Virtual power plant baseline load data prediction method and system based on deep learning

Abstract

The present application belongs to the field of power load prediction, and particularly relates to a virtual power plant baseline load data prediction method and system based on deep learning. The method divides the virtual power plant into demand response type and energy supply type virtual power plants according to different types of virtual power plant operation characteristics by clustering analysis on historical load data, and respectively constructs the first and second baseline load prediction models. The input features of the training set data of the demand response type and energy supply type virtual power plants are respectively differentially weighted by using the attention mechanism. The weight of the price factor in the input features of the training data set of the demand response type virtual power plant is enhanced, and the weight of the weather factor in the input features of the training data set of the energy supply type virtual power plant is enhanced, thereby improving the accuracy of the baseline load prediction of the virtual power plant and helping to improve the effectiveness of the virtual power plant evaluation.

Description

Technical Field

[0001] This invention belongs to the field of power load forecasting, specifically relating to a method and system for forecasting virtual power plant baseline load data based on deep learning. Background Technology

[0002] A virtual power plant (VPP) is an effective organizational structure that aggregates multiple distributed power sources and demand-response loads to participate in grid operation. Typically, a VPP leverages advanced control, measurement, and communication technologies to aggregate various distributed power sources such as wind farms and photovoltaic power plants, as well as various controllable loads and load-side resources such as electric vehicles. This allows for information exchange among the various sources, enabling them to jointly participate in the energy regulation and transmission of the large power grid and improve energy efficiency. Therefore, based on the types of resources within the VPP, it can be divided into energy supply-type VPPs and demand-response VPPs. The main factors influencing the participation of VPPs in grid regulation include meteorological factors and electricity price factors in the region where the VPP is located. The former is more sensitive to weather factors, while the latter is more sensitive to electricity price factors.

[0003] Virtual power plant baseline load is an important reference for the effectiveness of user participation in grid peak shaving and frequency regulation demand response. In the evaluation of the effectiveness of power demand response projects, customer baseline load (CBL) forecasting can provide a quantitative basis for evaluating power demand response and the degree of user load adjustment. However, existing models that use neural networks to forecast customer baseline load do not consider the characteristics of different types of virtual power plants, resulting in inaccurate forecasting results and low evaluation effectiveness. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to address the shortcomings of the above-mentioned technical solutions by providing a method and system for predicting baseline load data of virtual power plants based on deep learning, which can improve the accuracy of virtual power plant baseline load prediction and enhance the effectiveness of virtual power plant evaluation.

[0005] To achieve the above objectives, according to one aspect of the present invention, a method for predicting baseline load data of a virtual power plant based on deep learning is provided, comprising the following steps:

[0006] Step (1): Collect historical load data of all virtual power plants in the area and preprocess the historical load data;

[0007] Specifically, data acquisition involves extracting the data needed for load forecasting from complex raw data for use in data analysis and modeling. Data preprocessing includes tasks such as correcting bad data and data standardization.

[0008] Step (2): Perform k-means clustering analysis on the historical load data, and classify the virtual power plants into demand response virtual power plants and energy supply virtual power plants according to the operating characteristics of different types of virtual power plants, and record the category labels;

[0009] Step (3): Merge the meteorological data, electricity price data and historical load data of the virtual power plants in the affected area;

[0010] Specifically, the data fusion processing involves adding meteorological data and electricity price data to the load data, then using a standard fusion algorithm to convert the data into the same format and reorganize the data structure to construct data that can be directly input into the model for training.

[0011] Step (4): Perform dimensionality reduction on the fused dataset using principal component analysis;

[0012] Specifically, principal component analysis (PCA) includes standardizing the original variables, calculating the correlation coefficient matrix, calculating eigenvalues ​​and eigenvectors, and determining the principal components. Using PCA to reduce the dimensionality of the fused dataset effectively reduces data complexity. The principle of PCA is easy to understand: it projects the original matrix onto a new orthogonal coordinate system, and by sequentially selecting the directions of the new coordinate axes, maximizes the projection of the matrix onto these axes, i.e., maximizes the variance of the principal component sample values. Preserving the projections of the data in these directions reduces the data dimensionality.

[0013] Step (5): Divide the training set data and test set data for demand-response virtual power plants and energy supply virtual power plants respectively;

[0014] Specifically, the dataset is split into two independent parts: one part is used as the training set and the other part is used as the test set. For example, 70% of the data can be used as the training set and 30% as the test set, or 80% of the data can be used as the training set and 20% as the test set. There is no specific limit to the proportion.

[0015] Step (6): Construct a first baseline load forecasting model and a second baseline load forecasting model for demand-response virtual power plants and energy supply virtual power plants respectively. The first baseline load forecasting model and the second baseline load forecasting model respectively include an input layer, a feature extraction layer, an attention layer, a feature fusion layer and a feature output layer connected in series.

[0016] Step (7): Using the attention mechanism, differential weighting is applied to the input features of the training data of the demand response virtual power plant and the energy supply virtual power plant respectively;

[0017] Step (8): Train the first baseline load forecasting model and the second baseline load forecasting model on the training set, and verify the trained first baseline load forecasting model and the second baseline load forecasting model on the test set;

[0018] Step (9): Based on the classification label of the virtual power plant to be predicted, input the relevant data into the corresponding baseline load prediction model, perform prediction through the trained baseline load prediction model, and output the baseline load prediction value;

[0019] Step (10): After the end of the day's operation, obtain the actual load value for the day, calculate the deviation between the actual load value and the baseline load forecast value for a certain period, and take the average value of the deviation as the compensation value. Correct the baseline load forecast value according to the compensation value.

[0020] It is easy to understand that due to the uncertainty of load response, there will be a deviation between the baseline load forecast and the actual load. For example, the response period of the load is from 10:00 to 12:00 on a given day, and the non-response periods are from 0:00 to 10:00 and from 12:00 to 24:00. In this case, the deviation between the actual load value and the baseline load forecast in the 0:00 to 10:00 and 12:00 to 24:00 periods can be calculated based on the actual load value of the day. The average value of the deviation can be taken as the compensation value, and the baseline load forecast can be corrected based on the compensation value.

[0021] Preferably, the k-means clustering analysis in step (2) uses the distance between points as a similarity evaluation index, and the distance algorithm uses Euclidean distance to calculate the distance:

[0022]

[0023] Where, x i It is the i-th variable value of sample x; y i It is the value of the i-th variable in sample y.

[0024] Preferably, the meteorological data in step (3) includes the highest temperature, lowest temperature, average temperature, sunshine, wind speed, humidity, and precipitation; the electricity price data includes peak electricity price, off-peak electricity price, and normal electricity price.

[0025] Preferably, in step (7), the input features of the training datasets for demand-response virtual power plants and energy-supply virtual power plants are weighted differently using an attention mechanism. Specifically, the electricity price factor in the input features of the training dataset for demand-response virtual power plants is weighted, and the weather factor in the input features of the training dataset for energy-supply virtual power plants is weighted.

[0026] Preferably, the weighted calculation formula for the input features of the training dataset is as follows:

[0027] y t =A(x) t ,p t-1 ,q t-1 );

[0028]

[0029] x't=qt j xt j ;

[0030] Among them, y t Let A be the attention score matrix at time t; A(.) is the matrix used to calculate y. t Neural networks; x t p is the initial value input at time t; t-1 The predicted value at time t-1; q t-1 q represents the attention weight value at time t-1; tj For q t The attention weight value of the j-th feature, y tj For y t The attention score of the j-th feature, where n is the number of input features, x tj For x t The initial value of the j-th feature, x' t For x t Weighted features.

[0031] According to another aspect of the present invention, a deep learning-based virtual power plant baseline load data prediction system is also provided, which includes the following modules:

[0032] The data acquisition module is used to collect historical load data of all virtual power plants in the area and to preprocess the historical load data.

[0033] The clustering analysis module is used to perform k-means clustering analysis on the historical load data, classify virtual power plants into demand response virtual power plants and energy supply virtual power plants according to the operating characteristics of different types of virtual power plants, and record the category labels.

[0034] The fusion processing module is used to fuse meteorological data, electricity price data, and historical load data of virtual power plants within the affected area.

[0035] The principal component analysis module uses principal component analysis to reduce the dimensionality of the fused dataset.

[0036] The data partitioning module partitions the data into training set data and test set data for demand-response virtual power plants and energy supply virtual power plants, respectively.

[0037] The model building module constructs a first baseline load forecasting model and a second baseline load forecasting model for demand-response virtual power plants and energy supply virtual power plants, respectively. The first baseline load forecasting model and the second baseline load forecasting model each include an input layer, a feature extraction layer, an attention layer, a feature fusion layer and a feature output layer connected in series.

[0038] The differential weighting module uses an attention mechanism to differentially weight the input features of the training data for demand-response virtual power plants and energy-supply virtual power plants, respectively.

[0039] The training and validation module is used to train the first baseline load prediction model and the second baseline load prediction model on the training set, and to validate the trained first baseline load prediction model and the second baseline load prediction model on the test set.

[0040] The prediction output module inputs relevant data into the corresponding baseline load prediction model based on the classification label of the virtual power plant to be predicted, performs prediction through the trained baseline load prediction model, and outputs the baseline load prediction value.

[0041] The prediction and correction module obtains the actual load value of the day after the end of the operating day, calculates the deviation between the actual load value and the baseline load prediction value for a certain period of time, takes the average value of the deviation as the compensation value, and corrects the baseline load prediction value based on the compensation value.

[0042] Preferably, the k-means clustering analysis in the clustering analysis module uses the distance between points as a similarity evaluation index, and the distance algorithm uses Euclidean distance for calculation.

[0043]

[0044] Where, x i It is the i-th variable value of sample x; y i It is the value of the i-th variable in sample y.

[0045] Preferably, the meteorological data in the fusion processing module includes the highest temperature, lowest temperature, average temperature, sunshine, wind speed, humidity, and precipitation; the electricity price data includes peak electricity price, off-peak electricity price, and normal electricity price.

[0046] Preferably, in the differential weighting module, an attention mechanism is used to differentially weight the input features of the training datasets for demand-response virtual power plants and energy-supply virtual power plants, respectively. Specifically, the weight of the electricity price factor in the input features of the demand-response virtual power plant training dataset is increased, and the weight of the weather factor in the input features of the energy-supply virtual power plant training dataset is increased.

[0047] Preferably, the weighted calculation formula for the input features of the training dataset is as follows:

[0048] y t =A(x) t ,p t-1 ,q t-1 );

[0049]

[0050] x't=qt j xt j ;

[0051] Among them, y t Let A be the attention score matrix at time t; A(.) is the matrix used to calculate y. t Neural networks; x t p is the initial value input at time t; t-1 The predicted value at time t-1; q t-1 q represents the attention weight value at time t-1; tj For q t The attention weight value of the j-th feature, y tj For y t The attention score of the j-th feature, where n is the number of input features, x tj For x t The initial value of the j-th feature, x' t For x t Weighted features.

[0052] Based on the above technical solution, the present invention provides a method and system for predicting virtual power plant baseline load data based on deep learning, which has the following technical effects:

[0053] This invention utilizes k-means clustering analysis on historical load data to categorize virtual power plants into demand-response and energy-supply virtual power plants based on their operational characteristics. A first baseline load forecasting model and a second baseline load forecasting model are constructed for each. An attention mechanism is used to differentially weight the input features of the training data for the demand-response and energy-supply virtual power plants, respectively. The weight of electricity price factors in the input features of the demand-response virtual power plant training dataset is enhanced, as is the weight of weather factors in the input features of the energy-supply virtual power plant training dataset. Finally, based on the attention mechanism, a neural network is applied to predict the baseline load data for both demand-response and energy-supply virtual power plants, improving the accuracy of virtual power plant baseline load forecasting and contributing to the effectiveness of virtual power plant evaluation. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0055] Figure 1 A flowchart illustrating a deep learning-based method for predicting baseline load data of a virtual power plant, as provided in an embodiment of the present invention. Detailed Implementation

[0056] 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 of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0057] The following section will first combine the appendix. Figure 1 The concepts involved in this application will be explained. It should be noted that the following explanation of each concept is only to make the content of this application easier to understand and does not imply any limitation on the scope of protection of this application.

[0058] Example 1: In this example, to achieve the above objective, as follows... Figure 1 As shown, a deep learning-based method for predicting baseline load data of a virtual power plant is provided, which includes the following steps:

[0059] Step (1): Collect historical load data of all virtual power plants in the area and preprocess the historical load data;

[0060] Specifically, data acquisition involves extracting the data needed for load forecasting from complex raw data for use in data analysis and modeling. Data preprocessing includes tasks such as correcting bad data and data standardization.

[0061] Step (2): Perform k-means clustering analysis on the historical load data, and classify the virtual power plants into demand response virtual power plants and energy supply virtual power plants according to the operating characteristics of different types of virtual power plants, and record the category labels;

[0062] Step (3): Merge the meteorological data, electricity price data and historical load data of the virtual power plants in the affected area;

[0063] Specifically, the data fusion processing involves adding meteorological data and electricity price data to the load data, then using a standard fusion algorithm to convert the data into the same format and reorganize the data structure to construct data that can be directly input into the model for training.

[0064] Step (4): Perform dimensionality reduction on the fused dataset using principal component analysis;

[0065] Specifically, principal component analysis (PCA) includes standardizing the original variables, calculating the correlation coefficient matrix, calculating eigenvalues ​​and eigenvectors, and determining the principal components. Using PCA to reduce the dimensionality of the fused dataset effectively reduces data complexity. The principle of PCA is easy to understand: it projects the original matrix onto a new orthogonal coordinate system, and by sequentially selecting the directions of the new coordinate axes, maximizes the projection of the matrix onto these axes, i.e., maximizes the variance of the principal component sample values. Preserving the projections of the data in these directions reduces the data dimensionality.

[0066] Step (5): Divide the training set data and test set data for demand-response virtual power plants and energy supply virtual power plants respectively;

[0067] Specifically, the dataset is split into two independent parts: one part is used as the training set and the other part is used as the test set. For example, 70% of the data can be used as the training set and 30% as the test set, or 80% of the data can be used as the training set and 20% as the test set. There is no specific limit to the proportion.

[0068] Step (6): Construct a first baseline load forecasting model and a second baseline load forecasting model for demand-response virtual power plants and energy supply virtual power plants respectively. The first baseline load forecasting model and the second baseline load forecasting model respectively include an input layer, a feature extraction layer, an attention layer, a feature fusion layer and a feature output layer connected in series.

[0069] Step (7): Using the attention mechanism, differential weighting is applied to the input features of the training data of the demand response virtual power plant and the energy supply virtual power plant respectively;

[0070] Step (8): Train the first baseline load forecasting model and the second baseline load forecasting model on the training set, and verify the trained first baseline load forecasting model and the second baseline load forecasting model on the test set;

[0071] Step (9): Based on the classification label of the virtual power plant to be predicted, input the relevant data into the corresponding baseline load prediction model, perform prediction through the trained baseline load prediction model, and output the baseline load prediction value;

[0072] Step (10): After the end of the day's operation, obtain the actual load value for the day, calculate the deviation between the actual load value and the baseline load forecast value for a certain period, and take the average value of the deviation as the compensation value. Correct the baseline load forecast value according to the compensation value.

[0073] It is easy to understand that due to the uncertainty of load response, there will be a deviation between the baseline load forecast and the actual load. For example, the response period of the load is from 10:00 to 12:00 on a given day, and the non-response periods are from 0:00 to 10:00 and from 12:00 to 24:00. In this case, the deviation between the actual load value and the baseline load forecast in the 0:00 to 10:00 and 12:00 to 24:00 periods can be calculated based on the actual load value of the day. The average value of the deviation can be taken as the compensation value, and the baseline load forecast can be corrected based on the compensation value.

[0074] Preferably, the k-means clustering analysis in step (2) uses the distance between points as a similarity evaluation index, and the distance algorithm is calculated using Euclidean distance:

[0075]

[0076] Where, x i It is the i-th variable value of sample x; y i It is the value of the i-th variable in sample y.

[0077] Preferably, the meteorological data in step (3) includes the highest temperature, lowest temperature, average temperature, sunshine, wind speed, humidity, and precipitation; the electricity price data includes peak electricity price, off-peak electricity price, and normal electricity price.

[0078] Preferably, in step (7), the input features of the training datasets for demand-response virtual power plants and energy-supply virtual power plants are weighted differently using an attention mechanism. Specifically, the electricity price factor in the input features of the training dataset for demand-response virtual power plants is weighted, and the weather factor in the input features of the training dataset for energy-supply virtual power plants is weighted.

[0079] Preferably, the weighted calculation formula for the input features of the training dataset is as follows:

[0080] y t =A(x t ,p t-1 ,q t-1 );

[0081]

[0082] x't=qt j xt j ;

[0083] Among them, y t Let A be the attention score matrix at time t; A(.) is the matrix used to calculate y. t Neural networks; x t p is the initial value input at time t; t-1 The predicted value at time t-1; q t-1 q represents the attention weight value at time t-1; tj For q t The attention weight value of the j-th feature, y tj For y t The attention score of the j-th feature, where n is the number of input features, x tj For x t The initial value of the j-th feature, x' t For x t Weighted features.

[0084] Example 2: In this example, a deep learning-based virtual power plant baseline load data prediction system is provided, which includes the following modules:

[0085] The data acquisition module is used to collect historical load data of all virtual power plants in the area and to preprocess the historical load data.

[0086] The clustering analysis module is used to perform k-means clustering analysis on the historical load data, classify virtual power plants into demand response virtual power plants and energy supply virtual power plants according to the operating characteristics of different types of virtual power plants, and record the category labels.

[0087] The fusion processing module is used to fuse meteorological data, electricity price data, and historical load data of virtual power plants within the affected area.

[0088] The principal component analysis module uses principal component analysis to reduce the dimensionality of the fused dataset.

[0089] The data partitioning module partitions the data into training set data and test set data for demand-response virtual power plants and energy supply virtual power plants, respectively.

[0090] The model building module constructs a first baseline load forecasting model and a second baseline load forecasting model for demand-response virtual power plants and energy supply virtual power plants, respectively. The first baseline load forecasting model and the second baseline load forecasting model each include an input layer, a feature extraction layer, an attention layer, a feature fusion layer and a feature output layer connected in series.

[0091] The differential weighting module uses an attention mechanism to differentially weight the input features of the training data for demand-response virtual power plants and energy-supply virtual power plants, respectively.

[0092] The training and validation module is used to train the first baseline load prediction model and the second baseline load prediction model on the training set, and to validate the trained first baseline load prediction model and the second baseline load prediction model on the test set.

[0093] The prediction output module inputs relevant data into the corresponding baseline load prediction model based on the classification label of the virtual power plant to be predicted, performs prediction through the trained baseline load prediction model, and outputs the baseline load prediction value.

[0094] The prediction and correction module obtains the actual load value of the day after the end of the operating day, calculates the deviation between the actual load value and the baseline load prediction value for a certain period of time, takes the average value of the deviation as the compensation value, and corrects the baseline load prediction value based on the compensation value.

[0095] Preferably, the k-means clustering analysis in the clustering analysis module uses the distance between points as a similarity evaluation index, and the distance algorithm uses Euclidean distance for calculation.

[0096]

[0097] Where, x i It is the i-th variable value of sample x; y i It is the value of the i-th variable in sample y.

[0098] Preferably, the meteorological data in the fusion processing module includes the highest temperature, lowest temperature, average temperature, sunshine, wind speed, humidity, and precipitation; the electricity price data includes peak electricity price, off-peak electricity price, and normal electricity price.

[0099] Preferably, in the differential weighting module, an attention mechanism is used to differentially weight the input features of the training datasets for demand-response virtual power plants and energy-supply virtual power plants, respectively. Specifically, the weight of the electricity price factor in the input features of the demand-response virtual power plant training dataset is increased, and the weight of the weather factor in the input features of the energy-supply virtual power plant training dataset is increased.

[0100] Preferably, the weighted calculation formula for the input features of the training dataset is as follows:

[0101] y t =A(x) t ,p t-1 ,q t-1 );

[0102]

[0103] x't=qt j xt j ;

[0104] Among them, y t Let A be the attention score matrix at time t; A(.) is the matrix used to calculate y. t Neural networks; x t p is the initial value input at time t; t-1 The predicted value at time t-1; q t-1 q represents the attention weight value at time t-1; tj For q t The attention weight value of the j-th feature, y tj For y t The attention score of the j-th feature, where n is the number of input features, x tj For x t The initial value of the j-th feature, x' t For x t Weighted features.

[0105] The technical solution of this invention improves the accuracy of baseline load forecasting for virtual power plants and helps to enhance the effectiveness of virtual power plant assessment.

[0106] Those skilled in the art will understand that the embodiments described herein can be provided as methods, apparatus (devices), or computer program products. Therefore, this document may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. This includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0107] This document is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments herein. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0108] The embodiments and / or implementation methods described above are merely preferred embodiments and / or implementation methods for implementing the technology of the present invention, and are not intended to limit the implementation methods of the technology of the present invention in any way. Any person skilled in the art can make some modifications or alterations to other equivalent embodiments without departing from the scope of the technical means disclosed in the content of the present invention, but they should still be regarded as the technology or embodiments that are substantially the same as the present invention.

Claims

1. A method for predicting baseline load data of a virtual power plant based on deep learning, characterized in that, Includes the following steps: Step (1): Collect historical load data of all virtual power plants in the area and preprocess the historical load data; Step (2): Perform k-means clustering analysis on the historical load data, and classify the virtual power plants into demand response virtual power plants and energy supply virtual power plants according to the operating characteristics of different types of virtual power plants, and record the category labels; Step (3): Merge the meteorological data, electricity price data and historical load data of the virtual power plants in the affected area; Step (4): Perform dimensionality reduction on the fused dataset using principal component analysis; Step (5): Divide the training set data and test set data for demand-response virtual power plants and energy supply virtual power plants respectively; Step (6): Construct a first baseline load forecasting model and a second baseline load forecasting model for demand-response virtual power plants and energy supply virtual power plants respectively. The first baseline load forecasting model and the second baseline load forecasting model respectively include an input layer, a feature extraction layer, an attention layer, a feature fusion layer and a feature output layer connected in series. Step (7): Using the attention mechanism, differential weighting is applied to the input features of the training data of the demand response virtual power plant and the energy supply virtual power plant respectively; Step (8): Train the first baseline load forecasting model and the second baseline load forecasting model on the training set, and verify the trained first baseline load forecasting model and the second baseline load forecasting model on the test set; Step (9): Based on the classification label of the virtual power plant to be predicted, input the relevant data into the corresponding baseline load prediction model, perform prediction through the trained baseline load prediction model, and output the baseline load prediction value; Step (10): After the end of the day's operation, obtain the actual load value for the day, calculate the deviation between the actual load value and the baseline load forecast value for a certain period, and take the average value of the deviation as the compensation value. Correct the baseline load forecast value according to the compensation value.

2. The method for predicting baseline load data of a virtual power plant based on deep learning according to claim 1, characterized in that, The k-means clustering analysis in step (2) uses the distance between points as a similarity evaluation index, and the distance algorithm uses Euclidean distance to calculate the distance: Where, x i y is the i-th variable value of sample x, where n represents the total number of variables; i It is the value of the i-th variable in sample y.

3. The method for predicting baseline load data of a virtual power plant based on deep learning according to claim 1, characterized in that, The meteorological data in step (3) includes the highest temperature, lowest temperature, average temperature, sunshine, wind speed, humidity, and precipitation; the electricity price data includes peak electricity price, off-peak electricity price, and normal electricity price.

4. The method for predicting baseline load data of a virtual power plant based on deep learning according to claim 1, characterized in that, In step (7), the attention mechanism is used to differentially weight the input features of the training datasets of demand-response virtual power plants and energy supply virtual power plants respectively. Specifically, the weight of the electricity price factor in the input features of the training dataset of demand-response virtual power plants is increased, and the weight of the weather factor in the input features of the training dataset of energy supply virtual power plants is increased.

5. The deep learning-based virtual power plant baseline load data prediction method according to claim 4, characterized in that, The weighted calculation formula for the input features of the training dataset is as follows: y t =A(x t ,p t-1 ,q t-1 ); x' t =q tj x tj ; Among them, y t Let A be the attention score matrix at time t; A(.) is the matrix used to calculate y. t Neural networks; x t p is the initial value input at time t; t-1 The predicted value at time t-1; q t-1 q represents the attention weight value at time t-1; tj For q t The attention weight value of the j-th feature, y tj For y t The attention score of the j-th feature, where n is the number of input features, x tj For x t The initial value of the j-th feature, x' t For x t Weighted features.

6. A deep learning-based virtual power plant baseline load data prediction system, characterized in that, Includes the following modules: The data acquisition module is used to collect historical load data of all virtual power plants in the area and to preprocess the historical load data. The clustering analysis module is used to perform k-means clustering analysis on the historical load data, classify virtual power plants into demand response virtual power plants and energy supply virtual power plants according to the operating characteristics of different types of virtual power plants, and record the category labels. The fusion processing module is used to fuse meteorological data, electricity price data, and historical load data of virtual power plants within the affected area. The principal component analysis module uses principal component analysis to reduce the dimensionality of the fused dataset. The data partitioning module partitions the data into training set data and test set data for demand-response virtual power plants and energy supply virtual power plants, respectively. The model building module constructs a first baseline load forecasting model and a second baseline load forecasting model for demand-response virtual power plants and energy supply virtual power plants, respectively. The first baseline load forecasting model and the second baseline load forecasting model each include an input layer, a feature extraction layer, an attention layer, a feature fusion layer and a feature output layer connected in series. The differential weighting module uses an attention mechanism to differentially weight the input features of the training data for demand-response virtual power plants and energy-supply virtual power plants, respectively. The training and validation module is used to train the first baseline load prediction model and the second baseline load prediction model on the training set, and to validate the trained first baseline load prediction model and the second baseline load prediction model on the test set. The prediction output module inputs relevant data into the corresponding baseline load prediction model based on the classification label of the virtual power plant to be predicted, performs prediction through the trained baseline load prediction model, and outputs the baseline load prediction value. The prediction and correction module obtains the actual load value of the day after the end of the operating day, calculates the deviation between the actual load value and the baseline load prediction value for a certain period of time, takes the average value of the deviation as the compensation value, and corrects the baseline load prediction value based on the compensation value.

7. The deep learning-based virtual power plant baseline load data prediction system according to claim 6, characterized in that, The k-means clustering analysis in the clustering analysis module uses the distance between points as a similarity evaluation index, and the distance algorithm uses Euclidean distance for calculation. Where, x i y is the i-th variable value of sample x, where n represents the total number of variables; i It is the value of the i-th variable in sample y.

8. The deep learning-based virtual power plant baseline load data prediction system according to claim 6, characterized in that, The meteorological data in the fusion processing module includes the highest temperature, lowest temperature, average temperature, sunshine, wind speed, humidity, and precipitation; the electricity price data includes peak electricity price, off-peak electricity price, and normal electricity price.

9. The deep learning-based virtual power plant baseline load data prediction system according to claim 6, characterized in that, In the differential weighting module, an attention mechanism is used to differentially weight the input features of the training datasets for demand-response virtual power plants and energy-supply virtual power plants, respectively. Specifically, the weight of the electricity price factor in the input features of the demand-response virtual power plant training dataset is increased, and the weight of the weather factor in the input features of the energy-supply virtual power plant training dataset is increased.

10. The deep learning-based virtual power plant baseline load data prediction system according to claim 9, characterized in that, The weighted calculation formula for the input features of the training dataset is as follows: y t =A(x t ,p t-1 ,q t-1 ); x' t =q tj x tj ; Among them, y t Let A be the attention score matrix at time t; A(.) is the matrix used to calculate y. t Neural networks; x t p is the initial value input at time t; t-1 The predicted value at time t-1; q t-1 q represents the attention weight value at time t-1; tj For q t The attention weight value of the j-th feature, y tj For y t The attention score of the j-th feature, where n is the number of input features, x tj For x t The initial value of the j-th feature, x' t For x t Weighted features.

Images