A virtual power plant energy storage aggregation scheduling method based on data knowledge distillation

Abstract

The application relates to the technical field of virtual power plant energy storage aggregation scheduling, and provides a virtual power plant energy storage aggregation scheduling method based on data knowledge distillation, which comprises the following steps: collecting the positions and communication data of each distributed power supply of a virtual power plant, performing data knowledge distillation on the communication data, and establishing a communication characteristic vector; extracting adjacent distributed power supplies, calculating the power supply difference degree of the adjacent distributed power supplies, and calculating the first characteristic value of the distributed power supplies; determining the first change moment of the distributed power supplies, calculating the second characteristic value of the distributed power supplies; combining the first characteristic value, calculating the prey weight, scheduling and optimizing the energy storage aggregation of all the distributed power supplies of the virtual power plant according to the prey weight, obtaining an optimal scheduling method, and realizing the energy storage aggregation scheduling of the virtual power plant. The application can provide a scheduling scheme with smaller scheduling burden.

Description

Technical Field

[0001] This application relates to the field of virtual power plant energy storage aggregation and scheduling technology, specifically to a virtual power plant energy storage aggregation and scheduling method based on data knowledge distillation. Background Technology

[0002] Virtual power plants (VPPs) can aggregate distributed resources such as power generation, grid, load, and storage on the distribution side. By aggregating and scheduling energy storage systems, they effectively improve the operational regulation capabilities of the power system. Simultaneously, when faced with massive and variable user-side data, VPPs can utilize technologies such as data knowledge distillation to extract key information from the data, achieving efficient and accurate energy storage aggregation and scheduling. When the energy storage system of a technology-based VPP participates in scheduling in conjunction with distributed power sources and loads, it needs to execute corresponding control strategies through a control center to coordinate the power flow of different energy sources and loads. Specifically, the central control center collects real-time information on distributed power sources and performs unified scheduling, while the energy storage system responds in real-time to the power command values ​​corresponding to the peak shaving and valley filling plans issued by the central control center, improving the overall output accuracy of the VPP.

[0003] Existing dispatching schemes can obtain the active and reactive power of each power source through processing at the dispatch center within a distributed energy storage system of a virtual power plant. Distributed power sources then receive these active and reactive power data, enabling the dispatching of power resources. However, in distributed energy storage systems, information exchange between power sources occurs only with adjacent power sources, and different power sources have varying information delays. This may prevent the virtual power plant from accurately obtaining the real-time status information of each power source during dispatching, leading to erroneous dispatching. Furthermore, differences in the number and distance of adjacent power sources between different power sources increase the dispatching burden and latency during long-distance or high-demand dispatching, preventing the dispatching scheme from achieving timely energy storage aggregation and dispatching of the virtual power plant, and potentially increasing power losses. Summary of the Invention

[0004] This application provides a virtual power plant energy storage aggregation and scheduling method based on data knowledge distillation to solve the problem of heavy scheduling burden in existing scheduling schemes when virtual power plants perform energy storage aggregation and scheduling due to the influence of different distances between adjacent power sources and drastic changes in the state of the power sources themselves. The specific technical solution adopted is as follows: One embodiment of this application provides a virtual power plant energy storage aggregation and scheduling method based on data knowledge distillation, the method comprising the following steps: The location of each distributed power source in the virtual power plant and the different types of communication data of the distributed power sources at different acquisition times are collected. Data knowledge distillation is performed based on the communication data to establish the communication feature vector of each distributed power source at each acquisition time. Extract all adjacent distributed power sources of each distributed power source in the virtual power plant. Based on the positional differences between the distributed power source and all its adjacent distributed power sources and the differences in the values ​​of the same type of communication data, calculate the power difference degree between the distributed power source and each adjacent distributed power source at each acquisition time. Based on the power difference degree between the distributed power source and all its adjacent distributed power sources at the same acquisition time, calculate the first feature value of the distributed power source at the same acquisition time. Based on the similarity values ​​of the communication feature vectors of the distributed power source at different acquisition times, the first change time of the distributed power source is determined. Based on the chronological relationship between the acquisition time and the first change time of the distributed power source, the second feature value of the distributed power source at the acquisition time is calculated. Based on the first and second characteristic values ​​of the same distributed power source at the same acquisition time, the prey weight of the same distributed power source at the same acquisition time is calculated. Based on the prey weight, the energy storage aggregation of all distributed power sources in the virtual power plant is scheduled and optimized to obtain the optimal scheduling method and realize the energy storage aggregation scheduling of the virtual power plant.

[0005] Furthermore, the specific method for establishing the communication feature vector is as follows: The communication data is cleaned and converted into string data. Knowledge distillation is performed on the encoded string data to obtain the feature vector of each distributed power source at each acquisition time. The feature vector is recorded as the communication feature vector.

[0006] Furthermore, the specific calculation method for the power supply difference is as follows: Communication data includes active power, reactive power, voltage, current, state of charge (SOC), and battery temperature; Calculate the power difference between the distributed power source and each adjacent distributed power source at each data acquisition time.

[0007] Furthermore, the formula for calculating the power supply difference is: in, Indicates the first The distributed power source is in the first State characteristic values ​​at each acquisition time; Indicates the first The distributed power source is in the first Active power at each acquisition moment; Indicates the first The distributed power source is in the first The first collection time and the first The maximum active power value for all data collection times within one month prior to the current data collection time; Indicates the first The distributed power source is in the first State of charge (SOC) at each acquisition time; Indicates the first The distributed power source is in the first Battery temperature at each data acquisition moment; Indicates the first The distributed power source is in the first The first collection time and the first The maximum battery temperature at all data collection times within one month prior to the current data collection time; Indicates the first The distributed power source and the first The distributed power source is in the first Power supply differences at each acquisition time; Represents the natural constant; Indicates the first The distributed power source and the first The distance between distributed power sources; Indicates the first The maximum distance between a distributed power source and all its neighboring distributed power sources; Indicates the first The number of all collection times included in the month preceding the current collection time; Indicates the first The distributed power source is in the first The state feature values ​​at each acquisition time.

[0008] Furthermore, the specific calculation method for the first characteristic value of the distributed power source at the same acquisition time is as follows: The sum of the power differences between the distributed power source and all its neighboring distributed power sources at the same data acquisition time is denoted as the sum of the power differences of the distributed power source at the same data acquisition time. The product of the number of all distributed power sources adjacent to the distributed power source and the sum of the power differences of the distributed power sources at the time of data acquisition is denoted as the first characteristic value of the distributed power source at the time of data acquisition.

[0009] Furthermore, the method for determining the first change moment of the distributed power source is as follows: The similarity between the communication feature vectors of the distributed power source at the acquisition time and the previous adjacent acquisition time is denoted as the communication feature similarity of the distributed power source at the acquisition time. Determine the similarity threshold; The earliest acquisition time among the acquisition times where the communication feature similarity corresponding to the distributed power source is greater than the similarity judgment threshold is recorded as the first change time of the distributed power source.

[0010] Furthermore, the specific method for determining the similarity threshold is as follows: The sum of the mean and three times the variance of the communication feature similarity of the distributed power source at the time of acquisition and all acquisition times within one month prior to the acquisition time is denoted as the similarity judgment threshold of the distributed power source at the acquisition time.

[0011] Furthermore, the formula for calculating the second characteristic value of the distributed power source at the acquisition time is: in, Indicates the first The distributed power source is in the first The second characteristic value at each acquisition time; Indicates that the distributed power source is in the first... Similarity of communication features at each acquisition time; Indicates that the distributed power source is in the first... The first data collection time and the first change time The similarity of their communication feature vectors; Indicates that the distributed power source is in the first... The similarity threshold for each collection time; Indicates the first The data collection time is the [number]th [time]. The first moment of change for a distributed power source; Indicates the first The data collection time is at the [number]th acquisition time. After the first change moment of a distributed power source.

[0012] Furthermore, the method for calculating the prey weight of the same distributed power source at the same acquisition time is as follows: The sum of the normalized mean of the first and second feature values ​​of the same distributed power source at the same acquisition time and the number 1 is denoted as the prey weight of the same distributed power source at the same acquisition time.

[0013] Furthermore, the method for optimizing the scheduling of all distributed power storage aggregations in the virtual power plant based on prey weights to obtain the optimal scheduling method includes the following specific methods: The product of the position of the distributed power source and the prey weight of the distributed power source at the acquisition time during the iteration process of the whale optimization algorithm is denoted as the first product of the distributed power source at the acquisition time and in this iteration process. The absolute value of the difference between the first product of the distributed power source at the acquisition time and in the iteration process and the position of the optimal solution in the iteration process is denoted as the relative distance between the distributed power source and the prey at the acquisition time and in the iteration process. The sum of the relative distances between all the distributed power sources of the virtual power plant and the prey at the same acquisition time and in the same iteration process is used as the objective function. The whale optimization algorithm is used to minimize the objective function to obtain the optimal scheduling method.

[0014] The beneficial effects of this application are: This application considers that in a distributed energy storage system of a virtual power plant, each distributed power source only has a local topological relationship with its adjacent distributed power sources. Furthermore, the number of adjacent distributed power sources and the distance between them both affect the power state and latency changes of the distributed power sources. Therefore, this application evaluates the degree of impact of distributed power sources on the latency of the virtual power plant's energy storage aggregation and scheduling at the time of data acquisition. It obtains the first characteristic value of each distributed power source in the virtual power plant at each data acquisition time. The larger the first characteristic value of a distributed power source at the data acquisition time, the more it should be prioritized for energy storage aggregation and scheduling in the virtual power plant. When the communication data of a distributed power source changes drastically at the data acquisition time, timely scheduling adjustments are even more necessary. Firstly, based on the similarity values ​​of the communication characteristic vectors of the distributed power sources at different data acquisition times, the first change time of the distributed power source is determined. The first change time is the time when the distributed power source changes. When the communication data of the distributed power source changes drastically at the time of data acquisition, it is even more necessary to promptly adjust the scheduling of the distributed power source. Based on the chronological relationship between the acquisition time and the first change time of the distributed power source, the necessity of adjusting the distributed power source at the acquisition time is evaluated. The second characteristic value of the distributed power source at the acquisition time is calculated. Then, based on the first and second characteristic values ​​of the same distributed power source at the same acquisition time, the prey weight of the same distributed power source at that same acquisition time is calculated. Based on the prey weight, the scheduling optimization of all distributed power sources in the virtual power plant is performed to obtain the optimal scheduling method, realizing the energy storage aggregation scheduling of the virtual power plant. This addresses the problem of high scheduling burden caused by the different distances between adjacent power sources and drastic changes in the power source's own state during energy storage aggregation scheduling in the virtual power plant, thus reducing the scheduling burden and providing a more reasonable and less costly scheduling scheme. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A schematic diagram of a virtual power plant energy storage aggregation and scheduling method based on data knowledge distillation, provided as an embodiment of this application; Figure 2 This is a flowchart illustrating the power difference acquisition process provided in one embodiment of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] Please see Figure 1 The diagram illustrates a flowchart of a virtual power plant energy storage aggregation and scheduling method based on data knowledge distillation, according to an embodiment of this application. The method includes the following steps: Step S001: Collect the location of each distributed power source in the virtual power plant and the different types of communication data of the distributed power sources at different collection times. Perform data knowledge distillation based on the communication data to establish the communication feature vector of each distributed power source at each collection time.

[0019] Acquire the location of each distributed power source in the virtual power plant and the communication data of the distributed power sources at different acquisition times.

[0020] Preferably, as an embodiment of this application, communication data of each distributed power source of the virtual power plant is acquired at all collection times within one month. The collection time interval of the communication data is 15 seconds. The communication data includes electrical parameters and battery status parameters. The electrical parameters include active power, reactive power, voltage and current. The battery status parameters include state of charge (SOC) and battery temperature. The location of the distributed power source is the latitude and longitude of the distributed power source.

[0021] Specifically, taking communication data stored in a JSON file as an example, the JSON file is cleaned by removing non-critical data such as distributed power source IDs. The `json.dumps` function in Python is used to convert the communication data of each distributed power source in the JSON file at different acquisition times into string data. Encoding techniques are then used to encode all string data. A data knowledge distillation model is used to perform knowledge distillation on the encoded string data to obtain a low-dimensional, dense feature vector for each distributed power source at each acquisition time. This feature vector is then recorded as the communication feature vector, which represents the communication data in the form of a feature vector. Encoding techniques such as TF-IDF vectorization and one-hot encoding can be used, and the data knowledge distillation model can use algorithms such as AT and VID.

[0022] Among them, string conversion technology, encoding technology and data knowledge distillation model are all well-known technologies and will not be elaborated further.

[0023] At this point, the location of each distributed power source in the virtual power plant, as well as the communication data and communication feature vectors of the distributed power sources at different acquisition times, are obtained.

[0024] Step S002: Extract all adjacent distributed power sources of each distributed power source in the virtual power plant. Based on the location differences between the distributed power source and all its adjacent distributed power sources and the value differences of the same type of communication data, calculate the power difference degree between the distributed power source and each adjacent distributed power source at each acquisition time. Based on the power difference degree between the distributed power source and all its adjacent distributed power sources at the same acquisition time, calculate the first feature value of the distributed power source at the same acquisition time.

[0025] In a distributed energy storage system within a virtual power plant, each distributed power source only has local topological relationships with its adjacent distributed power sources. This means that information exchange occurs only between adjacent distributed power sources. After the status information of adjacent distributed power sources is exchanged, it is transmitted to the virtual power plant's dispatch center, resulting in a relatively smooth charging and discharging power flow for the energy storage units. However, the number of adjacent distributed power sources and the distance between them vary. A larger number of adjacent distributed power sources leads to a more complex local topological relationship, and greater distances between adjacent distributed power sources result in longer transmission times for information exchange. Therefore, when the status of a distributed power source with a more complex local topological relationship and greater distances from its neighbors changes, the changes in its power state and latency are more drastic, making it prone to sudden changes.

[0026] For each distributed power source in the virtual power plant, all adjacent distributed power sources are extracted. Based on the location differences between the distributed power source and all its adjacent distributed power sources, as well as the differences in the values ​​of the same type of communication data, the power source difference degree between the distributed power source and each of its adjacent distributed power sources at each data acquisition time is calculated. The formulas for calculating the power source difference degree are as follows: in, Indicates the first The distributed power source is in the first State characteristic values ​​at each acquisition time; Indicates the first The distributed power source is in the first Active power at each acquisition moment; Indicates the first The distributed power source is in the first The first collection time and the first The maximum active power value for all data collection times within one month prior to the current data collection time; Indicates the first The distributed power source is in the first State of charge (SOC) at each acquisition time; Indicates the first The distributed power source is in the first Battery temperature at each data acquisition moment; Indicates the first The distributed power source is in the first The first collection time and the first The maximum battery temperature at all data collection times within one month prior to the current data collection time; Indicates the first The distributed power source and the first The distributed power source is in the first Power supply differences at each acquisition time; Represents the natural constant; Indicates the first The distributed power source and the first The distance between distributed power sources is calculated in this embodiment using Haversine distance, based on the location of each distributed power source. Indicates the first The maximum distance between a distributed power source and all its neighboring distributed power sources; Indicates the first The number of all collection times included in the month preceding the current collection time; Indicates the first The distributed power source is in the first The state feature values ​​at each acquisition time.

[0027] The flowchart for obtaining power supply difference is as follows: Figure 2 As shown.

[0028] Based on the power difference between the distributed power source and all its neighboring distributed power sources at the same data acquisition time, the first characteristic value of the distributed power source at the same data acquisition time is calculated. The sum of the power differences between the distributed power source and all its neighboring distributed power sources at the same data acquisition time is denoted as the sum of power differences of the distributed power source at the same data acquisition time. The product of the number of all distributed power sources adjacent to the distributed power source and the sum of power differences of the distributed power source at the data acquisition time is denoted as the first characteristic value of the distributed power source at the data acquisition time. The formula for calculating the first characteristic value is: in, Indicates the first The distributed power source is in the first The first feature value at each acquisition time; Indicates the first The distributed power source and all its neighboring distributed power sources in the th... The sum of power supply differences at each acquisition time; Indicates the relationship with the first The number of different distributed power sources adjacent to each other.

[0029] The larger the first characteristic value of the distributed power source at the time of data acquisition, the greater the impact of the distributed power source on the latency of the virtual power plant in energy storage aggregation and scheduling. Therefore, when performing energy storage aggregation and scheduling on the virtual power plant, the distributed power source should be the key scheduling target.

[0030] Thus, the first characteristic value of each distributed power source in the virtual power plant at each acquisition moment is obtained.

[0031] Step S003: Based on the similarity values ​​of the communication feature vectors of the distributed power source at different acquisition times, determine the first change time of the distributed power source. Based on the chronological relationship between the acquisition time and the first change time of the distributed power source, calculate the second feature value of the distributed power source at the acquisition time.

[0032] When the communication data of the distributed power source changes drastically at the time of data acquisition, it is even more necessary to schedule and adjust the distributed power source in a timely manner.

[0033] The first change moment of the distributed power source is determined by the similarity values ​​of the communication feature vectors of the distributed power source at different acquisition times.

[0034] Preferably, as an embodiment of this application, the similarity between the communication feature vectors of the distributed power source at the acquisition time and the adjacent acquisition time before the acquisition time is recorded as the communication feature similarity of the distributed power source at the acquisition time; the sum of the mean of the communication feature similarities of the distributed power source at the acquisition time and three times the variance of all acquisition times within one month before the acquisition time is recorded as the similarity judgment threshold of the distributed power source at the acquisition time; the earliest acquisition time among the acquisition times where the communication feature similarity of the distributed power source is greater than the similarity judgment threshold is recorded as the first change time of the distributed power source.

[0035] It is understandable that the construction of the similarity judgment threshold conforms to 3 in principle.

[0036] Based on the chronological relationship between the data acquisition time and the first change time of the distributed power source, the second characteristic value of the distributed power source at the data acquisition time is calculated. The formula for calculating the second characteristic value is: in, Indicates the first The distributed power source is in the first The second characteristic value at each acquisition time; Indicates that the distributed power source is in the first... Similarity of communication features at each acquisition time; Indicates that the distributed power source is in the first... The first data collection time and the first change time The similarity of their communication feature vectors; Indicates that the distributed power source is in the first... The similarity threshold for each collection time; Indicates the first The data collection time is the [number]th [time]. The first moment of change for a distributed power source; Indicates the first The data collection time is at the [number]th acquisition time. After the first change moment of a distributed power source.

[0037] Preferably, this embodiment uses cosine similarity as a method for measuring the similarity of communication feature vectors. The calculation of cosine similarity is a well-known technique and will not be described in detail here. As another implementation, based on the purpose of measuring the similarity of two communication feature vectors, the implementer may use other methods in the prior art, such as Spearman correlation coefficient, to obtain the similarity of two communication feature vectors. This application does not impose any special restrictions.

[0038] Understandably, before the first change moment of the distributed power source is identified, this situation belongs to the third segment of the formula for calculating the second eigenvalue, i.e. The situation.

[0039] At this point, the second characteristic value of each distributed power source in the virtual power plant at the time of acquisition is obtained.

[0040] Step S004: Based on the first and second characteristic values ​​of the same distributed power source at the same acquisition time, calculate the prey weight of the same distributed power source at the same acquisition time. Based on the prey weight, perform scheduling optimization on the energy storage aggregation of all distributed power sources in the virtual power plant to obtain the optimal scheduling method and realize the energy storage aggregation scheduling of the virtual power plant.

[0041] The sum of the normalized mean of the first and second feature values ​​of the same distributed power source at the same acquisition time and the number 1 is denoted as the prey weight of the same distributed power source at the same acquisition time.

[0042] It should be noted that this embodiment uses the maximum-minimum normalization method to calculate the normalized value. In practical applications, implementers may use other methods of existing technology, such as the Z-Score standard normalization method or the sigmoid function, to calculate the normalized value, and no limitation is made here.

[0043] Each distributed power source in the virtual power plant is considered a whale in the whale optimization algorithm. The solution space dimension of the whale optimization algorithm is the total number of distributed power sources included in the virtual power plant. The product of the position of the distributed power source and the prey weight of the distributed power source at the acquisition time during the iteration process of the whale optimization algorithm is denoted as the first product of the distributed power source at the acquisition time and in this iteration process. The absolute value of the difference between the first product of the distributed power source at the acquisition time and in the iteration process and the position of the optimal solution in the iteration process is denoted as the relative distance between the distributed power source and the prey at the acquisition time and in the iteration process. The sum of the relative distances between all the distributed power sources in the virtual power plant and the prey at the same acquisition time and in the same iteration process is used as the objective function. The whale optimization algorithm is used to minimize the objective function to obtain the optimal solution.

[0044] In this embodiment, the population size of the whale optimization algorithm is set to 30, the maximum number of iterations is set to 200, and the lower and upper bounds of the solution space are set to -15 and 15, respectively. Using the whale optimization algorithm to obtain the optimal solution is a well-known technique and will not be described in detail here.

[0045] It is important to understand that the relative distance between the distributed power source and the prey at the time of acquisition and during the iteration process is the distance that the whale corresponding to the distributed power source will move towards the prey in the next iteration process. When the relative distance with the prey is greater, the local topological relationship of the distributed power source in the solution space dimension of the whale optimization algorithm is stronger, the state information changes more greatly, and the compensation for movement should be increased to find the optimal solution faster.

[0046] Understandably, the optimal solution is the optimal scheduling method when a virtual power plant aggregates all distributed power sources for energy storage. During energy storage aggregation scheduling, the scheduling system can decompose the optimal solution into executable instructions. For example, an executable instruction could be: discharge a certain distributed power source at a power of 30kW for 2 hours. Before issuing the executable instructions, the scheduling system performs a safety check to ensure that the executable instructions are within the safe operating parameters of the energy storage devices, such as the maximum charging and discharging power. After the executable instructions are issued, the energy storage unit receives the instructions, and the local controller executes the instructions, controlling the power conversion system to complete the charging and discharging operation.

[0047] This completes the aggregation and scheduling of energy storage in the virtual power plant.

[0048] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. A virtual power plant energy storage aggregation and scheduling method based on data knowledge distillation, characterized in that, The method includes the following steps: The location of each distributed power source in the virtual power plant and the different types of communication data of the distributed power sources at different acquisition times are collected. Data knowledge distillation is performed based on the communication data to establish the communication feature vector of each distributed power source at each acquisition time. Extract all adjacent distributed power sources of each distributed power source in the virtual power plant. Based on the positional differences between the distributed power source and all its adjacent distributed power sources and the differences in the values ​​of the same type of communication data, calculate the power difference degree between the distributed power source and each adjacent distributed power source at each acquisition time. Based on the power difference degree between the distributed power source and all its adjacent distributed power sources at the same acquisition time, calculate the first feature value of the distributed power source at the same acquisition time. Based on the similarity values ​​of the communication feature vectors of the distributed power source at different acquisition times, the first change time of the distributed power source is determined. Based on the chronological relationship between the acquisition time and the first change time of the distributed power source, the second feature value of the distributed power source at the acquisition time is calculated. Based on the first and second characteristic values ​​of the same distributed power source at the same acquisition time, the prey weight of the same distributed power source at the same acquisition time is calculated. Based on the prey weight, the energy storage aggregation of all distributed power sources in the virtual power plant is scheduled and optimized to obtain the optimal scheduling method and realize the energy storage aggregation scheduling of the virtual power plant. The formula for calculating the power supply difference is: Communication data includes active power, reactive power, voltage, current, state of charge (SOC), and battery temperature; in, Indicates the first The distributed power source is in the first State characteristic values ​​at each acquisition time; Indicates the first The distributed power source is in the first Active power at each acquisition moment; Indicates the first The distributed power source is in the first The first collection time and the first The maximum active power value for all data collection times within one month prior to the current data collection time; Indicates the first The distributed power source is in the first State of charge (SOC) at each acquisition time; Indicates the first The distributed power source is in the first Battery temperature at each data acquisition moment; Indicates the first The distributed power source is in the first The first collection time and the first The maximum battery temperature at all data collection times within one month prior to the current data collection time; Indicates the first The distributed power source and the first The distributed power source is in the first Power supply differences at each acquisition time; Represents the natural constant; Indicates the first The distributed power source and the first The distance between distributed power sources; Indicates the first The maximum distance between a distributed power source and all its neighboring distributed power sources; Indicates the first The number of all collection times included in the month preceding the current collection time; Indicates the first The distributed power source is in the first State characteristic values ​​at each acquisition time; The method for determining the first change moment of the distributed power source is as follows: The similarity between the communication feature vectors of the distributed power source at the acquisition time and the previous adjacent acquisition time is denoted as the communication feature similarity of the distributed power source at the acquisition time. Determine the similarity threshold; The earliest collection time among the collection times where the communication feature similarity corresponding to the distributed power source is greater than the similarity judgment threshold is recorded as the first change time of the distributed power source. The formula for calculating the second characteristic value of the distributed power source at the time of data acquisition is: in, Indicates the first The distributed power source is in the first The second characteristic value at each acquisition time; Indicates that the distributed power source is in the first... Similarity of communication features at each acquisition time; Indicates that the distributed power source is in the first... The similarity between the communication feature vectors at each acquisition time and the first change time; Indicates that the distributed power source is in the first... The similarity threshold for each collection time; Indicates the first The data collection time is the [number]th [time]. The first moment of change for a distributed power source; Indicates the first The data collection time is at the [number]th acquisition time. After the first change moment of a distributed power source.

2. The virtual power plant energy storage aggregation and scheduling method based on data knowledge distillation according to claim 1, characterized in that, The specific method for establishing the communication feature vector is as follows: The communication data is cleaned and converted into string data. Knowledge distillation is performed on the encoded string data to obtain the feature vector of each distributed power source at each acquisition time. The feature vector is recorded as the communication feature vector.

3. The virtual power plant energy storage aggregation and scheduling method based on data knowledge distillation according to claim 1, characterized in that, The specific calculation method for the power supply difference is as follows: Calculate the power difference between the distributed power source and each adjacent distributed power source at each data acquisition time.

4. The virtual power plant energy storage aggregation and scheduling method based on data knowledge distillation according to claim 1, characterized in that, The specific calculation method for the first characteristic value of the distributed power source at the same acquisition time is as follows: The sum of the power differences between the distributed power source and all its neighboring distributed power sources at the same data acquisition time is denoted as the sum of the power differences of the distributed power source at the same data acquisition time. The product of the number of all distributed power sources adjacent to the distributed power source and the sum of the power differences of the distributed power sources at the time of data acquisition is denoted as the first characteristic value of the distributed power source at the time of data acquisition.

5. The virtual power plant energy storage aggregation and scheduling method based on data knowledge distillation according to claim 1, characterized in that, The specific method for determining the similarity threshold is as follows: The sum of the mean and three times the variance of the communication feature similarity of the distributed power source at the time of acquisition and all acquisition times within one month prior to the acquisition time is denoted as the similarity judgment threshold of the distributed power source at the acquisition time.

6. The virtual power plant energy storage aggregation and scheduling method based on data knowledge distillation according to claim 1, characterized in that, The method for calculating the prey weight of the same distributed power source at the same data collection time is as follows: The sum of the normalized mean of the first and second feature values ​​of the same distributed power source at the same acquisition time and the number 1 is denoted as the prey weight of the same distributed power source at the same acquisition time.

7. The virtual power plant energy storage aggregation and scheduling method based on data knowledge distillation according to claim 1, characterized in that, The method for optimizing the scheduling of all distributed power storage aggregations in the virtual power plant based on prey weights to obtain the optimal scheduling method includes the following specific methods: The product of the position of the distributed power source and the prey weight of the distributed power source at the acquisition time during the iteration process of the whale optimization algorithm is denoted as the first product of the distributed power source at the acquisition time and in this iteration process. The absolute value of the difference between the first product of the distributed power source at the acquisition time and in the iteration process and the position of the optimal solution in the iteration process is denoted as the relative distance between the distributed power source and the prey at the acquisition time and in the iteration process. The sum of the relative distances between all the distributed power sources of the virtual power plant and the prey at the same acquisition time and in the same iteration process is used as the objective function. The whale optimization algorithm is used to minimize the objective function to obtain the optimal scheduling method.

Images