A method and system for generating a renewable energy output sequence based on a moving average diffusion

Abstract

The application belongs to the technical field of power system analysis and artificial intelligence, and relates to a renewable energy output sequence generation method and system based on moving average diffusion, and the technical points are as follows: S1, collecting historical new energy station output data to construct a new energy output sequence; S2, performing normalization processing to obtain a normalized output sequence; S3, performing forward diffusion on the normalized output sequence based on a diffusion transition matrix and using a moving average window; S4, performing adaptive noise scheduling to complete forward noise addition; S5, constructing a reverse diffusion generation model and training the model; and S6, calculating the optimal diffusion step of the model according to a moving average transition matrix, and reconstructing the resolution of the new energy output sequence based on the model. The application implements a diffusion process in the moving average window of the new energy output sequence, adds noise to the low-frequency information according to a matching noise scheduling coefficient, removes the noise, improves the integrity of the data, and reduces the data error.

Description

Technical Field

[0001] This invention relates to the fields of power system analysis and artificial intelligence technology, and specifically discloses a method and system for generating renewable energy output sequences based on moving average diffusion. Background Technology

[0002] As the proportion of renewable energy sources such as wind power and photovoltaics in the power system continues to increase, the fluctuations and stochastic characteristics of renewable energy output place higher demands on power system operation and dispatch. Therefore, accurately constructing renewable energy output sequence models is of great significance for renewable energy power prediction, probabilistic scenario generation, and grid safety operation. Diffusion models are widely used in time series generation and prediction tasks. Existing renewable energy output sequence diffusion models mostly adopt denoising diffusion probabilistic models, which complete data degradation by gradually adding Gaussian noise to the original sequence, and then use neural networks to learn the reverse denoising process, thereby realizing the generation and prediction of renewable energy output sequences.

[0003] However, traditional diffusion models employ a consistent diffusion strategy for all frequency components, while the effective time-series information in renewable energy output sequences is primarily dominated by low-frequency components. During diffusion, low-frequency information decays rapidly, causing the premature loss of key time-series features in renewable energy output sequences, thereby reducing the accuracy of model generation and prediction. Furthermore, differences in information distribution between different diffusion stages can easily lead to model training instability and increased generation errors.

[0004] Therefore, the present invention aims to provide a method and system for generating renewable energy output sequences based on moving average diffusion, in order to solve the above-mentioned problems. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for generating renewable energy output sequences based on moving average diffusion. By diffusion within a moving average window of the renewable energy output sequence, low-frequency information is denoised / de-denoised using a matching noise scheduling coefficient, thus solving the problem of rapid attenuation of low-frequency information during the diffusion process of renewable energy output sequences in traditional diffusion models. The specific plan is as follows: A method for generating renewable energy output sequences based on moving average diffusion, the method comprising the following steps: S1. Collect historical power output data from renewable energy power plants and construct renewable energy power output sequences; S2. The new energy output sequence is normalized to obtain the normalized new energy output sequence. S3. The normalized new energy output sequence is based on the diffusion transfer matrix and forward diffusion is performed using a moving average window. S4. In the process of forward diffusion of the normalized new energy output sequence based on the diffusion transfer matrix and using a moving average window, forward noise is added through adaptive noise scheduling. S5. Construct a reverse diffusion generation model and train the reverse diffusion generation model based on several new energy output sequences in the forward diffusion process. S6. Calculate the optimal diffusion step of the reverse diffusion generation model based on the diffusion transfer matrix and the moving average transfer matrix. The reverse diffusion generation model performs reverse diffusion on the new energy output sequence based on the optimal diffusion step.

[0006] Furthermore, in step S1, power output data from renewable energy power plants is collected, and an initial renewable energy power output sequence is constructed based on this data. .

[0007] Furthermore, the initial renewable energy output sequence Normalization is performed by applying Z-score standardization to the renewable energy output sequence to obtain the normalized initial renewable energy output sequence. .

[0008] Furthermore, the initial renewable energy output sequence Proceed forward diffusion until forward diffusion reaches the first... Step to get ,in This indicates the total number of forward diffusion steps; The forward diffusion process is as follows:

[0009] Represents the initial new energy output sequence In the The forward diffusion process yields the new energy output sequence. ; No. The forward diffusion yields the new energy output sequence. for:

[0010] in, Indicates the first Step diffusion transition matrix; Indicates the first The noise scheduling coefficient corresponding to the forward diffusion step; Represents the identity matrix; Indicates standard Gaussian noise; This indicates that the mean is 0 and the covariance matrix is... Gaussian distribution; for:

[0011] in, This represents a diagonal matrix generating function; Indicates the first The cumulative forward diffusion coefficient corresponding to the forward diffusion step; During the forward diffusion process, the initial new energy output sequence and new energy power output sequence The moving average window is diffused forward, where the set of moving average windows is:

[0012] in, Indicates by Moving average window The set that constitutes; Indicates the first A moving average window; Indicates the moving average window number; Indicates the number of moving average windows; Represents the set of natural numbers; express Divisible by the length of the new energy power output sequence .

[0013] Furthermore, in the process of forward diffusion of the normalized new energy output sequence based on the diffusion transfer matrix and using a moving average window, a set of moving average operators is employed. The first in Moving average convolution operator , for the Moving average window Convolution operation is performed on the new energy output sequence in the data: or

[0014]

[0015] in, Indicates the first Each moving average convolution operator corresponds to the matrix after the convolution operation is expanded. This represents the convolution unwinding operator; Indicates the number of rows. The number of columns is The space of real matrices; right Interpolation is performed along the forward diffusion direction:

[0016] in, Indicates the first The moving average transition matrix corresponding to each moving average convolution operator; Indicates the interpolation operator; The set of transition matrices corresponding to different moving average convolution operators Interpolation is performed along the forward diffusion direction to obtain the total number of forward diffusion steps. The set of diffusion transfer matrices at time:

[0017] in, Indicates by A set consisting of diffusion transition matrices; Indicates by A set consisting of moving average transition matrices.

[0018] Furthermore, by calculating the first The corresponding percentage decrease in mean standard deviation during the forward diffusion process Calculate the noise scheduling coefficient :

[0019]

[0020] in, This indicates the distribution of new energy output sequence data. The expected value of the following mathematical expression This represents the distribution of new energy output sequence data; This represents the function for calculating standard deviation. When the During the forward diffusion process, , At that time, that is Then it stops spreading forward. This is the new energy output sequence in the final forward diffusion state.

[0021] Furthermore, in the process of forward diffusion of the normalized new energy output sequence using a moving average window, for The The new energy output sequence for the forward diffusion process is as follows:

[0022] in, In the first New energy output sequence of forward diffusion process With the initial new energy output sequence Conditions, No. Conditional distribution of forward diffusion states; For the first New energy output sequence of forward diffusion process ; For the first New energy output sequence of forward diffusion process ; in:

[0023] Indicates the first Step diffusion transition matrix; Indicates the first The noise scheduling coefficient corresponding to the forward diffusion step; Indicates the first The random noise coefficient corresponding to the step-back diffusion; Indicates a Gaussian distribution; The initial new energy output sequence Conditional distribution under conditions for:

[0024] in, Indicates from step 1 to step 2. The set of forward diffusion states of a step.

[0025] Furthermore, a backdiffusion generation model is constructed based on a denoising neural network. This model is then used to generate the new energy output sequence in the final forward diffusion state. Reverse diffusion occurs; The reverse diffusion generation model for:

[0026] in, This represents the conditional probability distribution with the neural network output as the conditional input. This represents a denoising neural network. Indicates the parameters of the denoising neural network; This indicates conditional input information; Training the backdiffusion generation model:

[0027] in, This represents the model training loss function; The mathematical expectation is represented by the model training loss function, which is used to train the backdiffusion generative model until the model training loss function reaches the preset convergence condition, at which point the training of the backdiffusion generative model is stopped.

[0028] In the reverse diffusion, by calculating with the first Moving average transition matrix The corresponding optimal diffusion step is used for reverse diffusion:

[0029] in, Indicates the first Moving average transition matrix The corresponding optimal diffusion step; This represents the variable that corresponds to the minimum value of the objective function; Indicates the first One original moving average transition matrix; This represents the set of diffusion transition matrices corresponding to the total diffusion step; Initial new energy output sequence By combining the moving average transition matrix, a low-resolution new energy power output sequence is constructed. :

[0030] Low-resolution renewable energy output sequence Normalization is performed to obtain the normalized low-resolution renewable energy output sequence. ; by As the initial state for backdiffusion, a fully trained backdiffusion generation model is used with the optimal diffusion step. Perform reverse diffusion:

[0031]

[0032]

[0033] in, Indicates the first Step-back diffusion state; Indicates the first Step-back diffusion state; Represents the optimal diffusion step sequence; For the first The diffusion transition matrix corresponding to the step; Indicates the first The diffusion transition matrix corresponding to the step; Indicates the first The noise scheduling coefficient corresponding to the step; Indicates the first The noise scheduling coefficient corresponding to the step; Indicates the first The random noise coefficient corresponding to the back diffusion step. When the reverse diffusion process reaches the preset reverse diffusion termination step, a noise-free normalized new energy output sequence is obtained and output.

[0034] The present invention also provides a renewable energy output sequence generation system based on moving average diffusion, the system being capable of executing the above-described renewable energy output sequence generation method based on moving average diffusion; The system includes a data acquisition and processing module, a forward diffusion module, an adaptive noise scheduling module, a back diffusion module, and a back diffusion resolution reconstruction module. The system includes a data acquisition and processing module, a forward diffusion module, an adaptive noise scheduling module, a back diffusion module, and a back diffusion resolution reconstruction module. The data acquisition and processing module collects historical power generation data from renewable energy power plants, constructs a renewable energy power generation sequence, and normalizes the renewable energy power generation sequence to obtain a normalized renewable energy power generation sequence. The forward diffusion module performs forward diffusion on the normalized new energy output sequence based on the diffusion transfer matrix and using a moving average window. The adaptive noise scheduling module adjusts the noise scheduling coefficient in each forward diffusion step by calculating the percentage decrease in the average standard deviation in each forward diffusion step. The reverse diffusion module constructs a reverse diffusion generation model based on a denoising neural network, and trains the reverse diffusion generation model based on several new energy output sequences in the forward diffusion process. The reverse diffusion resolution reconstruction module calculates the optimal diffusion step of the reverse diffusion generation model based on the diffusion transfer matrix and the moving average transfer matrix. The reverse diffusion generation model performs reverse diffusion on the new energy output sequence based on the optimal diffusion step.

[0035] Compared with existing technologies, the beneficial effects of this solution are: The method / system of this invention implements adaptive noise diffusion for different frequency information. When the new energy power output sequence is forward diffused within a moving average window, the noise scheduling coefficient for adding noise is adaptively adjusted based on the new energy power output sequence within the moving average window. During the reverse diffusion process, the noise scheduling coefficient from the forward diffusion stage is trained / learned to perform adaptive noise reduction data processing on the new energy power output sequence. This ensures that the processed new energy power output sequence retains information in each frequency band, improving the integrity of the information and the accuracy of data processing, while reducing data errors. This overcomes the problem of rapid attenuation of low-frequency information during the diffusion of new energy power output sequences in traditional diffusion models. Attached Figure Description

[0036] Figure 1 This is a flowchart of the method in this invention.

[0037] Figure 2 This is a flowchart of the system's workflow in this invention. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0039] like Figures 1-2 As shown, the method / system of this invention implements adaptive noise diffusion for different frequency information. During forward diffusion of the new energy output sequence, the noise scheduling coefficient is dynamically adjusted based on the sequence characteristics within the current moving average window to achieve refined noise addition processing. During reverse diffusion, the adaptive noise scheduling coefficient from the forward diffusion is inherited to perform adaptive noise reduction processing on the noisy sequence. This invention can accurately reconstruct and fully retain the frequency band information of the original data. Through the adaptive noise scheduling mechanism, it significantly improves the information integrity and overall processing accuracy of the generated sequence, enhancing the practicality and reliability of the model. It can effectively suppress additional data errors introduced by model simplification or improper noise processing, thereby overcoming the key technical problem of traditional diffusion models causing rapid attenuation and difficulty in effectively retaining low-frequency information due to fixed noise scheduling when processing new energy output sequences. This method ensures that the generated sequence is statistically closer to the distribution of the original data.

[0040] The details are as follows: A method for generating renewable energy output sequences based on moving average diffusion, comprising: S1. Collect historical power output data of renewable energy power plants and construct renewable energy power output sequences.

[0041] Collect historical or real-time power output data from wind farms, photovoltaic power plants, or integrated renewable energy power plants to construct a renewable energy power output sequence dataset. The renewable energy power output sequence can be hourly, minute-level, or second-level time series data, and time scale standardization and data preprocessing can be performed according to application requirements. An initial renewable energy power output sequence is constructed from the processed renewable energy power plant power output data. , , Indicates the length of the new energy output sequence. express A dimensional real vector space.

[0042] S2. The output sequence of new energy sources is standardized using Z-Score, i.e., first centered (subtracting the mean). Then scale (divide by standard deviation) ), thus obtaining the normalized initial new energy output sequence .

[0043]

[0044] in, This represents the normalized sequence of new energy power output. This represents the function for calculating the mean. This represents the function for calculating standard deviation.

[0045] Through the above normalization process, the normalized new energy output sequence has zero mean and unit variance characteristics, thereby ensuring that the subsequent moving average diffusion process can proceed stably within a unified data distribution space.

[0046] S3, Initial New Energy Output Sequence Proceed forward diffusion until forward diffusion reaches the first... Step to get ,in This indicates the total number of forward diffusion steps (total forward diffusion steps).

[0047] In the In forward diffusion, when the moving average kernel size equals the length of the new energy output sequence, the following should be true: , Thus satisfying .in, This indicates the final forward diffusion state.

[0048] The forward diffusion process is as follows:

[0049] Indicating the initial renewable energy output sequence Under the conditions, the first The new energy output sequence is obtained by forward diffusion. ; That is, the first The forward diffusion yields the new energy output sequence. for:

[0050] in, , , Indicates the first Step diffusion transition matrix; express 3D real matrix space; , Indicates the first The noise scheduling coefficient corresponding to the forward diffusion step. Represents the set of real numbers; Represents the identity matrix; Indicates standard Gaussian noise; This indicates that the mean is 0 and the covariance matrix is... Gaussian distribution; for:

[0051] in, This represents a diagonal matrix generating function; Indicates the first The cumulative forward diffusion coefficient corresponding to the forward diffusion step; During the forward diffusion process, the initial new energy output sequence and new energy power output sequence When forward diffusion is applied using moving average windows, the set of moving average windows is:

[0052] in, Indicates by Moving average window The set that constitutes; Indicates the first A moving average window; Indicates the moving average window number; Indicates the number of moving average windows; Represents the set of natural numbers; express Divisible by the length of the new energy power output sequence .

[0053] In the process of forward diffusion of the normalized new energy output sequence based on the diffusion transfer matrix and using a moving average window, a set of moving average operators is employed. The first in Moving average convolution operator , for the Moving average window Convolution operation is performed on the new energy output sequence in the data: or

[0054]

[0055] in, Indicates the first Each moving average convolution operator corresponds to the matrix after the convolution operation is expanded. This represents the convolution unwinding operator; Represent the space of real matrices; Indicates the number of rows. The number of columns is The space of real matrices; To maintain the unchanged dimension of the renewable energy output sequence during the forward diffusion process, the following is performed: Interpolation is performed along the forward diffusion direction to obtain the corresponding moving average transition matrix:

[0056] in, Indicates the first The moving average transition matrix corresponding to each moving average convolution operator; This represents the interpolation operator, used to interpolate the expanded matrix along the time step direction. 3D matrix.

[0057] The set of transition matrices corresponding to different moving average convolution operators Interpolation along the forward diffusion direction aims to smoothly connect the discrete transition matrices defined at different time scales, thereby constructing a continuous and complete forward diffusion path. The final result of this operation is the total number of forward diffusion steps. The set of diffusion transfer matrices at time:

[0058] in, Indicates by A set of diffusion transition matrices that fully defines the entire evolution process from the initial state to the final state of forward diffusion; Indicates the first The forward diffusion transition matrix accurately characterizes the probability transition properties of the sequence state from one step to the next under a specific forward diffusion step size. Indicates by A set consisting of moving average transition matrices.

[0059] Through interpolation, the discrete transition relationships of different moving average windows are effectively integrated to form a unified set of transition matrices that is continuous in the time dimension and can cover the preset total forward diffusion steps, laying a key mathematical foundation for the subsequent noise addition and generation process.

[0060] S4. In the process of forward diffusion of the normalized new energy output sequence based on the diffusion transfer matrix and using a moving average window, forward noise is added through adaptive noise scheduling.

[0061] Traditional forward diffusion models typically employ noise scheduling strategies based on the transfer coefficient to maintain variance stability during the forward diffusion process. The noise scheduling coefficient is usually defined as:

[0062] However, in the forward diffusion process of the moving average constructed in this invention, the variance of the new energy output sequence gradually decreases due to the moving average operation, so the degree of variance reduction corresponding to different new energy output sequence datasets varies.

[0063] Therefore, the new energy output sequence dataset is first calculated in the [number]th [year]. The percentage decrease in mean standard deviation during the forward diffusion process:

[0064] in, Indicates the first The percentage decrease in the mean standard deviation corresponding to the forward diffusion step; This indicates the distribution of new energy output sequence data. The expected value of the following mathematical expression This represents the distribution of new energy output sequence data; This represents the function for calculating standard deviation.

[0065] Noise scheduling coefficients are constructed based on the rate of decrease of the average standard deviation:

[0066] In the process of forward diffusion of the normalized new energy output sequence using a moving average window, for The The new energy output sequence of the forward diffusion process must satisfy the following joint distribution family:

[0067] in, This indicates that, given an initial new energy output sequence Joint distribution under certain conditions; Indicates from step 1 to step 2. The set of forward diffusion states for the step; This represents the total number of forward diffusion steps. .

[0068] That is, for :

[0069] in, Indicates a given forward diffusion state With the initial new energy output sequence Conditions, No. Conditional distribution of forward diffusion states; Indicates the first Step diffusion transition matrix; Indicates the first Step diffusion transition matrix; Indicates the first The noise scheduling coefficient corresponding to the forward diffusion step; Indicates the first The random noise coefficient corresponding to the step-back diffusion; This represents a Gaussian distribution.

[0070] The mean term in the above conditional distribution is used to ensure that the joint distribution and the corresponding marginal distribution remain consistent at each forward diffusion step (forward diffusion process), that is... This ensures that For any forward diffusion step Both are valid.

[0071] S5. Constructing a backdiffusion generation model based on a denoising neural network. :

[0072] in, This represents the conditional probability distribution with the neural network output as the conditional input. This represents a denoising neural network. Indicates the parameters of the denoising neural network; This represents conditional input information; for conditional generation tasks, the conditional input... As input to a denoising neural network, it is used to achieve condition generation tasks such as predicting the output sequence of new energy sources.

[0073] The reverse diffusion generation model is trained based on the new energy output sequences corresponding to different diffusion steps in the forward diffusion process:

[0074] in, This represents the model training loss function; The mathematical expectation is represented by the backdiffusion generation model, which is trained using the model training loss function. When the training loss function meets the preset convergence condition, the backdiffusion generation model completes training.

[0075] S6. Calculate the optimal backdiffusion step of the backdiffusion generation model based on the moving average transition matrix, and perform backdiffusion based on the optimal backdiffusion step.

[0076] Due to the diffusion transfer matrix set The set of moving average transition matrices Interpolation is obtained, therefore, the back diffusion process is accelerated by finding the back diffusion step (optimal back diffusion step) that is closest to the original moving average transition matrix.

[0077] For the Moving average transition matrix The corresponding backdiffusion step search method is as follows:

[0078] in, Indicates the first Moving average transition matrix The corresponding optimal reverse diffusion step; This represents the variable that corresponds to the minimum value of the objective function; This represents the set of corresponding diffusion transfer matrices.

[0079] Construct an optimal back-diffusion step set based on several optimal back-diffusion steps. As an accelerated backdiffusion sampling step, since the selected backdiffusion step is only related to the factor of the new energy output sequence length, this strategy is defined as a factor-based backdiffusion sampling strategy.

[0080] when In this case, the reverse diffusion process can be regarded as a numerical solution process of ordinary differential equations discretized by Euler.

[0081] Furthermore, when the interpolation operator is in When performing uniform interpolation, the backdiffusion sampling process is equivalent to numerically solving the above ordinary differential equation using a larger step size.

[0082] Let the moving average window corresponding to the low-resolution new energy output sequence be defined. satisfy At that time, the corresponding low-resolution new energy output sequence is constructed using the moving average transition matrix:

[0083] in, This represents the low-resolution renewable energy output sequence. Through the above normalization process, the normalized low-resolution renewable energy output sequence is obtained. .

[0084] According to the formula Determine the corresponding optimal reverse diffusion step Subsequently, the order As the initial state for backdiffusion, a fully trained backdiffusion generation model is used, with the optimal backdiffusion step... Proceed with a gradual reverse diffusion:

[0085] in, Indicates the first Step-back diffusion state; Indicates the first Step-back diffusion state; Represents the optimal diffusion step sequence; Indicates the first The moving average transition matrix corresponding to each step; Indicates the first The moving average transition matrix corresponding to each step; Indicates the first The noise scheduling coefficient corresponding to the step; Indicates the first The noise scheduling coefficient corresponding to the step; Indicates the first The random noise coefficient corresponding to the back diffusion step. The diffusion transfer matrix and noise scheduling coefficients in the reverse diffusion process are inherited from the transfer matrix and noise scheduling coefficients of the corresponding diffusion step in the forward diffusion process. , .

[0086] When the back diffusion process reaches the preset back diffusion termination step, a noise-free normalized new energy power output sequence is obtained, enabling super-resolution reconstruction of the collected historical new energy power station power output data or real-time new energy power station power output data.

[0087] The present invention also provides a renewable energy output sequence generation system based on moving average diffusion. The data acquisition and processing module collects historical renewable energy power plant output data, constructs a renewable energy output sequence, and normalizes the renewable energy output sequence to make the renewable energy output sequence have zero mean and unit variance characteristics, thus obtaining the normalized renewable energy output sequence. The forward diffusion module is based on the diffusion transfer matrix. It performs forward diffusion on the normalized new energy output sequence using a moving average window. It constructs a set of diffusion transfer matrices by interpolating the moving average convolution kernel and the transfer matrix, thereby realizing the gradual diffusion of the new energy output sequence from a high-resolution state to a low-resolution state. The adaptive noise scheduling module adjusts the noise scheduling coefficient in each forward diffusion step by calculating the percentage decrease in the average standard deviation in each forward diffusion step, in order to maintain variance consistency during the diffusion process. The back diffusion module constructs a back diffusion generation model based on a denoising neural network, and trains the back diffusion generation model based on the new energy output sequence corresponding to different diffusion steps in the forward diffusion process. The reverse diffusion resolution reconstruction module calculates the optimal diffusion step of the reverse diffusion generation model based on the diffusion transfer matrix and the moving average transfer matrix. Based on the optimal diffusion step, the reverse diffusion generation model performs reverse diffusion on the new energy output sequence to achieve resolution reconstruction of the new energy output sequence.

[0088] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for generating renewable energy output sequences based on moving average diffusion, characterized in that, The method includes the following steps: S1. Collect historical power output data from renewable energy power plants and construct renewable energy power output sequences; S2. The new energy output sequence is normalized to obtain the normalized new energy output sequence. S3. The normalized new energy output sequence is based on the diffusion transfer matrix and forward diffusion is performed using a moving average window. S4. In the process of forward diffusion of the normalized new energy output sequence based on the diffusion transfer matrix and using a moving average window, forward noise is added through adaptive noise scheduling. S5. Construct a reverse diffusion generation model and train the reverse diffusion generation model based on several new energy output sequences in the forward diffusion process. S6. Calculate the optimal diffusion step of the reverse diffusion generation model based on the diffusion transfer matrix and the moving average transfer matrix. The reverse diffusion generation model performs reverse diffusion on the new energy output sequence based on the optimal diffusion step.

2. The method for generating renewable energy output sequences based on moving average diffusion according to claim 1, characterized in that, In step S1, power output data from renewable energy power plants is collected, and an initial renewable energy power output sequence is constructed based on the power output data. .

3. The method for generating renewable energy output sequences based on moving average diffusion according to claim 2, characterized in that, Initial new energy power output sequence Normalization is performed by applying Z-score standardization to the renewable energy output sequence to obtain the normalized initial renewable energy output sequence. .

4. The method for generating renewable energy output sequences based on moving average diffusion according to claim 3, characterized in that, Initial new energy power output sequence Proceed forward diffusion until forward diffusion reaches the first... Step to get ,in This indicates the total number of forward diffusion steps; The forward diffusion process is as follows: Represents the initial new energy output sequence In the The forward diffusion process yields the new energy output sequence. ; No. The forward diffusion yields the new energy output sequence. for: in, Indicates the first Step diffusion transition matrix; Indicates the first The noise scheduling coefficient corresponding to the forward diffusion step; Represents the identity matrix; Indicates standard Gaussian noise; This indicates that the mean is 0 and the covariance matrix is... Gaussian distribution; for: in, This represents a diagonal matrix generating function; Indicates the first The cumulative forward diffusion coefficient corresponding to the forward diffusion step; During the forward diffusion process, the initial new energy output sequence and new energy power output sequence The moving average window is diffused forward, where the set of moving average windows is: in, Indicates by Moving average window The set that constitutes; Indicates the first A moving average window; Indicates the moving average window number; Indicates the number of moving average windows; Represents the set of natural numbers; express Divisible by the length of the new energy power output sequence .

5. The method for generating renewable energy output sequences based on moving average diffusion according to claim 4, characterized in that, In the process of forward diffusion of the normalized new energy output sequence based on the diffusion transfer matrix and using a moving average window, a set of moving average operators is employed. The first in Moving average convolution operator , for the Moving average window Convolution operation is performed on the new energy output sequence in the data: or in, Indicates the first Each moving average convolution operator corresponds to the matrix after the convolution operation is expanded. This represents the convolution unwinding operator; Indicates the number of rows. The number of columns is The space of real matrices; right Interpolation is performed along the forward diffusion direction: in, Indicates the first The moving average transition matrix corresponding to each moving average convolution operator; Indicates the interpolation operator; The set of transition matrices corresponding to different moving average convolution operators Interpolation is performed along the forward diffusion direction to obtain the total number of forward diffusion steps. The set of diffusion transfer matrices at time: in, Indicates by A set consisting of diffusion transition matrices; Indicates by A set consisting of moving average transition matrices.

6. The method for generating renewable energy output sequences based on moving average diffusion according to claim 5, characterized in that, By calculating the first The corresponding percentage decrease in mean standard deviation during the forward diffusion process Calculate the noise scheduling coefficient : in, This indicates the distribution of new energy output sequence data. The expected value of the following mathematical expression This represents the distribution of new energy output sequence data; This represents the function for calculating standard deviation. When the During the forward diffusion process, , At that time, that is Then it stops spreading forward. This is the new energy output sequence in the final forward diffusion state.

7. The method for generating renewable energy output sequences based on moving average diffusion according to claim 6, characterized in that, In the process of forward diffusion of the normalized new energy output sequence using a moving average window, for The The new energy output sequence for the forward diffusion process is as follows: in, In the first New energy output sequence of forward diffusion process With the initial new energy output sequence Conditions, No. Conditional distribution of forward diffusion states; For the first New energy output sequence of forward diffusion process ; For the first New energy output sequence of forward diffusion process ; in: Indicates the first Step diffusion transition matrix; Indicates the first The noise scheduling coefficient corresponding to the forward diffusion step; Indicates the first The random noise coefficient corresponding to the step-back diffusion; Indicates a Gaussian distribution; The initial new energy output sequence Conditional distribution under conditions for: in, Indicates from step 1 to step 2. The set of forward diffusion states of a step.

8. The method for generating renewable energy output sequences based on moving average diffusion according to claim 6, characterized in that, A backdiffusion generation model is constructed based on a denoising neural network. This model is then used to analyze the new energy output sequence in the final forward diffusion state. Reverse diffusion occurs; The reverse diffusion generation model for: in, This represents the conditional probability distribution with the neural network output as the conditional input. This represents a denoising neural network. Indicates the parameters of the denoising neural network; This indicates conditional input information; Training the backdiffusion generation model: in, This represents the model training loss function; The mathematical expectation is represented by the model training loss function, which is used to train the backdiffusion generative model until the model training loss function reaches the preset convergence condition, at which point the training of the backdiffusion generative model is stopped.

9. A method for generating renewable energy output sequences based on moving average diffusion according to claim 8, characterized in that, In the reverse diffusion, by calculating with the first Moving average transition matrix The corresponding optimal diffusion step is used for reverse diffusion: in, Indicates the first Moving average transition matrix The corresponding optimal diffusion step; This represents the variable that corresponds to the minimum value of the objective function; Indicates the first One original moving average transition matrix; This represents the set of diffusion transition matrices corresponding to the total diffusion step; Initial new energy output sequence By combining the moving average transition matrix, a low-resolution new energy power output sequence is constructed. : Low-resolution renewable energy output sequence Normalization is performed to obtain the normalized low-resolution renewable energy output sequence. ; by As the initial state for backdiffusion, a fully trained backdiffusion generation model is used with the optimal diffusion step. Perform reverse diffusion: in, Indicates the first Step-back diffusion state; Indicates the first Step-back diffusion state; Represents the optimal diffusion step sequence; For the first The diffusion transition matrix corresponding to the step; Indicates the first The diffusion transition matrix corresponding to the step; Indicates the first The noise scheduling coefficient corresponding to the step; Indicates the first The noise scheduling coefficient corresponding to the step; Indicates the first The random noise coefficient corresponding to the back diffusion step. When the reverse diffusion process reaches the preset reverse diffusion termination step, a noise-free normalized new energy output sequence is obtained and output.

10. A renewable energy output sequence generation system based on moving average diffusion according to claim 1, characterized in that, The system can execute a method for generating renewable energy output sequences based on moving average diffusion as described in any one of claims 1-9; The system includes a data acquisition and processing module, a forward diffusion module, an adaptive noise scheduling module, a back diffusion module, and a back diffusion resolution reconstruction module. The data acquisition and processing module collects historical power generation data from renewable energy power plants, constructs a renewable energy power generation sequence, and normalizes the renewable energy power generation sequence to obtain a normalized renewable energy power generation sequence. The forward diffusion module performs forward diffusion on the normalized new energy output sequence based on the diffusion transfer matrix and using a moving average window. The adaptive noise scheduling module adjusts the noise scheduling coefficient in each forward diffusion step by calculating the percentage decrease in the average standard deviation in each forward diffusion step. The reverse diffusion module constructs a reverse diffusion generation model based on a denoising neural network, and trains the reverse diffusion generation model based on several new energy output sequences in the forward diffusion process. The reverse diffusion resolution reconstruction module calculates the optimal diffusion step of the reverse diffusion generation model based on the diffusion transfer matrix and the moving average transfer matrix. The reverse diffusion generation model performs reverse diffusion on the new energy output sequence based on the optimal diffusion step.

Images