A virtual power plant capacity prediction method and system

Abstract

The present invention relates to the field of data processing technology, and in particular to a method and system for predicting capacity of a virtual power plant. The method obtains a historical capacity factor matrix; obtains eigenvalues ​​of the historical capacity factor matrix; obtains a denoised capacity factor matrix according to the eigenvalue distribution of the historical capacity factor matrix; analyzes the lag relationship between each sequence in the denoised capacity factor matrix and a denoised historical capacity information sequence to obtain a marginal characteristic lag matrix; then analyzes the correlation between each sequence in the marginal characteristic lag matrix and the denoised historical capacity information sequence to obtain a modified marginal characteristic lag matrix; completes the prediction of real-time capacity according to the modified marginal characteristic lag matrix, and makes the predicted capacity result more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method and system for predicting the capacity of a virtual power plant. Background Art

[0002] With the development of society and the continuous reform of the global power market, electric energy has become a topic of concern to everyone. Accurately predicting the capacity of virtual power plants can provide an important basis for energy planning and operational decisions. By understanding future capacity levels, energy planners and operators can formulate reasonable resource allocation and scheduling strategies to meet power demand, ensure the stability of energy supply, and optimize the operating efficiency of the system. At the same time, as an integration platform for distributed energy, the capacity forecast of virtual power plants is crucial to achieving supply and demand balance and maintaining grid stability.

[0003] Normally, the methods for predicting the capacity of virtual power plants are mostly based on the fluctuation curve of the capacity itself and its power production related factors. However, the capacity of virtual power plants is often affected by many factors, and conventional technologies usually cannot take into account the influence of the time sequence between sequences and cannot eliminate the influence of accidental factors when predicting. Summary of the invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a method and system for predicting the capacity of a virtual power plant. The technical solutions adopted are as follows:

[0005] In a first aspect, an embodiment of the present invention provides a method for predicting capacity of a virtual power plant, the method comprising the following steps:

[0006] Obtain a historical capacity factor matrix; the historical capacity factor matrix is ​​composed of a historical capacity information sequence, a new energy power generation ratio sequence, an average sunshine intensity change sequence in the area where the virtual power plant is located, a power grid load sequence, and a wind power change sequence in the area where the virtual power plant is located;

[0007] Obtain the eigenvalues ​​of the historical capacity factor matrix according to the historical capacity factor matrix; obtain the minimum characteristic truncation parameters of the historical capacity factor matrix according to the eigenvalues ​​of the historical capacity factor matrix; obtain the denoised capacity factor matrix according to the minimum characteristic truncation parameters of the historical capacity factor matrix; obtain the optimal lag order of each sequence in the denoised capacity factor matrix according to the lag relationship between each sequence in the denoised capacity factor matrix and the denoised historical capacity information sequence; obtain the marginal characteristic lag matrix according to the optimal lag order of each sequence in the denoised capacity factor matrix; obtain the weight coefficient of each sequence according to the correlation between each sequence in the marginal characteristic lag matrix and the denoised historical capacity information sequence; obtain the modified marginal characteristic lag matrix according to the weight coefficient of each sequence in the marginal characteristic lag matrix;

[0008] The production capacity is predicted based on the modified marginal characteristic lag matrix.

[0009] Furthermore, the eigenvalues ​​of the historical capacity factor matrix are obtained according to the historical capacity factor matrix, specifically:

[0010] Perform singular value decomposition on the historical capacity factor matrix to obtain the eigenvalues ​​of the historical capacity factor matrix.

[0011] Furthermore, the minimum characteristic truncation parameter of the historical capacity factor matrix is ​​obtained according to the characteristic value of the historical capacity factor matrix, specifically:

[0012] Arrange the eigenvalues ​​of the historical capacity factor matrix from large to small to obtain an eigenvalue sequence, calculate the k value that satisfies the requirement that the sum of the first k eigenvalues ​​accounts for more than 90% of the sum of all eigenvalues, and take the minimum value of the k value as the minimum characteristic cutoff parameter.

[0013] Furthermore, the step of obtaining a denoised capacity factor matrix according to the minimum characteristic cutoff parameter of the historical capacity factor matrix includes:

[0014] The eigenvalue sequence is truncated into two parts using the minimum eigenvalue truncation parameter, the eigenvalues ​​whose sequence numbers are less than or equal to the minimum eigenvalue truncation parameter in the eigenvalue sequence are retained, and the eigenvalues ​​whose sequence numbers are greater than the minimum eigenvalue truncation parameter in the eigenvalue sequence are set to zero, and the truncated eigenvalue matrix is ​​obtained;

[0015] The truncated eigenvalue matrix is ​​multiplied by its corresponding eigenvalue matrix to obtain the denoised capacity factor matrix.

[0016] Furthermore, the optimal lag order of each sequence in the denoised capacity factor matrix is ​​obtained according to the lag relationship between each sequence in the denoised capacity factor matrix and the denoised historical capacity information sequence, specifically:

[0017] Obtain the optimal alignment path between the denoised historical capacity information sequence and other sequences in the denoised capacity factor matrix according to a dynamic programming algorithm;

[0018] The time difference between corresponding elements in the optimal alignment path is taken as the lag order; for each lag order, the dedimensionalized denoised historical capacity information sequence is moved forward by a corresponding number of steps to achieve alignment; the sum of the Euclidean distances between the two aligned sequences is calculated; and the lag order that minimizes the sum is recorded as the optimal lag order.

[0019] Furthermore, the marginal characteristic lag matrix is ​​obtained according to the optimal lag order of each sequence in the denoised capacity factor matrix, and the acquisition method is:

[0020] According to the optimal lag order of each sequence in the denoising capacity factor matrix, the corresponding sequence is shifted back as a whole by a number of steps equal to the optimal lag order to obtain a sequence with adjusted lag order;

[0021] The matrix composed of the denoised historical capacity information sequence in the denoised capacity factor and the sequence adjusted by the lag order is recorded as the marginal characteristic lag matrix.

[0022] Furthermore, the weight coefficient of each sequence is obtained according to the correlation between each sequence in the marginal characteristic lag matrix and the denoised historical capacity information sequence, specifically including:

[0023] Grey correlation analysis is used to calculate the correlation coefficients between each sequence except the denoised historical capacity information sequence in the marginal characteristic lag matrix and the denoised historical capacity information sequence; the ratio of the correlation coefficient to the sum of all correlation coefficients is used as the weight coefficient of each sequence in the marginal characteristic lag matrix.

[0024] Furthermore, the step of obtaining a modified marginal characteristic lag matrix according to the weight coefficients of each sequence in the marginal characteristic lag matrix specifically includes:

[0025] The matrix obtained by multiplying the weight coefficient of each sequence in the marginal characteristic lag matrix with the corresponding sequence in the marginal characteristic lag matrix is ​​recorded as the modified marginal characteristic lag matrix.

[0026] Furthermore, the production capacity is predicted based on the modified marginal characteristic lag matrix, and the specific steps include:

[0027] The corrected marginal feature lag matrix is ​​used as the input of the long short-term memory network LSTM, and the predicted capacity is used as the output of the long short-term memory network LSTM.

[0028] In a second aspect, an embodiment of the present invention further provides a virtual power plant capacity prediction system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the steps of any one of the above methods when executing the computer program.

[0029] The present invention has at least the following beneficial effects:

[0030] The present invention mainly predicts the real-time capacity of the virtual power plant by obtaining factors related to capacity information. Specifically, the sequences of each factor are combined into a feature matrix, and then the singular value decomposition method is used to remove the interference of irregular factors in the data, so that the prediction results are not affected by random factors, and the denoised capacity factor matrix is ​​obtained. Then, the sequences of the denoised capacity factor matrix are lagged through dynamic time warping, so that the causal relationship between the sequences can be taken into account, and the changes in capacity information can be grasped more accurately.

[0031] At the same time, each sequence in the denoised capacity factor matrix is ​​analyzed according to the lag relationship to obtain the optimal lag order between each sequence, and the marginal characteristic lag matrix is ​​obtained by combining the denoised historical capacity information sequence. The marginal characteristic lag matrix is ​​corrected by using the correlation between each sequence and the denoised historical capacity information sequence, and the influence of different factors on capacity is considered to further improve the accuracy of the capacity forecast results. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0033] Figure 1 A flowchart of a method for predicting capacity of a virtual power plant provided by an embodiment of the present invention;

[0034] Figure 2 Flowchart for obtaining the modified marginal characteristic lag matrix. DETAILED DESCRIPTION

[0035] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of a virtual power plant capacity prediction method and system proposed by the present invention, its specific implementation method, structure, features and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0036] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0037] The specific scheme of a virtual power plant capacity prediction method and system provided by the present invention is described in detail below with reference to the accompanying drawings.

[0038] See also Figure 1 , which shows a flowchart of a method for predicting capacity of a virtual power plant provided by an embodiment of the present invention, the method comprising the following steps:

[0039] Step S001: Acquire data information related to the real-time production capacity of the virtual power plant.

[0040] The data information related to the real-time capacity of the virtual power plant includes: historical capacity information of the target area in the past year, the proportion of new energy power generation, and the total power load information. Then, historical meteorological data is obtained from the China Meteorological Science Data Sharing Service Network (CMA). Among them, capacity-related information is obtained from the official website of the target area power grid, and other energy-related information can be obtained from the official website of the target area statistics bureau. According to market rules, the change in capacity is usually slow, so here we use every It should be noted that the time interval for resampling can be adjusted by the implementer according to the actual situation. is 6.

[0041] The historical capacity information sequence is recorded as , the proportion of new energy power generation is , the average sunshine intensity variation sequence of the area where the virtual power plant is located is , the grid load sequence is , the wind power change sequence in the area where the virtual power plant is located is .

[0042] Step S002: According to the real-time capacity-related data information of the virtual power plant, data denoising is performed, and its lag and correlation are analyzed to construct a modified marginal characteristic lag matrix.

[0043] Through the above steps, we can obtain historical data related to real-time production capacity. In order to integrate different time series data to build a comprehensive data set and use the sliding window method to build training samples, we can improve data consistency, enrich feature information, increase the number of training samples, and provide more comprehensive input for predictive modeling, thereby improving the accuracy and reliability of prediction results.

[0044] Since the above data are all sampled at a unified time interval and the sampling range is the data of the most recent year, the data of each series are unified in timestamp. , the proportion of new energy power generation , average sunshine intensity change series , Grid load sequence and wind speed change sequence Spliced ​​into a historical capacity factor matrix , specifically:

[0045]

[0046] In the formula, Represents the historical capacity factor matrix, which contains historical data information on the main factors affecting capacity changes. , , , , They respectively represent the historical production capacity information sequence, the proportion of new energy power generation sequence, the average sunshine intensity change sequence, the power grid load sequence, and the wind power change sequence.

[0047] 1) Remove random noise factors from the data.

[0048] Since historical capacity information will be affected by many unexpected factors during data acquisition, and we do not consider the impact of unexpected events when predicting real-time capacity, it is necessary to filter the noise data of the historical capacity factor matrix. Here, the singular value decomposition (SVD) method is used to calculate the eigenvalues ​​in the historical capacity factor matrix for analysis, as follows;

[0049]

[0050] in, represents the historical capacity factor matrix, which is decomposed into the form of three matrix products through singular value decomposition; is a diagonal matrix, representing the eigenvalues ​​of the historical capacity factor matrix; and are the characteristic matrices corresponding to the eigenvalues.

[0051] The larger the eigenvalue, the more important the corresponding eigenvector. In order to eliminate the influence of random factors in the data, this embodiment adopts the method of truncation of eigenvalues ​​to remove the noise data in the original matrix. It should be noted that the implementer can choose the method of removing noise data according to the actual situation. In this embodiment, the diagonal matrix The eigenvalues ​​in are sorted from large to small and are recorded as , and then retain no more than 90% of the eigenvalues ​​according to the sum of the eigenvalues. In order to retain the first 90% of the eigenvalues, the minimum feature truncation parameter is calculated by the following formula to satisfy:

[0052]

[0053] In the formula, represents the minimum feature cutoff parameter, Indicates The size of the eigenvalues, represents the number of all eigenvalues, Before The sum of the eigenvalues, represents the sum of all eigenvalues.

[0054] When the conditions are met After the value, keep the previous eigenvalues ​​and set the rest to 0.

[0055] Finally, the truncated eigenvalue matrix The corresponding feature matrix and Multiplying them together will give us the noise-filtered historical capacity factor matrix , where the historical production capacity information sequence, new energy power generation ratio sequence, average sunshine intensity change sequence, grid load sequence, and wind power change sequence after SVD noise filtering are recorded as , , , , .

[0056] At this point, the denoised historical capacity factor matrix is ​​obtained .

[0057] 2) Based on the denoised historical capacity factor matrix Constructing the Marginal Characteristics Lag Matrix .

[0058] When analyzing the relationship between the proportion of new energy power generation, the average sunshine intensity change sequence, the grid load sequence, and the wind power change sequence and the capacity, some of these factors may have a lag in the impact on the capacity. Therefore, this embodiment considers constructing a lag matrix to eliminate the impact of the lag of sequence information. The application of the lag matrix can also help capture the lag effect of these factors on the capacity, better understand the dynamic relationship between them, and provide a useful reference for power market policy formulation and capacity forecasting.

[0059] For the above sequences, calculate their differences from the denoised historical capacity information sequence Next, for the denoised renewable energy power generation ratio series and denoising historical capacity information sequence The hysteresis relationship between them is described in detail.

[0060] Considering that the measurement scales of the two sequences may be different, the sequences are first dedimensionalized, as follows:

[0061]

[0062] In the formula, represents the number of the de-dimensionalized denoised historical capacity information sequence data, Represents the first data, Represents the mean of the denoised historical capacity information series.

[0063] At this point, the dimensionless and denoised historical capacity information sequence is obtained: , and similarly, we get the dimensionless denoised renewable energy power generation ratio series .

[0064] Considering that the change in the proportion of new energy power generation at the current time will lead to changes in future production capacity, if the proportion of new energy power generation at the current time increases, the future production capacity will increase. This embodiment uses the idea of ​​dynamic time warping to analyze the lag relationship between the de-dimensionalized denoised historical production capacity information sequence and the de-dimensionalized denoised new energy power generation proportion sequence:

[0065] A) Use dynamic programming algorithm to obtain the dimensionless and denoised renewable energy power generation ratio sequence and the dimensionless denoised historical capacity information sequence The distance matrix records the distance measure between each time point, which can reflect the similarity between sequences. Here, the distance is calculated using the Euclidean distance. Then, based on the distance matrix, the path with the smallest sum of distances is selected as the optimal alignment path through the dynamic programming algorithm. Among them, the dynamic programming algorithm is an existing well-known technology and will not be repeated in this embodiment. The implementer can select a suitable distance measure according to the actual situation. In this embodiment, the Euclidean distance is used.

[0066] B) The time difference between the elements corresponding to the best alignment path is used as the lag order, the two sequences are aligned according to different lag orders, and the sum of the Euclidean distances of each of the two aligned sequences is further calculated. It should be noted that the lag order can be selected by the implementer according to the actual situation. In this embodiment, the lag order with the smallest sum of the Euclidean distances is selected as the best lag order between the two sequences.

[0067] Next, the de-dimensionalized and denoised renewable energy generation ratio sequence is adjusted through the optimal lag order, as follows:

[0068]

[0069] In the formula, It means that the dimensionless denoised renewable energy power generation ratio series is obtained through the optimal lag order The i-th data after adjustment, represents the optimal lag order for the two series, represents the number of units in the de-dimensionalized denoised renewable energy generation ratio sequence. data. When The value is the dimensionless and denoised average sunshine intensity in the area where the virtual power plant is located.

[0070] Similarly, the denoised average sunshine intensity change sequence can be obtained respectively , denoising grid load sequence and denoised wind speed variation sequence and denoising historical capacity information sequence The lag relationship between them is obtained, and the corresponding sequence adjusted according to the optimal lag order is obtained , , . And use this to construct the marginal characteristic lag matrix ,as follows:

[0071]

[0072] In the formula, represents the marginal characteristic lag matrix, represents the denoised historical capacity information sequence, , , , They respectively represent the average sunshine intensity in the area where the virtual power plant is located, the renewable energy power generation ratio sequence, the grid load sequence, and the wind speed change in the area where the virtual power plant is located, adjusted according to the optimal lag order, and are recorded as the average sunshine intensity in the area where the virtual power plant is located with lag denoising, the renewable energy power generation ratio sequence, the grid load sequence, and the wind speed change in the area where the virtual power plant is located with lag denoising.

[0073] At this point, the marginal characteristic lag matrix of the characteristic sequence is obtained.

[0074] 3) Modify the marginal characteristic lag matrix according to the degree of correlation between different sequences and production capacity.

[0075] In the marginal characteristic lag matrix obtained above, different influencing factor sequences have different degrees of influence on the capacity sequence. Therefore, we need to assign different weights to them for distinction.

[0076] Specifically, the grey correlation analysis is used to calculate the correlation between each lag sequence and the denoised historical capacity information sequence as the weight of each lag sequence. It should be noted that the grey correlation analysis is a well-known technology and will not be described in detail in this embodiment.

[0077] Then convert it into the corresponding weight coefficient as follows:

[0078]

[0079]

[0080]

[0081]

[0082] In the formula, , , , They respectively represent the weight coefficients corresponding to the lag denoising renewable energy power generation proportion sequence, the lag denoising average sunshine intensity change sequence, the lag denoising grid load sequence and the lag denoising wind power change sequence; , , , They respectively represent the correlation coefficients between the lagged denoised renewable energy power generation proportion series, the lagged denoised average sunshine intensity change series, the lagged denoised power grid load series and the lagged denoised wind power change series and the denoised historical capacity information series.

[0083] Next, the marginal feature lag matrix above is transformed according to the weight coefficient After correction, the modified marginal characteristic lag matrix R is obtained as follows:

[0084]

[0085] In the formula, , , , They respectively represent the weight coefficients corresponding to the lag denoising renewable energy power generation proportion sequence, the lag denoising average sunshine intensity change sequence, the lag denoising grid load sequence and the lag denoising wind power change sequence; Represents the denoised historical capacity information sequence; , , , They represent the renewable energy generation proportion sequence, average sunshine intensity change sequence, grid load sequence, and wind power change sequence adjusted according to the optimal lag order. , , , They are recorded as the lag denoising renewable energy generation proportion sequence, the lag denoising average sunshine intensity change sequence, the lag denoising grid load sequence and the lag denoising wind power change sequence. The acquisition process of the modified marginal characteristic lag matrix is ​​as follows: Figure 2 shown.

[0086] At this point, we have obtained the modified marginal characteristic lag matrix.

[0087] Step S003: According to the modified marginal feature lag matrix, the long short-term memory network LSTM model is used to perform capacity forecasting.

[0088] The production capacity is predicted in real time based on the modified marginal characteristic lag matrix obtained in the above steps.

[0089] Here, the long short-term memory network LSTM is used to predict the production capacity, the optimizer is Adam, the loss function is the mean square error function, and the model is optimized in the direction of reducing the loss during the training process. It should be noted that the long short-term memory network LSTM is an existing well-known technology and will not be described in detail in this embodiment.

[0090] Specifically, the training samples are constructed according to the modified marginal feature lag matrix: a) the input sequence is determined to be all the sequences of the modified marginal feature lag matrix, and the output sequence is the predicted capacity sequence; b) the window size of a sample is determined to be 28, that is, one week of data, and the 4 consecutive data in the predicted capacity sequence immediately following the window are used as sample labels, that is, the capacity change of one day is predicted; c) by setting the step size to 1, sliding on the sequence, and obtaining a sample each time sliding. The implementer can adjust the step size according to the actual situation.

[0091] Then, the training samples are divided into training set, validation set and test set, where the size of validation set and test set are both 10% of the total samples. It should be noted that the size of validation set and test set can be adjusted by the implementer according to the actual situation.

[0092] At this point, the real-time capacity prediction function of the virtual power plant is completed.

[0093] Based on the same inventive concept as the above method, an embodiment of the present invention also provides a virtual power plant capacity prediction system, including a memory, a processor, and a computer program stored in the memory and running on the processor, and when the processor executes the computer program, the steps of any one of the above-mentioned virtual power plant capacity prediction methods are implemented.

[0094] In summary, the embodiment of the present invention mainly predicts the real-time capacity of the virtual power plant by obtaining factors related to the capacity information. Specifically, each factor sequence is combined into a feature matrix, and then the singular value decomposition method is used to remove the interference of irregular factors in the data, so that the prediction results are not affected by random factors, and the denoised capacity factor matrix is ​​obtained. Then, each sequence of the denoised capacity factor matrix is ​​lagged through dynamic time warping, so that the causal relationship between the sequences can be taken into account, and the changes in capacity information can be grasped more accurately.

[0095] At the same time, each sequence in the denoised capacity factor matrix is ​​analyzed according to the lag relationship to obtain the optimal lag order between each sequence, and the marginal characteristic lag matrix is ​​obtained by combining the denoised historical capacity information sequence. The marginal characteristic lag matrix is ​​corrected by using the correlation between each sequence and the denoised historical capacity information sequence, and the influence of different factors on capacity is considered to further improve the accuracy of the capacity forecast results.

[0096] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The above is a description of a specific embodiment of this specification. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0097] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0098] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting the capacity of a virtual power plant, characterized in that: The method comprises the following steps: Obtain a historical capacity factor matrix; the historical capacity factor matrix is ​​composed of a historical capacity information sequence, a new energy power generation ratio sequence, an average sunshine intensity change sequence in the area where the virtual power plant is located, a power grid load sequence, and a wind power change sequence in the area where the virtual power plant is located; Obtain the eigenvalues ​​of the historical capacity factor matrix according to the historical capacity factor matrix; obtain the minimum characteristic truncation parameters of the historical capacity factor matrix according to the eigenvalues ​​of the historical capacity factor matrix; obtain the denoised capacity factor matrix according to the minimum characteristic truncation parameters of the historical capacity factor matrix; obtain the optimal lag order of each sequence in the denoised capacity factor matrix according to the lag relationship between each sequence in the denoised capacity factor matrix and the denoised historical capacity information sequence; obtain the marginal characteristic lag matrix according to the optimal lag order of each sequence in the denoised capacity factor matrix; obtain the weight coefficient of each sequence according to the correlation between each sequence in the marginal characteristic lag matrix and the denoised historical capacity information sequence; obtain the modified marginal characteristic lag matrix according to the weight coefficient of each sequence in the marginal characteristic lag matrix; The production capacity is predicted based on the modified marginal characteristic lag matrix.

2. A virtual power plant capacity prediction method according to claim 1, characterized in that: The method of obtaining the eigenvalues ​​of the historical capacity factor matrix according to the historical capacity factor matrix is ​​specifically as follows: Perform singular value decomposition on the historical capacity factor matrix to obtain the eigenvalues ​​of the historical capacity factor matrix.

3. A virtual power plant capacity prediction method according to claim 1, characterized in that: The method of obtaining the minimum characteristic cutoff parameter of the historical capacity factor matrix according to the characteristic value of the historical capacity factor matrix is ​​specifically: Arrange the eigenvalues ​​of the historical capacity factor matrix from large to small to obtain an eigenvalue sequence, calculate the k value that satisfies the requirement that the sum of the first k eigenvalues ​​accounts for more than 90% of the sum of all eigenvalues, and take the minimum value of the k value as the minimum characteristic cutoff parameter.

4. A virtual power plant capacity prediction method according to claim 3, characterized in that: The step of obtaining the denoised capacity factor matrix according to the minimum characteristic cutoff parameter of the historical capacity factor matrix includes: The eigenvalue sequence is truncated into two parts using the minimum eigenvalue truncation parameter, the eigenvalues ​​whose sequence numbers are less than or equal to the minimum eigenvalue truncation parameter in the eigenvalue sequence are retained, and the eigenvalues ​​whose sequence numbers are greater than the minimum eigenvalue truncation parameter in the eigenvalue sequence are set to zero, and the truncated eigenvalue matrix is ​​obtained; The truncated eigenvalue matrix is ​​multiplied by its corresponding eigenvalue matrix to obtain the denoised capacity factor matrix.

5. A virtual power plant capacity prediction method according to claim 1, characterized in that: The optimal lag order of each sequence in the denoised capacity factor matrix is ​​obtained according to the lag relationship between each sequence in the denoised capacity factor matrix and the denoised historical capacity information sequence, specifically: Obtain the optimal alignment path between the denoised historical capacity information sequence and other sequences in the denoised capacity factor matrix according to a dynamic programming algorithm; The time difference between corresponding elements in the optimal alignment path is taken as the lag order; for each lag order, the dedimensionalized denoised historical capacity information sequence is moved forward by a corresponding number of steps to achieve alignment; the sum of the Euclidean distances between the two aligned sequences is calculated; and the lag order that minimizes the sum is recorded as the optimal lag order.

6. A virtual power plant capacity prediction method according to claim 1, characterized in that: The marginal characteristic lag matrix is ​​obtained according to the optimal lag order of each sequence in the denoised capacity factor matrix, and the acquisition method is: According to the optimal lag order of each sequence in the denoising capacity factor matrix, the corresponding sequence is shifted back as a whole by a number of steps equal to the optimal lag order to obtain a sequence with adjusted lag order; The matrix composed of the denoised historical capacity information sequence in the denoised capacity factor and the sequence adjusted by the lag order is recorded as the marginal characteristic lag matrix.

7. A virtual power plant capacity prediction method according to claim 1, characterized in that: The step of obtaining the weight coefficient of each sequence according to the correlation between each sequence in the marginal characteristic lag matrix and the denoised historical capacity information sequence specifically includes: Grey correlation analysis is used to calculate the correlation coefficients between each sequence except the denoised historical capacity information sequence in the marginal characteristic lag matrix and the denoised historical capacity information sequence; the ratio of the correlation coefficient to the sum of all correlation coefficients is used as the weight coefficient of each sequence in the marginal characteristic lag matrix.

8. A virtual power plant capacity prediction method according to claim 1, characterized in that: The step of obtaining the modified marginal characteristic lag matrix according to the weight coefficients of each sequence in the marginal characteristic lag matrix specifically includes: The matrix obtained by multiplying the weight coefficient of each sequence in the marginal characteristic lag matrix with the corresponding sequence in the marginal characteristic lag matrix is ​​recorded as the modified marginal characteristic lag matrix.

9. A virtual power plant capacity prediction method according to claim 1, characterized in that: The specific steps of completing the prediction of production capacity according to the modified marginal characteristic lag matrix include: The corrected marginal feature lag matrix is ​​used as the input of the long short-term memory network LSTM, and the predicted capacity is used as the output of the long short-term memory network LSTM.

10. A virtual power plant capacity prediction system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 9 are implemented.

Images