A multi-region wind power short-term scenario generation method and system

Abstract

The application discloses a kind of multi-region wind power short-term scene generation method and system, comprising: collecting the meteorological data and historical wind power output data of multi-region wind farm, and carry out cleaning and standardization processing;Using pre-trained encoding network to carry out feature embedding to meteorological feature, to evaluate the dynamic adjacency matrix between regions, capture spatial correlation;Based on the dynamic adjacency matrix calculated by Spearman correlation coefficient, reflect the dynamic correlation of wind power output of different regions;Utilize variational autoencoder to map high-dimensional wind power output curve to low-dimensional hidden space feature vector, and reconstruct high-dimensional output sequence by decoder;Condition hidden space diffusion model is constructed, hidden space sample is gradually added to normal distribution, and denoising network is designed to gradually remove noise;The high-dimensional wind power output scene is generated by reconstructing the low-dimensional hidden space scene output by denoising network.The application improves the accuracy of generated scene, and provides effective support for economic dispatching and risk assessment of power system.

Description

Technical Field

[0001] This invention relates to the fields of renewable energy and smart grid technology, and in particular to a method and system for generating short-term wind power scenarios in multiple regions based on a conditional latent space diffusion model and dynamic graph representation. Background Technology

[0002] With the intensification of global climate change and the rise in demand for renewable energy, wind power has become an important component in addressing the energy transition. However, wind power output is affected by the chaos of the atmospheric system, exhibiting significant randomness, volatility, and intermittency. As wind farms are rapidly connected to the grid, the uncertainty of short-term wind power output poses significant challenges to power system balance analysis, economic dispatch, and safe and stable operation. Furthermore, influenced by atmospheric motion, the uncertainty of short-term wind power output across multiple regions exhibits spatial correlations, and mismatches between output predictions can easily lead to absorption difficulties, hindering the overall improvement of wind power utilization. To address the uncertainty of wind power output, scenario-based stochastic programming methods have been widely applied in problems such as grid economic dispatch. Higher scenario generation accuracy results in solutions to the corresponding stochastic optimization problems that are closer to the actual optimal value. Therefore, accurately constructing multi-regional short-term wind power output scenarios is crucial for solving day-ahead and intraday rolling economic dispatch problems under large-scale wind power integration, improving wind power absorption levels, and ensuring the safe and stable operation of the power system.

[0003] Most current research on scene generation employs decoupled modeling methods, which offer good interpretability, but the fitting process strictly relies on regression of the assumed function parameters. Limited by the function's form, it struggles to comprehensively characterize the complex temporal, spatial, and meteorological correlations of wind power output, often resulting in significant deviations in the generated scene results. Furthermore, to model the spatial correlation of wind power output across multiple regions, existing research typically models the static spatial correlation of wind power output, achieving good results under long-term stable conditions. However, the spatial correlation of wind power output is not only related to the distance between sites but also closely linked to real-time environmental conditions such as wind speed, direction, and temperature. Short-term fluctuations in wind power mean that its spatial correlation is dynamically changing, making it difficult to accurately reflect the correlation of short-term changes in wind power output across multiple regions under weather influences using a constant spatial correlation.

[0004] Definitions of abbreviations and key terms:

[0005] Dynamic Graph Neural Network (DGNN): A dynamic graph neural network is an extension of the traditional graph neural network (GNN) for processing graph data that changes over time. DGNNs can capture dynamic changes in the graph structure, such as the addition or deletion of nodes or changes in edges.

[0006] Representation Learning for Dynamic Graphs: Dynamic graph representation refers to the generation of effective representations for dynamic graphs through learning algorithms. This representation captures the graph's structure, node features, and dynamic characteristics that change over time.

[0007] Numerical Weather Prediction (NWP) data is generated using numerical weather prediction models to forecast future weather conditions. NWP utilizes mathematical models and computer simulations of atmospheric physical processes, analyzing current meteorological observation data to generate weather forecasts for future time periods.

[0008] Generative Adversarial Networks (GANs) consist of two main parts: a generator and a discriminator. The generator is responsible for generating data samples from random noise, while the discriminator attempts to distinguish these generated samples from real samples. These two models are trained simultaneously through a "zero-sum game" process.

[0009] Variational Auto-Encoder (VAE): A VAE consists of an encoder and a decoder. The encoder is responsible for transforming the input data into a probability distribution (usually a Gaussian distribution), while the decoder samples from this distribution and attempts to reconstruct the original input.

[0010] Diffusion Model (DM): The diffusion model is an emerging generative model that solves the posterior representation problem through an explicit diffusion process and has shown excellent performance in various tasks. Summary of the Invention

[0011] This invention addresses the shortcomings of existing technologies by providing a method and system for generating short-term wind power scenarios in multiple regions.

[0012] To achieve the above-mentioned objectives, the technical solution adopted by the present invention is as follows:

[0013] A method for generating short-term wind power scenarios in multiple regions includes the following steps:

[0014] S1: Collect meteorological data and historical wind power output data from wind farms in multiple regions. Clean and standardize the collected data to remove noise and outliers.

[0015] S2: Feature embedding of the meteorological feature NWP for each region is performed using a pre-trained encoding network. The embedding results are used to evaluate the dynamic adjacency matrix between regions to capture spatial correlations between regions.

[0016] S3: Based on the Spearman correlation coefficient of wind power output in each region, the dynamic adjacency matrix is ​​calculated in real time to reflect the dynamic correlation between wind power output in different regions.

[0017] S4: Using a variational autoencoder (VAE), the high-dimensional wind power output curve is mapped to a low-dimensional latent space feature vector. The trend features of wind power output are extracted by the encoder E, and the latent space features are reconstructed into a high-dimensional output sequence using the decoder D.

[0018] S5: Construct a conditional latent space diffusion model, including a diffusion process and a denoising process. Define the diffusion process by gradually adding noise to the latent space samples until they are normally distributed. Design a denoising network that uses conditional information as a covariate to gradually remove noise and recover the latent space samples.

[0019] S6: A denoising network is trained using residual connections and spatiotemporal convolutional layers to improve its ability to learn spatiotemporally relevant features. The training objective is to minimize the KL divergence between the posterior and prior distributions to ensure the effectiveness of the denoising process.

[0020] S7: The low-dimensional latent space scene output by the denoising network is reconstructed to generate a high-dimensional wind power output scene. The generated scene is then used for applications in the economic dispatch and risk assessment of the power system.

[0021] Furthermore, for each region i, the input meteorological data is represented as follows: T represents the time step, and C represents the feature dimension.

[0022] Using a pre-trained encoding network f e For x i Embedding is performed to generate a high-dimensional representation of the features.

[0023] The encoding network is a convolutional neural network structure, which includes: an input layer, intermediate layers, and an output layer;

[0024] Input layer: Adjust the number of channels to accommodate subsequent convolutional layers;

[0025] Intermediate layers: Multiple convolutional layers are used, and the receptive field is increased through residual connection structures to prevent network degradation;

[0026] Output layer: Skip connections are used to combine the results of each convolutional layer, achieving the fusion of features at multiple time scales, and finally obtaining the embedding result.

[0027] Furthermore, the encoding network is trained using L1 loss to perform deterministic prediction regression:

[0028]

[0029] Among them, y i For the actual wind power output of region i, This is the embedded result.

[0030] Furthermore, embedding is performed on all regions to obtain a set of feature embeddings: Based on this, a dynamic adjacency matrix is ​​constructed.

[0031] Furthermore, the sub-steps of S4 are as follows:

[0032] S41: High-dimensional wind power output sequence Input to encoder E.

[0033] Encoder E generates the mean vector and standard deviation vector Thus, the probability distribution p in the latent space is defined. θ (z|y).

[0034] S42: Sample latent space variables from the above probability distribution

[0035] S43: Input the latent space variable z into the decoder D to generate the reconstructed wind power output sequence.

[0036] S44: Calculate the VAE loss function, including KL divergence and reconstruction error:

[0037]

[0038] After introducing the weight coefficient λ, the updated loss function is:

[0039]

[0040] In the formula: ε(·) and The mapping functions for the encoder and decoder are respectively; q φ (z|y) represents the posterior distribution of the encoder output, and represents the distribution of the latent variable z given the input y. θ (z|y) represents the reconstruction distribution generated by the decoder, which represents the probability of data recovered from the latent variable z.

[0041] S45: Map the wind power sequence of each region using an encoder to obtain the latent space set Z0.

[0042] S46: The latent space features are used in the downstream tasks of the diffusion model to reconstruct the low-dimensional scene output by the denoising network into a high-dimensional wind power output scene, ensuring the accurate characterization of detailed fluctuation features.

[0043] Furthermore, the diffusion process is as follows:

[0044] The diffusion process gradually adds noise to the latent space sample Z0 until it reaches Z. N This will gradually lead to a transformation towards a standard normal distribution, as follows:

[0045]

[0046] In the formula: β∈(0,1) represents the scaling factor at the nth step; ∈ represents noise samples, β n denoted by the scaling factor at step n, which controls the degree of noise addition, and ∈ represents the random noise added during the diffusion process.

[0047] The Markov chain process is represented as:

[0048]

[0049] As n increases, the latent space samples tend to follow a standard normal distribution.

[0050] Furthermore, the denoising process can be represented as:

[0051]

[0052] The goal of denoising is to minimize the KL divergence between the posterior and prior distributions, which is expressed as:

[0053]

[0054] In the formula, l ELBO This indicates the lower bound of evidence in variational inference. Let ||Z0| be the expectation of all possible Z0 and noise ∈ sampled from the standard normal distribution, ||Z0| - ∈ Z0|. θ (Z n ,n,C)||2 represents the reconstruction error.

[0055] Furthermore, the denoising network is designed to enhance dynamic spatial relevance by combining residual connections, skip connections, and graph convolutions.

[0056] The graph convolutional layer is represented as:

[0057]

[0058] In the formula, H l-1 and H l W represents the input and output of the l-th layer, respectively. dThe parameters of the adaptive matrix, for The degree matrix; W and b are the network learnable parameters of GCN; ⊙ represents the Hadamard product; σ is the activation function.

[0059] This invention also discloses a multi-regional wind power short-term scenario generation system, which can be used to implement the above-mentioned multi-regional wind power short-term scenario generation method, specifically including:

[0060] Data collection module: Collects meteorological data and historical wind power output data from wind farms in multiple regions, and performs data cleaning and standardization to eliminate noise and outliers;

[0061] Feature embedding module: Uses a pre-trained encoding network to embed meteorological features (NWP) for each region, and uses the embedding results to evaluate the dynamic adjacency matrix between regions in order to capture the spatial correlation between regions;

[0062] Dynamic adjacency matrix calculation module: Based on the Spearman correlation coefficient of wind power output in each region, the dynamic adjacency matrix is ​​calculated in real time to reflect the dynamic correlation between wind power output in different regions.

[0063] Latent space mapping module: The high-dimensional wind power output curve is mapped to a low-dimensional latent space feature vector using a variational autoencoder (VAE). The encoder extracts the trend features of wind power output and the decoder reconstructs the high-dimensional output sequence.

[0064] Conditional Latent Space Diffusion Model Module: Constructs a conditional latent space diffusion model that includes a diffusion process and a denoising process. The diffusion process is defined to gradually add noise to the latent space samples until they are normally distributed. A denoising network is designed to gradually remove noise and restore the latent space samples by using conditional information as a covariate.

[0065] Denoising network training module: The denoising network is trained using a structure of residual connections and spatiotemporal convolutional layers to improve its ability to learn spatiotemporally related features. The training objective is to minimize the KL divergence between the posterior and prior distributions.

[0066] Scene generation module: The low-dimensional latent space scene output by the denoising network is reconstructed to generate a high-dimensional wind power output scene, and the generated scene is finally used for the economic dispatch and risk assessment of the power system.

[0067] The present invention also discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-mentioned method for generating short-term wind power scenarios in multiple regions.

[0068] The present invention also discloses a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method for generating short-term multi-regional wind power scenarios.

[0069] Compared with the prior art, the advantages of the present invention are as follows:

[0070] 1. Efficient Modeling: By utilizing a conditional latent space diffusion model, this invention significantly reduces the computational complexity of high-dimensional wind power output scenarios. This model solves the posterior representation problem through an explicit diffusion process, ensuring that the generated scenario not only conforms to the trend of wind power output patterns but also retains realistic detailed fluctuation characteristics.

[0071] 2. Dynamic Graph Representation: A dynamic graph neural network is introduced, utilizing a dynamic adjacency matrix to capture the real-time spatial correlation between wind power outputs in multiple regions. This method improves the accuracy of the denoising network in representing the dynamic correlations of wind power in multiple regions, making the generated scenario more consistent with actual wind power output conditions.

[0072] 3. Optimize feature extraction: By using supervised feature embedding, the model is guided to effectively extract key features, avoiding feature filtering in each step of the denoising network, thus focusing on the mapping modeling task and significantly improving the performance of the denoising model.

[0073] 4. Detail reconstruction capability: The variational autoencoder (VAE) is used to map the high-dimensional modeling task to the low-dimensional latent space, which enhances the diffusion model's ability to characterize the detailed fluctuation features of wind power output scenarios, and further improves the model's learning ability and generation effect. Attached Figure Description

[0074] Figure 1 This is an architecture diagram of the multi-regional wind power short-term scenario generation method according to an embodiment of the present invention;

[0075] Figure 2 This is a flowchart of feature embedding and dynamic graph representation in an embodiment of the present invention;

[0076] Figure 3 This is a flowchart of the VAE latent space embedding process according to an embodiment of the present invention;

[0077] Figure 4 This is a flowchart of the diffusion and denoising process of the conditional latent space diffusion model in an embodiment of the present invention;

[0078] Figure 5 This is a diagram of the denoising network structure in an embodiment of the present invention. Detailed Implementation

[0079] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and examples.

[0080] This invention provides a method for generating short-term wind power scenarios in multiple regions, implemented through conditional latent space diffusion model technology and dynamic graph representation methods, such as... Figure 1 As shown, it mainly includes three parts: feature embedding and dynamic graph representation, latent space mapping, and conditional latent space diffusion model.

[0081] 1) Feature embedding and dynamic graph representation.

[0082] Using a pre-trained encoding network f e The NWP input features for each region are embedded, and the results are based on deterministic prediction. Evaluate the real-time dynamic adjacency matrix Will As a conditional embedding of the diffusion model, to utilize the denoising network f D This study explores the dynamic spatial relationships of wind power in multiple regions and eliminates the need for filtering high-dimensional NWP features in each denoising step of the diffusion model, thus reducing the computational burden. Multiple independent embedding networks are used to perform deterministic prediction and regression of high-dimensional NWP features of wind power in each region. The Spearman correlation coefficients among the deterministic prediction results of wind power in multiple regions are used as corresponding elements of the dynamic adjacency matrix, thereby achieving real-time measurement of the dynamic correlation between different regions. A schematic diagram of feature embedding and dynamic graph representation is shown below. Figure 2 As shown.

[0083] When inputting data to the NWP of the i-th region During embedding, the input layer convolutional neural network is used to adjust x. i The number of channels is determined to obtain a high-dimensional representation of the features. The intermediate layers utilize a structure of residual connections between multiple convolutional layers to increase the receptive field to the input data, preventing network degradation. The output layer then performs skip connections between the results of each convolutional layer to achieve feature fusion across multiple time scales. After these steps, x is finally obtained. i Feature embedding results In this embodiment, L1 loss is used to train the embedded network for deterministic prediction regression, as shown in the following equation:

[0084]

[0085] In the formula: y i For the actual wind power output of region i, The embedding results are used for deterministic predictions of wind power in region i.

[0086] Repeating the above steps will yield the set of feature embeddings for all regions. Furthermore, based on By calculating the Spearman correlation coefficients between regions, a dynamic adjacency matrix characterizing the real-time spatial correlation of wind power can be obtained. This method enables the diffusion model to explicitly guide the model to extract key NWP features in a supervised manner, decomposing the tasks of model feature extraction and scene mapping. This avoids feature filtering in each step of the denoising network, allowing the denoising network to focus on mapping and modeling the uncertainty of the prediction results.

[0087] 2) Latent space mapping.

[0088] The pre-trained VAE serves as the upstream module of the diffusion model. The encoder ε encodes and maps the high-dimensional wind power output curve Y of each region to obtain a low-dimensional latent space feature vector Z0. The decoder D then reconstructs the high-resolution scene result S from Z0 through decoding, thereby improving the learning ability of the diffusion model. The VAE model is used to perform latent space mapping and reconstruction of the wind power output sequence. The task of learning the wind power output trend is assigned to the diffusion model, while the task of learning detailed fluctuations is assigned to the VAE decoder. A schematic diagram of the VAE latent space embedding is shown below. Figure 3 As shown.

[0089] The encoder ε and decoder D of the VAE model consist of two independent network structures, where the encoder ε takes the wind power output sequence as input. Output mean vector and standard deviation vector Then from the probability distribution p θ The latent space variables can be obtained by sampling in (z|y). The decoder then performs a decoding mapping on z to obtain the reconstructed wind power output sequence. The training objective of VAE is the posterior distribution p θ (z|y) and prior distribution q φ The KL divergence between (z|y) is minimized. After transforming the problem into optimization of the lower bound of the maximum likelihood function using variational inference, the VAE loss function can be expressed as:

[0090]

[0091] In the formula: the first term is the posterior probability distribution p of the latent variables fitted by the encoding network. θ The first term is the KL (Kullback-Leibler) divergence between (z|y) and the prior standard normal distribution p(z); the second term is the reconstruction error loss, which describes the difference between the generated scene and the historical scene.

[0092] KL divergence focuses on assessing the overall trend of wind power output, but lacks characterization of specific fluctuation information, resulting in overly smooth time series that cannot reconstruct detailed information about wind power fluctuations. Therefore, this embodiment introduces a weighting coefficient λ to balance the reconstruction error loss and the KL divergence weight. The improved loss function is shown below:

[0093]

[0094] In the formula: ε(·) and λ represents the mapping functions for the encoder and decoder, respectively; λ is the weight coefficient.

[0095] By mapping the wind power sequences of each region into the latent space using the encoder ε, a latent space set Z0 can be obtained. This transforms the high-dimensional mapping modeling of the diffusion model into a low-dimensional latent space, which helps improve the learning ability of the diffusion model. The decoder... As a downstream model of the diffusion model, it reconstructs the low-dimensional scene output by the denoising network into a high-dimensional time series scene, which can accurately depict the detailed fluctuation characteristics of the wind power output scene.

[0096] 3) Conditional latent space diffusion model.

[0097] It includes a diffusion process and a denoising process. The diffusion process progressively adds noise to the latent space feature vector Z0 of each region until it reaches Z. N The denoising process uses C as a covariate and leverages a denoising network to fuse spatiotemporal prior knowledge to achieve the desired result for Z. N Gradual denoising for the decoder The restored scenario provides a reliable latent space scenario. A conditional latent space diffusion model is used to model the joint probability distribution of wind power conditions across multiple regions. The conditional latent space diffusion model consists of two inverse Markov chain processes: diffusion and denoising, as follows... Figure 4 As shown.

[0098] The diffusion process gradually adds noise to the latent space sample Z0 over N steps, eventually causing the sample to tend towards a normal distribution. Suppose that the time series sample after the nth diffusion step is represented as Z. n Then from Z n-1 To Z n The diffusion process can be represented as:

[0099]

[0100] In the formula: β∈(0,1) represents the scaling factor at step n; ∈ represents noise samples.

[0101] Z n Z0 can be represented by an n-order Markov chain process:

[0102]

[0103] make Then Z n The relationship with Z0 can be expressed as:

[0104]

[0105] From the above formula, it can be seen that as n gradually increases, Approaching 0, eventually Z n It will converge to a prior distribution, which is usually chosen as the standard normal distribution. As a prior distribution.

[0106] The denoising process is the inverse of the diffusion process, and it affects the prior distribution q(Z). N The process involves progressive denoising to ultimately restore the latent space sample distribution q(Z0|C). A neural network is used to analyze the conditional distribution p of the denoising process. θ (Z n-1 |Z n C) Fitting is performed, and the denoising process can be represented as an nth-order Markov chain process:

[0107]

[0108] The training objective of the denoising process is to minimize the posterior distribution p. θ (Z n-1 |Z n ,C) and prior distribution q(Z) n-1 |Z n The KL divergence of C). However, this prior distribution is usually difficult to obtain directly, so it can be transformed by Bayes' theorem, converting the denoising process into fitting the noisy samples, which can be expressed as:

[0109]

[0110] In the formula: ∈ θ =f D (Z n (n, C) represents the output of the denoising network, which determines how much noise needs to be removed in the current denoising step; f D (·) is the mapping function of the denoising network.

[0111] The denoising network takes denoised samples, denoising steps n, feature embeddings, and a dynamic adjacency matrix A as input, and outputs the noise to be removed in that step. The loss function for training the denoising network can be expressed as:

[0112]

[0113] p θ (Z n-1 |Z n C)

[0114] To achieve the posterior distribution p θ (Z n-1 |Z n C) Effective Fit: This embodiment designs a denoising network structure that considers the dynamic spatial correlation of wind power in multiple regions, such as... Figure 5 As shown, the denoising network employs a network structure with residual and skip connections. Specifically, the feature encoding layer and the Linear layer map the discrete denoising step n to a continuous vector space, and the Conv convolutional layer implements the mapping of Z... n and Feature extraction and fusion are performed by combining Temporal Convolutional Network (TCN) layers with DGNN layers to form a spatiotemporal convolutional layer, which learns the spatiotemporal correlation features of the input data, and finally obtains the nth step for sample Z. n The denoising result ∈ θ .

[0115] Furthermore, to improve the learning performance of the DGNN network, this embodiment employs a learnable adaptive matrix W. d For dynamic adjacency matrix The adaptive adjustment process can be represented as follows:

[0116]

[0117] Where: H l-1 and H l These are the input and output of the l-th graph convolutional layer, respectively; W d The parameters of the adaptive matrix; for The degree matrix; W and b are the network learnable parameters of GCN; ⊙ represents the Hadamard product; σ is the activation function.

[0118] In another embodiment of the present invention, a multi-regional wind power short-term scenario generation system is provided. This system can be used to implement the above-described multi-regional wind power short-term scenario generation method, specifically including: modules

[0119] Data collection module: Collects meteorological data and historical wind power output data from wind farms in multiple regions, and performs data cleaning and standardization to eliminate noise and outliers;

[0120] Feature embedding module: Uses a pre-trained encoding network to embed meteorological features (NWP) for each region, and uses the embedding results to evaluate the dynamic adjacency matrix between regions in order to capture the spatial correlation between regions;

[0121] Dynamic adjacency matrix calculation module: Based on the Spearman correlation coefficient of wind power output in each region, the dynamic adjacency matrix is ​​calculated in real time to reflect the dynamic correlation between wind power output in different regions.

[0122] Latent space mapping module: The high-dimensional wind power output curve is mapped to a low-dimensional latent space feature vector using a variational autoencoder (VAE). The encoder extracts the trend features of wind power output and the decoder reconstructs the high-dimensional output sequence.

[0123] Conditional Latent Space Diffusion Model Module: Constructs a conditional latent space diffusion model that includes a diffusion process and a denoising process. The diffusion process is defined to gradually add noise to the latent space samples until they are normally distributed. A denoising network is designed to gradually remove noise and restore the latent space samples by using conditional information as a covariate.

[0124] Denoising network training module: The denoising network is trained using a structure of residual connections and spatiotemporal convolutional layers to improve its ability to learn spatiotemporally related features. The training objective is to minimize the KL divergence between the posterior and prior distributions.

[0125] Scene generation module: The low-dimensional latent space scene output by the denoising network is reconstructed to generate a high-dimensional wind power output scene, and the generated scene is finally used for the economic dispatch and risk assessment of the power system.

[0126] In another embodiment of the present invention, a terminal device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used in the operation of a multi-regional wind power short-term scenario generation method.

[0127] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory). This computer-readable storage medium is a memory device in a terminal device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and extended storage media supported by the terminal device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device.

[0128] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the method for generating short-term wind power scenarios in a multi-regional area as described in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by a processor.

[0129] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0130] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0131] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0132] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0133] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the implementation methods of the present invention, and should be understood that the scope of protection of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of the present invention.

Claims

1. A method for generating short-term wind power scenarios in multiple regions, characterized in that: Includes the following steps: S1: Collect meteorological data and historical wind power output data from wind farms in multiple regions; clean and standardize the collected data to eliminate noise and outliers; S2: Use a pre-trained encoding network to embed the meteorological feature NWP for each region; use the embedding results to evaluate the dynamic adjacency matrix between regions to capture the spatial correlation between regions. S3: Based on the Spearman correlation coefficient of wind power output in each region, a dynamic adjacency matrix is ​​calculated in real time to reflect the dynamic correlation between wind power output in different regions. S4: Using a variational autoencoder (VAE), the high-dimensional wind power output curve is mapped to a low-dimensional latent space feature vector; the trend features of wind power output are extracted through encoder E, and decoder D is used to reconstruct the latent space features into a high-dimensional output sequence. S5: Construct a conditional latent space diffusion model, including a diffusion process and a denoising process. Define the diffusion process to gradually add noise to the latent space samples until they are normally distributed. Design a denoising network to gradually remove noise and restore the latent space samples, using conditional information as a covariate. S6: The denoising network is trained using a structure of residual connections and spatiotemporal convolutional layers to improve its ability to learn spatiotemporally related features; The training objective is to minimize the KL divergence between the posterior and prior distributions to ensure the effectiveness of the denoising process. S7: Reconstruct the low-dimensional latent space scene output by the denoising network to generate a high-dimensional wind power output scene; use the generated scene for the application of economic dispatch and risk assessment of the power system; In the multi-regional wind power short-term scenario generation method, a pre-trained encoding network is used. The NWP input features for each region are embedded, and the results are based on deterministic prediction. Evaluate the real-time dynamic adjacency matrix ,Will As a conditional embedding of the diffusion model, to utilize denoising networks This study explores the dynamic spatial relationships of wind power in multiple regions and eliminates the need for filtering high-dimensional NWP features in each denoising step of the diffusion model, thus reducing the computational load of the diffusion model. Multiple independent embedding networks are used to perform deterministic prediction and regression of high-dimensional NWP features of wind power in each region, and the Spearman correlation coefficient between the deterministic prediction results of wind power in multiple regions is used as the corresponding element of the dynamic adjacency matrix, thereby realizing real-time measurement of the dynamic correlation between different regions. When inputting data to the NWP of the i-th region During embedding, the input layer convolutional neural network is used to adjust... The number of channels is used to obtain a high-dimensional representation of the features; The intermediate layers utilize residual connections across multiple convolutional layers to increase the receptive field to the input data, preventing network degradation. The output layer performs skip connections between the results of each convolutional layer, achieving feature fusion across multiple time scales. Ultimately, this yields... Feature embedding results The L1 loss is used to train the embedded network for deterministic prediction regression, as shown in the following equation: ; In the formula: For the actual wind power output of region i, Embedded results for deterministic wind power prediction in region i; Repeat the above steps to obtain the feature embedding set of all regions. And then according to Calculate the Spearman correlation coefficient between regions to obtain the dynamic adjacency matrix characterizing the real-time spatial correlation of wind power. .

2. The method for generating short-term wind power scenarios in multiple regions according to claim 1, characterized in that: For each region The input meteorological data is represented as T represents the time step, and C represents the feature dimension; Using pre-trained encoding networks right Embedding is performed to generate a high-dimensional representation of the features; The encoding network is a convolutional neural network structure, which includes: an input layer, intermediate layers, and an output layer; Input layer: Adjust the number of channels to accommodate subsequent convolutional layers; Intermediate layers: Multiple convolutional layers are used, and the receptive field is increased through residual connection structures to prevent network degradation; Output layer: Skip connections are used to combine the results of each convolutional layer, achieving the fusion of features at multiple time scales, and finally obtaining the embedding result. .

3. The method for generating short-term wind power scenarios in multiple regions according to claim 1, characterized in that: The sub-steps of S4 are as follows: S41: High-dimensional wind power output sequence Input to encoder E; Encoder E generates the mean vector and standard deviation vector Thus, the probability distribution of the latent space is defined. ; S42: Sample latent space variables from the above probability distribution ; S43: Input the latent space variable z into the decoder D to generate the reconstructed wind power output sequence. ; S44: Calculate the VAE loss function, including KL divergence and reconstruction error: ； Introducing weighting coefficients Then, the loss function is updated as follows: ； In the formula: and These are the mapping functions for the encoder and decoder, respectively; Let represent the posterior distribution of the encoder output, and let z represent the distribution of the latent variable z given the input y. Let z represent the reconstructed distribution generated by the decoder, and let z represent the probability of data recovered from the latent variable z. S45: Map the wind power sequence of each region using an encoder to obtain the latent space set. ; S46: The latent space features are used in the downstream tasks of the diffusion model to reconstruct the low-dimensional scene output by the denoising network into a high-dimensional wind power output scene, ensuring the accurate characterization of detailed fluctuation features.

4. The method for generating short-term wind power scenarios in multiple regions according to claim 1, characterized in that: The diffusion process is as follows: The diffusion process gradually transforms the hidden space samples Add noise to This will gradually lead to a transformation towards a standard normal distribution, as follows: ； In the formula: This represents the scaling factor at step n; For noise samples, Indicates the first The scaling factor of the step controls the degree of noise addition. This represents random noise added during the diffusion process; The Markov chain process is represented as: ； As n increases, the latent space samples tend to follow a standard normal distribution. .

5. The method for generating short-term wind power scenarios in multiple regions according to claim 4, characterized in that: The noise reduction process is represented as follows: ； The goal of denoising is to minimize the KL divergence between the posterior and prior distributions, which is expressed as: ； In the formula, This indicates the lower bound of evidence in variational inference. Represents all possible and noise sampled from a standard normal distribution Expectations This indicates the reconstruction error.

6. The method for generating short-term wind power scenarios in multiple regions according to claim 5, characterized in that: The denoising network is designed to enhance dynamic spatial relevance by combining residual connections, skip connections, and graph convolution. The graph convolutional layer is represented as: ; In the formula, and The first Layer inputs and outputs, The parameters of the adaptive matrix, for The degree matrix; b represents the network learnable parameters of GCN; ⊙ represents the Hadamard product; This is the activation function.

7. A multi-regional wind power short-term scenario generation system, characterized in that: This system can be used to implement the multi-regional wind power short-term scenario generation method according to any one of claims 1 to 6, specifically including: Data collection module: Collects meteorological data and historical wind power output data from wind farms in multiple regions, and performs data cleaning and standardization to eliminate noise and outliers; Feature embedding module: Uses a pre-trained encoding network to embed the meteorological feature NWP of each region, and uses the embedding results to evaluate the dynamic adjacency matrix between regions in order to capture the spatial correlation between regions. Dynamic adjacency matrix calculation module: Based on the Spearman correlation coefficient of wind power output in each region, the dynamic adjacency matrix is ​​calculated in real time to reflect the dynamic correlation between wind power output in different regions. Latent space mapping module: The high-dimensional wind power output curve is mapped to a low-dimensional latent space feature vector using a variational autoencoder (VAE). The encoder extracts the trend features of wind power output, and the decoder reconstructs the high-dimensional output sequence. Conditional Latent Space Diffusion Model Module: Constructs a conditional latent space diffusion model that includes a diffusion process and a denoising process. The diffusion process is defined to gradually add noise to the latent space samples until they are normally distributed. A denoising network is designed to gradually remove noise and restore the latent space samples by using conditional information as a covariate. Denoising network training module: The denoising network is trained using a structure of residual connections and spatiotemporal convolutional layers to improve its ability to learn spatiotemporally related features. The training objective is to minimize the KL divergence between the posterior and prior distributions. Scene generation module: The low-dimensional latent space scene output by the denoising network is reconstructed to generate a high-dimensional wind power output scene, and the generated scene is finally used for the economic dispatch and risk assessment of the power system.

8. A computer-readable storage medium, characterized in that: It stores a computer program that, when executed by a processor, implements the multi-regional wind power short-term scenario generation method according to any one of claims 1 to 6.

Images