage and frequency. A particle swarm optimization algorithm is used to calculate the output ratio of each resource.

[0124] For example, at a wind speed of 10 m / s, wind power can account for 60% of the total output, photovoltaic power 30%, and energy storage 10%, in order to balance cost and stability.

[0125] For example, after acquiring real-time power output data, time series analysis is used to extract trends. Suppose that wind power output increases from 800 MW to 950 MW within one hour, exceeding the technical parameter limit (upper limit 900 MW). Using linear interpolation, the wind power output is adjusted to 890 MW, and energy storage increases output by 10 MW, maintaining stable total output. This adjustment ensures that resources operate within a safe range while maintaining the overall power output balance of the virtual power plant.

[0126] In this embodiment, Monte Carlo simulation is used to analyze the uncertainty of the power output allocation scheme. Assuming a wind speed prediction error of ±5%, 1000 simulations are conducted to determine the power output uncertainty range as ±50 MW. Combined with grid load demand data, for example, if the load demand fluctuates between 500-700 MW and exceeds a threshold (±10%), the power output ratio is adjusted using a weighted average method, with wind power weight reduced by 5% and energy storage weight increased by 3%, to cope with sudden load changes.

[0127] For example, when extracting short-term output feature vectors using the sliding window method, assuming a 10-minute window, we analyze the trend of photovoltaic output decreasing from 200 MW to 180 MW. If this change causes the grid frequency to deviate by more than 0.2 Hz, the output is redistributed using a quadratic programming method, reducing the photovoltaic output to 170 MW and supplementing it with 15 MW of energy storage to ensure the safe operation of the grid.

[0128] In this embodiment, the data fusion method generates virtual power plant operating status characteristics.

[0129] For example, by integrating real-time output, voltage, and frequency data from wind power, solar power, and energy storage, a comprehensive operating status vector is generated. Stability indicators, such as output fluctuation rate below 5%, are derived through analysis. Economic indicators are obtained through cost accounting; for example, the total operating cost is 50 yuan per megawatt-hour, lower than the 60 yuan per megawatt-hour of traditional power plants, demonstrating economic advantages.

[0130] Understandably, the above methods, through multi-level optimization and dynamic adjustment, ensure efficient dispatch of the virtual power plant in complex operating environments. Each step is implemented based on real-time data, balancing economic efficiency and stability, and adapting to the variable demands of the power grid and resources.

[0131] Step S105: Formulate real-time dispatch instructions according to the optimal resource allocation scheme. If the predicted load demand increase exceeds 5%, prioritize the use of energy storage resources for rapid response. If the wind power output prediction deviation rate is greater than 10%, start controllable loads to participate in balance regulation. Ensure that the virtual power plant operation status always meets the grid dispatch requirements through a closed-loop feedback control mechanism.

[0132] Specifically, real-time operational data of the virtual power plant is acquired, and a real-time state feature vector of the virtual power plant is generated through data fusion technology. Based on the real-time state feature vector, a preset load demand forecasting model is used to calculate the predicted load demand for a future period. After obtaining the predicted load demand, the load demand increase is calculated by comparing it with historical data. If the load demand increase exceeds a preset threshold, a priority ranking algorithm is used to determine the order of energy storage resource allocation. Based on the energy storage resource allocation order, scheduling instructions for energy storage resources are generated to quickly respond to changes in load demand. Real-time wind power output data is acquired, and the wind power output forecast deviation rate is calculated by comparing it with the wind power output forecasting model. If the wind power output forecast deviation rate is greater than a preset threshold, the participation level of controllable loads is determined through the controllable load management platform. Based on the participation level of controllable loads, scheduling instructions for controllable loads are generated for balancing and adjustment. The real-time operating status of the virtual power plant is acquired, and resource scheduling instructions are adjusted through a closed-loop feedback control algorithm. Based on the adjusted resource scheduling instructions, the operating status of the virtual power plant is updated to ensure that the grid dispatching requirements are met.

[0133] Specifically, data fusion technology is particularly crucial when acquiring real-time operational data of virtual power plants.

[0134] For example, by fusing data from different resources, such as real-time output data from wind power, energy storage, and controllable loads, a comprehensive real-time state feature vector is generated. For instance, if a virtual power plant has a wind power output of 50MW, an energy storage charging / discharging state of -10MW (discharging), and a controllable load of 20MW at 15:00, after data fusion, the real-time state feature vector might be [50, -10, 20], reflecting the overall operating status of the virtual power plant.

[0135] In this embodiment, a preset load demand forecasting model is used to predict load demand over a future period. For example, the ARIMA model is used to predict the load demand for the next 24 hours, resulting in a predicted value sequence [100, 105, 110, ...] MW. After obtaining the predicted load demand values, they are compared with historical data to calculate the increase in load demand. For example, if the current load demand is 90 MW, and the predicted load demand for the next hour is 105 MW, the increase is 15 MW. If the preset threshold is 10 MW, the increase exceeds the threshold, triggering the mobilization of energy storage resources.

[0136] It should be noted that the order in which energy storage resources are deployed is determined by a priority ranking algorithm. For example, energy storage resources with fast response times and low costs, such as lithium battery energy storage systems, are deployed first, followed by pumped-storage hydroelectric power stations, which have slower response times. Scheduling instructions are generated based on the deployment order of energy storage resources. For instance, 10MW of lithium battery energy storage may be deployed, and 5MW of pumped-storage hydroelectric power station may be deployed, to quickly respond to changes in load demand.

[0137] Specifically, calculating the wind power output prediction deviation rate is crucial. For example, if the wind power output prediction model predicts an output of 60MW at 15:00, but the actual output is 50MW, the prediction deviation rate is (60-50) / 60 = 16.67%. If the preset threshold is 10%, the deviation rate will exceed the threshold, triggering the participation of controllable loads. For instance, through the controllable load management platform, if the controllable load participation rate is determined to be 30%, then 30MW of controllable loads will be used for balancing and adjustment.

[0138] In this embodiment, a closed-loop feedback control algorithm is used to adjust resource scheduling commands. For example, the initial scheduling command is 50MW wind power output, 10MW energy storage discharge, and 20MW controllable load. Based on the real-time operating status of the virtual power plant, the closed-loop feedback control algorithm adjusts the scheduling command to 55MW wind power output, 8MW energy storage discharge, and 22MW controllable load to ensure that the grid scheduling requirements are met.

[0139] Preferably, after updating the operating status of the virtual power plant, it is ensured that the grid dispatch requirements are met. For example, if the grid dispatch requires the virtual power plant to output 100MW at 15:00, the adjusted operating status would be 55MW wind power output, 8MW energy storage discharge, and 22MW controllable load, for a total output of 55 + 8 + 22 = 85MW. If the requirements are not met, the resource dispatch instructions are further adjusted until they are satisfied.

[0140] In this embodiment, the virtual power plant can quickly respond to changes in load demand, balance fluctuations in wind power output, and ensure the safe and stable operation of the power grid through the above-described method. Simultaneously, by prioritizing the use of energy storage resources and controllable loads, the operating costs of the virtual power plant are reduced, improving its economic efficiency. For example, when load demand increases, priority is given to using energy storage resources, avoiding high-priced electricity purchases; when wind power output fluctuates, balancing and regulation are performed through controllable loads, preventing wind curtailment.

[0141] For example, in a virtual power plant, after adopting the above method, the load demand response time was shortened from 10 minutes to 5 minutes, the wind power output prediction deviation rate was controlled within 5%, and the operating cost of the virtual power plant was reduced by 10%, significantly improving its economic efficiency. At the same time, the operational stability of the virtual power plant was ensured, and the grid dispatch requirements were met.

[0142] Understandably, the above methods, through data fusion, load demand forecasting, energy storage resource allocation, wind power output forecasting deviation rate control, controllable load participation, and closed-loop feedback control, achieve optimized operation of the virtual power plant. This method not only improves the economics of the virtual power plant but also enhances its operational stability, providing a strong guarantee for the safe and stable operation of the power grid.

[0143] Step S106: The sliding time window method is used to continuously monitor the deviation between the actual operating data and the predicted values ​​of each resource. When the cumulative deviation exceeds the preset error threshold, the aggregation parameter correction program is automatically triggered. The resource characteristic parameters and correlation coefficients are updated through an online learning algorithm to obtain the corrected aggregation mechanism parameters for the next round of prediction calculation.

[0144] Specifically, a sliding time window method is used to continuously collect operational data for each resource. The collected data is cleaned by a data preprocessing module to obtain a standardized operational dataset. By comparing the standardized operational dataset with the prediction model, the deviation between the actual operational data and the predicted values ​​for each resource is calculated, resulting in a deviation dataset. If the cumulative deviation value in the deviation dataset exceeds a preset error threshold, the deviation analysis module identifies the source of the deviation and determines the resource characteristic parameters that need to be corrected. Based on the determined resource characteristic parameters, an online learning algorithm is used to update the parameter weights, resulting in a corrected aggregate parameter set.

[0145] Specifically, the sliding time window method, which continuously collects operational data from various resources, forms the basis for the dynamic management of virtual power plants. For example, in a virtual power plant scenario, the system uses a 10-minute time window to collect real-time operational data from distributed photovoltaic (PV) systems, energy storage devices, and controllable loads, including PV output power, energy storage state of charge, and load power consumption. This data is uploaded to the cloud once per second via sensors and smart meters, ensuring high timeliness.

[0146] It should be noted that the advantage of a sliding time window lies in its ability to dynamically update data, avoiding scheduling deviations caused by time lags, thus providing a real-time foundation for subsequent analysis. The collected data is cleaned through a data preprocessing module to obtain a standardized operational dataset.

[0147] Specifically, data cleaning handles outliers and missing values. For example, photovoltaic output data may contain negative values ​​or abnormally high peak values ​​due to sensor malfunctions. The cleaning module will repair the data using historical averages or interpolation methods to ensure the integrity of the dataset.

[0148] In this embodiment, assuming that the output data of a photovoltaic power station is missing at a certain moment, the system can use linear interpolation based on the output trend of the previous 5 minutes to complete the data and generate a standardized output dataset. This method can effectively reduce noise interference and provide reliable input for subsequent deviation analysis. By comparing the standardized operating dataset with the prediction model, the deviation value between the actual operating data and the predicted value of each resource is calculated to obtain the deviation dataset. For example, the load prediction model of the virtual power plant predicts that the total load in a certain hour period is 1000 kW, while the actual collected data is 1050 kW, with a deviation rate of 5%. The deviation dataset will record the deviation of all resources, such as photovoltaic output deviation, energy storage response deviation, etc.

[0149] It should be noted that generating the deviation dataset helps to quickly locate weak points in system operation, providing a basis for precise scheduling. If the cumulative deviation value in the deviation dataset exceeds a preset error threshold, the deviation analysis module identifies the source of the deviation and determines the resource characteristic parameters that need to be corrected.

[0150] In this embodiment, assuming a preset error threshold of 8%, and a cumulative deviation reaching 10% within a certain time period, the deviation analysis module, through historical data comparison, discovers that the photovoltaic power output prediction model has a large deviation due to not considering the impact of sudden weather changes. The system will mark the parameters of the photovoltaic power output prediction model, such as the weather impact weight, as needing correction. This method can quickly pinpoint the root cause of the problem and improve the system's adaptability. Based on the determined resource characteristic parameters, an online learning algorithm is used to update the parameter weights, resulting in a corrected aggregate parameter set. For example, regarding the photovoltaic power output prediction deviation, the system dynamically adjusts the weather impact weight from 0.3 to 0.5 using an online learning algorithm, and regenerates the prediction model by combining it with real-time meteorological data.

[0151] It should be noted that the advantage of online learning is that it eliminates the need for offline model retraining, enabling real-time parameter optimization and ensuring that the virtual power plant can still operate efficiently in complex environments.

[0152] Preferably, the corrected aggregate parameter set is immediately applied to the next scheduling instruction generation, thereby improving the accuracy of load balancing.

[0153] In this embodiment, the virtual power plant achieves closed-loop management throughout the entire process using the aforementioned method. For example, if a virtual power plant in a certain region detects a surge in load demand during peak hours, the system captures real-time data through a sliding time window. After data analysis, it is found that the deviation is mainly due to insufficient wind power output. The deviation analysis module identifies that the parameters of the wind power prediction model need adjustment, and the online learning algorithm then optimizes the parameters, generates new scheduling instructions, and prioritizes the use of energy storage resources to respond to load demand. This end-to-end management ensures the operational stability of the virtual power plant while improving resource utilization efficiency.

[0154] Step S107: Recalculate the overall operating status evaluation index of the virtual power plant using the corrected aggregation mechanism parameters, including key performance parameters such as prediction accuracy, regulation capacity, and economic benefits. Establish an operating status database to record historical optimization results and form a continuously improving intelligent distributed energy aggregation and prediction system.

[0155] Specifically, the system analyzes the trend of virtual power plant operation status changes through a set of state change features, extracts key influencing factors using principal component analysis, and obtains a key factor set. If the factor values ​​in the key factor set exceed a preset threshold, the factor screening module determines the operating parameters that need optimization, generating an optimized parameter set. Based on the optimized parameter set, the virtual power plant's resource allocation strategy is adjusted, and a linear regression algorithm is used to update the resource scheduling weights, resulting in an updated scheduling weight set. Real-time resource scheduling instructions are generated using the updated scheduling weight set and output to the virtual power plant control system to obtain a scheduling execution dataset. If the execution deviation in the scheduling execution dataset exceeds a preset threshold, the deviation analysis module identifies the source of the deviation and determines the control parameter set that needs correction. The virtual power plant operation status model is updated based on the corrected control parameter set, and the model parameters are optimized using a support vector machine algorithm, resulting in an optimized operation status model. The optimized operation status model generates a predicted value sequence for the virtual power plant, outputs it to the operation status database, and updates the historical optimization dataset.

[0156] For example, when analyzing the trend of virtual power plant operation status changes through state change feature sets, one can first collect operational data of various resources in the virtual power plant, such as the output of power generation equipment, the charging and discharging status of energy storage equipment, and real-time load demand data. The collected data typically includes time-series information; for example, the output change of a wind turbine generator over the past 24 hours might show an increase from 50 kW to 80 kW, followed by a decrease to 60 kW. This trend reflects the dynamic nature of resource operation status. Principal component analysis algorithms can then be used to extract key influencing factors.

[0157] Understandably, principal component analysis transforms multiple related variables into a few principal components through dimensionality reduction.

[0158] For example, wind speed, temperature, and equipment aging might be combined into a single key factor, representing the combined impact of environmental and equipment conditions. The resulting set of key factors might include factor A representing power generation efficiency and factor B representing load volatility. Assuming a preset threshold of 10%, if factor A reaches 12%, it indicates that power generation efficiency significantly deviates from expectations.

[0159] In this embodiment, if the factor values ​​in the key factor set exceed a preset threshold, the factor screening module can be used to determine the operating parameters that need to be optimized.

[0160] For example, if factor A exceeds the limit, analysis might reveal that the operating parameters of the power generation equipment, such as speed or power settings, need adjustment. The filtering module will generate an optimized parameter set based on historical data and the current operating status, such as adjusting the wind turbine's speed from 800 rpm to 850 rpm to improve output efficiency. When adjusting the virtual power plant's resource allocation strategy based on the optimized parameter set, a linear regression algorithm can be used to update the resource scheduling weights.

[0161] For example, analysis of historical data revealed that the discharge weight of energy storage devices during peak hours should be increased from 0.6 to 0.75 to better meet load demand. The updated scheduling weight set was then used to generate real-time resource scheduling instructions, such as instructions requiring energy storage devices to discharge at maximum power between 17:00 and 19:00, outputting to the virtual power plant control system.

[0162] For example, after obtaining the scheduling execution dataset, if it is found that the execution deviation exceeds the preset threshold, such as the actual discharge power being only 85% of the command value, the deviation reaches 15%, exceeding the 10% threshold, then the deviation analysis module identifies the source of the deviation.

[0163] Understandably, the deviation may stem from response delays in the energy storage device or performance degradation due to device aging. After determining the set of control parameters requiring correction, such as adjusting the energy storage device's response time from 5 seconds to 3 seconds, or increasing the output ratio of the standby generator set, a support vector machine algorithm is used to optimize the model parameters when updating the virtual power plant operating state model based on the corrected control parameter set.

[0164] For example, by analyzing historical deviation data and current operating data, the weight parameters related to load forecasting in the model can be adjusted to make the model more adaptable to demand fluctuations during peak periods. The optimized operating state model then generates a series of predicted values ​​for a virtual power plant, such as predicting that load demand will increase from 1,000 kW to 1,200 kW in the next hour.

[0165] In this embodiment, after the predicted value sequence is output to the running status database, the historical optimization dataset is updated.

[0166] For example, the database records changes in adjusted generator speed, energy storage device discharge power, and load forecasts. This data provides a basis for subsequent analysis.

[0167] Understandably, updating the historical optimization dataset helps the system learn continuously.

[0168] For example, when the system identifies recurring deviations, such as low response efficiency of energy storage devices in low-temperature environments, control parameters can be further optimized to generate more reasonable dispatch instructions. In this way, the operating status of the virtual power plant can be continuously improved. This involves optimizing model parameters for the support vector machine algorithm.

[0169] For example, by analyzing the operating data of the past week, the prediction error pattern during peak load periods can be identified, and the model parameters can be adjusted to make the predicted value sequence closer to the actual needs.

[0170] This embodiment also provides 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.

[0171] This embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method.

[0172] 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 method for virtual power plant distributed energy aggregation and prediction, characterized in that, Includes the following steps: Historical operational data of distributed energy resources are acquired, and the historical operational data are processed using multidimensional feature extraction algorithms and time series analysis methods to obtain a characteristic difference matrix; Based on the characteristic difference matrix, the time-series complementarity coefficient is calculated for the correlation degree between resources, and the intra-group complementarity strength value is obtained by using the Pearson correlation coefficient method. Based on the intra-group complementarity strength value, the high-priority resource combination is determined. A dynamic aggregation weight allocation mechanism is established for high-priority resource combinations. The weight coefficients of each resource in the aggregation process are dynamically adjusted according to real-time meteorological conditions and load demand forecast results. The overall output forecast value of the virtual power plant is calculated through a weighted fusion algorithm, and the corresponding uncertainty range is generated. After obtaining the overall output prediction value, a multi-objective optimization model is established with the objective functions of maximizing economic efficiency and optimizing operational stability. The optimal resource allocation scheme is obtained by using the particle swarm optimization algorithm. Real-time dispatch instructions are formulated based on the optimal resource allocation scheme, and the virtual power plant operation status is always met by the grid dispatching requirements through a closed-loop feedback control mechanism. A sliding time window method is used to continuously monitor the deviation between the actual operating data and the predicted values ​​of each resource. When the cumulative deviation exceeds the preset error threshold, the aggregation parameters are automatically corrected, and the corrected aggregation mechanism parameters are used for the next round of prediction calculation. The overall operating status evaluation index of the virtual power plant is recalculated using the modified aggregation mechanism parameters. An operating status database is established to record historical optimization results, resulting in a continuously improving intelligent distributed energy aggregation and prediction system.

2. The method according to claim 1, characterized in that, The obtained characteristic difference matrix includes: Historical operational data of distributed energy resources are acquired, and a multi-dimensional feature extraction algorithm is used to extract power characteristics from the historical operational data to obtain an initial characteristic dataset; the historical operational data includes wind power, photovoltaic, energy storage, and controllable loads; The initial characteristic dataset is processed by time series analysis to extract power fluctuation amplitude, response time constant and regulation range, and obtain standardized output characteristic parameters. The output characteristic parameters are normalized using a standardization method to eliminate dimensional differences and obtain a standardized characteristic dataset. Based on the standardized characteristic dataset, a characteristic difference matrix is ​​constructed, and the output characteristic differences among distributed energy resources are calculated to obtain the characteristic difference matrix.

3. The method according to claim 1, characterized in that, The determination of high-priority resource combinations includes: Based on the characteristic difference matrix, the key feature vectors of each distributed energy resource are extracted by principal component analysis to obtain the dimensionality-reduced feature dataset. Based on the dimensionality-reduced feature dataset, wind power, photovoltaic power, energy storage and controllable load are grouped using the k-means clustering algorithm to obtain resource collaborative clustering results; Based on the resource collaborative clustering results, the temporal complementarity coefficient of resources within each group is calculated, and the Pearson correlation coefficient method is used to obtain the complementarity strength value within the group; If the complementary strength value within the group exceeds the preset threshold, the coordination strength between energy storage and controllable load within the group is calculated using the grey relational analysis method to obtain the coordination coefficient within the group. Based on the coordination coefficient within the group, a weighted average method is used to integrate the complementary strength value and the coordination coefficient within the group to obtain the comprehensive synergy score for each group. By comprehensively evaluating collaboration scores, the collaboration priorities of each group are ranked, and high-priority resource combinations are determined.

4. The method according to claim 1, characterized in that, The calculation yields the predicted overall output of the virtual power plant, and generates the corresponding uncertainty range, including: Based on real-time meteorological conditions, wind power and photovoltaic power output data are obtained, and time series analysis is used to obtain the output change trend of each resource. If the output change trend exceeds the preset fluctuation threshold, short-term feature data is extracted using the sliding window method to obtain the resource output feature vector; Based on the resource output feature vector, the support vector regression algorithm is used to predict the future output value of each resource, and the single resource output prediction result is obtained. By using the load demand forecast results, the real-time load demand curve of the virtual power plant is obtained, and the range of load demand fluctuations is determined. If the deviation between the load demand fluctuation range and the single resource output prediction result exceeds the preset threshold, the linear interpolation method is used to adjust the dynamic weight coefficient of each resource to obtain the optimized weight allocation scheme. Based on the optimized weight allocation scheme, a weighted fusion algorithm is used to calculate the overall power output prediction value of the virtual power plant, and obtain the comprehensive power output prediction result.

5. The method according to claim 1, characterized in that, The optimal resource allocation scheme includes: After obtaining the overall power output prediction value, a multi-objective optimization model is constructed. The objective function is defined as maximizing economic efficiency and optimizing operational stability. The constraints are set as the technical parameter limits of each resource and the requirements for safe operation of the power grid. The particle swarm optimization algorithm is used to calculate the optimal power output allocation ratio and resource configuration scheme of each resource. Based on the optimal output allocation ratio of each resource, real-time output data of each resource is obtained. Through time series analysis, the output change trend of each resource is extracted to determine whether the output of each resource meets the technical parameter limits. If the resource output exceeds the technical parameter limit, the output allocation ratio is adjusted by linear interpolation to obtain a corrected output allocation scheme. Based on the revised power allocation scheme, real-time operating status data of the virtual power plant is obtained. The uncertainty of each resource power allocation scheme is analyzed using the Monte Carlo simulation method to determine the range of uncertainty in the operating status. Based on the uncertainty range of the operating status, real-time load demand data of the power grid is obtained to determine whether the load demand fluctuation exceeds the preset threshold. If the load demand fluctuation exceeds the preset threshold, the output allocation ratio of each resource is adjusted by weighted average method to obtain an optimized resource scheduling scheme. Based on the optimized resource scheduling scheme, real-time power output monitoring data of each resource is obtained. Short-term power output feature vectors are extracted using the sliding window method to determine whether the power output of each resource meets the requirements for safe operation of the power grid. If the output of resources does not meet the requirements for safe operation of the power grid, the allocation ratio of each resource output is recalculated using the quadratic programming method to obtain the adjusted scheduling scheme. Based on the adjusted scheduling plan, obtain comprehensive operation data of the virtual power plant, generate overall operation status characteristics of the virtual power plant through data fusion methods, and determine the operation stability indicators of the virtual power plant. Based on the operational stability indicators of the virtual power plant, economic evaluation data is obtained. Through cost accounting methods, the economic indicators of the virtual power plant are calculated to obtain a comprehensive optimized operation plan.

6. The method according to claim 1, characterized in that, The method of ensuring that the virtual power plant's operating state always meets the grid dispatch requirements through a closed-loop feedback control mechanism includes: Acquire real-time operating data of the virtual power plant and generate a real-time status feature vector of the virtual power plant through data fusion technology; Based on the real-time status feature vector, a preset load demand prediction model is used to calculate the predicted load demand value for a future period of time. After obtaining the load demand forecast, the increase in load demand is calculated by comparing it with historical data; If the increase in load demand exceeds a preset threshold, the order of energy storage resource allocation will be determined by a priority sorting algorithm. Based on the order of energy storage resource allocation, dispatch instructions for energy storage resources are generated to quickly respond to changes in load demand; Obtain real-time wind power output data and calculate the wind power output prediction deviation rate by comparing it with the wind power output prediction model; If the wind power output prediction deviation rate is greater than the preset threshold, the participation level of controllable loads will be determined through the controllable load management platform. Based on the participation level of controllable loads, scheduling instructions for controllable loads are generated for balancing and adjustment. The system acquires the real-time operating status of the virtual power plant and adjusts resource scheduling instructions through a closed-loop feedback control algorithm. Update the virtual power plant's operating status according to the adjusted resource scheduling instructions to ensure that the grid scheduling requirements are met.

7. The method according to claim 1, characterized in that, The process of adjusting the aggregation parameters includes: The sliding time window method is used to continuously collect the operation data of each resource. The collected data is cleaned by the data preprocessing module to obtain a standardized operation dataset. By comparing the standardized operational dataset with the prediction model, the deviation between the actual operational data and the predicted values ​​of each resource is calculated, and the deviation dataset is obtained. If the cumulative deviation value in the deviation dataset exceeds the preset error threshold, the deviation analysis module will identify the source of the deviation and determine the resource characteristic parameters that need to be corrected. Based on the determined resource characteristic parameters, an online learning algorithm is used to update the parameter weights, resulting in a corrected aggregate parameter set.

8. The method according to claim 1, characterized in that, The continuously improved intelligent distributed energy aggregation and prediction system includes: The overall operating status evaluation index of the virtual power plant is recalculated using the revised aggregation mechanism parameters. By analyzing the trend of virtual power plant operation status changes through state change feature set analysis, key influencing factors are extracted using principal component analysis algorithm to obtain key factor set; If the factor values ​​in the key factor set exceed the preset threshold, the factor screening module determines the operating parameters that need to be optimized and generates an optimization parameter set. The resource allocation strategy of the virtual power plant is adjusted according to the optimized parameter set, and the resource scheduling weights are updated using a linear regression algorithm to obtain the updated scheduling weight set. Real-time resource scheduling instructions are generated by updating the scheduling weight set and output to the virtual power plant control system to obtain the scheduling execution dataset. If the execution deviation in the scheduled execution dataset exceeds the preset threshold, the deviation analysis module identifies the source of the deviation and determines the set of control parameters that need to be corrected. The virtual power plant operation state model is updated based on the revised control parameter set, and the model parameters are optimized using the support vector machine algorithm to obtain the optimized operation state model. The optimized operating status model generates a sequence of predicted values ​​for the virtual power plant, which is then output to the operating status database to update the historical optimization dataset.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-8.

10. 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 1-8.

Images