A short-term forecasting method for flexibility demand of renewable energy generation

Abstract

The present invention relates to a short-term forecasting method for flexibility demand of renewable energy power generation, comprising: calculating and obtaining historical data of flexibility demand of renewable energy power generation, obtaining basic flexibility demand of renewable energy power generation with 15 minutes as a basic period within the forecast day; calculating fluctuation coefficient of flexibility demand of renewable energy power generation in each period within the forecast day; and calculating flexibility demand of renewable energy power generation within the forecast day. The present invention is based on the analysis of optimization mechanism of traditional multi-objective grey wolf algorithm. The present invention adopts iterative chaotic sequence to initialize individuals of the algorithm, thereby ensuring the dispersion of individuals; proposes to divide the alpha wolf population and eliminate individuals, thereby improving the individual diversity of the algorithm; adopts piecewise sinusoidal convergence operator to replace the original linear decreasing convergence operator, divides the convergence process of the algorithm into two parts, namely search and encirclement, and optimizes them respectively, thereby improving optimization efficiency in the early stage of the algorithm and convergence accuracy in the later stage.

Description

Technical Field

[0001] The invention relates to the technical field of renewable energy power generation demand forecasting, in particular to a short-term forecasting method for flexibility demand of renewable energy power generation. Background Art

[0002] In recent years, intermittent renewable energy generation represented by wind power and photovoltaics has developed rapidly in my country. However, the uncertainty of wind / solar output has led to a significant increase in the operating risks of the traditional power system with coal-fired power as the main power source. Coal-fired power units start slowly and have minimum technical output constraints. Wind power and photovoltaics rely heavily on the power regulation of coal-fired power units to smooth out the inherent output volatility of wind power and photovoltaics, resulting in the following problems: if the coal-fired power is started too much, it will cause a large amount of wind and solar power to be abandoned; conversely, it may cause the regulation to fail to keep up with the rapid fluctuations of renewable energy, causing the risk of power outages.

[0003] In the past, when new energy sources had not yet been developed on a large scale, power balance problems were mainly caused by load fluctuations. Since the rotational inertia of the traditional power system during stable operation can effectively suppress system imbalance, its load fluctuation law is strong and the random disturbance is small. Although the deterministic backup configuration is relatively extensive, it is sufficient to smooth out load fluctuations. However, with the continuous increase in the penetration rate of new energy power generation in the system, the uncertainty of the power system has also increased. The measures to provide grid redundancy in the traditional power grid are not only expensive but also inflexible, and cannot meet the needs of safe and stable operation of the power system. In order to ensure the balance of supply and demand during operation, it is necessary for the power system to have a certain degree of resilience and response capabilities, the so-called flexibility, to eliminate or reduce the negative impact of uncertain factors as much as possible and ensure the safe and stable operation of the power system.

[0004] Therefore, there is an urgent need for a flexibility demand forecasting method for renewable energy power generation to provide a more scientific and engineering-practical solution for the flexibility resource balance during power system planning and design, operation control, and optimal scheduling. Summary of the invention

[0005] The object of the present invention is to provide a method for short-term forecasting of flexibility demand of renewable energy power generation.

[0006] To achieve the above object, the present invention adopts the following technical solution: a short-term forecasting method for flexibility demand of renewable energy power generation, the method comprising the following steps in sequence:

[0007] (1) Select the renewable energy power generation power of the 30 historical days before the forecast date as historical data, with a data sampling period of 15 minutes, correct the abnormal data in the historical data, and calculate the historical data of the flexibility demand of renewable energy power generation

[0008] (2) Historical data on flexibility demand based on renewable energy generation Forecast the basic demand for renewable energy generation flexibility in 15-minute intervals during the forecast day

[0009] (3) Select the historical forecast data of the flexibility demand of renewable energy generation for the 30 historical days before the forecast date and historical data on flexibility demand for renewable energy generation Taking 15 minutes as the basic period, a flexibility demand forecast error matrix is ​​formed to calculate the flexibility demand fluctuation coefficient of renewable energy power generation in each period within the forecast day.

[0010] (4) Based on the flexibility of renewable energy generation and the flexibility demand fluctuation coefficient of renewable energy generation Calculate forecasted intra-day renewable energy generation flexibility needs

[0011] The step (1) specifically comprises the following steps:

[0012] (1a) Obtain the renewable energy generation power n historical days before the forecast date Where i represents the specific date of the historical day, i = 1, 2, ..., n, t represents the specific time period of the day, because the sampling period is 15 minutes, so t = 1, 2, ..., 96; n = 30;

[0013] (1b) The renewable energy generation power of n historical days before the forecast date is obtained The abnormal data in is corrected as follows:

[0014] (1b1) For isolated single point data of 0 or single point missing data, the average value method is used to process it. Suppose the renewable energy power generation power of a certain historical day i is for: If the data x at a certain time of the day If it is zero, then correct for x belongs to t;

[0015] (1b2) The method of comparing the overall load change is used to judge the load distortion data. That is, if the change of the data at a certain moment exceeds N times the average level at the same moment, it means that it is distorted data, that is, bad data. The load data change level is analyzed to obtain the basic average change δ at each moment of each day in n historical days. i :

[0016]

[0017] Where, t=1,2,...,96; i=1,2,...,n;

[0018] Then find the average change δ of n historical days:

[0019]

[0020] Taking δ as the basic judgment condition, if the change exceeds N times, the data is considered abnormal, that is:

[0021]

[0022] The abnormal data is processed by the method in step (1b1) to obtain the revised renewable energy power generation of n historical days before the forecast date.

[0023] (1c) Power generation based on renewable energy Calculate the historical data of the flexibility demand of renewable energy generation n historical days before the forecast date And construct the historical data matrix FlexP of flexibility demand for renewable energy generation:

[0024]

[0025] The step (2) specifically comprises the following steps:

[0026] (2a) The improved multi-objective grey wolf optimization algorithm IMOGWO is used to optimize the parameters of the least squares support vector machine LSSVM:

[0027] (2a1) Selecting the radial basis kernel function as the kernel function of LSSVM and initializing the position of the wolf pack, that is, initializing the relevant parameters of LSSVM, including the number of wolf packs, the number of alpha wolves, the size of the archive, and the maximum number of iterations;

[0028] (2a2) Calculate the individual fitness of gray wolves, select the alpha wolf population and eliminate individuals, and update the archive;

[0029] (2a3) Calculate the coefficient vector A:

[0030] A=2a(R ram -1)e=2(1-T / T max )(R ram -1)e (6)

[0031] Among them: a is the convergence operator, R ram is a random number between 0 and 1; e is a unit vector, T is the number of iterations, and Tmax is the maximum number of iterations;

[0032] Determine whether the bi-norm of the coefficient vector A is less than 1. If ||A||<1, it is determined that the wolf pack has locked on to the prey and is encircling and preying on it. The position of the subordinate population is updated according to the following formula;

[0033]

[0034] Where: X i is the position of the individual, X b is the position of the alpha wolf, D is the distance between the individual and the alpha wolf, L is the contraction vector; rand(0,1) represents a random number of 0 or 1; Represents dot multiplication operation; R λ is a constant, usually 1.5; R u and R v is a random number, which obeys the 0-1 normal distribution; Γ() is the standard gamma function;

[0035] If ||A||≥1, it is judged that the wolf pack is still in the stage of searching for prey, and the position of the subordinate population is updated according to the following formula:

[0036]

[0037] Where: X z To correct the individual; X w is the worst individual position; p * is the dynamic decreasing factor;

[0038] (2a4) Calculate the individual fitness of the wolf pack to obtain the non-dominated solution. At the same time, determine whether the archive is full. If it is full, replace the solution of the most crowded segment in the archive with the non-dominated solution obtained in the previous step, i.e., the position of the subordinate population; otherwise, directly include the non-dominated solution in the archive;

[0039] (2a5) Update parameters a and A and determine whether the number of iterations T is greater than T max , if yes, select the final solution from the archive; otherwise, return to step (2a3) for the next iteration;

[0040] (2a6) Using IMOGWO to optimize the parameters in LSSVM, and obtaining each LSSVM model optimized by IMOGWO;

[0041] (2b) The renewable energy generation power of n historical days before the forecast date Normalization is performed to eliminate the loss of data details caused by excessive load data fluctuations. Before entering LSSVM, these data are normalized. The specific formula is as follows:

[0042]

[0043] Where, t=1,2,...,96; i=1,2,...,30;

[0044] (2c) Using the variational mode analysis VMD algorithm, k intrinsic mode components IMF are obtained:

[0045] (2c1) The variational problem is to minimize the sum of the estimated bandwidths of the eigenmode components. The constraint is that the sum of the eigenmode components is the original signal. The formula is as follows:

[0046]

[0047] Where: {u k} is the set of intrinsic mode components IMF; {ω k} is the set of center frequencies; δ(t) is the pulse signal; K is the preset number of decompositions; f(t) is the original signal;

[0048] (2c2) The penalty factor α and the Lagrange multiplier λ(t) are used to transform the constrained variational problem into an unconstrained problem:

[0049]

[0050] (2c3) The unconstrained problem is solved by the alternating direction multiplier method to achieve effective separation of signal frequencies, where the iterative update formulas for the eigenmode component and the center frequency are:

[0051]

[0052]

[0053] Where: is the kth eigenmode component with center frequency ω at the n+1th iteration; is the center frequency of the kth eigenmode component at the n+1th iteration, that is, the center of gravity of the power spectrum of the current modal component. Update λ according to the following formula:

[0054]

[0055] According to formula (15), we can determine u k Whether the convergence conditions are met, if not, the update operation is continuously performed until the convergence conditions are met:

[0056]

[0057] (2c4) Input the k IMF subsequences obtained in step (2c) into each LSSVM model after the IMOGWO optimization described in step (2a), and superimpose and reconstruct the predicted values ​​of each IMF subsequence to obtain the basic demand for renewable energy flexibility with a basic period of 15 minutes within the final forecast day.

[0058] The step (3) specifically refers to: according to the historical forecast data of the flexibility demand of renewable energy power generation in the past 30 days before the forecast date, and historical data on flexibility demand for renewable energy generation Calculate the flexibility resource prediction error matrix Cpef:

[0059]

[0060] in: P N Rated installed capacity for electricity generation from renewable energy sources;

[0061] According to the flexible resource forecast error matrix Coef, the flexibility demand fluctuation coefficient of renewable energy generation in each period of the forecast day is calculated The specific calculation is as follows:

[0062]

[0063] The step (4) specifically refers to: according to the basic demand for renewable energy power generation flexibility and the coefficient of demand fluctuation of renewable energy generation flexibility The flexibility requirements for renewable energy generation during the forecasted day are calculated as follows:

[0064]

[0065] It can be seen from the above technical scheme that the beneficial effects of the present invention are: first, based on the analysis of the optimization mechanism of the traditional multi-objective gray wolf algorithm, the present invention adopts iterative chaotic sequence to initialize the individual of the algorithm, thereby ensuring the dispersion of the individuals; it proposes to divide the alpha wolf population and eliminate individuals, thereby improving the individual diversity of the algorithm; it adopts a piecewise sinusoidal convergence operator to replace the original linear decreasing convergence operator, and divides the convergence process of the algorithm into two parts, search and encirclement, for optimization respectively, thereby improving the optimization efficiency of the algorithm in the early stage and the convergence accuracy in the later stage; second, the present invention proposes an improvement strategy for the problem of blindness in setting the correlation coefficient in VMD, and introduces a self-reconstruction strategy and the mutual information correlation principle to determine the values ​​of the modal number k and the penalty factor; third, the present invention obtains the flexibility demand fluctuation coefficient of renewable energy power generation in each time period within the forecast day by calculation, thereby correcting the short-term forecast of the flexibility demand of renewable energy power generation to further improve the forecast accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 is a flow chart of the method of the present invention;

[0067] Figure 2 It is the calculation flow chart of IMOGWO algorithm;

[0068] Figure 3 It is the IMOGWO optimized LSSVM flow chart;

[0069] Figure 4 It is the VMD operation flow chart;

[0070] Figure 5 It is the IMOGWO-VMD-LSSVM prediction flow chart. DETAILED DESCRIPTION

[0071] like Figure 1 A short-term forecasting method for flexibility demand of renewable energy generation is shown, the method comprising the following steps in sequence:

[0072] (1) Select the renewable energy power generation power of the 30 historical days before the forecast date as historical data, with a data sampling period of 15 minutes, correct the abnormal data in the historical data, and calculate the historical data of the flexibility demand of renewable energy power generation

[0073] (2) Historical data on flexibility demand based on renewable energy generation Forecast the basic demand for renewable energy generation flexibility in 15-minute intervals during the forecast day

[0074] (3) Select the historical forecast data of the flexibility demand of renewable energy generation for the 30 historical days before the forecast date and historical data on flexibility demand for renewable energy generation Taking 15 minutes as the basic period, a flexibility demand forecast error matrix is ​​formed to calculate the flexibility demand fluctuation coefficient of renewable energy power generation in each period within the forecast day.

[0075] (4) Based on the flexibility of renewable energy generation and the flexibility demand fluctuation coefficient of renewable energy generation Calculate forecasted intra-day renewable energy generation flexibility needs

[0076] The step (1) specifically comprises the following steps:

[0077] (1a) Obtain the renewable energy generation power n historical days before the forecast date Where i represents the specific date of the historical day, i = 1, 2, ..., n, t represents the specific time period of the day, because the sampling period is 15 minutes, so t = 1, 2, ..., 96; n = 30;

[0078] (1b) The renewable energy generation power of n historical days before the forecast date is obtained The abnormal data in is corrected as follows:

[0079] (1b1) For isolated single point data of 0 or single point missing data, the average value method is used to process it. Suppose the renewable energy power generation power of a certain historical day i is for: If the data x at a certain time of the day If it is zero, then correct for x belongs to t;

[0080] (1b2) The method of comparing the overall load change is used to judge the load distortion data. That is, if the change of the data at a certain moment exceeds N times the average level at the same moment, it means that it is distorted data, that is, bad data. The load data change level is analyzed to obtain the basic average change δ at each moment of each day in n historical days. i :

[0081]

[0082] Where, t=1,2,...,96; i=1,2,...,n;

[0083] Then find the average change δ of n historical days:

[0084]

[0085] Taking δ as the basic judgment condition, if the change exceeds N times, the data is considered abnormal, that is:

[0086]

[0087] The abnormal data is processed by the method in step (1b1) to obtain the revised renewable energy power generation of n historical days before the forecast date.

[0088] (1c) Power generation based on renewable energy Calculate the historical data of the flexibility demand of renewable energy generation n historical days before the forecast date And construct the historical data matrix FlexP of flexibility demand for renewable energy generation:

[0089]

[0090]

[0091] like Figure 2 , 3 As shown, the step (2) specifically includes the following steps:

[0092] (2a) The improved multi-objective grey wolf optimization algorithm IMOGWO is used to optimize the parameters of the least squares support vector machine LSSVM:

[0093] (2a1) Selecting the radial basis kernel function as the kernel function of LSSVM and initializing the position of the wolf pack, that is, initializing the relevant parameters of LSSVM, including the number of wolf packs, the number of alpha wolves, the size of the archive, and the maximum number of iterations;

[0094] (2a2) Calculate the individual fitness of gray wolves, select the alpha wolf population and eliminate individuals, and update the archive;

[0095] (2a3) Calculate the coefficient vector A:

[0096] A=2a(R ram -1)e=2(1-T / T max )(R ram -1)e (6)

[0097] Among them: a is the convergence operator, R ram is a random number between 0 and 1; e is a unit vector, T is the number of iterations, and Tmax is the maximum number of iterations;

[0098] Determine whether the bi-norm of the coefficient vector A is less than 1. If ||A||<1, it is determined that the wolf pack has locked on to the prey and is encircling and preying on it. The position of the subordinate population is updated according to the following formula;

[0099]

[0100] Where: X i is the position of the individual, X b is the position of the alpha wolf, D is the distance between the individual and the alpha wolf, L is the contraction vector; rand(0,1) represents a random number of 0 or 1; Represents dot multiplication operation; R λ is a constant, usually 1.5; R u and R v is a random number, which follows a 0-1 normal distribution; Γ() is a standard gamma function;

[0101] If ||A||≥1, it is judged that the wolf pack is still in the stage of searching for prey, and the position of the subordinate population is updated according to the following formula:

[0102]

[0103] Where: X z To correct the individual; X w is the worst individual position; p * is the dynamic decreasing factor;

[0104] (2a4) Calculate the individual fitness of the wolf pack to obtain the non-dominated solution. At the same time, determine whether the archive is full. If it is full, replace the solution of the most crowded segment in the archive with the non-dominated solution obtained in the previous step, i.e., the position of the subordinate population; otherwise, directly include the non-dominated solution in the archive;

[0105] (2a5) Update parameters a and A and determine whether the number of iterations T is greater than T max , if yes, then select the final solution from the archive; otherwise, return to step (2a3) for the next iteration;

[0106] (2a6) Using IMOGWO to optimize the parameters in LSSVM, and obtaining each LSSVM model after IMOGWO optimization;

[0107] (2b) The renewable energy generation power of n historical days before the forecast date Normalization is performed to eliminate the loss of data details caused by excessive load data fluctuations. Before entering LSSVM, these data are normalized. The specific formula is as follows:

[0108]

[0109] Where, t=1,2,...,96; i=1,2,...,30;

[0110] like Figure 4 As shown in (2c), k intrinsic mode components IMF are obtained by using the variational mode analysis VMD algorithm:

[0111] (2c1) The variational problem is to minimize the sum of the estimated bandwidths of the eigenmode components. The constraint is that the sum of the eigenmode components is the original signal. The formula is as follows:

[0112]

[0113] Where: {u k} is the set of intrinsic mode components IMF; {ω k} is the set of center frequencies; δ(t) is the pulse signal; K is the preset number of decompositions; f(t) is the original signal;

[0114] (2c2) The penalty factor α and the Lagrange multiplier λ(t) are used to transform the constrained variational problem into an unconstrained problem:

[0115]

[0116] (2c3) The unconstrained problem is solved by the alternating direction multiplier method to achieve effective separation of signal frequencies, where the iterative update formulas for the eigenmode component and the center frequency are:

[0117]

[0118]

[0119] Where: is the kth eigenmode component with center frequency ω at the n+1th iteration; is the center frequency of the kth eigenmode component at the n+1th iteration, that is, the center of gravity of the power spectrum of the current modal component. Update λ according to the following formula:

[0120]

[0121] According to formula (15), we can determine u k Whether the convergence conditions are met, if not, the update operation is continuously performed until the convergence conditions are met:

[0122]

[0123] like Figure 5 As shown in (2c4), the k IMF subsequences obtained in step (2c) are input into each LSSVM model after the IMOGWO optimization described in step (2a), and the predicted values ​​of each IMF subsequence are superimposed and reconstructed to obtain the basic demand for renewable energy flexibility with a basic period of 15 minutes within the final forecast day.

[0124] 1) Establish a VMD simulation model, initialize the correlation coefficient, and decompose the data into k subsequences;

[0125] 2) Performing fast Fourier transform processing on each subsequence to obtain a series of processed time domain images;

[0126] 3) If no decomposition has occurred, the k=k+1 operation is performed. If decomposition has occurred, the initial value of k is k=k-1;

[0127] 4) Merge the obtained k subsequences to obtain the final modal number;

[0128] 5) After determining the number of modes, determine the penalty factor α, determine the search range and search step of the penalty factor α, calculate the mutual trust value corresponding to the penalty factor α, and select the penalty factor α corresponding to the maximum value of the mutual trust value as the final setting value;

[0129] 6) The final decomposed subsequence data is normalized and input into each LSSVM model optimized by IMOGWO;

[0130] 7) Superimpose and reconstruct the predicted values ​​of each subsequence to obtain the basic demand for renewable energy flexibility

[0131] The step (3) specifically refers to: according to the historical forecast data of the flexibility demand of renewable energy power generation in the past 30 days before the forecast date, and historical data on flexibility demand for renewable energy generation Calculate the flexibility resource prediction error matrix Coef:

[0132]

[0133] in: P N Rated installed capacity for electricity generation from renewable energy sources;

[0134] According to the flexible resource forecast error matrix Coef, the flexibility demand fluctuation coefficient of renewable energy generation in each period of the forecast day is calculated The specific calculation is as follows:

[0135]

[0136] The step (4) specifically refers to: based on the basic demand for renewable energy power generation flexibility and the coefficient of demand fluctuation of renewable energy generation flexibility The flexibility requirements for renewable energy generation during the forecasted day are calculated as follows:

[0137]

[0138] In summary, the present invention is based on the analysis of the optimization mechanism of the traditional multi-objective grey wolf algorithm. The present invention adopts iterative chaotic sequence to initialize the individual of the algorithm, thereby ensuring the dispersion of the individuals; it proposes to divide the alpha wolf population and eliminate individuals, thereby improving the individual diversity of the algorithm; it adopts a piecewise sinusoidal convergence operator to replace the original linear decreasing convergence operator, and divides the convergence process of the algorithm into two parts, namely, search and encirclement, for optimization respectively, thereby improving the optimization efficiency in the early stage of the algorithm and the convergence accuracy in the later stage; the present invention proposes an improvement strategy for the problem of blindness in setting the correlation coefficient in VMD, and introduces a self-reconstruction strategy and the mutual information correlation principle to determine the values ​​of the modal number k and the penalty factor; the present invention obtains the flexibility demand fluctuation coefficient of renewable energy power generation in each time period within the forecast day by calculation, thereby correcting the short-term forecast of the flexibility demand of renewable energy power generation to further improve the forecast accuracy.

Claims

1. A short-term forecasting method for flexibility demand of renewable energy power generation, characterized by: The method comprises the following steps in order: (1) Select the renewable energy power generation power of the 30 historical days before the forecast date as historical data, with a data sampling period of 15 minutes, correct the abnormal data in the historical data, and calculate the historical data of the flexibility demand of renewable energy power generation (2) Historical data on flexibility demand based on renewable energy generation Forecast the basic demand for renewable energy generation flexibility in 15-minute intervals during the forecast day (3) Select the historical forecast data of the flexibility demand of renewable energy generation for the 30 historical days before the forecast date and historical data on flexibility demand for renewable energy generation Taking 15 minutes as the basic period, a flexibility demand forecast error matrix is ​​formed to calculate the flexibility demand fluctuation coefficient of renewable energy power generation in each period within the forecast day. (4) Based on the flexibility of renewable energy generation and the flexibility demand fluctuation coefficient of renewable energy generation Calculate forecasted intra-day renewable energy generation flexibility needs The step (2) specifically comprises the following steps: (2a) The improved multi-objective grey wolf optimization algorithm IMOGWO is used to optimize the parameters of the least squares support vector machine LSSVM: (2a1) Selecting the radial basis kernel function as the kernel function of LSSVM and initializing the position of the wolf pack, that is, initializing the relevant parameters of LSSVM, including the number of wolf packs, the number of alpha wolves, the size of the archive, and the maximum number of iterations; (2a2) Calculate the individual fitness of gray wolves, select the alpha wolf population and eliminate individuals, and update the archive; (2a3) Calculate the coefficient vector A: A=2a(R ram -1)e=2(1-T / T max )(R ram -1)e (6) Among them: a is the convergence operator, R ram is a random number between 0 and 1; e is a unit vector, T is the number of iterations, and Tmax is the maximum number of iterations; Determine whether the bi-norm of the coefficient vector A is less than 1. If ||A||<1, it is determined that the wolf pack has locked onto the prey and is encircling and preying on it. The position of the subordinate population is updated according to the following formula; Where: X i is the position of the individual, X b is the position of the alpha wolf, D is the distance between the individual and the alpha wolf, L is the contraction vector; rand(0,1) represents a random number of 0 or 1; Represents dot multiplication operation; R λ is a constant, usually 1.5; R u and R v is a random number, which obeys the 0-1 normal distribution; Γ() is the standard gamma function; If ||A||≥1, it is judged that the wolf pack is still in the stage of searching for prey, and the position of the subordinate population is updated according to the following formula: Where: X z To correct the individual; X w is the worst individual position; p * is the dynamic decreasing factor; (2a4) Calculate the individual fitness of the wolf pack to obtain the non-dominated solution. At the same time, determine whether the archive is full. If it is full, replace the solution of the most crowded segment in the archive with the non-dominated solution obtained in the previous step, i.e., the position of the subordinate population; otherwise, directly include the non-dominated solution in the archive; (2a5) Update parameters a and A and determine whether the number of iterations T is greater than T max , if yes, select the final solution from the archive; otherwise, return to step (2a3) for the next iteration; (2a6) Using IMOGWO to optimize the parameters in LSSVM, and obtaining each LSSVM model optimized by IMOGWO; (2b) The renewable energy generation power of n historical days before the forecast date Normalization is performed to eliminate the loss of data details caused by excessive load data fluctuations. Before entering LSSVM, these data are normalized. The specific formula is as follows: Where, t=1,2,...,96; i=1,2,...,30; (2c) Using the variational mode analysis VMD algorithm, k intrinsic mode components IMF are obtained: (2c1) The variational problem is to minimize the sum of the estimated bandwidths of the eigenmode components. The constraint is that the sum of the eigenmode components is the original signal. The formula is as follows: Where: {u k } is the set of intrinsic mode components IMF; {ω k } is the set of center frequencies; δ(t) is the pulse signal; K is the preset number of decompositions; f(t) is the original signal; (2c2) The penalty factor α and the Lagrange multiplier λ(t) are used to transform the constrained variational problem into an unconstrained problem: (2c3) The unconstrained problem is solved by the alternating direction multiplier method to achieve effective separation of signal frequencies, where the iterative update formulas for the eigenmode component and the center frequency are: Where: is the kth eigenmode component with center frequency ω at the n+1th iteration; is the center frequency of the kth eigenmode component at the n+1th iteration, that is, the center of gravity of the power spectrum of the current modal component. Update λ according to the following formula: According to formula (15), we can determine u k Whether the convergence conditions are met, if not, the update operation is continuously performed until the convergence conditions are met: (2c4) Input the k IMF subsequences obtained in step (2c) into each LSSVM model after the IMOGWO optimization described in step (2a), and superimpose and reconstruct the predicted values ​​of each IMF subsequence to obtain the basic demand for renewable energy flexibility with a basic period of 15 minutes within the final forecast day. The step (3) specifically refers to: according to the historical forecast data of the flexibility demand of renewable energy power generation in the past 30 days before the forecast date, and historical data on flexibility demand for renewable energy generation Calculate the flexibility resource prediction error matrix Coef: in: P N Nominal installed capacity for renewable energy generation; According to the flexible resource forecast error matrix Coef, the flexibility demand fluctuation coefficient of renewable energy generation in each period of the forecast day is calculated The specific calculation is as follows: The step (4) specifically refers to: according to the basic demand for renewable energy power generation flexibility and the coefficient of demand fluctuation of renewable energy generation flexibility The flexibility requirements for renewable energy generation during the forecasted day are calculated as follows:

2. The method for short-term forecasting of flexibility demand of renewable energy power generation according to claim 1, characterized in that: The step (1) specifically comprises the following steps: (1a) Obtain the renewable energy generation power n historical days before the forecast date Where i represents the specific date of the historical day, i = 1, 2, ..., n, t represents the specific time period of the day, because the sampling period is 15 minutes, so t = 1, 2, ..., 96; n = 30; (1b) The renewable energy generation power of n historical days before the forecast date is obtained The abnormal data in is corrected as follows: (1b1) For isolated single point data of 0 or single point missing data, the average value method is used to process it. Suppose the renewable energy power generation power of a certain historical day i is for: If the data x at a certain time of the day If it is zero, then correct for x belongs to t; (1b2) The method of comparing the overall load change is used to judge the load distortion data. That is, if the change of the data at a certain moment exceeds N times the average level at the same moment, it means that it is distorted data, that is, bad data. The load data change level is analyzed to obtain the basic average change δ at each moment of each day in n historical days. i : Where, t=1,2,...,96; i=1,2,...,n; Then find the average change δ of n historical days: Taking δ as the basic judgment condition, if the change exceeds N times, the data is considered abnormal, that is: The abnormal data is processed by the method in step (1b1) to obtain the revised renewable energy power generation of n historical days before the forecast date. (1c) Power generation based on renewable energy Calculate the historical data of the flexibility demand of renewable energy generation n historical days before the forecast date And construct the historical data matrix FlexP of flexibility demand for renewable energy generation:

Images