Wind power probability prediction method and system based on diffusion model and non-stationary signal enhancement

Abstract

The present invention discloses a method and system for predicting wind power probability based on a diffusion model and non-stationary signal enhancement. The method includes: using a diffusion model to model the uncertainty of historical wind power, sampling the probability distribution characteristics of wind power through a diffusion process and a denoising process; using historical wind power sequences and future weather forecast data to grasp the non-stationary fluctuation trend of future wind power, constructing a non-stationary signal enhancement method, as a kernel solver of the diffusion model denoising process, and rolling training the diffusion model based on a shared parameter mechanism; based on the diffusion model obtained by training with non-stationary signal enhancement as the kernel solver, a multi-step probability prediction result of wind power is obtained using a double-layer sampling method along the diffusion step length and time series reasoning. The present invention can be deployed in a power grid monitoring and dispatching department or a wind farm station to provide accurate multi-time scale wind power probability prediction results to meet the needs of power grid monitoring and economic safety regulation.

Description

Technical Field

[0001] The present invention belongs to the technical field of renewable energy development and utilization, and in particular relates to a method and system for predicting wind power probability based on a diffusion model and non-stationary signal enhancement. Background Art

[0002] In response to global climate change and achieving sustainable development of the global economy and society, the construction of smart grids including new energy is playing an increasingly important role. As the cost of wind power installation continues to decline, my country's wind power generation is showing a rapid development trend.

[0003] However, the non-stationary characteristics of wind power show the time series fluctuation of wind power mean and variance, which is an important factor causing grid frequency fluctuation and hindering local consumption. In addition, the modeling of wind power uncertainty is also lacking. Therefore, it is necessary to consider the non-stationary fluctuation of wind power and implement accurate wind power probability prediction to provide data support for safe and stable grid dispatching.

[0004] Wind power probability prediction is mainly divided into parameter prediction and non-parameter prediction. Although the non-parameter prediction method has the advantage of high computational efficiency, it is difficult to fit the probability density function of wind power as a whole, and it is easy to cause the error accumulation problem of multi-step prediction, which limits the improvement of the accuracy and reliability of wind power probability prediction. Therefore, it is difficult to grasp the time series non-stationary characteristics of wind power and accurately predict the probability distribution of wind power. Summary of the invention

[0005] Purpose of the invention: In view of the difficulties in the above-mentioned wind power probability prediction, the present invention proposes a wind power probability prediction method and system based on diffusion model and non-stationary signal enhancement, which effectively improves the hourly wind power probability prediction performance.

[0006] Technical solution: In order to achieve the above-mentioned invention object, the wind power probability prediction method based on diffusion model and non-stationary signal enhancement proposed in the present invention includes the following steps:

[0007] The uncertainty of historical wind power is modeled using a diffusion model. The diffusion model includes a diffusion process and a denoising process. The diffusion process continuously adds Gaussian noise to the original wind power until its characteristics are completely destroyed. The denoising process uses a kernel solver to fit the Gaussian distribution parameters of each step of the diffusion process to obtain the probabilistic modeling results of historical wind power.

[0008] The non-stationary fluctuation trend of future wind power is grasped by using the key parameters of historical wind power stability and future wind speed forecast information, and a non-stationary signal enhancement method is constructed. The non-stationary signal enhancement method is used as the kernel solver of the diffusion model denoising process, and the diffusion model is trained in a rolling manner based on a shared parameter mechanism.

[0009] Based on the trained diffusion model with non-stationary signal enhancement as the kernel solver, the multi-step probability prediction results of wind power are obtained by using the double-layer sampling method along the diffusion step and time series reasoning.

[0010] Furthermore, the diffusion process is expressed as:

[0011]

[0012] Among them, q(x n |x n-1 ) represents the Gaussian distribution that needs to be added in the n-1th step, x n Represents the characteristics of the nth step of the diffusion process; Indicates the mean The standard deviation is β n Gaussian distribution of I; β n represents the noise ratio of the n-th step diffusion process, and I represents the unit matrix;

[0013] By setting α n =1-β n , we get the direct analytical solution method for the diffusion process from step 0 to step n:

[0014]

[0015] in The back propagation probability of the diffusion model is obtained by the Bayesian method:

[0016]

[0017]

[0018] in, and are the expectation and variance of the back propagation probability of the diffusion model, q(x n |x n-1 ), q(x n ) and q(x n-1 ) are calculated from the diffusion process.

[0019] Furthermore, the noise ratio β of the diffusion process n It is expressed as:

[0020]

[0021]

[0022] Where s is a positive real number and N represents the step size of the diffusion process.

[0023] Furthermore, the denoising process of the diffusion model is to make the denoising probability pθ (x n-1 |x n ) as close as possible to the back propagation probability q(x n-1 |x n ,x 0 ), the denoising probability is expressed as θ , standard deviation is Σ θ Gaussian distribution The KL divergence is used to establish the likelihood function to measure the difference between the learnable denoising probability and the actual diffusion probability:

[0024]

[0025] Among them, C is a constant and has nothing to do with the model parameter θ; represents the variational bound operator;

[0026] Then the mean of the denoising probability is:

[0027]

[0028] Among them, ∈ θ (x n ,n) is a learnable diffusion model solver. The diffusion model is solved by training solver ∈ θ (x n ,n) realize end-to-end training. After the training is completed, the wind power probability prediction result can be obtained by Gaussian noise sampling.

[0029] Furthermore, the non-stationary signal enhancement method specifically includes:

[0030] The expectation and variance of the time series section are used to characterize the non-stationary characteristics of wind power fluctuations, and the historical wind power series is summarized as a stationary series:

[0031]

[0032] Among them, Q′ is the query vector of the historical wind power sequence after stabilization, μ Q and σ Q are the expectation and variance of the historical wind power series respectively;

[0033] The improved multi-head attention mechanism is used to calculate the similarity relationship between the stabilized wind power sequences, and the calculation method of the destabilized attention mechanism is obtained:

[0034]

[0035]

[0036]

[0037] Among them, fMLP is a learnable linear layer, d k represents the dimension of Q′, represents the weather forecast wind speed data, K' and V' are the key vector and value vector of the stabilized historical wind power sequence, τ and Δ represent the key parameters for measuring the stability of wind power, which are predicted by the historical wind power sequence and weather forecast wind speed data.

[0038] Furthermore, the non-stationary signal enhancement method is used as the kernel solver of the diffusion model denoising process, and the rolling training of the diffusion model based on the shared parameter mechanism includes:

[0039] Incorporating hidden states to characterize the temporal relationship of wind power fluctuations, the hidden state h t The update is based on the previous hidden state h t -1 , denoised historical wind power series And the weather forecast information at time t It is derived and expressed as:

[0040]

[0041] Among them, H Datt () represents the non-stationary signal enhancement method, represents the value of diffusion step size n at time t, f FFN,θ represents the linear forward propagation function, The transition variable representing the hidden state update process;

[0042] The non-stationary enhancement method is used as the kernel solver of the diffusion model denoising process. The minimum negative log-likelihood function is constructed by the hidden state transfer variable as the loss function to train the diffusion model that integrates the non-stationary signal enhancement method. During training, the training parameters are shared among various denoising processes. The training process of the diffusion model that integrates the non-stationary signal enhancement method is expressed as:

[0043]

[0044] Where ∈ represents the standard Gaussian distribution.

[0045] Furthermore, the double-layer sampling method along the diffusion step and the temporal reasoning specifically includes:

[0046] The denoising feature of the denoising process at the same time is determined by the final value of the sampling diffusion process Get the prediction result after denoising at step n-1 at time t It is expressed as:

[0047]

[0048] in, Represents the trained model;

[0049] Iterate from step n-1 to step 0 to get the wind power probability prediction result at time t

[0050] Based on the hidden state h t , wind power measurement sequence And weather forecast information at time t+1 The hidden state h is updated by the diffusion model obtained by training and integrating the non-stationary signal enhancement method t+1 , and then realize the rolling solution of the model to obtain the multi-step probability prediction results of wind power Where T is the time step that needs to be predicted.

[0051] The present invention also provides a wind power probability prediction system based on a diffusion model and non-stationary signal enhancement, comprising:

[0052] The historical data simulation module uses a diffusion model to model the uncertainty of historical wind power. The diffusion model includes a diffusion process and a denoising process. The diffusion process continuously adds Gaussian noise to the original wind power until its characteristics are completely destroyed. The denoising process uses a kernel solver to fit the Gaussian distribution parameters of each diffusion process to obtain the probabilistic modeling results of historical wind power.

[0053] The model training module uses the key parameters of historical wind power stability and future wind speed forecast information to grasp the non-stationary fluctuation trend of future wind power, constructs a non-stationary signal enhancement method, uses the non-stationary signal enhancement method as the kernel solver of the diffusion model denoising process, and trains the diffusion model in a rolling manner based on a shared parameter mechanism;

[0054] The model prediction module, based on the trained diffusion model with non-stationary signal enhancement as the core solver, uses the double-layer sampling method along the diffusion step and time series reasoning to obtain the multi-step probability prediction results of wind power.

[0055] The present invention also provides a computer device, comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the programs are executed by the processors, the steps of the wind power probability prediction method based on the diffusion model and non-stationary signal enhancement as described above are implemented.

[0056] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the wind power probability prediction method based on the diffusion model and non-stationary signal enhancement as described above are implemented.

[0057] Beneficial effects: The present invention aims at the two major limitations of the current research on wind power probability prediction: the difficulty in accurately modeling the uncertainty of wind power and the problem of over-stability in prediction. A wind power probability prediction method based on diffusion model and non-stationary signal enhancement is proposed. The uncertainty of wind power is modeled by diffusion model, so that the model does not rely on the prior knowledge of wind power probability distribution, and an accurate wind power probability density function is obtained through gradual diffusion and denoising process; and non-stationary signal enhancement is designed as the kernel solver of the diffusion model denoising process to capture the non-stationary characteristics of wind power time series fluctuations, solve the problem of over-stability in wind power prediction, introduce meteorological forecast data and non-stationary parameters to correct the prediction results in real time, which can effectively improve the wind power fluctuation fitting ability and reduce error accumulation. The present invention provides wind power probability prediction results 10 minutes to 2.5 hours in advance, which can be applied to the centralized prediction platform of the power grid or the on-site prediction system of the wind farm, can effectively foresee the non-stationary fluctuation of wind power, improve the ability of wind power on-site consumption, and thus meet the needs of economic and safe regulation of the power grid, with good social and economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a schematic diagram of the prediction method flow of the present invention;

[0059] Figure 2 This is a schematic diagram of the non-stationary signal enhancement method proposed by the present invention;

[0060] Figure 3 This is a schematic diagram of the parameter sharing training method proposed by the present invention;

[0061] Figure 4 This is a wind power probability prediction effect diagram in an embodiment of the present invention;

[0062] Figure 5 This is a diagram showing the effect of multi-step wind power prediction in an embodiment of the present invention. DETAILED DESCRIPTION

[0063] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0064] Reference Figure 1 The present invention proposes a wind power probability prediction method based on a diffusion model and non-stationary signal enhancement, comprising the following steps:

[0065] Step S1, using the diffusion model to model the uncertainty of historical wind power. The diffusion model includes a diffusion process and a denoising process. The diffusion process continuously adds Gaussian noise to the original wind power until its characteristics are completely destroyed. The denoising process uses a kernel solver to fit the Gaussian distribution parameters of each diffusion process to obtain the probabilistic modeling result of historical wind power.

[0066] In the present invention, the diffusion model is also called a generative diffusion model. In order to accurately model the uncertainty of wind power, the generative diffusion model samples the probability distribution characteristics of wind power through two stages: diffusion and denoising.

[0067] (1) The diffusion process can be expressed by continuously adding Gaussian noise to the original wind power until its characteristics are completely destroyed.

[0068] Shown as:

[0069]

[0070] Among them, q(x n |x n-1 ) represents the Gaussian distribution that needs to be added in the n-1th step. n Represents the characteristics of the nth step of the diffusion process. Indicates the mean The standard deviation is β n Gaussian distribution of I. β n represents the noise ratio of the n-th diffusion process, and I represents the unit matrix.

[0071] Therefore, by setting α n =1-β n , we can get the direct analytical solution method of the diffusion process from step 0 to step n:

[0072]

[0073] in,

[0074] Next, the back propagation probability of the diffusion model can also be obtained by the Bayesian method:

[0075]

[0076]

[0077] in, and are the expectation and variance of the back propagation probability of the diffusion model, q(x n |x n-1 ), q(x n ) and q(x n-1 ) can be calculated by the diffusion process. By setting an N-step diffusion process, the original wind power sequence can be denoised into Gaussian noise.

[0078] Considering that the degree of noise added in the diffusion process increases with the diffusion step length, the noise ratio of the diffusion process β n It can be expressed as:

[0079]

[0080]

[0081] Where s is a positive real number and N represents the step size of the diffusion process.

[0082] (2) The denoising process of the diffusion model is to make the denoising probability p that needs to be fitted θ (x n-1 |x n ) as close as possible to the back propagation probability q(x n-1 |x n ,x 0 ), the denoising probability can be expressed as θ , standard deviation is σ θ Gaussian distribution In order to accurately measure the difference between the learnable denoising probability and the actual diffusion probability, the KL divergence is used to establish the likelihood function:

[0083]

[0084] Among them, C is a constant and has nothing to do with the model parameter θ. represents the variational bound operator.

[0085] Therefore, the mean of the denoising probability can be derived as:

[0086]

[0087] Among them, ∈ θ (x n ,n) is a learnable diffusion model solver. The diffusion model is solved by training the solver ∈ θ (x n ,n) realize end-to-end training. After the training is completed, the wind power probability prediction result can be obtained by Gaussian noise sampling.

[0088] Step S2, using the key parameters of historical wind power stability and future wind speed forecast information to grasp the non-stationary fluctuation trend of future wind power, constructing a non-stationary signal enhancement method, using the non-stationary signal enhancement method as the kernel solver of the diffusion model denoising process, and rolling training the diffusion model based on the shared parameter mechanism.

[0089] like Figure 2 As shown, the non-stationary signal enhancement method specifically includes:

[0090] First, considering that the non-stationary characteristics of wind power fluctuations can be characterized by the expectation and variance of the time series section, the historical wind power series query vector is summarized as a stationary series:

[0091]

[0092] Where Q′ is the stabilized wind power sequence, μ x and σ x are the expectation and variance of the historical wind power series respectively. The same can be extended to key vectors and value vectors:

[0093] Next, since the non-stationary characteristics of future wind power are difficult to be completely inferred from the historical wind power sequence, incorporating the weather forecast wind speed information can enhance the prediction model's ability to grasp the non-stationary characteristics of wind power. Therefore, the improved multi-head attention mechanism can be used to calculate the similarity relationship between the stabilized wind power sequences, and the destabilization attention mechanism calculation method H is obtained. Datt ():

[0094]

[0095]

[0096]

[0097] Among them, f MLP is a learnable linear layer, d k is the dimension of the stationary feature Q′, represents the weather forecast wind speed data, τ and Δ represent the key parameters for measuring the stability of wind power, which are predicted by the historical wind power series and weather forecast wind speed data.

[0098] According to one implementation, a non-stationary wind power enhancement module is established in step S2. The non-stationary wind power enhancement module is connected to the historical wind power sequence and the future weather forecast data. The non-stationary features in the historical wind power sequence are retained through a de-stationary attention mechanism to alleviate the common over-stationary problem.

[0099] Combining the advantages of diffusion model and non-stationary signal enhancement method, the diffusion model with non-stationary signal enhancement method as the core solver includes model training process and sampling solution process.

[0100] The training process of the scattered model with non-stationary signal enhancement method as the core solver is as follows Figure 3 First, in order to enrich the feature representation learning ability, hidden states are incorporated to represent the temporal relationship of wind power fluctuations. Hidden state h t The update is based on the hidden state h at the previous moment t-1 , the denoised historical wind power series And the weather forecast information at time t It can be deduced and expressed as:

[0101]

[0102] in, represents the value of diffusion step size n at time t, f FFN,θ represents the linear forward propagation function, Transition variable representing the hidden state update process.

[0103] Secondly, the minimum negative log-likelihood function can be constructed from the hidden state transfer variables and used as the loss function to train the diffusion-non-stationary signal modeling model, which can be expressed as:

[0104]

[0105] Next, since the essence of diffusion model fitting is to train the kernel solver ∈ θ The non-stationary enhancement method is used as the kernel solver of the diffusion model denoising process, so that the model can take into account the non-stationary fluctuation characteristics and uncertainty of wind power. During training, the training parameters are shared among various denoising processes. The model training process integrating the non-stationary signal enhancement method can be expressed as:

[0106]

[0107] Where ∈ represents the standard Gaussian distribution.

[0108] Step S3, based on the diffusion model obtained through training with non-stationary signal enhancement as the core solver, a multi-step probability prediction result of wind power is obtained by using a double-layer sampling method along the diffusion step and time series reasoning. This is the sampling solution process.

[0109] The double-layer sampling method along the diffusion step and temporal reasoning specifically includes:

[0110] First, the denoising characteristics of the denoising process at the same time can be obtained by sampling the final value of the diffusion process Therefore, the prediction result after denoising at step n-1 at time t is It can be expressed as:

[0111]

[0112] in, Represents the trained model.

[0113] Then, by iterating from step n-1 to step 0, the wind power probability prediction result at time t can be obtained: Then, based on the hidden state h t , wind power measurement sequence And weather forecast information at time t+1 The non-stationary signal enhancement module obtained by training updates the hidden state h t+1 , and then realize the rolling solution of the prediction model to obtain the multi-step probability prediction results of wind power Where T is the time step that needs to be predicted.

[0114] The above describes the specific implementation process of the method of the present invention for wind power probability prediction. The present invention also selects a data set provided by the United States Renewable Energy Laboratory and selects historical wind power measurement information of a wind farm in Michigan, USA from 2010 to 2012 for model verification. The wind power measurement time resolution is 10 minutes.

[0115] According to the multi-step probability prediction results of photovoltaic power, three error analysis evaluation indicators are used to evaluate the prediction performance of the model, namely, prediction interval bandwidth (PIAW), prediction interval coverage (PICP), and comprehensive indicator Winkler score (WS), which are expressed as follows:

[0116]

[0117]

[0118]

[0119] Among them, y t Represents the measured value of wind power at time t. and Respectively, they represent the lower and upper bounds of the probability interval with confidence level α at time t. s Indicates the number of samples in the test set.

[0120] For the dataset of the National Renewable Energy Laboratory, data from January 2010 to December 2011 were selected as training samples, and data from January 2012 to December 2012 were selected as test samples. The probability prediction error of wind power one hour in advance using the method of the present invention is shown in Table 1. In order to verify the gain of the non-stationary signal enhancement method for wind power prediction, Table 2 compares the original diffusion model and the diffusion model with a recurrent neural network (RNN) as the core solver, and shows the effectiveness of the non-stationary signal enhancement method in solving the over-stationary problem based on error analysis and stationarity test (Augmented Dickey-Fuller, ADF). In addition, in order to intuitively reflect the prediction effect, Figure 4 The wind power probability prediction results obtained by using the method of the present invention are shown. Figure 5 The performance of the model of the present invention in a longer prediction time scale is shown. From Table 1 and Table 2, Figure 4 , Figure 5 It can be seen that the present invention can accurately realize the probability prediction of wind power 10 minutes to 2.5 hours in advance, meeting the real-time dispatching requirements of the power grid.

[0121] Table 1 Error indicators of test samples based on the dataset of the US Renewable Energy Laboratory

[0122] PIAW PICP WS 90% confidence 1.901±1.062 83.537±11.937 2.246±1.031 80% confidence 1.297±0.613 78.636±21.476 1.892±0.575

[0123] Table 2 Performance gain of non-stationary signal enhancement based on the NREL dataset

[0124]

[0125]

[0126] In summary, the wind power probability prediction method based on diffusion model and non-stationary signal enhancement proposed in the present invention can significantly improve the accuracy of wind power probability prediction 10 minutes to 2.5 hours in advance. The present invention can be applied to the centralized prediction system of new energy or the on-site prediction system of wind farms, guiding relevant units to adjust power generation plans according to the photovoltaic power probability prediction results, reduce spare capacity, promote on-site consumption of new energy, and meet the needs of power grid monitoring and economic safety regulation.

[0127] Based on the same technical concept as the method embodiment, the present invention also provides a wind power probability prediction system based on a diffusion model and non-stationary signal enhancement, comprising:

[0128] The historical data simulation module uses a diffusion model to model the uncertainty of historical wind power. The diffusion model includes a diffusion process and a denoising process. The diffusion process continuously adds Gaussian noise to the original wind power until its characteristics are completely destroyed. The denoising process uses a kernel solver to fit the Gaussian distribution parameters of each diffusion process to obtain the probabilistic modeling results of historical wind power.

[0129] The model training module uses the key parameters of historical wind power stability and future wind speed forecast information to grasp the non-stationary fluctuation trend of future wind power, constructs a non-stationary signal enhancement method, uses the non-stationary signal enhancement method as the kernel solver of the diffusion model denoising process, and trains the diffusion model in a rolling manner based on a shared parameter mechanism;

[0130] The model prediction module, based on the trained diffusion model with non-stationary signal enhancement as the core solver, uses the double-layer sampling method along the diffusion step and time series reasoning to obtain the multi-step probability prediction results of wind power.

[0131] It should be understood that the wind power probability prediction system based on diffusion model and non-stationary signal enhancement in the embodiment of the present invention can implement all the technical solutions in the above-mentioned method embodiment, and the functions of its various functional modules can be specifically implemented according to the methods in the above-mentioned method embodiments. The specific implementation process can refer to the relevant description in the above-mentioned embodiments, which will not be repeated here.

[0132] The present invention also provides a computer device, comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the programs are executed by the processors, the steps of the wind power probability prediction method based on the diffusion model and non-stationary signal enhancement as described above are implemented.

[0133] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the wind power probability prediction method based on the diffusion model and non-stationary signal enhancement as described above are implemented.

[0134] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, devices (systems), computer equipment or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0135] The present invention is described with reference to a flowchart of a method according to an embodiment of the present invention. It should be understood that each process in the flowchart and a combination of processes in the flowchart can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process in the flowchart. Figure 1 A device that specifies functions in a process or multiple processes.

[0136] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A function specified in a process or multiple processes.

[0137] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 The steps of a specified function in a process or multiple processes.

Claims

1. A wind power probability prediction method based on diffusion model and non-stationary signal enhancement, characterized in that: The following steps are involved: The uncertainty of historical wind power is modeled using a diffusion model. The diffusion model includes a diffusion process and a denoising process. The diffusion process continuously adds Gaussian noise to the original wind power until its characteristics are completely destroyed. The denoising process uses a kernel solver to fit the Gaussian distribution parameters of each step of the diffusion process to obtain the probabilistic modeling results of historical wind power. The non-stationary fluctuation trend of future wind power is grasped by using the key parameters of historical wind power stability and future wind speed forecast information, and a non-stationary signal enhancement method is constructed. The non-stationary signal enhancement method is used as the kernel solver of the diffusion model denoising process, and the diffusion model is trained in a rolling manner based on a shared parameter mechanism. Based on the trained diffusion model with non-stationary signal enhancement as the core solver, the multi-step probability prediction results of wind power are obtained by using the double-layer sampling method along the diffusion step and time series reasoning. Among them, the non-stationary signal enhancement method specifically includes: The expectation and variance of the time series section are used to characterize the non-stationary characteristics of wind power fluctuations, and the historical wind power series is summarized as a stationary series: Among them, Q′ is the query vector of the historical wind power sequence after stabilization, μ Q and σ Q are the expectation and variance of the historical wind power series respectively; The improved multi-head attention mechanism is used to calculate the similarity relationship between the stabilized wind power sequences, and the calculation method of the destabilized attention mechanism is obtained: Among them, f MLP is a learnable linear layer, d k represents the dimension of Q′, represents the weather forecast wind speed data, K' and V' are the key vector and value vector of the stabilized historical wind power sequence, τ and Δ represent the key parameters for measuring the stability of wind power, which are predicted by the historical wind power sequence and the weather forecast wind speed data; The non-stationary signal enhancement method is used as the kernel solver of the diffusion model denoising process, and the diffusion model is trained based on the shared parameter mechanism, including: Incorporating hidden states to characterize the temporal relationship of wind power fluctuations, the hidden state h t The update is based on the previous hidden state h t-1 , denoised historical wind power series And the weather forecast information at time t It is derived and expressed as: Among them, H Datt () represents the non-stationary signal enhancement method, represents the value of diffusion step size n at time t, f FFN,θ represents the linear forward propagation function, The transition variable representing the hidden state update process; The non-stationary enhancement method is used as the kernel solver of the diffusion model denoising process. The minimum negative log-likelihood function is constructed by the hidden state transfer variable as the loss function to train the diffusion model that integrates the non-stationary signal enhancement method. During training, the training parameters are shared among various denoising processes. The training process of the diffusion model that integrates the non-stationary signal enhancement method is expressed as: Where E represents the variational bound operator, ∈ represents the standard Gaussian distribution, θ is the model parameter, ∈ θ A learnable solver for diffusion models.

2. The method according to claim 1, characterized in that The diffusion process is expressed as: Among them, q(x n |x n-1 ) represents the Gaussian distribution that needs to be added in the n-1th step, x n Represents the characteristics of the nth step of the diffusion process; Indicates the mean The standard deviation is β n Gaussian distribution of I; β n represents the noise ratio of the n-th step diffusion process, and I represents the unit matrix; By setting α n =1-β n , we get the direct analytical solution method for the diffusion process from step 0 to step n: in The back propagation probability of the diffusion model is obtained by the Bayesian method: in, and are the expectation and variance of the back propagation probability of the diffusion model, q(x n |x n-1 ), q(x n ) and q(x n-1 ) are calculated from the diffusion process.

3. The method according to claim 2, characterized in that Noise ratio β of the diffusion process n It is expressed as: Where s is a positive real number and N represents the step size of the diffusion process.

4. The method according to claim 2, characterized in that: The denoising process of the diffusion model is to make the denoising probability p θ (x n-1 |x n ) as close as possible to the back propagation probability q(x n-1 |x n ,x0), the denoising probability is expressed as θ , standard deviation is Σ θ Gaussian distribution The KL divergence is used to establish the likelihood function to measure the difference between the learnable denoising probability and the actual diffusion probability: Among them, C is a constant and has nothing to do with the model parameter θ; represents the variational bound operator; Then the mean of the denoising probability is: Among them, ∈ θ (x n ,n) is a learnable diffusion model solver. The diffusion model is solved by training solver ∈ θ (x n ,n) realize end-to-end training. After the training is completed, the wind power probability prediction result can be obtained by Gaussian noise sampling.

5. The method according to claim 1, characterized in that The double-layer sampling method along the diffusion step and temporal reasoning specifically includes: The denoising feature of the denoising process at the same time is determined by the final value of the sampling diffusion process Get the prediction result after denoising at step n-1 at time t It is expressed as: in, Represents the trained model; Iterate from step n-1 to step 0 to get the wind power probability prediction result at time t Based on the hidden state h t , wind power measurement sequence And weather forecast information at time t+1 The hidden state h is updated by the diffusion model obtained by training and integrating the non-stationary signal enhancement method t+1 , and then realize the rolling solution of the model to obtain the multi-step probability prediction results of wind power Where T is the time step that needs to be predicted.

6. A wind power probability prediction system based on diffusion model and non-stationary signal enhancement, characterized in that: include: The historical data simulation module uses a diffusion model to model the uncertainty of historical wind power. The diffusion model includes a diffusion process and a denoising process. The diffusion process continuously adds Gaussian noise to the original wind power until its characteristics are completely destroyed. The denoising process uses a kernel solver to fit the Gaussian distribution parameters of each diffusion process to obtain the probabilistic modeling results of historical wind power. The model training module uses the key parameters of historical wind power stability and future wind speed forecast information to grasp the non-stationary fluctuation trend of future wind power, constructs a non-stationary signal enhancement method, uses the non-stationary signal enhancement method as the kernel solver of the diffusion model denoising process, and trains the diffusion model in a rolling manner based on a shared parameter mechanism; The model prediction module uses the diffusion model obtained through training with non-stationary signal enhancement as the core solver, and obtains the multi-step probability prediction results of wind power by using the double-layer sampling method along the diffusion step and time series reasoning; Among them, the non-stationary signal enhancement method specifically includes: The expectation and variance of the time series section are used to characterize the non-stationary characteristics of wind power fluctuations, and the historical wind power series is summarized as a stationary series: Among them, Q′ is the query vector of the historical wind power sequence after stabilization, μ Q and σ Q are the expectation and variance of the historical wind power series respectively; The improved multi-head attention mechanism is used to calculate the similarity relationship between the stabilized wind power sequences, and the calculation method of the destabilized attention mechanism is obtained: Among them, f MLP is a learnable linear layer, d k represents the dimension of Q′, represents the weather forecast wind speed data, K' and V' are the key vector and value vector of the stabilized historical wind power sequence, τ and Δ represent the key parameters for measuring the stability of wind power, which are predicted by the historical wind power sequence and the weather forecast wind speed data; The non-stationary signal enhancement method is used as the kernel solver of the diffusion model denoising process, and the diffusion model is trained based on the shared parameter mechanism, including: Incorporating hidden states to characterize the temporal relationship of wind power fluctuations, the hidden state h t The update is based on the previous hidden state h t-1 , denoised historical wind power series And the weather forecast information at time t It is derived and expressed as: Among them, H Datt () represents the non-stationary signal enhancement method, represents the value of diffusion step size n at time t, f FFN,θ represents the linear forward propagation function, The transition variable representing the hidden state update process; The non-stationary enhancement method is used as the kernel solver of the diffusion model denoising process. The minimum negative log-likelihood function is constructed by the hidden state transfer variable as the loss function to train the diffusion model that integrates the non-stationary signal enhancement method. During training, the training parameters are shared among various denoising processes. The training process of the diffusion model that integrates the non-stationary signal enhancement method is expressed as: in, represents the variational bound operator, ∈ represents the standard Gaussian distribution, θ is the model parameter, ∈ θ A learnable solver for diffusion models.

7. A computer device, characterized in that: include: one or more processors; Memory; as well as One or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the programs are executed by the processors, the steps of the wind power probability prediction method based on the diffusion model and non-stationary signal enhancement as described in any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the wind power probability prediction method based on diffusion model and non-stationary signal enhancement are implemented as described in any one of claims 1 to 5.

Images