Method for participating in energy-frequency regulation market by wind power system based on cloud energy storage

Abstract

The application relates to the technical field of power market, and discloses a method for a wind power system based on cloud energy storage to participate in an energy-frequency market, which comprises the following steps: performing time-frequency synchronous decomposition noise reduction on original wind power and multi-dimensional feature data and predicting wind power; determining optimal harmonic moments and performing probability density estimation on prediction error data; constructing a prediction error fluctuation domain, integrating a multiple uncertainty set; based on the multiple uncertainty set, constructing a master-slave game optimization model of a wind power plant and a cloud energy storage operator and a wind power system multi-objective optimization model considering cloud energy storage and power quality, and solving the optimal capacity and lease price of the leased cloud energy storage and the energy-frequency market participation strategy of the wind power system. The application introduces the concept of cloud energy storage into the wind power plant, which can not only reduce the additional investment cost of building supporting energy storage facilities for the wind power plant, but also reduce the deviation penalty cost and improve the frequency regulation performance benefit.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power market, in particular to a method for wind power system participating in energy-frequency market based on cloud energy storage. BACKGROUND

[0002] At present, the penetration rate of new energy represented by wind power is increasing, and new energy will bear more frequency modulation tasks, and it has become a trend for new energy power generators to participate in energy-frequency market bidding competition. However, the volatility and uncertainty of wind power output make it face the risk of punishment due to bidding deviation in market participation, which weakens its market competitiveness. As a high-quality flexible resource, energy storage can effectively alleviate the wind power deviation problem. However, building supporting energy storage for wind farms alone will greatly increase the investment cost, and there is a risk of long cost recovery period or no recovery.

[0003] With the rise of cloud energy storage business model, leasing cloud energy storage has become a practical solution, which not only effectively improves the utilization rate of energy storage and reduces the cost of wind farms, but also promotes the consumption of new energy and promotes the competitiveness of wind farms in the market. Leasing cloud energy storage also provides favorable conditions for wind power and energy storage system to jointly participate in energy-frequency market, further optimizing the participation mode of wind farms in the power market.

[0004] From the above problems, the uncertainty of wind power output may lead to high risk cost of wind farms, and wind power prediction as the basis for participating in power market bidding directly affects the bidding strategy and income of wind farms. Therefore, it is necessary to use more advanced combined prediction model to improve the accuracy and stability of wind power prediction results. In addition, the pricing strategy of wind power leasing cloud energy storage and the strategy of wind storage joint participation in energy-frequency market are not perfect, especially the influence of multiple uncertainty factors is not fully considered, so it is necessary to improve them.

[0005] At present, no effective solution has been proposed for the problems in the related art. SUMMARY

[0006] In view of the problems in the related art, the present application proposes a method for wind power system participating in energy-frequency market based on cloud energy storage to overcome the above technical problems existing in the prior art.

[0007] To this end, the specific technical solutions adopted by the present application are as follows:

[0008] A method for wind power system participating in energy-frequency market based on cloud energy storage, comprising the following steps:

[0009] S1, obtaining original wind power and multi-dimensional feature data, using an improved masking signal method combined with a robust local mean decomposition method to perform time-frequency synchronous decomposition and denoising on the original wind power and multi-dimensional feature data, and using a Transformer model based on local information enhancement and sparse attention mechanism to predict wind power;

[0010] S2, determining the optimal harmonic moment number of the harmonic transformation-based density estimation method, performing probability density estimation on the prediction error data of power and price, and constructing a multiple uncertainty set;

[0011] S3, based on the multiple uncertainty set, constructing a master-slave game optimization model of the wind farm and the cloud energy storage operator and a wind power system multi-objective optimization model considering cloud energy storage and power quality, and solving the optimal capacity and lease price of the leased cloud energy storage and the energy-frequency market participation strategy of the wind power system.

[0012] As preferred, the obtaining original wind power and multi-dimensional feature data, using an improved masking signal method combined with a robust local mean decomposition method to perform time-frequency synchronous decomposition and denoising on the original wind power and multi-dimensional feature data, and using a Transformer model based on local information enhancement and sparse attention mechanism to predict wind power comprises the following steps:

[0013] S11, obtaining original wind power and multi-dimensional feature data, decomposing the original wind power and multi-dimensional feature data into a plurality of optimal component product function components and residual signals based on a robust local mean decomposition method; using a Spearman correlation coefficient to remove components containing interference signals, and using approximate entropy to select components containing effective features;

[0014] S12, using an improved masking signal method to mask the selected components, and reconstructing the processed optimal component product function components to obtain denoised wind power signal components; using a Transformer model based on local information enhancement and sparse attention mechanism to predict wind power in combination with the denoised wind power signal components.

[0015] As preferred, the obtaining original wind power and multi-dimensional feature data, decomposing the original wind power and multi-dimensional feature data into a plurality of optimal component product function components and residual signals based on a robust local mean decomposition method; using a Spearman correlation coefficient to remove components containing interference signals, and using approximate entropy to select components containing effective features comprises the following steps:

[0016] S111, obtaining original wind power and multi-dimensional feature data, using a mirror extension method to process the boundary and search for all local extreme points in the original wind power signal; obtaining a local mean function and a local envelope function by a sliding average method, and calculating a zero mean signal and a pure frequency signal;

[0017] The calculation formula of the zero-mean signal is:

[0018] k 11 (t) = x(t) - m 11 (t)

[0019] The calculation formula of the pure frequency signal is:

[0020] s 11 (t) = k 11 (t) / a 11 (t)

[0021] In the formula, k 11 (t) is the zero-mean signal, x(t) is the wind power signal to be decomposed, m 11 (t) is the local mean function, s 11 (t) is the processed pure frequency signal, and a 11 (t) is the local envelope function.

[0022] S112, define the target function, extract the first product function from the original wind power signal to obtain the residual signal, and regard the residual signal as a new signal to repeat the iterative extraction of all optimal component product function components until the residual becomes a constant or a monotonic function.

[0023] The expression of the target function is:

[0024] f = RMS[z(t)] + EK[z(t)]

[0025]

[0026] In the formula, f is the target function value, RMS is the root mean square, EK is the empirical kurtosis calculation formula, z(t) is the zero baseline envelope signal, is the arithmetic mean of z(t), N s is the total number of signals.

[0027] S113, use the Spearman correlation coefficient to remove components containing interference signals, and use approximate entropy to select components containing effective features.

[0028] As a preferred, the selected components are masked using the improved masking signal method, and the wind power signal components after noise reduction are obtained by reconstructing the processed optimal component product function components; a Transformer model based on local information enhancement and sparse attention mechanism is used to predict wind power combined with the wind power signal components after noise reduction, including the following steps:

[0029] S121, calculate a masking mean value signal, select a number of components in the robust local mean value decomposition result for masking processing in combination with the Spearman correlation coefficient and the approximate entropy; perform robust local mean value decomposition on the masking processing result to obtain a signal component after masking processing, and perform final signal reconstruction according to the signal component after masking processing to obtain a wind power signal component after noise reduction;

[0030] The calculation formula of the masking mean value signal is:

[0031]

[0032] The expression of the masking processing is:

[0033]

[0034] The expression of the signal component after masking processing is:

[0035]

[0036] In the formula, s i (t) is the i th masking signal, is the average amplitude, is the average of the instantaneous frequency, PF i+ (t), PF i- (t) are signal components after plus mean and minus mean processing, respectively, PF i (t) is the original component, IMS PF i is the signal component after masking processing, pf i+ , pf i- are the first components decomposed and output by PF i+ (t) and PF i- (t), respectively;

[0037] S122, perform standardization processing on the wind power signal component after noise reduction, use a fully connected layer to perform high-dimensional representation on the wind power signal component after standardization processing, and extract local information of enhanced data by stacking multiple one-dimensional convolutions to obtain a signal after local enhancement;

[0038] S123, take a sample obtained after multiple convolution operations on the signal after local enhancement as an input of an encoder layer, use position encoding to mark position information of each sample, input the result after position encoding to an encoder layer of a Transformer model, and perform layer normalization processing on data in the encoder;

[0039] S124, the multi-head sparse attention mechanism is used to replace the full connection attention mechanism in the standard Transformer model to capture the global information of the sequence, integrate the global and local information of the wind power data, and map and learn the components of the wind power through the intermediate layers of the network, and finally output the wind power prediction result.

[0040] As preferred, the determination of the optimal harmonic moment number of the harmonic transform-based density estimation method, the probability density estimation of the prediction error data of power and price, and the construction of the multiple uncertainty set include the following steps:

[0041] S21, respectively calculate the wind power, energy market capacity clearing price, frequency regulation market capacity clearing price, and mileage clearing price prediction error, determine the optimal harmonic moment number of the harmonic transform-based density estimation method using the weighted average algorithm, and perform probability density estimation on the prediction error data of power and price;

[0042] S22, use the cumulative probability density function and the pre-set confidence to select the upper and lower limits of the prediction error, construct the prediction error fluctuation domain of the wind power, energy market capacity clearing price, frequency regulation market capacity clearing price, and mileage clearing price, and build a boundary adaptive optimization multiple uncertainty set.

[0043] As preferred, the respective calculation of the wind power, energy market capacity clearing price, frequency regulation market capacity clearing price, and mileage clearing price prediction error, the determination of the optimal harmonic moment number of the harmonic transform-based density estimation method using the weighted average algorithm, and the probability density estimation on the prediction error data of power and price include the following steps:

[0044] S211, respectively calculate the wind power, energy market capacity clearing price, frequency regulation market capacity clearing price, and mileage clearing price prediction error, and the prediction error is the difference between the predicted value and the true value;

[0045] S212, initialize the parameters, construct the fitness function of the optimal harmonic moment number based on relative entropy; in each iteration, calculate the weighted average position of the current harmonic moment number, and take the weighted average position of the current harmonic moment number as the representative position of the entire search space; adopt different moving exploration strategies, respectively aiming at exploring a wide solution space and fine-tuning the optimal harmonic moment number;

[0046] The expression of the fitness function of the optimal harmonic moment number based on relative entropy is:

[0047]

[0048] The calculation formula of the weighted average position is:

[0049]

[0050] where Fitness(h) is the fitness function of the best harmonic moment based on relative entropy, f(h) is the probability density function based on harmonic transform, is the probability density function of kernel density estimation, h is the harmonic moment to be optimized, N Candidate is the number of selected candidate solutions, P is the population size, it is the current iteration number, Max It is the maximum number of iterations, Sum Fitness is the sum of all fitness values of the selected candidate, h i is the i-th candidate optimal harmonic moment, h Miu is the weighted average position;

[0051] S213, based on the random constant and the dynamic parameter related to the iteration number, the moving strategy is selected and adjusted to achieve the balance between exploration and development, the best harmonic moment is obtained, and the prediction error of the wind power probability, the energy market capacity clearing price, the frequency modulation market capacity clearing price and the mileage clearing price is estimated by using the density estimation method based on the harmonic transform;

[0052] wherein the expression of the probability density function is:

[0053]

[0054] The expression of the probability density function based on the harmonic transform is:

[0055]

[0056] wherein f e (e) is the probability density function obtained by inverse Mellin transform, α is a fixed value, taken as 2, δ is a hypothetical value, assuming that the complex number ω = α + βi, then δ = 2β is assumed, e is the prediction error, f(e) is the probability density function based on the harmonic transform, is the harmonic transform.

[0057] As preferred, the upper and lower limits of the prediction error are selected by using the cumulative probability density function and the pre-set confidence, the prediction error fluctuation domain of the wind power, the energy market capacity clearing price, the frequency modulation market capacity clearing price and the mileage clearing price is constructed, and the boundary adaptive optimization multiple uncertainty set is constructed, including the following steps:

[0058] S221, the upper and lower limits of the prediction error are selected by using the cumulative probability density function and the pre-set confidence, and the fluctuation domain of the uncertain variable prediction is constructed;

[0059] wherein the expression of the cumulative probability density function is:

[0060]

[0061] The confidence interval is:

[0062] e lower = F -1 (0.025)

[0063] e upper = F -1 (0.975)

[0064] The expression of the prediction error fluctuation domain is:

[0065]

[0066] In the formula, F(e) is a cumulative probability density function, e is a prediction error, e lower , and e upper are the lower limit and the upper limit of the prediction error fluctuation respectively, F -1 is the inverse function of the cumulative distribution function, is a predicted value;

[0067] S222, integrate the prediction error fluctuation domains of the wind power, energy market capacity clearing price, frequency regulation market capacity clearing price, and mileage clearing price into a multiple uncertainty set;

[0068] In the formula, the expression of the multiple uncertainty set is:

[0069]

[0070] In the formula, are the wind power, energy market capacity clearing price, frequency regulation market capacity clearing price, and mileage clearing price respectively, are the upper and lower limits of the wind power prediction fluctuation at time τ respectively, are the upper and lower limits of the energy market capacity clearing price at time t respectively, are the upper and lower limits of the frequency regulation market capacity clearing price at time t respectively, are the upper and lower limits of the mileage clearing price at time t respectively.

[0071] As preferred, based on the multiple uncertainty set, a master-slave game optimization model of the wind farm and the cloud energy storage operator and a multi-objective optimization model of the wind power system considering cloud energy storage and power quality are constructed, and the best capacity and lease price of the leased cloud energy storage and the energy-frequency market participation strategy of the wind power system are solved, including the following steps:

[0072] S31, establish a master-slave game optimization model of the wind farm and the cloud energy storage operator considering multiple uncertainties, and solve the best capacity and lease price of the leased cloud energy storage by taking the maximum benefit of the subject cloud energy storage operator and the minimum cost of the slave wind farm as the target;

[0073] S32, a wind power system multi-objective optimization model considering cloud energy storage and power quality is established, and the optimized cloud energy storage leasing price and leasing capacity are combined to maximize the benefits of the wind power system and optimize the power quality at the grid connection point, so as to solve the energy-frequency modulation market participation strategy of the wind power system.

[0074] Preferably, the master-slave game optimization model of the wind farm and the cloud energy storage operator comprises a game subject cloud energy storage operator optimization model and a game slave wind power cluster robust optimization model.

[0075] The objective function expression of the game subject cloud energy storage operator optimization model is:

[0076]

[0077] The objective function expression of the game slave wind power cluster robust optimization model is:

[0078]

[0079]

[0080] In the formula, I c is the cloud energy storage operator income, is the rented energy storage income, is the frequency modulation market bidding income, is the energy market bidding income, is the cloud energy storage operation cost, are the cloud energy storage power and capacity sales prices respectively, are the cloud energy storage charging and discharging power and capacity respectively, and Δt is the day-ahead market time scale, is the frequency modulation performance index, is the cloud energy storage bidding capacity in the frequency modulation market, are the capacity clearing price and mileage clearing price of the frequency modulation market respectively, and η is the mileage call rate, is the capacity clearing price of the energy market, is the cloud energy storage bidding capacity in the energy market, τ is a certain time in the real-time market, and t is a certain time in the day-ahead market, are the cloud energy storage capacity cost and mileage cost respectively, and C w is the total cost of the wind farm, are the wind power leased cloud energy storage cost, wind power deviation penalty cost, and opportunity cost respectively, are the wind power bidding income in the frequency modulation market and the energy market respectively, and λ1 and λ2 are the declaration penalty coefficients of the energy market and the frequency modulation market respectively, respectively are the day-ahead bidding capacity of wind power in energy market and frequency regulation market, respectively are the real-time bidding capacity of wind storage in energy market and frequency regulation market, Δτ is the time scale of real-time market, T is the total time, is the bidding price of frequency capacity, is the bidding price of frequency mileage, φ(t) is the time-of-use price of electricity energy market, w.fm.c is the declared capacity of wind power cluster, P w.fm.c.max is the upper limit of declared capacity of wind power cluster allowed by frequency regulation market, K W (t) is the frequency performance index value, P w.e.c (t), R w,e,c (t) are respectively the bidding capacity and price of wind power cluster in energy market.

[0081] Preferably, the objective function of the wind power system multi-objective optimization model considering cloud storage and power quality comprises optimal wind power system revenue and optimal grid-connected wind power system electric power quality;

[0082] wherein the expression of optimal wind power system revenue is:

[0083]

[0084] the expression of optimal grid-connected wind power system electric power quality is:

[0085]

[0086] in the formula, F1 is the wind power system revenue function, is the total revenue of energy and frequency regulation market, is the wind power deviation penalty cost, is the optimized wind power lease cloud storage cost, is the opportunity cost of the second stage, H is a set of 24 time periods in one day, respectively are the day-ahead bidding capacity of wind power in energy market and frequency regulation market of the second stage, respectively are the real-time bidding capacity of wind storage in energy market and frequency regulation market of the second stage, is the wind power frequency performance index, respectively are the day-ahead bidding capacity of wind power in energy market and frequency regulation market of the second stage, respectively are the real-time bidding capacity of wind storage in energy market and frequency regulation market of the second stage, w.fm.c.o is the declared capacity of wind power cluster of the second stage, F2 is the grid-connected wind power system electric power quality function, M is the total number of time nodes in a year, P w (t) is the output power of wind farm at time t, Q(t) is the comprehensive evaluation value of electric power quality at time t.

[0087] The beneficial effects of the present application are:

[0088] 1) The present application can effectively remove noise in data while preserving the main features of the signal by combining improved masking signal method (IMS) with robust local mean decomposition method (RLMD) for time-frequency synchronous decomposition and noise reduction of wind power and multi-dimensional feature data. In addition, the introduction of local information enhancement module and sparse attention mechanism Transformer model can enhance the model's capture of local change characteristics in wind power time series data while preserving the learning of global dependencies. The combined prediction model (IMS-RLMD-LSA-Transformer) improves the accuracy and robustness of wind power prediction by fusing time-frequency noise reduction and deep learning feature extraction, providing a more reliable data foundation for subsequent market participation strategies.

[0089] 2) The present application adopts a new meta-heuristic optimization algorithm based on weighted average position concept to determine the optimal harmonic moment number of the density estimation method based on harmonic transformation, which has adaptive characteristics and can effectively balance bias and variance to improve estimation accuracy and reliability. The probability density estimation of prediction error provides a more accurate way to better handle errors and uncertainties in the optimization process, avoiding the limitations of traditional methods. Using the generated volatility domain, a boundary adaptive optimization multiple uncertainty set is constructed to provide more accurate optimization solutions for wind power systems under different complex and dynamic scenarios.

[0090] 3) The present application can optimize the lease price and capacity between wind farms and cloud energy storage operators by establishing a master-slave game optimization model between wind farms and cloud energy storage operators. This game optimization model not only balances the interests of wind farms and cloud energy storage operators, but also effectively reduces the uncertainty caused by wind power fluctuations, improving the overall efficiency of the system. Using robust optimization methods, combined with a multi-objective optimization model considering cloud energy storage and power quality, the system's energy market and frequency regulation market participation strategy can be optimized under a certain degree of conservatism. Ensuring that even in the face of prediction errors and market fluctuations, the system can still achieve the optimal market participation strategy, thereby improving the stability and efficiency of market participation. BRIEF DESCRIPTION OF DRAWINGS

[0091] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0092] Figure 1is a flowchart of a method for a wind power system based on cloud energy storage to participate in an energy-frequency market according to an embodiment of the present application. DETAILED DESCRIPTION

[0093] To further illustrate the embodiments, the present application provides drawings which are part of the disclosure of the present application, mainly used to illustrate the embodiments, and can be used to explain the operation principle of the embodiments in conjunction with the related description of the specification. Those skilled in the art should understand other possible implementations and advantages of the present application by referring to these contents. The components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0094] According to an embodiment of the present application, a method for a wind power system based on cloud energy storage to participate in an energy-frequency market is provided.

[0095] The present application will be further described in conjunction with the drawings and specific embodiments. As shown in the drawings, the method for a wind power system based on cloud energy storage to participate in an energy-frequency market according to an embodiment of the present application includes the following steps: Figure 1

[0096] S1, obtaining original wind power and multi-dimensional feature data, using an improved masking signal method combined with a robust local mean decomposition method to perform time-frequency synchronous decomposition and noise reduction on the original wind power and multi-dimensional feature data, and using a Transformer model based on local information enhancement and sparse attention mechanism to predict wind power;

[0097] In this embodiment, first, the RLMD method (Robust Local Mean Decomposition) is used to decompose the obtained original wind power and multi-dimensional feature data into multiple components, the Spearman correlation coefficient is used to remove components containing interference signals, and then the approximate entropy is used to select components containing effective features. Second, the IMS method (i.e., the improved masking signal method) is used to mask the selected components, and the PF (optimal component product function) components after processing are reconstructed to obtain the denoised signal. Using the sparse attention mechanism, the global correlation between the virtual sample and the wind power data is calculated to capture the long-distance dependence relationship. At the same time, by calculating the correlation of adjacent samples, the local dependence characteristics are further extracted, and the global and local information of the data is comprehensively integrated. After feature extraction and global-local information fusion, the model maps and learns the wind power components through the intermediate layer of the network, and finally realizes the prediction of the wind power.

[0098] ​The method comprises the following steps of:

[0099] S11, obtaining original wind power and multi-dimensional feature data, decomposing the original wind power and multi-dimensional feature data into a plurality of optimal component product function components and residual signals based on a robust local mean decomposition method; removing components containing interference signals by using a Spearman correlation coefficient, and selecting components containing effective features by using approximate entropy;

[0100] Specifically, the method comprises the following steps of obtaining original wind power and multi-dimensional feature data, decomposing the original wind power and multi-dimensional feature data into a plurality of optimal component product function components and residual signals based on a robust local mean decomposition method; removing components containing interference signals by using a Spearman correlation coefficient, and selecting components containing effective features by using approximate entropy.

[0101] (1) obtaining original wind power and multi-dimensional feature data, decomposing the obtained original wind power and multi-dimensional feature data into a plurality of components by using RLMD, and the steps of the RMLD algorithm are as follows:

[0102] 1) initial signal processing and local extreme point identification:

[0103] Let the wind power signal to be decomposed be x(t), and search for all local extreme points n in x(t) by using a mirror extension method i ;

[0104]

[0105] In the formula, m i is the mean value of two adjacent extreme points n i and n i+1 , a i is a local envelope estimation value;

[0106] 2) obtain a local mean function and a local envelope function by using a sliding average method, and calculate a zero mean signal and a pure frequency signal;

[0107] The calculation formula of the zero mean signal is as follows:

[0108] k 11 (t)=x(t)-m 11 (t)

[0109] The calculation formula of the pure frequency signal is as follows:

[0110] s 11 (t)=k 11(t) / a 11 (t)

[0111] where k 11 (t) is a zero-mean signal, x(t) is a wind power signal to be decomposed, m 11 (t) is a local mean function, s 11 (t) is a processed pure frequency signal, a 11 (t) is a local envelope function;

[0112] 3) define a target function, the expression of the target function is:

[0113] f = RMS[z(t)] + EK[z(t)]

[0114]

[0115] where f is a target function value, RMS is a root mean square, EK is an empirical kurtosis calculation formula, z(t) is a zero baseline envelope signal, is an arithmetic mean of z(t), N s is a total number of signals;

[0116] When the iteration condition is met, the calculated pure frequency modulation signal s1(t) and the envelope signal a 1j (t) are multiplied to obtain a first product function;

[0117] s1(t) = s 1j (t)

[0118]

[0119] where s 1j (t) is a pure frequency modulation signal of the jth iteration, PF1(t) is a first optimal component product function of the envelope signal;

[0120] 4) extract the first product function PF1(t) from the wind power signal x(t) to obtain a residual signal u1(t), then take the residual signal u1(t) as a new signal to repeat the iteration step to extract all optimal component product function components until u j (t) becomes a constant or a monotonic function; finally, the original signal x(t) is decomposed into p product function PF components and a residual u j (t);

[0121]

[0122] where p is a number of product function components, u j (t) is a residual of the jth iteration;

[0123] (2) Using the Spearman correlation coefficient to remove components containing interference signals and using approximate entropy to select components containing effective features.

[0124] The Spearman correlation coefficient is a non-parametric correlation measure used to assess the correlation between the ranks or orders of two sets of data, and the calculation method is as follows:

[0125]

[0126] In the formula, r s is the Spearman correlation coefficient, d i is the rank of the difference between the ranks of the two variables, and n is the number of data pairs;

[0127] Approximate entropy is a statistical tool for measuring the complexity and regularity of time series data, and the calculation principle is as follows:

[0128]

[0129] In the formula, ApEn is the approximate entropy, Φm(r) is the field probability of m-dimensional embedding, m is the embedding dimension, and r is the threshold;

[0130] S12, using the improved masking signal method to mask the selected components, and reconstructing the optimal component product function component after processing to obtain the denoised wind power signal component; using a Transformer model based on local information enhancement and sparse attention mechanism to predict wind power in combination with the denoised wind power signal component.

[0131] In this embodiment, the improved masking signal method refers to first removing components containing interference noise signals using the Spearman correlation coefficient before using the masking signal method, and then selecting components containing effective features using approximate entropy; secondly, the selected components are masked using the masking signal method, which improves the efficiency and accuracy of masking. The existing masking signal method needs to manually solve different masking signals for each component obtained from signal decomposition, which has a large amount of calculation and cannot achieve the effect of noise reduction.

[0132] Specifically, the use of the improved masking signal method to mask the selected components and the reconstruction of the optimal component product function component after processing to obtain the denoised wind power signal component; using a Transformer model based on local information enhancement and sparse attention mechanism to predict wind power in combination with the denoised wind power signal component includes the following steps:

[0133] (1) Using the IMS method to mask the selected components, and reconstructing the PF component after processing to obtain the denoised signal;

[0134] 1) Calculate the masking mean signal, and the calculation formula of the masking mean signal is:

[0135]

[0136] In the formula, s i (t) is the i th masking signal, is the average amplitude, is the mean value of the instantaneous frequency;

[0137] 2) Select a number of components in the robust local mean decomposition result for masking processing in combination with the Spearman correlation coefficient and the approximate entropy, and the expression of the masking processing is:

[0138]

[0139] In the formula, PF i+ (t) and PF i- (t) are the signal components after the plus mean and minus mean processing respectively, and PF i (t) is the original component;

[0140] 3) The RLMD decomposition is performed again on PF i+ (t) and PF i- (t) respectively, and the signal components after the masking processing are obtained, and the expression of the signal components after the masking processing is:

[0141]

[0142] In the formula, IMS PF i is the signal component after the masking processing, and pf i+ and pf i- are the first components decomposed and output by PF i+ (t) and PF i- (t) respectively;

[0143] 4) The final signal reconstruction is performed according to the signal components after the masking processing, and the wind power signal components after the noise reduction are obtained, and the expression of the reconstructed signal is:

[0144]

[0145] In the formula, irx(t) is the reconstructed signal;

[0146] (2) The Transformer model based on the local information enhancement and the sparse attention mechanism is adopted to predict the wind power in combination with the wind power signal components after the noise reduction;

[0147] Specifically, the noise-reduced wind power signal component is input into the model, and by stacking multiple layers of one-dimensional convolution (Conv1D), each layer uses a different size of convolution kernel to capture the local variation features in the wind power data from multiple scales, and to strengthen the expression of local information. The pre-layer normalization is used instead of the traditional post-layer normalization. The sparse attention mechanism is used to calculate the global correlation between the virtual sample and the wind power data, and to capture the long-distance dependence relationship. At the same time, by calculating the correlation between adjacent samples, the local dependence characteristics are further extracted, and the global and local information of the wind power data is comprehensively integrated. After feature extraction and global-local information fusion, the model maps and learns the components of the wind power through the intermediate layers of the network, and finally outputs the wind power prediction result.

[0148] 1) Local information enhancement based on Conv1D, that is, the noise-reduced wind power signal component is standardized, the standardized wind power signal component is represented in high dimension by using a fully connected layer, and the local information of the enhanced data is extracted by stacking multiple layers of one-dimensional convolution to obtain the locally enhanced signal;

[0149] First, the input wind power data component X = [x1, x2, x3, …, xN] is standardized by Z-score, and the standardized wind power data component X is obtained. N Z-score ∈R N×M , R is a set of real numbers, and M is the number of features;

[0150]

[0151] In the formula, X Z-score is the Z-standardized wind power data, x is the average value of all samples, S is the standard deviation, and x i is the i-th sample, i ∈ [1, N], and N is the total number of samples;

[0152] Then, the fully connected layer is used, and X Z-score is mapped to d model dimension to obtain its high-dimensional representation Secondly, multiple layers of Conv1D are stacked to extract and enhance the local information of the data, and

[0153] X Conv = X Z-score-d + max (0, BN (Conv1D n (X Z-score-d ))

[0154] In the formula, X Conv is the locally enhanced signal, X Z-score-d is the high-dimensional representation of the original data after standardization, n is the number of convolution layers, and BN is the batch normalization function.​

[0155] Finally, X Conv After the sample obtained by the multi-layer convolution operation is taken as the input of the encoder layer, the position information of each sample is marked using position encoding, and the result after position encoding is input to the encoder layer of the Transformer model. In the encoder, the data is subjected to layer normalization processing, as shown in the formula:

[0156] X Conv ′ = LayerNorm(ReLU(X Conv ))

[0157] In the formula, X Conv ′ is the signal after layer normalization processing, LayerNorm is layer normalization processing, and ReLU is an activation function.

[0158] 2) Global information extraction based on sparse attention mechanism

[0159] When X Conv ′ is integrated into the encoder layer, the multi-head sparse attention mechanism is used to replace the full connection attention mechanism in the standard Transformer model to capture the global information of the sequence. For the vector sequence H = X Conv ,H = [h1,...h i ...,h N ], representing the high-dimensional embedding of the i-th sample data. The formula for updating the sample data is:

[0160]

[0161] MHSAttention(q,H) = Concat(head1,...head a ...,head h )·W O

[0162] head a = Attention(qW Q ,KW K ,VW V )

[0163]

[0164] In the formula, representing the context information of the i-th sample data, Concat represents a connection function, respectively, the left and right adjacent states of the current sample, S t-1 respectively, the previous state, initial value and previous state of the virtual data, For each sample data state, MHSAttention is a multi-head coefficient attention mechanism, q represents the high-dimensional embedding of each sample data, h represents the number of multi-head, head a represents the a-th head in the calculation result of the multi-head sparse attention mechanism, W Q ,W K ,W V ,W O are different learnable parameters, Q, K, V, and O are different value quantities, d K is the projection space dimension, softmax is a normalization function, and T is the length of the input sequence.

[0165] Introducing represents sample data, and the initial state is S 0 = average(H) and the initial value H 0 = H, and the update virtual data process is:

[0166] S t = MultiAttention(S t-1 , Concat[S t-1 ; H t ])

[0167] S t = LayerNorm(ReLU(St))

[0168] In the formula, S t is the final state of the sample data, MultiAttention is a dot product attention mechanism, H t is the final state of the virtual data, S t-1 is the state of the previous moment.

[0169] Then enter the feedforward neural network (FFN) to increase the high-dimensional number of data, filter out useful information, and then reduce to a low-dimensional number to maintain the consistency of the dimensions before and after the FFN operation.

[0170] H t = LayerNorm(ReLU(H t ))

[0171] FFN = max(0, H t · W1 + b1) · W2 + b2

[0172] In the formula, FFN is a feedforward neural network; W1, W2, b1, and b2 are the weights and offsets of the two linear transformations of FFN, respectively.

[0173] Finally, d modelThe data is reduced to one dimension by a fully connected layer to obtain the final wind power prediction result X trm ∈R N×1 ;

[0174]

[0175] wherein X trm is the final wind power prediction result.

[0176] S2, determine the optimal harmonic moment number of the harmonic transform-based density estimation method, perform probability density estimation on the prediction error data of power and price, and construct a multiple uncertainty set;

[0177] The method comprises the following steps:

[0178] S21, respectively calculate the wind power, energy market capacity clearing price, frequency regulation market capacity clearing price, and mileage clearing price prediction error, determine the optimal harmonic moment number of the harmonic transform-based density estimation method using a weighted average algorithm, and perform probability density estimation on the prediction error data of power and price;

[0179] Specifically, the method comprises the following steps:

[0180] (1) Select relative entropy as the fitness function, and determine the optimal harmonic moment number of the harmonic transform-based density estimation method using a weighted average algorithm;

[0181] 1) Initialization stage:

[0182] h i,j =rand·(h maxj -h minj )+h minj

[0183] wherein h i,j is the position of the i-th solution in the j-th dimension, rand is a random number between 0 and 1, h max,j , and h min,j are the lower and upper limit values of the harmonic moment number, respectively;

[0184] Define the optimal harmonic moment number fitness function based on relative entropy:

[0185]

[0186] where Fitness(h) is the fitness function based on the best harmonic moments of relative entropy, f(h) is the probability density function based on the harmonic transform, is the probability density function of the kernel density estimation, h is the harmonic moment to be optimized;

[0187] 2) Calculate the weighted average position:

[0188]

[0189] where N Candidate is the number of selected candidate solutions, P is the population size, it is the current iteration number, Max It is the maximum number of iterations, Sum Fitness is the sum of all fitness values of the selected candidates, h i is the i-th candidate optimal harmonic moment, h Miu is the weighted average position;

[0190] 3) Define the search phase:

[0191]

[0192] where f(it) is the candidate solution, A is a constant that controls the balance between exploration and exploitation phases;

[0193] 4) Search phase:

[0194]

[0195] where h i,j (it+1) is the i-th solution at the j-th position at the (it+1) iteration, S is the step size, is the position in the global best solution at iteration it;

[0196] h i1 (it+1) = rand · (Uh min - Lh min ) + Lh min

[0197] where Lh min , Uh min are the minimum values of the lower and upper limits in all dimensions, respectively;

[0198] 5) Exploitation phase:

[0199] The strategy simulates how the search agent population moves with a high probability to the search space to exploit new global optimal values;

[0200]

[0201] where w 11 , w 12 , w 13 , w 21 , w 22 , w 31 , w 32 are random numbers between 0 and 1 respectively, is the individual best search position, h i,2 (it+1), h i,3 (it+1), h i,4 (it+1) are the position update formulas under three development strategies respectively, is the global best search position, is the global best position.

[0202] (2) Calculate the prediction error of wind power, energy market capacity clearing price, frequency modulation market capacity clearing price and mileage clearing price respectively, and estimate the probability density of the prediction error data of power and price based on the density estimation method of harmonic transform;

[0203] 1) Compare the prediction results with the true data, and calculate the prediction error:

[0204]

[0205] wherein, is the prediction value at time t, y(t) is the true value at time t, and e(t) is the prediction error at time t;

[0206] 2) Estimate the probability density of the prediction error data based on the density estimation method of harmonic transform;

[0207] According to the Euler formula and the inverse Mellin transform, the derived probability density function is obtained, and the expression of only the real part is:

[0208]

[0209] wherein, f e (e) is the probability density function obtained by the inverse Mellin transform, α is a fixed value, which is 2 in the embodiment, and e is the prediction error; δ is a hypothetical value, assuming that the complex number ω = α + βi, then the hypothetical value δ = 2β;

[0210] Define the harmonic transform as:

[0211]

[0212] wherein, is the harmonic transform;

[0213] The probability density function based on the harmonic transform is:

[0214]

[0215] In the formula, f(e) is a probability density function based on harmonic transformation.

[0216] S22, selecting upper and lower limits of the prediction error by using the cumulative probability density function and the preset confidence, constructing a prediction error fluctuation range of the wind power, energy market capacity clearing price, frequency regulation market capacity clearing price and mileage clearing price, and building a boundary adaptive optimization multiple uncertainty set.

[0217] Specifically, the step of selecting upper and lower limits of the prediction error by using the cumulative probability density function and the preset confidence, constructing a prediction error fluctuation range of the wind power, energy market capacity clearing price, frequency regulation market capacity clearing price and mileage clearing price, and building a boundary adaptive optimization multiple uncertainty set includes the following steps:

[0218] (1) According to the cumulative probability density function, the upper and lower limits of the prediction error are selected by using a confidence of 0.95 to construct a prediction fluctuation range of the uncertain variable;

[0219] The expression of the cumulative probability density function is:

[0220]

[0221] The confidence interval is selected:

[0222] e lower =F -1 (0.025)

[0223] e upper =F -1 (0.975)

[0224] In the formula, F(e) is the cumulative probability density function, e is the prediction error, e lower , and e upper are the lower and upper limits of the prediction error fluctuation respectively, and F -1 is the inverse function of the cumulative distribution function;

[0225] The prediction fluctuation range is constructed:

[0226]

[0227] In the formula, is the prediction value at time t;

[0228] (2) The prediction error fluctuation ranges of the wind power, energy market capacity clearing price, frequency regulation market capacity clearing price and mileage clearing price are integrated into a multiple uncertainty set;

[0229] wherein, the expression of the multiple uncertainty set is:

[0230]

[0231] wherein, respectively, the wind power, the energy market capacity clearing price, the frequency regulation market capacity clearing price and the mileage clearing price, respectively, the upper and lower limits of the wind power prediction fluctuation at time τ, respectively, the upper and lower limits of the energy market capacity clearing price at time t, respectively, the upper and lower limits of the frequency regulation market capacity clearing price at time t, respectively, the upper and lower limits of the mileage clearing price at time t.

[0232] S3, based on the multiple uncertainty set, a master-slave game optimization model of the wind farm and the cloud energy storage operator and a wind power system multi-objective optimization model considering cloud energy storage and power quality are constructed, and the optimal capacity and leasing price of the leased cloud energy storage and the energy-frequency market participation strategy of the wind power system are solved.

[0233] In this embodiment, a master-slave game optimization model of the wind farm and the cloud energy storage operator considering multiple uncertainties is established, the maximum benefit of the subject cloud energy storage operator and the minimum cost of the slave wind farm are taken as the target, and the optimal capacity and leasing price of the leased cloud energy storage are solved; a wind power system multi-objective optimization model considering cloud energy storage and power quality is established, the maximum benefit of the wind power system and the optimal power quality of the grid connection point are taken as the double targets, the energy-frequency market participation strategy of the wind power system under a certain degree of conservation is solved, so as to improve the market participation income and enhance the robustness of the response to uncertainty.

[0234] wherein, the multiple uncertainty set is based on the following steps:

[0235] S31, a master-slave game optimization model of the wind farm and the cloud energy storage operator considering multiple uncertainties is established, and the maximum benefit of the subject cloud energy storage operator and the minimum cost of the slave wind farm are taken as the target, and the optimal capacity and leasing price of the leased cloud energy storage are solved;

[0236] Specifically, the master-slave game optimization model of the wind farm and the cloud energy storage operator includes a game subject cloud energy storage operator optimization model and a game slave wind cluster robust optimization model;

[0237] (1) Game subject cloud energy storage operator optimization model

[0238] 1) Objective function expression is:

[0239]

[0240]

[0241] where I c is the cloud energy storage operator revenue, is the rented energy storage revenue, is the frequency regulation market bidding revenue, is the energy market bidding revenue, is the cloud energy storage operation cost, are the cloud energy storage power and capacity sale price, respectively, are the cloud energy storage charging and discharging power and capacity, respectively, and Δt is the day-ahead market time scale, is the frequency regulation performance index, is the cloud energy storage bidding capacity in the frequency regulation market, are the capacity and mileage clearing price of the frequency regulation market, respectively, and η is the mileage call rate, is the capacity clearing price of the energy market, is the cloud energy storage bidding capacity in the energy market, τ is a certain time in the real-time market, and t is a certain time in the day-ahead market, are the cloud energy storage capacity cost and mileage cost, respectively;

[0242] 2) Constraint conditions:

[0243] Cloud energy storage aggregator operating capacity constraint:

[0244]

[0245] where is the rated capacity of the cloud energy storage, are the upper and lower limits of the state of charge of the energy storage, respectively;

[0246] Cloud energy storage initial state and final state constraint:

[0247]

[0248] where are the initial state and final state of charge of the cloud energy storage, respectively;

[0249] Cloud energy storage configurable actual energy storage constraint:

[0250]

[0251] where F is the set of all distributed energy storages, θ is a 0-1 decision variable, and C cl(i) is the capacity of the i-th distributed energy storage.

[0252] (2) Robust optimization model for wind power cluster

[0253] 1) Objective function expression is:

[0254]

[0255] In the formula, C w is the total cost of the wind farm, is the wind power lease cloud energy storage cost, wind power deviation penalty cost, opportunity cost, is the wind power bidding income in the frequency regulation market and the energy market, λ1, λ2 is the declaration penalty coefficient of the energy market and the frequency regulation market, is the day-ahead bidding capacity of wind power in the energy market and the frequency regulation market, is the real-time bidding capacity of wind storage in the energy market and the frequency regulation market, Δτ is the real-time market time scale, N T is the total time, is the frequency regulation capacity bidding price, is the frequency regulation mileage bidding price, φ(t) is the time-of-use electricity price, P w.fm.c is the declared capacity of the wind power cluster, P w.fm.c.max is the upper limit of the declared capacity of the wind power cluster in the frequency regulation market, K W (t) is the frequency regulation performance index value, P w.e.c (t), R w,e,c (t) is the capacity and price of the wind power cluster in the energy market bidding;

[0256] 2) Constraint conditions

[0257] The declared capacity constraint of the frequency regulation market for wind turbines is:

[0258] 0.03P fm.e ≤P w.fm.c (t)≤0.25P fm.e

[0259] In the formula, P w.fm.c (t) is the declared capacity of the wind power cluster, P fm.e is the rated capacity of the frequency regulation market;

[0260] The constraint that the wind power lease cloud energy storage needs to meet to improve the frequency regulation performance index is:

[0261] K min ≤a i r c (t)×100+b i ≤K w.max

[0262] wherein r c (t) is the proportion of renting cloud energy storage, a i , b i are the coefficients of the piecewise fitting function, K min is the minimum threshold of the comprehensive frequency modulation performance index of the frequency modulation market, K w.max is the maximum value of the comprehensive frequency modulation performance index;

[0263] Wind power system wind curtailment constraint:

[0264] P w.fm.c (t) + P w.e.c (t) ≥ (1-ε W.max )P w (t)

[0265] wherein ε W.max is the maximum allowable wind curtailment ratio, P w (t) is the wind power system capacity;

[0266] Wind power bidding capacity constraint in two markets:

[0267] 0 ≤ P w.fm.c (t) + P w.e.c (t) ≤ P w (t)

[0268] wherein P w.fm.c is the declared capacity of the wind power cluster.

[0269] S32, a wind power system multi-objective optimization model considering cloud energy storage and power quality is established, and the optimized cloud energy storage rental price and rental capacity are combined to maximize the benefits of the wind power system and optimize the power quality at the grid connection point, so as to solve the energy-frequency modulation market participation strategy of the wind power system.

[0270] Specifically, the objective function of the wind power system multi-objective optimization model considering cloud energy storage and power quality includes wind power system benefit optimization and wind power system grid-connected power quality optimization;

[0271] (1) The expression of wind power system benefit optimization is:

[0272]

[0273] wherein F1 is the wind power system benefit function, is the total benefit of energy and frequency modulation market, is the wind power deviation penalty cost, is the optimized wind power cloud energy storage cost, is the opportunity cost of the second stage, H is a set of 24 time periods in one day, are the day-ahead bidding capacities of the second stage wind power energy market and frequency modulation market, respectively, These refer to the real-time bidding capacity of wind and storage in the second phase of the energy market and the frequency regulation market, respectively. For wind power frequency regulation performance indicators, These represent the day-ahead bidding capacity for wind power in the second phase of the energy market and frequency regulation market. These represent the real-time bidding capacity of wind and storage in the second phase of the energy market and the frequency regulation market, respectively, P w.fm.c.o For the second phase of wind power cluster application capacity;

[0274] (2) The expression for the optimal power quality of grid-connected wind power systems is:

[0275]

[0276] In the formula, F2 is the power quality function of the wind power system connected to the grid, M is the total number of time nodes throughout the year, and P w Q(t) represents the output power of the wind farm at time t, and Q(t) represents the comprehensive power quality assessment value at time t.

[0277] (3) Constraints:

[0278] 1) The day-ahead bidding capacity for wind power is subject to this constraint:

[0279]

[0280] In the formula, For the predicted output power of wind power, These represent the bid capacity of wind power in the energy and frequency regulation markets as of today;

[0281] 2) Real-time bidding capacity for wind and storage:

[0282]

[0283] In the formula, These represent the real-time bidding capacity of wind and energy storage systems in the energy and frequency regulation markets. For real-time output power of wind power, For cloud energy storage discharge capacity;

[0284] 3) Constraints on wind power output and reserved frequency regulation capacity for participation in the energy market:

[0285]

[0286] In the formula, Let τ be the maximum wind power output at time τ.

[0287] 4) Maximum real-time frequency regulation capacity constraint for wind power:

[0288]

[0289] In the formula, δ represents the rated power of the wind power, and δ is the wind power grade limit coefficient.

[0290] 5) Maximum capacity constraint for leased energy storage within the Δk time period:

[0291]

[0292] In the formula, These represent the charging and discharging power capacities of cloud energy storage. Rated power of energy storage;

[0293] 6) Constraints for calculating power quality indicators:

[0294]

[0295] In the formula: P i (t), Q i (t) represents the active and reactive power injected into node i at time t, respectively. i (t), U j (t) represents the actual voltages of node i and node j at time t, respectively, G ij B ij θ ij Let i represent the conductance, susceptance, and power angle between node i and node j, respectively, and D represent the total number of nodes in the system.

[0296] In summary, by utilizing the technical solutions described above, this invention combines the Improved Signal Masking (IMS) method with Robust Local Means Decomposition (RLMD) to perform time-frequency synchronous decomposition and denoising of wind power and multi-dimensional feature data. This effectively removes noise from the data while preserving the main features of the signal. Furthermore, the Transformer model, by introducing a local information enhancement module and a sparse attention mechanism, enhances the model's ability to capture local variation features in wind power time-series data while retaining the learning of global dependencies. The combined prediction model (IMS-RLMD-LSA-Transformer), by fusing time-frequency denoising and deep learning feature extraction, improves the accuracy and robustness of wind power prediction, providing a more reliable data foundation for subsequent market participation strategies.

[0297] In addition, the application adopts a new meta-heuristic optimization algorithm based on the concept of weighted average position to determine the optimal harmonic moment of the density estimation method based on harmonic transformation, which has the characteristics of adaptability, can effectively balance the deviation and variance, and improve the estimation accuracy and reliability. It provides a more accurate way to estimate the probability density of prediction error, which can help better handle errors and uncertainties in the optimization process and avoid the limitations of traditional methods. By generating the volatility domain, a multiple uncertainty set with adaptive boundary optimization is constructed to provide more accurate optimization solutions for wind power systems under different complex and dynamic scenarios.

[0298] In addition, by establishing a master-slave game optimization model between the wind farm and the cloud energy storage operator, the application can optimize the lease price and capacity between the wind farm and the cloud energy storage. This game optimization model not only balances the interests of the wind farm and the cloud energy storage operator, but also effectively reduces the uncertainty caused by wind power fluctuations and improves the overall efficiency of the system. By using robust optimization methods and combining a multi-objective optimization model considering cloud energy storage and power quality, the participation strategy of the energy market and frequency modulation market of the system can be optimized under a certain degree of conservatism. Even in the face of prediction errors and market fluctuations, the system can still achieve the optimal market participation strategy, thereby improving the stability and efficiency of market participation.

[0299] The above only describes the preferred embodiments of the application and is not intended to limit the application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the application shall be included in the protection scope of the application.

Claims

1. A method for wind power systems based on cloud energy storage to participate in the energy-frequency regulation market, characterized in that, Includes the following steps: S 1. Obtain the original wind power and multi-dimensional feature data, use the improved masking signal method combined with the robust local mean decomposition method to perform time-frequency synchronous decomposition and noise reduction on the original wind power and multi-dimensional feature data, and use the Transformer model based on local information enhancement and sparse attention mechanism to predict wind power. S2. Determine the optimal harmonic moment number based on the density estimation method of harmonic transformation, perform probability density estimation on the prediction error data of power and price, and construct a set of multiple uncertainties. S3. Based on multiple uncertainty sets, construct a master-slave game optimization model between wind farms and cloud energy storage operators, and a multi-objective optimization model of wind power system considering cloud energy storage and power quality. Solve for the optimal capacity and leasing price of cloud energy storage and the energy-frequency regulation market participation strategy of wind power system. The process of determining the optimal harmonic moment number based on the density estimation method of harmonic transformation, estimating the probability density of power and price prediction error data, and constructing a set of multiple uncertainties includes the following steps: S21. Calculate the prediction errors of wind power, energy market capacity clearing price, frequency regulation market capacity clearing price and mileage clearing price respectively. Use the weighted average algorithm to determine the optimal harmonic moment number based on the density estimation method of harmonic transformation, and perform probability density estimation on the prediction error data of power and price. S22. Using the cumulative probability density function and pre-set confidence level to select the upper and lower limits of the prediction error, construct the prediction error fluctuation domains of wind power, energy market capacity clearing price, frequency regulation market capacity clearing price and mileage clearing price, and construct a set of multiple uncertainties with boundary adaptive optimization. The process of constructing a master-slave game optimization model between wind farms and cloud energy storage operators, based on multiple uncertainty sets, and a multi-objective optimization model of the wind power system considering cloud energy storage and power quality, and solving for the optimal capacity and leasing price of leased cloud energy storage and the energy-frequency regulation market participation strategy of the wind power system, includes the following steps: S31. Establish a master-slave game optimization model for wind farms and cloud energy storage operators that considers multiple uncertainties, and solve for the optimal capacity and rental price of cloud energy storage with the goal of maximizing the interests of the master cloud energy storage operator and minimizing the cost of the slave wind farm. S32. Establish a multi-objective optimization model for wind power systems that considers cloud energy storage and power quality. Combine the optimized cloud energy storage leasing price and leasing capacity with the dual objectives of maximizing wind power system profits and optimizing power quality at the grid connection point to solve the energy-frequency regulation market participation strategy for wind power systems.

2. The method for a wind power system based on cloud energy storage to participate in the energy-frequency regulation market according to claim 1, characterized in that, The process of acquiring raw wind power and multidimensional feature data, using an improved masking signal method combined with robust local mean decomposition to perform time-frequency synchronous decomposition and noise reduction on the raw wind power and multidimensional feature data, and then using a Transformer model based on local information enhancement and sparse attention mechanism to predict wind power includes the following steps: S11. Obtain the original wind power and multidimensional feature data, and decompose the original wind power and multidimensional feature data into several optimal component product function components and residual signals based on the robust local mean decomposition method; use Spearman correlation coefficient to remove components containing interference signals, and use approximate entropy to select components containing effective features. S12. The selected components are masked using an improved masking signal method, and the optimal component product function is reconstructed to obtain the denoised wind power signal components. The wind power is predicted by using a Transformer model based on local information enhancement and sparse attention mechanism, combined with the denoised wind power signal components.

3. A method for a wind power system based on cloud energy storage to participate in the energy-frequency regulation market according to claim 2, characterized in that, The process of acquiring raw wind power and multidimensional feature data, decomposing the raw wind power and multidimensional feature data into several optimal component product function components and residual signals based on the robust local mean decomposition method, removing components containing interference signals using the Spearman correlation coefficient, and selecting components containing effective features using approximate entropy includes the following steps: S111. Obtain the original wind power and multi-dimensional feature data, and use the mirror expansion method to process the boundary and search for all local extreme points in the original wind power signal; obtain the local mean function and local envelope function through the moving average method, and calculate the zero-mean signal and pure frequency signal. The formula for calculating the zero-mean signal is: k 11 (t)=x(t)-m 11 (t) The formula for calculating a pure frequency signal is: s 11 (t)=k 11 (t) / a 11 (t) In the formula, k 11 (t) is the zero-mean signal, x(t) is the wind power signal to be decomposed, and m 11 (t) is a local mean function, s 11 (t) is the processed pure frequency signal, a 11 (t) is the local envelope function; S112. Define the objective function, extract the first product function from the original wind power signal to obtain the residual signal, and treat the residual signal as a new signal to repeatedly perform iterative extraction of all the optimal component product function components until the residual becomes a constant or a monotonic function. The objective function is expressed as follows: f = RMS[z(t)] + EK[z(t)] In the formula, f is the objective function value, RMS is the root mean square, EK is the empirical kurtosis calculation formula, and z(t) is the zero baseline envelope signal. Let N be the arithmetic mean of z(t). s The total number of signals; S 113. Use the Spearman correlation coefficient to remove components containing interference signals, and use approximate entropy to select components containing effective features.

4. A method for a wind power system based on cloud energy storage to participate in the energy-frequency regulation market according to claim 2, characterized in that, The process involves masking selected components using an improved signal masking method and reconstructing the optimal component product function to obtain the denoised wind power signal components. The prediction of wind power using a Transformer model based on local information enhancement and sparse attention mechanisms, combined with the denoised wind power signal components, includes the following steps: S 121. Calculate the masked mean signal, and select several components from the robust local mean decomposition result for masking by combining the Spearman correlation coefficient and approximate entropy; perform robust local mean decomposition on the masking result to obtain the masked signal components, and perform final signal reconstruction based on the masked signal components to obtain the denoised wind power signal components. The formula for calculating the masking mean signal is as follows: The expression for masking is: The expression for the masked signal components is: In the formula, s i (t) represents the i-th masking signal. The average amplitude, PF is the mean of the instantaneous frequencies. i+ (t), PF i- (t) represents the signal components after average addition and subtraction, respectively, PF i (t) represents the original component, IMS_PF i For the masked signal components, pf i+ , pf i- PF i+ (t) and PF i- (t) is the first component of the decomposed output; S 122. Standardize the denoised wind power signal components, use a fully connected layer to perform high-dimensional representation of the standardized wind power signal components, and extract local information of the enhanced data by stacking multiple one-dimensional convolutions to obtain the locally enhanced signal. S123. The samples obtained by performing multi-layer convolution operations on the locally enhanced signal are used as the input of the encoder layer. The position information of each sample is marked by position encoding. The position encoded result is input into the encoder layer of the Transformer model, and the data is processed by layer normalization in the encoder. S 124. The multi-head sparse attention mechanism is used to replace the fully connected attention mechanism in the standard Transformer model to capture global information of the sequence, integrate global and local information of wind power data, and map and learn the components of wind power through the intermediate layer of the network, and finally output the wind power prediction result.

5. A method for a wind power system based on cloud energy storage to participate in the energy-frequency regulation market according to claim 1, characterized in that, The steps involved in calculating the prediction errors for wind power, energy market capacity clearing price, frequency regulation market capacity clearing price, and mileage clearing price, determining the optimal harmonic moment number using a weighted average algorithm based on the harmonic transformation density estimation method, and performing probability density estimation on the prediction error data for power and price include: S211. Calculate the prediction errors for wind power, energy market capacity clearing price, frequency regulation market capacity clearing price, and mileage clearing price, respectively, where the prediction error is the difference between the predicted value and the actual value. S212. Initialize parameters and construct a fitness function based on relative entropy for the optimal harmonic moment number. In each iteration, calculate the weighted average position of the current harmonic moment number and use the weighted average position of the current harmonic moment number as the representative position of the entire search space. Employ different moving exploration strategies, with the objectives of exploring a wide range of solution spaces and finely adjusting the optimal harmonic moment number, respectively. The fitness function based on the optimal harmonic moment number of relative entropy is expressed as follows: The formula for calculating the weighted average position is: In the formula, Fitness(h) is the fitness function based on the optimal harmonic moment number of relative entropy, and f(h) is the probability density function based on harmonic transformation. Let N be the probability density function for kernel density estimation, h be the harmonic moments to be optimized, and N be the probability density function for kernel density estimation. Candidate The number of selected candidate solutions, P is the population size, it is the current iteration number, and Max is the maximum number of solutions. It Sum is the maximum number of iterations. Fitness h is the sum of all fitness values ​​of the selected candidates. i Let h be the i-th candidate optimal harmonic moment number. Miu The position of the weighted average; S213. Based on random constants and dynamic parameters related to iteration number, a moving strategy is selected and adjusted to achieve a balance between exploration and development, obtain the optimal harmonic moment number, and use the density estimation method based on harmonic transformation to estimate the probability density of prediction errors for wind power probability, energy market capacity clearing price, frequency regulation market capacity clearing price and mileage clearing price. The expression for the probability density function is: The expression for the probability density function based on harmonic transform is: In the formula, f e (e) is the probability density function obtained by the inverse Merlin transform, α is a fixed value, δ is the assumed value, e is the prediction error, f(e) is the probability density function based on the harmonic transform, and H[] is the harmonic transform.

6. A method for a wind power system based on cloud energy storage to participate in the energy-frequency regulation market according to claim 1, characterized in that, The process of constructing the prediction error fluctuation domains for wind power, energy market capacity clearing price, frequency regulation market capacity clearing price, and mileage clearing price by using the cumulative probability density function and pre-set confidence level to select the upper and lower limits of the prediction error, and constructing a set of multiple uncertainties for boundary adaptive optimization, includes the following steps: S221. Using the cumulative probability density function and combining it with the pre-set confidence level to select the upper and lower limits of the prediction error, construct the prediction fluctuation domain of uncertain variables. The cumulative probability density function is expressed as follows: The confidence interval is: e lower =F -1 (0.025) e upper =F -1 (0.975) The expression for predicting the fluctuation range is: In the formula, F(e) is the cumulative probability density function, e is the prediction error, and e lower e upper These are the lower and upper limits of the prediction error fluctuation, respectively, F -1 It is the inverse function of the cumulative distribution function. The predicted value at time t; S222: Integrate the prediction error fluctuation domains of wind power, energy market capacity clearing price, frequency regulation market capacity clearing price, and mileage clearing price into a set of multiple uncertainties; The expression for the set of multiple uncertainties is: In the formula, These are wind power capacity, energy market capacity clearing price, frequency regulation market capacity clearing price, and mileage clearing price. These represent the upper and lower limits of the predicted fluctuation in wind power at time τ. These represent the upper and lower limits of the energy market capacity clearing price at time t. These represent the upper and lower limits of the frequency modulation market capacity clearing price at time t. These represent the upper and lower limits of the mileage clearing price at time t.

7. A method for a wind power system based on cloud energy storage to participate in the energy-frequency regulation market according to claim 1, characterized in that, The master-slave game optimization model between wind farms and cloud energy storage operators includes the optimization model of the main player, cloud energy storage operators, and the robust optimization model of the slave player, wind power clusters. The objective function expression of the optimization model for the cloud energy storage operator, the main player in the game, is as follows: The objective function expression of the robust optimization model for wind power clusters in the game theory approach is: In the formula, I c For the benefit of cloud energy storage operators, For the revenue from leasing energy storage, To generate revenue from FM market bidding, For revenue from bidding in the energy market, For cloud energy storage operation costs, These are the rental prices for cloud energy storage power and capacity, respectively, P t cl , These represent the charging and discharging power and capacity of cloud energy storage, respectively, with Δt representing the current market timescale. For frequency modulation performance indicators, For cloud energy storage bidding capacity in the frequency regulation market, These represent the capacity clearing price and mileage clearing price in the FM market, respectively, where η is the mileage call rate. To clear the energy market for capacity. Let τ represent the bidding capacity for cloud energy storage in the energy market, τ be the real-time market value at a specific moment, and t be the day-ahead market value at a specific moment. These represent the cloud energy storage capacity cost and mileage cost, respectively. w The total cost of the wind farm, These are the costs of wind power leasing and cloud energy storage, wind power deviation penalty costs, and opportunity costs. λ1 and λ2 represent the bidding revenue of wind power in the frequency regulation market and the bidding revenue in the electricity market, respectively, and the bidding penalty coefficients in the energy market and the frequency regulation market, respectively. These represent the day-ahead bidding capacity of wind power in the energy market and frequency regulation market, respectively. These represent the real-time bidding capacity of wind and storage in the energy market and the frequency regulation market, respectively, with Δτ representing the real-time market time scale. The bidding price for frequency modulation capacity, The bidding price for frequency regulation mileage is φ(t), where φ(t) is the time-of-use electricity price in the electricity market, and P is the mileage bidding price. w.fm.c For the declared capacity of wind power clusters, P w.fm.c.max To allow wind power clusters to apply for capacity limits in the frequency regulation market, K W (t) represents the frequency modulation performance index value, P w.e.c (t), R w，e，c (t) represents the capacity and price of the wind power cluster in the energy market bidding, respectively.

8. A method for a wind power system based on cloud energy storage to participate in the energy-frequency regulation market according to claim 7, characterized in that, The objective function of the multi-objective optimization model for wind power systems that considers cloud energy storage and power quality includes optimizing wind power system revenue and optimizing grid-connected power quality of wind power systems. The expression for the optimal return of the wind power system is: The expression for the optimal power quality of grid-connected wind power systems is: In the formula, F1 is the revenue function of the wind power system. For the total revenue of the energy and frequency modulation market, To incur penalties for deviations in wind power development. To optimize the cost of cloud energy storage for wind power leasing, Let H be the opportunity cost of the second phase, and H be the set of 24 time periods in a day. These are the day-ahead bidding caps for the second phase of the wind power energy market and the frequency regulation market, respectively. These refer to the real-time bidding capacity of wind and storage in the second phase of the energy market and the frequency regulation market, respectively. For wind power frequency regulation performance indicators, These represent the day-ahead bidding capacity for wind power in the second phase of the energy market and frequency regulation market. These represent the real-time bidding capacity of wind and storage in the second phase of the energy market and the frequency regulation market, respectively, P w.fm.c.o For the second phase of wind power cluster application capacity, F2 is the power quality function of the wind power system connected to the grid, M is the total number of time nodes throughout the year, and P is the power quality function of the wind power system connected to the grid. w Q(t) represents the output power of the wind farm at time t, and Q(t) represents the comprehensive power quality assessment value at time t.

Images