Distributed energy scheduling method and system based on virtual power plant

Abstract

The invention provides a distributed energy scheduling method and system based on a virtual power plant, and the scheduling method is also used for a scheduling system comprising an energy integration module, an energy power generation prediction module and the virtual power plant, and specifically comprises the steps: carrying out the grouping of distributed energy, and screening the target energy of each energy group; constructing a basic power generation prediction model of each energy group based on the historical power generation information of the corresponding target energy; correcting the basic power generation prediction model according to the historical power generation information and the historical scheduling information of each distributed energy source, obtaining a corresponding power generation prediction model and performing power generation prediction, obtaining a power generation prediction value of each distributed energy source, integrating the power generation prediction value of each distributed energy source by the virtual power plant, and constructing an energy scheduling strategy in combination with a scheduling target. According to the method, efficient and accurate independent modeling of each distributed energy source is realized through target energy source grouping screening, basic model establishment and historical information correction, the prediction accuracy can be effectively improved, and the accuracy of a scheduling strategy is further improved.

Description

Technical Field

[0001] This invention relates to the field of distributed energy dispatching technology, and in particular to a distributed energy dispatching method and system based on virtual power plants. Background Technology

[0002] In the early stages of distributed energy development, due to the small scale of grid integration and relatively relaxed requirements for grid dispatch accuracy, most distributed energy systems adopted a unified predictive model to identify corresponding dispatch capabilities. At this time, when virtual power plants execute distributed energy operations, because the number and type of distributed energy nodes are small, virtual power plants do not require complex and refined dispatch. They only need to simply integrate the prediction results output by the unified model and combine them with the basic power supply needs of the grid to formulate a coarse-grained dispatch strategy to meet operational requirements.

[0003] However, with the rapid development of the distributed energy industry, the number of nodes connected to the power grid has increased significantly, and the types of energy have become increasingly diversified. The scheduling requirements of virtual power plants have gradually upgraded from basic integration to precise coordination. This not only requires the aggregation of dispersed resources but also the achievement of scheduling goals such as smoothing out output fluctuations, participating in grid peak shaving and frequency regulation, and reducing overall energy costs through refined scheduling. However, the unified modeling methods commonly used in existing technologies cannot take into account the individual differences of the same type of distributed energy. The predicted values ​​output deviate significantly from the actual output of each distributed energy source, making it impossible for virtual power plants to accurately grasp the dispatchable potential of each distributed energy source. Consequently, the accuracy of the formulated scheduling strategies is low, which can easily lead to local power supply gaps or power curtailment, affecting the stability of grid operation. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies that use a unified modeling approach to predict the power generation of various distributed energy sources, which ignores the individual differences of distributed energy sources of the same type and results in low accuracy of the scheduling strategies formulated by virtual power plants based on the prediction results. This invention provides a distributed energy scheduling method and system based on virtual power plants. By selecting target energy sources in groups, establishing basic models, and correcting historical information, it achieves efficient and accurate independent modeling of each distributed energy source, effectively improving the accuracy of prediction results and thus improving the accuracy of the scheduling strategies constructed by virtual power plants.

[0005] The objective of this invention is achieved through the following technical solution: Distributed energy dispatching methods based on virtual power plants include: The distributed energy resources within the current scheduling area are grouped, and the target energy resources corresponding to each energy group are selected according to the grouping results. Based on the historical power generation information of the corresponding target energy, construct basic power generation prediction models for each energy group; The basic power generation prediction model is revised based on the historical power generation information and historical dispatch information of each distributed energy source to obtain the power generation prediction model of each distributed energy source. The power generation forecast values ​​of each distributed energy source are obtained based on the corresponding power generation forecast model. The virtual power plant integrates the power generation forecast values ​​of each distributed energy source and constructs an energy dispatch strategy in combination with the dispatch objectives.

[0006] To reflect the individual differences among distributed energy sources, each distributed energy source is modeled independently. During independent modeling, similar distributed energy sources are first grouped together based on their energy characteristics to identify target energy sources that accurately reflect the common patterns within the group. Corresponding basic power generation prediction models are then constructed based on these target energy sources, avoiding redundant modeling and training, reducing computational load, and improving the efficiency of independent modeling. Furthermore, the basic power generation prediction models are refined by incorporating historical power generation and dispatch information from each distributed energy source. This ensures that the established power generation prediction models for each distributed energy source retain both common patterns within the group and individual characteristics, guaranteeing the accuracy of the prediction results. Based on high-precision predictions, the virtual power plant can clearly understand the actual adjustable capacity of each distributed energy source, thereby accurately matching dispatch targets, ensuring the accuracy of the formulated dispatch strategies, and ultimately improving the stability of the power grid operation.

[0007] Furthermore, the step of grouping distributed energy resources within the current scheduling area and selecting target energy resources corresponding to each energy group according to the grouping results includes: Distributed energy resources are clustered based on historical output data and related environmental data to obtain several energy clusters; Calculate the center distance of each distributed energy source within each energy cluster, and score the data integrity of each distributed energy source by combining the corresponding historical output data; The scores for center distance and data integrity are weighted and summed to obtain a comprehensive score for each distributed energy source. The distributed energy source with the highest comprehensive score is selected as the target energy source for the corresponding energy group.

[0008] Furthermore, the construction of basic power generation prediction models for each energy group based on historical power generation information of the corresponding target energy source includes: The historical power generation data of the target energy source is preprocessed, and the corresponding related environmental data is retrieved and spatiotemporally aligned with it. Derived features are established based on the energy type of the target energy source, and a training sample set is constructed by combining aligned historical power generation data and related process data. Select the corresponding prediction model type based on the energy type of the target energy source, and train the selected prediction model based on the training sample set to construct the power generation prediction model for the target energy source. The power generation prediction model of the target energy is used as the basic power generation prediction model for the corresponding energy group.

[0009] Furthermore, the step of revising the basic power generation prediction model based on the historical power generation information and historical dispatch information of each distributed energy source to obtain the power generation prediction model for each distributed energy source includes: Based on the corresponding historical power generation information, obtain the deviation characteristics between each distributed energy source and the target energy source in the same group; Environmental variables are selected based on the corresponding distributed energy type, and key environmental variables are screened based on the correlation coefficient between environmental variables and deviation characteristics. Using key environmental variables as input and historical deviation rate as output, a multilayer perceptron model is trained to obtain a deviation prediction sub-model, which is then connected to the corresponding basic power generation prediction model. Based on the corresponding historical scheduling information, obtain the historical scheduling instruction parameters of each distributed energy source, and combine them with the corresponding actual execution results to establish and train the scheduling response sub-model; By connecting the scheduling response sub-model to the corresponding basic power generation prediction model, the power generation prediction models of each distributed energy source can be obtained.

[0010] Furthermore, the step of revising the basic power generation prediction model based on the historical power generation information and historical dispatch information of each distributed energy source to obtain the power generation prediction model for each distributed energy source also includes: Obtain the target forecast time period based on the predicted demand, and match the scheduling plan based on the target forecast time period; Based on the scheduling plan matching results, identify the operating scenarios of each distributed energy source within the target prediction time period; Based on the corresponding operating scenario, set correction weights for the deviation prediction sub-model and dispatch response sub-model of each distributed energy source when they are connected to the corresponding basic power generation prediction model.

[0011] Furthermore, the step of setting correction weights for the deviation prediction sub-model and scheduling response sub-model of each distributed energy source according to the corresponding operating scenario includes: When distributed energy resources are in a natural operating scenario without scheduling instructions, the correction weight of the deviation prediction sub-model is set higher than that of the scheduling response sub-model. When distributed energy resources are subject to scheduling interventions with scheduling instructions at all times, the correction weight of the scheduling response sub-model is set higher than that of the deviation prediction sub-model. When distributed energy resources are in a mixed operation scenario where scheduling instructions are present for some periods, the correction weights of the scheduling response sub-model and the deviation prediction sub-model are dynamically switched according to the execution time period of the scheduling instructions.

[0012] Furthermore, the virtual power plant integrates the power generation forecasts of various distributed energy sources and constructs an energy dispatch strategy based on dispatch objectives, including: The virtual power plant integrates the power generation forecasts of various distributed energy sources according to the forecast period, and performs resource attribute association processing on the integrated power generation forecasts based on the energy type of each distributed energy source. The virtual power plant obtains the current dispatch identity information and sets dispatch targets based on the power grid operation requirements; The objective function for energy scheduling is set according to the scheduling target. The constraints for energy scheduling are set according to the integrated power generation forecast and its resource attributes. The objective function is solved by linear programming algorithm. The energy scheduling strategy is constructed based on the solution results.

[0013] Furthermore, before solving the objective function using a linear programming algorithm, the following steps are also performed: Obtain the target forecast time period based on the predicted demand, and match the scheduling plan based on the target forecast time period; Based on the matching results, identify the prediction periods with scheduling instructions, set the corresponding policy constraints for the prediction periods according to the corresponding scheduling instructions, and add the policy constraints to the energy scheduling constraints.

[0014] A distributed energy dispatch system based on virtual power plants is used to execute the above-mentioned distributed energy dispatch method, including: The energy integration module is used to integrate distributed energy resources within the dispatch area and filter target energy sources of various energy types. The energy generation prediction module, connected to the energy integration module, is used to obtain the generation prediction model of each distributed energy source based on the target energy source of each energy type and the historical generation information and historical dispatch information corresponding to each distributed energy source, and to perform generation prediction based on the corresponding generation prediction model to obtain the generation prediction value of each distributed energy source. The virtual power plant, connected to the power generation forecasting module, is used to integrate the power generation forecasts of distributed energy resources and combine them with scheduling objectives to construct energy dispatching strategies.

[0015] Furthermore, the energy generation prediction module includes: The basic model building unit, connected to the energy integration module, is used to establish basic power generation models for each energy type based on the corresponding target energy. The model correction unit, connected to the basic model construction unit, is used to correct the corresponding basic power generation model based on the historical power generation information and historical dispatch information of each distributed energy source, and to obtain the power generation prediction model of each distributed energy source. The power generation prediction unit, connected to the model correction unit, is used to call the corresponding power generation prediction model to predict the power generation of each distributed energy source and obtain the predicted power generation value of each distributed energy source.

[0016] The beneficial effects of this invention are: (1) In order to reflect the individual differences of each distributed energy source, each distributed energy source is independently modeled. When modeling independently, distributed energy sources of the same type are first grouped into a cluster according to their energy characteristics to select target energy sources that can accurately reflect the common laws within the group. Based on the target energy sources, corresponding basic power generation prediction models are constructed to avoid repeated modeling and training, reduce the computational load of the modeling process, and improve the efficiency of independent modeling. Furthermore, the basic power generation prediction models are modified by combining the historical power generation information and historical dispatch information of each distributed energy source, so that the established power generation prediction models of each distributed energy source can retain the common laws within the group and have individual personalized characteristics, ensuring the accuracy of their prediction results. Based on the high-precision prediction values, the virtual power plant can clearly grasp the actual adjustable capacity of each distributed energy source, thereby accurately matching the dispatch target, ensuring the accuracy of the formulated dispatch strategy, and thus improving the stability of the power grid operation. (2) By clustering historical output and related environmental data, and combining center distance and data integrity weighted scores to select target energy, the high consistency of distributed energy characteristics within the energy group is ensured, as well as the typicality and data reliability of the target energy. In the process of building the basic power generation prediction model, by constructing derived features and adapting prediction models according to energy type, the basic power generation prediction model can accurately capture the core output patterns of similar energy sources, thus ensuring the representativeness and reliability of the constructed basic power generation prediction model. (3) Further, a deviation prediction sub-model is constructed by extracting deviation features and screening key environmental variables to offset the individual differences between each distributed energy source and the target energy source. At the same time, a scheduling response sub-model is constructed by combining historical scheduling information to eliminate the impact of scheduling instructions on the power generation of distributed energy sources. The correction weights of the two are dynamically set according to the scenario so that the power generation prediction model of each distributed energy source after correction can be adapted to the characteristics of the distributed energy source itself, thus ensuring the accuracy of the subsequent power generation prediction results. (4) Based on the high-precision power generation forecast obtained, the virtual power plant integrates the power generation forecast of each distributed energy source and associates the resource attributes. It sets the objective function by combining the scheduling identity and grid demand, and embeds the scheduling plan constraint of the target time period before solving to ensure that the scheduling strategy does not conflict with the existing plan. Finally, it achieves optimal resource allocation through linear programming, thus ensuring the accuracy and reliability of the scheduling strategy. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of a process of the present invention; Figure 2 This is a schematic diagram of a structure according to an embodiment of the present invention.

[0018] The module consists of: 1. Energy integration module; 2. Energy generation prediction module; 21. Basic model construction unit; 22. Model correction unit; 23. Power generation prediction unit; and 3. Virtual power plant. Detailed Implementation

[0019] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0020] Example: A distributed energy dispatching method based on virtual power plants, such as... Figure 1 As shown, it includes: The distributed energy resources within the current scheduling area are grouped, and the target energy resources corresponding to each energy group are selected according to the grouping results. Based on the historical power generation information of the corresponding target energy, construct basic power generation prediction models for each energy group; The basic power generation prediction model is revised based on the historical power generation information and historical dispatch information of each distributed energy source to obtain the power generation prediction model of each distributed energy source. The power generation forecast values ​​of each distributed energy source are obtained based on the corresponding power generation forecast model. The virtual power plant integrates the power generation forecast values ​​of each distributed energy source and constructs an energy dispatch strategy in combination with the dispatch objectives.

[0021] To address the issue of neglecting individual differences among energy sources within the same type in unified modeling methods, independent modeling is implemented for each distributed energy source to ensure the accuracy of its power generation prediction results. However, if independent modeling is carried out for each distributed energy source separately, steps such as historical data processing, model structure design, parameter training, and validation optimization must be performed during the modeling process for each distributed energy source. When there are a large number of distributed energy nodes in the scheduling area, this generates massive data processing tasks and model training computations, significantly extending the modeling cycle and causing the constructed prediction model to be unable to adapt to real-time scheduling requirements.

[0022] Therefore, this embodiment first groups the distributed energy resources in the current scheduling area to group nodes with similar characteristics into the same energy group. The basic model is built using only the target energy resources selected within the group as samples. Then, the individual is adapted through subsequent lightweight correction. This can reduce redundant calculations by using the common rules within the group and make up for individual differences by relying on the subsequent correction process. This significantly reduces the amount of computation while ensuring the accuracy of subsequent predictions.

[0023] Specifically, the step of grouping distributed energy resources within the current scheduling area and selecting target energy resources corresponding to each energy group according to the grouping results includes: Distributed energy resources are clustered based on historical output data and related environmental data to obtain several energy clusters; Calculate the center distance of each distributed energy source within each energy cluster, and score the data integrity of each distributed energy source by combining the corresponding historical output data; The scores for center distance and data integrity are weighted and summed to obtain a comprehensive score for each distributed energy source. The distributed energy source with the highest comprehensive score is selected as the target energy source for the corresponding energy group.

[0024] Before clustering, historical output data needs to be preprocessed to remove abnormal data such as equipment failures and extreme weather effects. Simultaneously, based on the energy type of each distributed energy source, corresponding associated environmental data is selected. For example, for photovoltaic distributed energy sources, parameters such as irradiance and ambient temperature are selected to determine associated environmental data; for wind power distributed energy sources, parameters such as wind speed, wind direction, and turbulence intensity are selected. In addition to the aforementioned renewable energy sources, energy storage distributed energy sources are also included. For energy storage distributed energy sources, parameters such as ambient temperature and pre-charge / discharge state of charge can be selected to obtain corresponding associated environmental data. After determining the corresponding historical output data and associated environmental data, they can be correlated using spatiotemporal alignment. Cluster analysis is then performed based on the correlated data, which effectively improves clustering accuracy compared to clustering methods based solely on geographical location or equipment type.

[0025] This embodiment specifically uses the K-means clustering algorithm to perform clustering processing on distributed energy resources. First, the elbow rule is used to determine the optimal number of clusters. Then, the corresponding historical output characteristics and associated environmental data are used as clustering variables. Through iterative calculation, the feature similarity of distributed energy resources within the same energy group is maximized, while the feature differences between energy groups are maximized. Finally, several energy groups with homogeneous characteristics are output.

[0026] After completing the corresponding clustering, for each energy group, the distance between each distributed energy source and the cluster center is calculated using the Euclidean distance formula, with the cluster center as the benchmark. The smaller the center distance, the more closely the characteristics of the distributed energy source match the common patterns within the group.

[0027] Meanwhile, the data integrity scores of each distributed energy source are scored by combining historical output data. Data coverage, data continuity, and data validity are used as scoring indicators, and the data integrity scores of each distributed energy source are obtained by weighted summation.

[0028] After determining the center distance and data integrity scores, weights for center distance and data integrity can be set according to actual needs. The center distance is then normalized, and a weighted sum is calculated based on the set weights and data integrity scores to determine the comprehensive score for each distributed energy source. Finally, within each energy group, the distributed energy source with the highest comprehensive score is selected as the target energy source, ensuring that the target energy source accurately represents the common patterns within the group and possesses a high-quality and highly reliable data foundation.

[0029] If there are ties in the overall score, the number of failures, maintenance frequency, etc. of the distributed energy sources with the same score can be further compared to select the more stable distributed energy source as the target energy source.

[0030] After identifying the target energy source in each energy group, a basic power generation prediction model representing the corresponding energy group can be constructed based on the historical power generation information of the target energy source. This model can then be used as the initial basic power generation prediction model for all distributed energy sources in the corresponding energy group. This avoids repetitive data processing and calculations during model construction, reduces the computational load of modeling, and effectively improves the efficiency of independent modeling for each distributed energy source.

[0031] The basic power generation prediction model for each energy group, constructed based on historical power generation information of the corresponding target energy, includes: The historical power generation data of the target energy source is preprocessed, and the corresponding related environmental data is retrieved and spatiotemporally aligned with it. Derived features are established based on the energy type of the target energy source, and a training sample set is constructed by combining aligned historical power generation data and related process data. Select the corresponding prediction model type based on the energy type of the target energy source, and train the selected prediction model based on the training sample set to construct the power generation prediction model for the target energy source. The power generation prediction model of the target energy is used as the basic power generation prediction model for the corresponding energy group.

[0032] First, the historical power generation data of the target energy source needs to undergo preprocessing, including outlier removal, data smoothing, and standardization. Then, the associated environmental data corresponding to the target energy source is scheduled and spatiotemporally aligned with the preprocessed historical power generation data. Specifically, the associated environmental data is matched with the historical power generation data based on the timestamps of the relevant data collection. If there is a time delay in the associated environmental data, synchronization is achieved through time offset correction. Simultaneously, the associated environmental data is matched with the target energy source based on the data collection location. If there are cross-spatial data collection instances, geographical location correction parameters such as attenuation coefficients are needed to adjust the associated environmental data, ensuring that the associated environmental data matches the actual operating environment of the target energy source.

[0033] Simultaneously, based on the differences in target energy types, targeted derived features are constructed to compensate for the shortcomings of the original data in characterizing power output patterns. Specifically, for photovoltaic (PV) target energy, derived features such as the rate of change in irradiance, temperature correction coefficient, and cloud cover duration are further established to enhance the dynamic impact of irradiance fluctuations, temperature effects, and cloud cover on PV power output. For wind power target energy, derived features such as wind speed gradient, wind direction stability, and turbulence intensity level are further established to enhance the impact of vertical wind speed distribution, wind direction changes, and turbulence on wind turbine output. For energy storage target energy, derived features such as the rate of change in state of charge (SOC), charge / discharge cycle count, and temperature-dependent capacity decay coefficient are further established to quantify the constraints of energy storage operating conditions on the charging and discharging capabilities of energy storage.

[0034] Based on this, the aligned historical power generation data is used as the model output label, and the original associated environmental data and derived features are used as input features to construct a training sample set in chronological order.

[0035] Based on the output characteristics and data patterns of the target energy source, an appropriate prediction model type is selected to accurately capture the factors influencing output. For example, for photovoltaic (PV) energy sources, whose output exhibits a strongly nonlinear relationship with sunlight and temperature, and is subject to instantaneous fluctuations due to cloud cover, a hybrid model consisting of convolutional neural networks (CNNs) and long short-term memory (LSTM) networks can be chosen as the corresponding prediction model. The CNN extracts local features from the associated environmental data, while the LSTM captures the temporal dependencies of the power generation data. For other types of target energy sources, an appropriate prediction model type can also be selected based on actual needs.

[0036] Once the prediction model type is determined, the model can be trained using the established training sample set, and the model parameters can be optimized and adjusted to obtain the power generation prediction model for the target energy source. This target energy source's power generation prediction model can then be directly used as the basic power generation prediction model for its energy cluster. Other distributed energy sources within the energy cluster only need to have deviation corrections and dispatch response corrections added to the basic power generation prediction model to quickly construct their own individual power generation prediction models, effectively improving modeling efficiency.

[0037] The step of revising the basic power generation prediction model based on the historical power generation information and historical dispatch information of each distributed energy source to obtain the power generation prediction model for each distributed energy source includes: Based on the corresponding historical power generation information, obtain the deviation characteristics between each distributed energy source and the target energy source in the same group; Environmental variables are selected based on the corresponding distributed energy type, and key environmental variables are screened based on the correlation coefficient between environmental variables and deviation characteristics. Using key environmental variables as input and historical deviation rate as output, a multilayer perceptron model is trained to obtain a deviation prediction sub-model, which is then connected to the corresponding basic power generation prediction model. Based on the corresponding historical scheduling information, obtain the historical scheduling instruction parameters of each distributed energy source, and combine them with the corresponding actual execution results to establish and train the scheduling response sub-model; By connecting the scheduling response sub-model to the corresponding basic power generation prediction model, the power generation prediction models of each distributed energy source can be obtained.

[0038] Based on the historical power generation data of the target energy within the same energy cluster, the historical power generation data of the distributed energy to be corrected within the cluster are retrieved simultaneously. The absolute and relative deviations for each time period are calculated. Then, according to the predetermined time periods such as daily and weekly, the mean and standard deviation of the relative deviation, the peak and valley time deviation values, etc. are calculated. At the same time, the abnormal deviation periods and the corresponding environmental scenarios are marked to construct deviation characteristics related to time period, deviation and scenario.

[0039] Next, select corresponding environmental variables based on the energy type of distributed energy. For example, for photovoltaic power, select environmental variables such as irradiance and ambient temperature; for wind power, select environmental variables such as wind speed and wind direction. For each candidate environmental variable, calculate the correlation coefficient between the environmental variable and the deviation characteristics, and select the environmental variables with high correlation coefficients as key environmental variables.

[0040] Using the selected key environmental variables as input features, the historical deviation rate of the distributed energy source to be corrected is used as the output label, and a multilayer perceptron model structure is set. The number of nodes in the input layer of the multilayer perceptron model is consistent with the number of key environmental variables. Two to three hidden layers can be set, with each layer having two to three times the number of nodes as the input layer. The output layer has one node to output the predicted deviation rate. The activation functions are ReLU and Linear functions to avoid gradient vanishing and ensure that the output conforms to the continuous value characteristic of the deviation rate.

[0041] The multilayer perceptron model is then trained using the previously obtained key environmental variables and historical deviation rates, resulting in the final deviation prediction sub-model for Tiger Hill.

[0042] When the deviation prediction sub-model is connected to the corresponding basic power generation prediction model, the basic power generation prediction model first outputs the initial predicted output of the distributed energy source to be corrected. The deviation prediction sub-model then simultaneously inputs the real-time key environmental variables of the distributed energy source to output the prediction deviation rate. The output prediction deviation rate is then used to adjust the initial predicted output of the basic power generation prediction model. That is, the corrected predicted output = initial predicted output × (1 - prediction deviation rate). If the prediction deviation rate is negative, the corrected predicted output = initial predicted output + |deviation rate × initial predicted output|.

[0043] Simultaneously, historical dispatch information of the distributed energy sources to be corrected is extracted, and the deviation between the actual output of the distributed energy sources after the dispatch instructions are issued and the natural output without instructions is obtained based on the dispatch instruction type, dispatch instruction parameters and actual execution results involved in the historical dispatch information.

[0044] In extracting the deviation between the actual output of distributed energy sources after the issuance of dispatch instructions and the natural output without instructions, the actual output before and after the issuance of dispatch instructions is first obtained based on historical dispatch information. Then, the command parameters such as the direction of output control, the control amplitude requirements, and the control implementation of the dispatch instructions are determined. The actual output before and after the issuance of dispatch instructions and the command parameters are then characterized.

[0045] Then, the gradient boosting decision tree model is selected as the scheduling response sub-model. The output deviation of the command intervention is used as the model output label. The corresponding loss function is set, and the scheduling response sub-model is trained based on the actual output before and after the issuance of the acquired feature-processed scheduling command and the command parameters. This allows the model to capture the influence of the scheduling command on the output and transform this influence into a correction rule, which is then embedded into the basic power generation model to eliminate the prediction bias that only predicts the output based on the natural environment and ignores the command intervention.

[0046] When the dispatch response sub-model is connected to the corresponding basic power generation prediction model, the basic power generation prediction model will first output the initial predicted output of the distributed energy to be corrected. The deviation prediction sub-model will then simultaneously input the existing dispatch instructions of the distributed energy to output the predicted value of the deviation of the output under the instructions. The initial predicted output will then be adjusted according to the predicted value of the deviation of the output under the instructions. That is, the corrected predicted output = the initial predicted output + the predicted value of the deviation of the output under the instructions.

[0047] The core factors affecting distributed energy output differ across operating scenarios. In natural operating scenarios without dispatch instructions, output deviations primarily stem from individual characteristics and environmental fluctuations, requiring a focus on correcting the deviation prediction sub-model. In scenarios with full-time dispatch intervention, output deviations are more attributable to the effectiveness of instruction execution, necessitating priority in correcting the dispatch response sub-model. Therefore, the correction priorities for the deviation prediction sub-model and the dispatch response sub-model differ across operating scenarios. To improve the correction process of the basic power generation prediction model to better align with actual dispatch needs and avoid correction deviations caused by adapting a single weight to all scenarios, the following steps are performed when constructing the deviation prediction sub-model and the dispatch response sub-model and integrating them into the basic power generation prediction model to obtain the power generation prediction models for each distributed energy source: Obtain the target forecast time period based on the predicted demand, and match the scheduling plan based on the target forecast time period; Based on the scheduling plan matching results, identify the operating scenarios of each distributed energy source within the target prediction time period; Based on the corresponding operating scenario, set correction weights for the deviation prediction sub-model and dispatch response sub-model of each distributed energy source when they are connected to the corresponding basic power generation prediction model.

[0048] The target forecast time period is determined based on the predicted demand. The matching condition is that the start timestamp of the target forecast time period is not earlier than the start time of the planned execution time period and the end timestamp is not later than the end time of the planned execution time period. The scheduling plans that overlap with the target time period are then filtered out. The relevant scheduling plan content is then filtered out from the scheduling plans based on parameters such as the device ID and name corresponding to the distributed energy.

[0049] Then, based on the matching scheduling plan, the operating scenarios for each distributed energy source are determined.

[0050] If, within the target prediction period, a distributed energy source has no matching scheduling plan, or if the content of the scheduling plan is unrelated to the output, then the distributed energy source can be considered to be in a natural operating scenario without scheduling instructions.

[0051] If a distributed energy source has scheduling plans related to its output in all time granularities within the target prediction period, or if the proportion of time granularities of scheduling plans related to output exceeds a predetermined value, then the distributed energy source can be considered to be in a scheduling intervention scenario with scheduling instructions at all times.

[0052] If a distributed energy source has both time granularities without scheduling instructions and time granularities with scheduling plans related to output within the target prediction time period, and the proportion of time granularities with scheduling plans related to output does not exceed a predetermined value, then the distributed energy source can be regarded as operating in a mixed scenario where scheduling instructions exist in some time periods.

[0053] Then, adjust the weights of the deviation prediction sub-model and the scheduling response sub-model according to the scenario type of the corresponding operation scenario.

[0054] The step of setting correction weights for the deviation prediction sub-model and scheduling response sub-model of each distributed energy source according to the corresponding operating scenario includes: When distributed energy resources are in a natural operating scenario without scheduling instructions, the correction weight of the deviation prediction sub-model is set higher than that of the scheduling response sub-model. When distributed energy resources are subject to scheduling interventions with scheduling instructions at all times, the correction weight of the scheduling response sub-model is set higher than that of the deviation prediction sub-model. When distributed energy resources are in a mixed operation scenario where scheduling instructions are present for some periods, the correction weights of the scheduling response sub-model and the deviation prediction sub-model are dynamically switched according to the execution time period of the scheduling instructions.

[0055] For dynamic switching of correction weights in mixed operation scenarios, the presence or absence of scheduling instructions within a time period can be used as the switching condition. During time periods with scheduling instructions, the correction weight settings under the scheduling intervention scenario are switched to, while during time periods without scheduling instructions, the correction weight settings under the natural operation scenario are switched to.

[0056] Once the power generation prediction model for all distributed energy sources is determined, the corresponding power generation prediction can be carried out based on the current real-time environmental data, the target prediction time period scheduling plan, etc., and the power generation prediction value of each distributed energy source can be obtained.

[0057] After obtaining the power generation forecasts for each distributed energy source, these forecasts can be integrated through a virtual power plant and combined with scheduling objectives to construct an energy dispatch strategy, including: The virtual power plant integrates the power generation forecasts of various distributed energy sources according to the forecast period, and performs resource attribute association processing on the integrated power generation forecasts based on the energy type of each distributed energy source. The virtual power plant obtains the current dispatch identity information and sets dispatch targets based on the power grid operation requirements; The objective function for energy scheduling is set according to the scheduling target. The constraints for energy scheduling are set according to the integrated power generation forecast and its resource attributes. The objective function is solved by linear programming algorithm. The energy scheduling strategy is constructed based on the solution results.

[0058] The virtual power plant extracts the generation forecast values ​​of all distributed energy sources in batches according to the target forecast period. It then determines whether the time granularity of each distributed energy source's generation forecast value is consistent with the target forecast period. If granularity differences exist, it further unifies the corresponding time granularity through mean aggregation or interpolation to ensure the consistency of the time dimension of the integrated data. Simultaneously, it extracts the confidence intervals of the corresponding generation forecast values, removes abnormal generation forecast values ​​whose confidence intervals exceed a preset threshold, and fills the gaps with the average forecast values ​​of similar distributed energy sources within the same energy group to ensure the reliability of the integrated data. Finally, it matches resource attributes to the integrated generation forecast values ​​based on the type of each distributed energy source. For example, for intermittent energy types such as photovoltaic and wind power, it associates resource attributes such as output fluctuation coefficients and grid connection technology constraints; for energy storage types, it associates resource attributes such as current SOC and charging / discharging power limits.

[0059] Virtual power plants assume different dispatch roles during grid dispatching, depending on demand, including grid auxiliary dispatching entities, electricity market participants, and user energy service providers. The core dispatching objectives differ for each role: grid auxiliary dispatching entities prioritize responding to grid supply and demand balance needs; electricity market participants balance the economics of power purchase and sale with market rule constraints; and user energy service providers prioritize minimizing overall energy costs for users while ensuring power supply reliability.

[0060] Therefore, the current dispatch identity information is first determined based on the interface between the virtual power plant and the power grid dispatch center. Then, the power grid operation demand for the target forecast period pushed by the power grid dispatch center is retrieved, such as load forecast, capacity demand, renewable energy consumption rate requirements, etc., to determine the specific dispatch target. The dispatch target is then quantified into executable indicators, and corresponding priority weights are added to the converted indicators.

[0061] Then, based on the converted execution indicators and their corresponding priority weights, the objective function for energy dispatch is set. Taking the role of grid auxiliary dispatch as an example, its quantified executable indicators are peak-shaving capacity compliance rate, frequency stability, renewable energy absorption rate, and grid loss rate. The objective function is to maximize the weighted sum of these executable indicators.

[0062] Then, set corresponding constraints based on the resource attributes, such as power balance constraints, renewable energy consumption constraints, intermittent energy constraints, and energy storage constraints.

[0063] Meanwhile, if a matching scheduling plan already exists within the target prediction period, its corresponding scheduling plan content needs to be converted into specific constraints to add constraint conditions, ensuring that the subsequent scheduling strategy will not conflict with the existing scheduling plan.

[0064] Specifically, the target forecast time period is obtained based on the predicted demand, and a scheduling plan is matched based on the target forecast time period; Based on the matching results, identify the prediction periods with scheduling instructions, set the corresponding policy constraints for the prediction periods according to the corresponding scheduling instructions, and add the policy constraints to the energy scheduling constraints.

[0065] After determining the objective function and constraints, they are input into the linear programming solver. The linear programming solver performs iterative calculations and outputs the optimal variable values ​​that make the objective function optimal, provided that all constraints are met. Examples of optimal values ​​include the grid-connected power of photovoltaic distributed energy sources and the charging and discharging time and power of energy storage distributed energy sources.

[0066] Based on the solution results and the correspondence between time granularity and energy type, the results are then broken down into specific scheduling instructions to determine the output scheduling content of each distributed energy source in each forecast period, thus forming the final energy scheduling strategy.

[0067] Finally, the energy dispatch strategy is pushed to the control terminals of each distributed energy source and the power grid dispatch center, which enables efficient coordinated control of each distributed energy source within the dispatch area.

[0068] Another aspect of this embodiment also provides a distributed energy dispatching system based on virtual power plants, such as... Figure 2 As shown, it includes: Energy integration module 1 is used to integrate distributed energy resources within the dispatch area and filter target energy sources of various energy types; The energy generation prediction module 2 is connected to the energy integration module. It is used to obtain the generation prediction model of each distributed energy source based on the target energy of each energy type and the historical generation information and historical dispatch information corresponding to each distributed energy source, and to perform generation prediction based on the corresponding generation prediction model to obtain the generation prediction value of each distributed energy source. Virtual power plant 3, connected to the power generation forecast module, is used to integrate the power generation forecast values ​​of distributed energy sources and construct energy dispatch strategies in conjunction with dispatch objectives.

[0069] The energy generation prediction model includes: The basic model building unit 21 is connected to the energy integration module and is used to establish basic power generation models for each energy type based on the corresponding target energy. The model correction unit 22 is connected to the basic model construction unit and is used to correct the corresponding basic power generation model based on the historical power generation information and historical dispatch information of each distributed energy source, so as to obtain the power generation prediction model of each distributed energy source. The power generation prediction unit 23 is connected to the model correction unit and is used to call the corresponding power generation prediction model to predict the power generation of each distributed energy source and obtain the power generation prediction value of each distributed energy source.

[0070] The energy integration module, the energy generation prediction module, and the basic model building unit, model correction unit, and power generation prediction unit contained in the energy generation prediction module are all high-performance data analysis and processing devices such as calculators, servers, and microcontrollers. They can be integrated into a virtual power plant and carry the relevant algorithms for cluster analysis, target energy screening, basic power generation model construction, basic power generation model correction, and power generation prediction. Moreover, each of the above modules and units has a corresponding data access port, which can retrieve the data required during the algorithm execution process.

[0071] Through the energy integration module, energy generation prediction module, and virtual power plant, accurate and efficient power generation prediction can be made for each distributed energy source within the dispatch area, thereby constructing an accurate energy dispatch strategy and ensuring the stability of the power grid operation within the dispatch area.

[0072] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any way. Other variations and modifications are possible without departing from the technical solutions described in the claims.

Claims

1. A distributed energy dispatching method based on virtual power plants, characterized in that, include: The distributed energy resources within the current scheduling area are grouped, and the target energy resources corresponding to each energy group are selected according to the grouping results. Based on the historical power generation information of the corresponding target energy, construct basic power generation prediction models for each energy group; The basic power generation prediction model is revised based on the historical power generation information and historical dispatch information of each distributed energy source to obtain the power generation prediction model of each distributed energy source. The power generation forecast values ​​of each distributed energy source are obtained based on the corresponding power generation forecast model. The virtual power plant integrates the power generation forecast values ​​of each distributed energy source and constructs an energy dispatch strategy in combination with the dispatch objectives.

2. The distributed energy dispatching method based on virtual power plants according to claim 1, characterized in that, The step of grouping distributed energy resources within the current scheduling area and selecting target energy resources corresponding to each energy group according to the grouping results includes: Distributed energy resources are clustered based on historical output data and related environmental data to obtain several energy clusters; Calculate the center distance of each distributed energy source within each energy cluster, and score the data integrity of each distributed energy source by combining the corresponding historical output data; The scores for center distance and data integrity are weighted and summed to obtain a comprehensive score for each distributed energy source. The distributed energy source with the highest comprehensive score is selected as the target energy source for the corresponding energy group.

3. The distributed energy dispatching method based on virtual power plants according to claim 1, characterized in that, The basic power generation prediction model for each energy group, constructed based on historical power generation information of the corresponding target energy, includes: The historical power generation data of the target energy source is preprocessed, and the corresponding related environmental data is retrieved and spatiotemporally aligned with it. Derived features are established based on the energy type of the target energy source, and a training sample set is constructed by combining aligned historical power generation data and related process data. Select the corresponding prediction model type based on the energy type of the target energy source, and train the selected prediction model based on the training sample set to construct the power generation prediction model for the target energy source. The power generation prediction model of the target energy is used as the basic power generation prediction model for the corresponding energy group.

4. The distributed energy dispatching method based on virtual power plants according to claim 1, characterized in that, The step of revising the basic power generation prediction model based on the historical power generation information and historical dispatch information of each distributed energy source to obtain the power generation prediction model for each distributed energy source includes: Based on the corresponding historical power generation information, obtain the deviation characteristics between each distributed energy source and the target energy source in the same group; Environmental variables are selected based on the corresponding distributed energy type, and key environmental variables are screened based on the correlation coefficient between environmental variables and deviation characteristics. Using key environmental variables as input and historical deviation rate as output, a multilayer perceptron model is trained to obtain a deviation prediction sub-model, which is then connected to the corresponding basic power generation prediction model. Based on the corresponding historical scheduling information, obtain the historical scheduling instruction parameters of each distributed energy source, and combine them with the corresponding actual execution results to establish and train the scheduling response sub-model; By connecting the scheduling response sub-model to the corresponding basic power generation prediction model, the power generation prediction models of each distributed energy source can be obtained.

5. The distributed energy dispatching method based on virtual power plants according to claim 4, characterized in that, The step of revising the basic power generation prediction model based on the historical power generation information and historical dispatch information of each distributed energy source to obtain the power generation prediction model for each distributed energy source also includes: Obtain the target forecast time period based on the predicted demand, and match the scheduling plan based on the target forecast time period; Based on the scheduling plan matching results, identify the operating scenarios of each distributed energy source within the target prediction time period; Based on the corresponding operating scenario, set correction weights for the deviation prediction sub-model and dispatch response sub-model of each distributed energy source when they are connected to the corresponding basic power generation prediction model.

6. The distributed energy dispatching method based on virtual power plants according to claim 5, characterized in that, The step of setting correction weights for the deviation prediction sub-model and dispatch response sub-model of each distributed energy source when they are connected to the corresponding basic power generation prediction model, based on the corresponding operating scenario, includes: When distributed energy resources are in a natural operating scenario without scheduling instructions, the correction weight of the deviation prediction sub-model is set higher than that of the scheduling response sub-model. When distributed energy resources are subject to scheduling interventions with scheduling instructions at all times, the correction weight of the scheduling response sub-model is set higher than that of the deviation prediction sub-model. When distributed energy resources are in a mixed operation scenario where scheduling instructions are present for some periods, the correction weights of the scheduling response sub-model and the deviation prediction sub-model are dynamically switched according to the execution time period of the scheduling instructions.

7. The distributed energy dispatching method based on virtual power plants according to claim 1, characterized in that, The virtual power plant integrates the power generation forecasts of various distributed energy sources and constructs an energy dispatch strategy based on dispatch objectives, including: The virtual power plant integrates the power generation forecasts of various distributed energy sources according to the forecast period, and performs resource attribute association processing on the integrated power generation forecasts based on the energy type of each distributed energy source. The virtual power plant obtains the current dispatch identity information and sets dispatch targets based on the power grid operation requirements; The objective function for energy scheduling is set according to the scheduling target. The constraints for energy scheduling are set according to the integrated power generation forecast and its resource attributes. The objective function is solved by linear programming algorithm. The energy scheduling strategy is constructed based on the solution results.

8. The distributed energy dispatching method based on virtual power plants according to claim 7, characterized in that, Before solving the objective function using a linear programming algorithm, the following steps are also performed: Obtain the target forecast time period based on the predicted demand, and match the scheduling plan based on the target forecast time period; Based on the matching results, identify the prediction periods with scheduling instructions, set the corresponding policy constraints for the prediction periods according to the corresponding scheduling instructions, and add the policy constraints to the energy scheduling constraints.

9. A distributed energy dispatching system based on virtual power plants, used to execute the distributed energy dispatching method based on virtual power plants as described in any one of claims 1 to 8, characterized in that, include: The energy integration module is used to integrate distributed energy resources within the dispatch area and filter target energy sources of various energy types. The energy generation prediction module, connected to the energy integration module, is used to obtain the generation prediction model of each distributed energy source based on the target energy of each energy type and the historical generation information and historical dispatch information corresponding to each distributed energy source, and to perform generation prediction based on the corresponding generation prediction model to obtain the generation prediction value of each distributed energy source. The virtual power plant, connected to the power generation forecasting module, is used to integrate the power generation forecasts of distributed energy resources and combine them with scheduling objectives to construct energy dispatching strategies.

10. The distributed energy dispatching system based on a virtual power plant according to claim 9, characterized in that, The energy generation prediction module includes: The basic model building unit, connected to the energy integration module, is used to establish basic power generation models for each energy type based on the corresponding target energy. The model correction unit, connected to the basic model construction unit, is used to correct the corresponding basic power generation model based on the historical power generation information and historical dispatch information of each distributed energy source, and to obtain the power generation prediction model of each distributed energy source. The power generation prediction unit, connected to the model correction unit, is used to call the corresponding power generation prediction model to predict the power generation of each distributed energy source and obtain the predicted power generation value of each distributed energy source.

Images