Double-model-driven virtual power plant electric power resource disposal method

Abstract

The invention discloses a double-model-driven virtual power plant power resource disposal method, which comprises the following steps of: decoupling energy resource data from a specific disposal scheme, judging the current real-time scheduling capability by utilizing an energy resource model, inputting the real-time scheduling capability and real-time operation state data into an energy disposal model, and performing real-time scheduling on the real-time scheduling capability and the real-time operation state data; and outputting a real-time disposal scheme of electric power energy scheduling. According to the method, the black box problem of a single model is avoided, the scheduling capability change caused by the change of the energy resource data can be clearly understood, the energy disposal model can be judged to ensure the feasibility of the disposal scheme, meanwhile, the resource scheduling capability judgment result can be visually verified, and the energy resource model can be independently optimized even if errors exist.

Description

Technical Field

[0001] The present invention relates to the field of electric power dispatching, and in particular to a dual-model driven virtual power plant electric power resource disposal method. Background Art

[0002] At present, the power resource disposal of virtual power plants usually utilizes various neural network models, and the relevant data involved include energy resource data, operating status data, etc. Although operating status data is the most important basis for virtual power plants to dispose of power energy, the disposal plan is based on energy resource data, and in the model training process, the logic of generating disposal plans is also based on energy resource data. Therefore, when the energy resource data changes, it is equivalent to a change in the basic composition of the virtual power plant. At this time, it may be difficult to cope with it by relying solely on the generalization ability of the original model, and it is easy to cause unreasonable power dispatch of the disposal plan.

[0003] Therefore, if changes in energy resource data change some of the original properties of the virtual power plant, these changes may cause the model performance to deteriorate and make it no longer accurately reflect the current energy resource status. At this time, the model must be retrained under the existing technology framework to maintain the expected accuracy and adaptability. Retraining requires a lot of computing resources and time costs.

[0004] In summary, when energy resource data changes, depending on the nature and magnitude of the change and the characteristics of the model itself, in order to maintain the original accuracy, it is usually necessary to retrain the model, which is costly and inefficient. Summary of the invention

[0005] In view of the problem that the energy resource data corresponding to the virtual power plant in the prior art may cause the original model to become invalid when it changes, resulting in unreasonable power resource disposal, the present invention provides a dual-model driven virtual power plant power resource disposal method, which decouples the energy resource data from the specific disposal plan, uses the energy resource model to judge the current real-time dispatching capacity, and then inputs the real-time dispatching capacity and real-time operating status data into the energy disposal model to output a real-time disposal plan for power energy dispatch. The present invention avoids the black box problem of a single model, can clearly understand the changes in dispatching capacity caused by changes in energy resource data, can be used for energy disposal model to make judgments to ensure the feasibility of the disposal plan, and can intuitively verify the resource dispatching capacity judgment results. Even if there are errors, the energy resource model can be optimized separately.

[0006] The following is the technical solution of the present invention.

[0007] A dual-model driven virtual power plant power resource disposal method comprises the following steps: S1: Obtain real-time operation status data and real-time energy resource data of the virtual power plant; S2: Input the real-time energy resource data into the energy resource model based on machine learning trained by the historical energy resource data and the corresponding historical dispatching capabilities to obtain the real-time dispatching capabilities of the virtual power plant; S3: Input the real-time dispatch capability and real-time operation status data into the energy disposal model based on deep learning trained by historical operation status data, historical dispatch capability and historical disposal plans, and output the real-time disposal plan of the virtual power plant.

[0008] In the present invention, the training of the energy disposal model is decoupled from the energy resource data, so that the generation of the final disposal plan has no direct correlation with the energy resource data, and the energy resource model takes over the energy resource data and converts it into a scheduling capability that is readable by the energy disposal model. Therefore, the direct impact of changes in energy resource data on the energy disposal model is eliminated, and the energy resource model performs buffering and conversion, so that the energy disposal model can focus more on the connection between scheduling capabilities and disposal plans, and will not pay too much attention to energy resource data, which leads to problems existing in the prior art.

[0009] Therefore, the present invention uses an energy resource model to output real-time scheduling capability based on real-time energy resource data, so that changes in the scheduling capability of the virtual power plant can be intuitively seen, and it also serves as one of the input conditions of the energy disposal model. Since the energy disposal model does not directly use energy resource data during training, but uses historical operating status data, historical scheduling capabilities, and historical disposal plans, the impact of the energy resource data is isolated to the energy resource model. As long as the output of the energy resource model is reliable, the output of the disposal plan can be guaranteed to be unaffected.

[0010] In the present invention, changes in energy resource data are judged by the energy resource model, so the intermediate process that has the greatest impact on the disposal plan is no longer a black box. According to the results, it can be clearly understood whether there is a problem in the link of judging the scheduling capability or in the link of generating the disposal plan. Therefore, even if an error occurs, it is no longer necessary to retrain all of them, thereby improving execution efficiency and accuracy.

[0011] In addition, the machine learning algorithm in the present invention is more suitable for the first half of the process of generating disposal solutions, i.e., the energy resource model of the present application, due to its strong interpretability, strong ability to process discrete data, high computational efficiency, and easy tuning; while the deep learning algorithm is more suitable for the second half of the process of generating disposal solutions, i.e., the energy disposal model of the present application, due to its advantages of strong nonlinear modeling ability, strong self-learning ability, strong generalization ability, and the ability to capture long-term dependencies. This combination can give full play to the advantages of each algorithm and improve the efficiency and accuracy of the disposal of power resources in the virtual power plant.

[0012] Preferably, in S2 and S3, the steps of acquiring training data required for the energy resource model and the energy disposal model include: A1: acquiring the operation records of the virtual power plant, extracting the historical disposal plans in the operation records, and associating the historical energy resource data and historical operation status data corresponding to each historical disposal plan; A2: Determine the historical dispatch capacity corresponding to each energy resource based on historical disposal plans and historical energy resource data.

[0013] Preferably, the historical energy resource data includes: The number, types, and operating parameters of distributed generation facilities; the number, types, and operating parameters of distributed energy storage systems; the number, types, and operating parameters of controllable loads.

[0014] Preferably, A1: obtaining the operation record of the virtual power plant, extracting the historical disposal plans in the operation record, and associating the historical energy resource data and historical operation status data corresponding to each historical disposal plan, including: Obtain operation records from the virtual power plant management system using API interfaces or data capture; Use data parsing algorithms to extract historical disposal plans from acquired operation records; According to the time of the historical disposal plan, the historical energy resource data and historical operation status data corresponding to the time are determined, and associated with the historical disposal plan. The associated data is verified and cleaned, and the associated data is stored in a relational database.

[0015] Preferably, A2: determining the historical dispatching capacity corresponding to each energy resource according to the historical disposal scheme and the historical energy resource data, comprises: Statistics and calculations are made on the dispatch performance of each energy resource under different historical disposal plans to obtain the historical dispatch priority and historical dispatch value of each energy resource; Statistics and calculations are made on the dispatch performance of different energy resources in different historical disposal plans to obtain the complementarity and substitutability between different energy resources and establish the complementary or substitutive relationship between different energy resources; Based on the historical scheduling priority and historical scheduling value of each energy resource, as well as the complementary or substitution relationship between different energy resources, a structured scheduling information database representing the historical scheduling capacity is constructed.

[0016] In the present invention, by counting and calculating the performance of each energy resource under different historical disposal schemes, the frequency, value, etc. of the energy resource being dispatched in history can be obtained, and then the historical dispatch priority and historical dispatch value can be determined to provide data support for subsequent resource allocation and scheduling strategies. At the same time, by counting and calculating the performance of different energy resources in different historical disposal schemes, the complementarity and substitutability between them can be obtained, where complementarity means that certain energy resources can enhance each other during scheduling, and substitutability means that in some cases, one energy resource can replace another energy resource to meet energy needs.

[0017] Preferably, the calculation method of the historical scheduling priority includes: According to the statistical results of the scheduling performance of each energy resource under different historical disposal plans, the historical scheduling times n of each energy resource are obtained. According to the number of historical disposal plans m, Y=n / m is calculated to obtain the historical scheduling rate Y, where the higher the historical scheduling rate Y is, the higher the scheduling priority is.

[0018] Preferably, the training process of the energy resource model includes: A primary energy resource model is established based on a decision tree or random forest algorithm, and the primary energy resource model is trained using historical energy resource data and corresponding historical scheduling capabilities to obtain a trained energy resource model.

[0019] Preferably, the training process of the energy disposal model includes: A primary energy disposal model is established based on the long short-term memory network algorithm. The primary energy disposal model is trained using historical operating status data, historical scheduling capabilities, and historical disposal plans to obtain a trained energy disposal model.

[0020] Preferably, the S2 further comprises: According to the data dimension of the real-time scheduling capability, a multi-dimensional vector parameter is set, and the real-time scheduling capability is recorded in the form of a multi-dimensional vector parameter; The cosine similarity of the multidimensional vector parameters at the previous and next moments is calculated. If the cosine similarity is less than the preset similarity, the energy resource model is updated.

[0021] The present invention also provides a dual-model driven virtual power plant power resource disposal system. When the dual-model driven virtual power plant power resource disposal system is in operation, the aforementioned dual-model driven virtual power plant power resource disposal method is executed.

[0022] The present invention also provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor calls the computer program in the memory, the steps of the above-mentioned dual-model driven virtual power plant power resource disposal method are implemented.

[0023] The present invention also provides a storage medium, in which computer executable instructions are stored. When the computer executable instructions are loaded and executed by a processor, the steps of the above-mentioned dual-model driven virtual power plant power resource disposal method are implemented.

[0024] The substantial effects of the present invention include: By introducing the dual-model architecture of the energy resource model and the energy disposal model, the present invention realizes the decoupling of energy resource data from specific disposal solutions. This design avoids the black box problem of a single model, making the impact of changes in energy resource data on disposal solutions controllable and predictable, and improving the stability and reliability of the system. The machine learning algorithm in the present invention is more suitable for the first half of the disposal solution generation process, that is, the energy resource model of the present application, because of its strong interpretability, strong ability to process discrete data, high computational efficiency and easy tuning; while the deep learning algorithm is more suitable for the second half of the disposal solution generation process, that is, the energy disposal model of the present application, because of its advantages such as strong nonlinear modeling ability, strong self-learning ability, strong generalization ability and the ability to capture long-term dependencies. This combination method can give full play to the advantages of each algorithm and improve the efficiency and accuracy of the virtual power plant power resource disposal.

[0025] The present invention uses the energy resource model to process real-time energy resource data and output real-time dispatching capabilities, providing accurate and reliable input conditions for the energy disposal model. Since the energy disposal model does not directly use energy resource data during training, but uses historical operating status data, historical dispatching capabilities, and historical disposal plans, it effectively isolates the direct impact of energy resource data on the disposal model, allowing the disposal model to focus more on the connection between dispatching capabilities and disposal plans.

[0026] In addition, by counting and calculating the performance of each energy resource under different historical disposal schemes, the present invention obtains the historical scheduling priority and historical scheduling value of the energy resource, providing strong data support for subsequent resource allocation and scheduling strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution will be clearly and completely described below in combination with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] It should be understood that in various embodiments of the present invention, the size of the sequence number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0030] It should be understood that in the present invention, "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0031] It should be understood that in the present invention, "plurality" refers to two or more than two. "And / or" is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "Contains A, B and C", "Contains A, B, C" means that A, B, and C are all included, "Contains A, B or C" means that one of A, B, and C is included, and "Contains A, B and / or C" means that any one, any two, or any three of A, B, and C are included.

[0032] The technical solution of the present invention is described in detail with specific embodiments below. The embodiments may be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0033] Embodiment 1: A dual-model driven virtual power plant power resource disposal method, such as Figure 1 As shown, the following steps are included: S1: Obtain real-time operation status data and real-time energy resource data of the virtual power plant; S2: Input the real-time energy resource data into the energy resource model based on machine learning trained by the historical energy resource data and the corresponding historical dispatching capabilities to obtain the real-time dispatching capabilities of the virtual power plant; S3: Input the real-time dispatch capability and real-time operation status data into the energy disposal model based on deep learning trained by historical operation status data, historical dispatch capability and historical disposal plans, and output the real-time disposal plan of the virtual power plant.

[0034] In this embodiment, the training of the energy disposal model is decoupled from the energy resource data, so that the generation of the final disposal plan has no direct connection with the energy resource data, and the energy resource model takes over the energy resource data and converts it into a scheduling capability that is readable by the energy disposal model. Therefore, the direct impact of changes in energy resource data on the energy disposal model is eliminated, and the energy resource model performs buffering and conversion, so that the energy disposal model can focus more on the connection between scheduling capabilities and disposal plans, and will not pay too much attention to energy resource data, which leads to problems existing in the prior art.

[0035] Therefore, this embodiment uses the energy resource model to output real-time scheduling capabilities based on real-time energy resource data, so that changes in the scheduling capabilities of the virtual power plant can be intuitively seen, and it also serves as one of the input conditions of the energy disposal model. Since the energy disposal model does not directly use energy resource data during training, but uses historical operating status data, historical scheduling capabilities, and historical disposal plans, the impact of the energy resource data is isolated to the energy resource model. As long as the output of the energy resource model is reliable, the output of the disposal plan can be guaranteed to be unaffected.

[0036] In this embodiment, the changes in energy resource data are judged by the energy resource model. Therefore, the intermediate process that has the greatest impact on the disposal plan is no longer a black box. According to the results, it can be clearly understood whether there is a problem in the link of judging the scheduling capacity or in the link of generating the disposal plan. Therefore, even if an error occurs, it is no longer necessary to retrain all of them, which improves the execution efficiency and accuracy.

[0037] The energy resource model and energy disposal model mentioned in S2 and S3 are important contents for implementing this embodiment. The training data acquisition steps required by them include: A1: Obtain the operation records of the virtual power plant, extract the historical disposal plans in the operation records, and associate the historical energy resource data and historical operation status data corresponding to each historical disposal plan, including: Use API interfaces or data capture to obtain operation records from the virtual power plant management system; for example, obtain the operation records of the virtual power plant in the past year in real time through the API interface, including daily power generation, energy storage and load data; The historical disposal plans are extracted from the acquired operation records by using data analysis algorithms; for example, the operation records may contain multiple daily or emergency dispatch plans that were actually adopted, and the data analysis algorithms can be used to identify and extract these emergency dispatch plans as historical disposal plans; According to the time of the historical disposal plan, the historical energy resource data and historical operation status data corresponding to the time are determined, and the historical disposal plan is associated with the data, and the associated data is verified and cleaned, and the associated data is stored in a relational database. For example, for an emergency dispatch plan adopted due to a sudden drop in wind power generation, it is necessary to associate the wind power generation data at that time, the charging and discharging status data of the energy storage system, the load data of the power grid, and other related information.

[0038] In this embodiment, the historical energy resource data includes: the number, type, and operating parameters of distributed generation facilities; the number, type, and operating parameters of distributed energy storage systems; and the number, type, and operating parameters of controllable loads. For example, a virtual power plant may contain 10 wind power generation facilities, 5 photovoltaic power generation facilities, 5 energy storage stations, 11 industrial production lines, etc. The number, type (such as wind power or photovoltaic) and operating parameters (such as rated power, power generation efficiency, etc.) of each facility are recorded in detail.

[0039] A2: Determine the historical dispatch capacity corresponding to each energy resource based on historical disposal plans and historical energy resource data, including: Statistics and calculations are made on the dispatch performance of each energy resource under different historical disposal plans to obtain the historical dispatch priority and historical dispatch value of each energy resource; Statistics and calculations are made on the dispatch performance of different energy resources in different historical disposal plans to obtain the complementarity and substitutability between different energy resources and establish the complementary or substitutive relationship between different energy resources; Based on the historical scheduling priority and historical scheduling value of each energy resource, as well as the complementary or substitution relationship between different energy resources, a structured scheduling information database representing the historical scheduling capacity is constructed.

[0040] The calculation method of the historical scheduling priority includes: According to the statistical results of the scheduling performance of each energy resource under different historical disposal plans, the historical scheduling times n of each energy resource are obtained. According to the number of historical disposal plans m, Y=n / m is calculated to obtain the historical scheduling rate Y, where the higher the historical scheduling rate Y is, the higher the scheduling priority is.

[0041] For example, a wind power generation facility was dispatched 5 times under different historical disposal plans, a photovoltaic power generation facility was dispatched 3 times, and a battery energy storage facility was dispatched 2 times to supplement power. If the total number of historical disposal plans m = 20, then the historical dispatch rate of the wind power generation facility Y = 5 / 20 = 0.25, the historical dispatch rate of the photovoltaic power generation facility Y = 3 / 20 = 0.15, and the historical dispatch rate of the battery energy storage facility Y = 2 / 20 = 0.1. Obviously, the historical dispatch priority of the wind power generation facility is the highest. The historical dispatch value can be understood as the total dispatch amount of each facility, for example 60MWh.

[0042] For another example, when statistics and calculations were made on the dispatch performance of different energy resources in different historical disposal plans, it was found that when wind power generation output was insufficient, battery energy storage was dispatched more frequently to meet demand, showing the complementarity between wind power generation and battery energy storage. Photovoltaic power generation has a higher output during the day, while wind power generation has a higher output during the night. They complement each other at different times, also showing complementarity. In order to meet the demand for electricity, a certain diesel generator set was dispatched more frequently to replace the photovoltaic and wind power generation facilities in the area, filling the power gap caused by insufficient photovoltaic and wind power generation. At this time, there is substitutability between the diesel generator set and other power generation facilities.

[0043] Among them, the completed structured dispatch information database contains at least the maximum output (MW), average output (MW), response time (seconds), regulation accuracy (%) and other parameters representing the regulation capacity of different power generation facilities, and marks the complementarity and substitutability of different facilities. Among them, parameters such as maximum output can often be directly obtained from energy resource data, while other parameters need to be calculated based on the actual working conditions of different facilities using energy resource models on a large amount of data.

[0044] In this embodiment, by counting and calculating the performance of each energy resource under different historical disposal schemes, the frequency, value, etc. of the energy resource being dispatched in history can be obtained, and then the historical dispatch priority and historical dispatch value can be determined to provide data support for subsequent resource allocation and scheduling strategies. At the same time, by counting and calculating the performance of different energy resources in different historical disposal schemes, the complementarity and substitutability between them can be obtained, where complementarity means that certain energy resources can enhance each other during dispatch, and substitutability means that in some cases, one energy resource can replace another energy resource to meet energy demand.

[0045] On this basis, the training process of the energy resource model includes: A primary energy resource model is established based on a decision tree or random forest algorithm, and the primary energy resource model is trained using historical energy resource data and corresponding historical scheduling capabilities to obtain a trained energy resource model.

[0046] The decision tree model builds the model by recursively dividing the data feature space, each node represents a decision point of a feature variable, and the leaf node represents the prediction result. The random forest model improves the accuracy and stability of the model by building multiple decision trees and combining their prediction results.

[0047] The decision tree algorithm represents the decision-making process in a tree structure. Each internal node represents a judgment on an attribute, each branch represents the output of a judgment result, and each leaf node represents a category. This structure makes the relationship between energy resource data and dispatch capability very intuitive, easy to understand and verify. Among decision tree algorithms, commonly used algorithms include ID3, C4.5, CART, etc. In the field of virtual power plants, since continuous data and missing value processing may be involved, C4.5 or CART algorithms can generally be used first. The decision tree model has good interpretability. The structure of the decision tree can be viewed through visualization tools to understand the decision-making process of the model. For example, by viewing the decision tree, you can clearly know how the type and number of a power station affect the dispatch capability.

[0048] Then, characteristic variables are extracted from the historical energy resource data, which should be able to reflect the output characteristics, availability, etc. of energy resources. Characteristic variables may include the type, quantity, connection relationship, average output, maximum output, minimum output, output volatility, etc. of energy resources.

[0049] During the training process, the model learns the mapping relationship between feature variables and historical dispatch capabilities, that is, how to predict the dispatch capabilities based on the characteristics of energy resources. The trained energy resource model is then evaluated using methods such as cross-validation to check the model's predictive performance. The model is tuned based on the evaluation results, such as adjusting the depth of the decision tree, the number of trees in the random forest, and other parameters to improve the model's accuracy and generalization ability.

[0050] In this embodiment, the machine algorithm can provide a more intuitive decision path, making the relationship between energy resource data and dispatching capacity more transparent, easy to understand and verify. For discrete data such as the number of power stations and the type of energy storage stations, the machine learning algorithm can better capture the laws behind them, so as to accurately judge the dispatching capacity. Compared with complex algorithms such as deep learning, machine learning algorithms usually have higher computing efficiency during training and prediction, and are suitable for scenarios with high real-time requirements. In addition, the machine learning algorithm has relatively few parameters, and the tuning process is relatively simple, and the model performance can be quickly adjusted according to actual needs. Therefore, the machine learning algorithm is more suitable for the first half of the process of generating a disposal plan, that is, the energy resource model of this application.

[0051] In addition, the training process of the energy disposal model includes: A primary energy disposal model is established based on the long short-term memory network algorithm. The primary energy disposal model is trained using historical operating status data, historical scheduling capabilities, and historical disposal plans to obtain a trained energy disposal model.

[0052] Among them, before training, it is necessary to design the structure of the LSTM model, including the input layer, hidden layer (one or more LSTM layers) and output layer. The input layer receives the preprocessed feature data, the hidden layer captures the long-term dependencies in the time series data through the LSTM unit, and the output layer generates the prediction results. And perform parameter initialization: randomly initialize the weights and biases of the LSTM model to prepare for the training process. And select a suitable loss function (such as mean square error MSE) to evaluate the difference between the model prediction results and the actual values, and select an appropriate optimizer (such as Adam) to update the model parameters to minimize the loss function, and use the training set data to iteratively train the LSTM model. In each iteration, the feature data is input into the model, the loss between the prediction results and the actual values ​​is calculated, and then the model parameters are updated through the back propagation algorithm. At the same time, the performance on the validation set is monitored to evaluate the generalization ability of the model and avoid overfitting.

[0053] Unlike the machine learning algorithms used in energy resource models, deep learning algorithms are good at processing complex nonlinear relationships and can capture the deep connections between operating status data, scheduling capacity data, etc., so as to formulate more accurate disposal plans. Through a large amount of training data, deep learning algorithms can automatically learn and optimize internal parameters and adapt to different operating scenarios without human intervention. The fully trained deep learning model has a strong generalization ability and can maintain stable performance output when facing new operating states. For time series data (such as load change trends), deep learning algorithms such as LSTM can capture the long-term dependencies between data and provide strong support for the formulation of long-term and effective disposal plans. Therefore, deep learning algorithms are more suitable for the second half of the disposal plan generation process, that is, the energy disposal model of this application.

[0054] Specifically, the LSTM model can capture the complex nonlinear relationship between operating status data, dispatching capacity data, etc., so as to formulate a more accurate disposal plan, and automatically adjust the network weights through the back-propagation algorithm and gradient descent method to achieve self-learning. During the training process, LSTM can continuously optimize internal parameters to adapt to different operating scenarios and data distributions. Therefore, the fully trained LSTM model has a strong generalization ability and can maintain stable performance output when facing new operating states. This enables it to maintain reliable disposal capabilities in a complex and changeable virtual power plant environment. At the same time, for time series data (such as load change trends, charging and discharging cycles of energy storage equipment, etc.), LSTM can capture long-term dependencies between data. This is crucial for formulating disposal plans that need to consider historical data and future trends. For example, when predicting future load changes, LSTM can comprehensively consider data from the past few days or even weeks to make more accurate predictions.

[0055] In other words, machine learning algorithms are suitable for building energy resource models due to their strong interpretability, ability to process discrete data, high computational efficiency, and ease of tuning; while deep learning algorithms are more suitable for building energy disposal models due to their strong nonlinear modeling capabilities, strong self-learning capabilities, strong generalization capabilities, and the ability to capture long-term dependencies. This combination can give full play to the advantages of each algorithm and improve the efficiency and accuracy of virtual power plant power resource disposal.

[0056] As a supplement to Example 1, S2 may further include: According to the data dimension of the real-time scheduling capability, a multi-dimensional vector parameter is set, and the real-time scheduling capability is recorded in the form of a multi-dimensional vector parameter; The cosine similarity of the multidimensional vector parameters at the previous and next moments is calculated. If the cosine similarity is less than the preset similarity, the energy resource model is updated.

[0057] In this embodiment, the update of the energy resource model adopts incremental learning based on knowledge distillation, including the following steps: using the original source resource model as the first teacher model, constructing and using the incremental energy resource data and the corresponding scheduling capabilities to train the second teacher model; Use the first teacher model to predict the old data, and use the second teacher model to predict the new data, and generate corresponding soft labels respectively; Build a student model and set a distillation loss function to measure the difference between the student model output and the soft labels of the two teacher models. For example, take the weighted average of the soft labels of the two teacher models as the target that the student model should imitate. Based on the distillation loss function, the soft labels of the two teacher models are used to guide the learning direction of the student model, and the parameters of the student model are updated to obtain the updated energy resource model.

[0058] This method retains the parameters of the old model and can be replaced or rolled back when necessary. Compared with direct retraining, it has higher controllability and transparency, making it easier to discover and adjust problems.

[0059] Embodiment 2: A dual-model driven virtual power plant power resource handling system. When the dual-model driven virtual power plant power resource handling system is running, a dual-model driven virtual power plant power resource handling method of Example 1 is executed.

[0060] Embodiment 3: An electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the processor calls the computer program in the memory, the steps of the above-mentioned dual-model driven virtual power plant power resource disposal method are implemented.

[0061] Embodiment 4: A storage medium stores computer executable instructions, which, when loaded and executed by a processor, implement the steps of the above-mentioned dual-model driven virtual power plant power resource disposal method.

[0062] In summary, the substantial effects of the above embodiments include: By introducing the dual-model architecture of the energy resource model and the energy disposal model, this embodiment achieves the decoupling of energy resource data from specific disposal solutions. This design avoids the black box problem of a single model, making the impact of changes in energy resource data on disposal solutions controllable and predictable, and improving the stability and reliability of the system.

[0063] This embodiment uses the energy resource model to process real-time energy resource data and output real-time dispatch capabilities, providing accurate and reliable input conditions for the energy disposal model. Since the energy disposal model does not directly use energy resource data during training, but uses historical operating status data, historical dispatch capabilities, and historical disposal plans, it effectively isolates the direct impact of energy resource data on the disposal model, allowing the disposal model to focus more on the connection between dispatch capabilities and disposal plans.

[0064] In addition, this embodiment also provides a rich training data acquisition step and model training process, ensuring that the energy resource model and energy disposal model can accurately reflect the actual operation and dispatching capabilities of the virtual power plant. By counting and calculating the performance of each energy resource under different historical disposal schemes, this embodiment obtains the historical dispatch priority and historical dispatch value of energy resources, providing strong data support for subsequent resource allocation and dispatching strategies.

[0065] Through the description of the above implementation methods, technical personnel in the relevant field can understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the specific device can be divided into different functional modules to complete all or part of the functions described above.

[0066] In the embodiments provided in the present application, it should be understood that the disclosed structures and methods can be implemented in other ways. For example, the embodiments of the structure described above are only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another structure, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, structures or units, which can be electrical, mechanical or other forms.

[0067] The units described as separate components may or may not be physically separated, and the components shown as units may be one physical unit or multiple physical units, that is, they may be located in one place or distributed in multiple different places. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0068] In addition, each functional unit in the embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0069] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium, including several instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor (processor) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0070] The above contents are only specific implementation methods of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A dual-model driven virtual power plant power resource disposal method, characterized in that: The following steps are involved: S1: Obtain real-time operation status data and real-time energy resource data of the virtual power plant; S2: Input the real-time energy resource data into the energy resource model based on machine learning trained by the historical energy resource data and the corresponding historical dispatching capabilities to obtain the real-time dispatching capabilities of the virtual power plant; S3: Input the real-time dispatch capability and real-time operation status data into the energy disposal model based on deep learning trained by historical operation status data, historical dispatch capability and historical disposal plans, and output the real-time disposal plan of the virtual power plant.

2. According to the dual-model driven virtual power plant power resource disposal method of claim 1, it is characterized in that: In S2 and S3, the steps for obtaining training data required for the energy resource model and the energy disposal model include: A1: Obtain the operation records of the virtual power plant, extract the historical disposal plans in the operation records, and associate the historical energy resource data and historical operation status data corresponding to each historical disposal plan; A2: Determine the historical dispatch capacity corresponding to each energy resource based on historical disposal plans and historical energy resource data.

3. According to the dual-model driven virtual power plant power resource disposal method of claim 1, it is characterized in that: The historical energy resource data include: The number, types, and operating parameters of distributed generation facilities; the number, types, and operating parameters of distributed energy storage systems; the number, types, and operating parameters of controllable loads.

4. A dual-model driven virtual power plant power resource disposal method according to claim 2, characterized in that: A1: obtaining the operation records of the virtual power plant, extracting the historical disposal plans in the operation records, and associating the historical energy resource data and historical operation status data corresponding to each historical disposal plan, including: Obtain operation records from the virtual power plant management system using API interfaces or data capture; Use data parsing algorithms to extract historical disposal plans from acquired operation records; According to the time of the historical disposal plan, the historical energy resource data and historical operation status data corresponding to the time are determined, and associated with the historical disposal plan. The associated data is verified and cleaned, and the associated data is stored in a relational database.

5. A dual-model driven virtual power plant power resource disposal method according to claim 2, characterized in that: A2: determining the historical dispatching capacity corresponding to each energy resource according to the historical disposal scheme and the historical energy resource data, including: performing statistics and calculations on the dispatching performance of each energy resource under different historical disposal schemes to obtain the historical dispatching priority and historical dispatching value of each energy resource; Statistics and calculations are made on the dispatch performance of different energy resources in different historical disposal plans to obtain the complementarity and substitutability between different energy resources and establish the complementary or substitutive relationship between different energy resources; Based on the historical scheduling priority and historical scheduling value of each energy resource, as well as the complementary or substitution relationship between different energy resources, a structured scheduling information database representing the historical scheduling capacity is constructed.

6. A dual-model driven virtual power plant power resource disposal method according to claim 5, characterized in that: The calculation method of the historical scheduling priority includes: According to the statistical results of the scheduling performance of each energy resource under different historical disposal plans, the historical scheduling times n of each energy resource are obtained. According to the number of historical disposal plans m, Y=n / m is calculated to obtain the historical scheduling rate Y, where the higher the historical scheduling rate Y is, the higher the scheduling priority is.

7. A dual-model driven virtual power plant power resource disposal method according to any one of claims 2-6, characterized in that: The training process of the energy resource model includes: A primary energy resource model is established based on a decision tree or random forest algorithm, and the primary energy resource model is trained using historical energy resource data and corresponding historical scheduling capabilities to obtain a trained energy resource model.

8. A dual-model driven virtual power plant power resource disposal method according to any one of claims 2-6, characterized in that: The training process of the energy disposal model includes: A primary energy disposal model is established based on the long short-term memory network algorithm. The primary energy disposal model is trained using historical operating status data, historical scheduling capabilities, and historical disposal plans to obtain a trained energy disposal model.

9. The method for handling power resources of a virtual power plant driven by dual models according to claim 1, characterized in that: The S2 further includes: According to the data dimension of the real-time scheduling capability, a multi-dimensional vector parameter is set, and the real-time scheduling capability is recorded in the form of a multi-dimensional vector parameter; The cosine similarity of the multidimensional vector parameters at the previous and next moments is calculated. If the cosine similarity is less than the preset similarity, the energy resource model is updated.

10. An electronic device, characterized in that: It includes a memory and a processor, wherein the memory stores a computer program, and when the processor calls the computer program in the memory, it implements the steps of a dual-model driven virtual power plant power resource disposal method as described in any one of claims 1 to 9.

11. A storage medium, characterized in that: The storage medium stores computer executable instructions, which, when loaded and executed by a processor, implement the steps of a dual-model driven virtual power plant power resource disposal method as described in any one of claims 1 to 9.

Images