Shahuo desert wind festival scene generation method and system based on conditional diffusion model

Abstract

The application discloses a method and system for generating a sandgao gale light festival scene based on a conditional diffusion model, and belongs to the technical field of power systems. First, the 24 solar terms are used as the division interval of the historical time sequence, and each solar term sample is given a label. Second, based on the feature extraction network of the solar term factor, the solar term factor is mapped into a set of feature vectors in the hidden space, effectively preserving the solar term regularity characteristics of the light and power output. Finally, a generative network based on small sample learning technology is introduced, and a solar term light scene generation network based on a conditional diffusion probability model C-DDPM is constructed, which converts the prior distribution of a single solar term feature vector into a joint probability distribution of the 24 solar terms. A denoising network architecture is used to capture the uncertainty and time sequence of the output, and after training is completed, the noise and corresponding solar term label can be input to generate a solar term scene. It is especially suitable for areas with scarce data samples, and the model can generate diversified light and scenery scenes through solar term conditions and noise input.

Description

Technical Field

[0001] This invention belongs to the field of power systems, specifically relating to a method and system for generating seasonal landscape scenes in the desert based on a conditional diffusion model. Background Technology

[0002] Constructing new energy bases in desert, Gobi, and arid regions is a crucial measure for my country to implement its "dual-carbon" strategy, with these bases becoming the main providers of electricity. However, due to the volatility and intermittency of wind and solar power output, even resource-rich desert and Gobi bases may experience consecutive days of low output, making it difficult to accurately represent meteorological fluctuations using short-term, daily scenarios. Therefore, accurately simulating continuous multi-day renewable energy output and constructing medium- to long-term time-series scenarios will directly impact the stable operation of the power system and planning decisions.

[0003] Among existing scene generation methods, generative models are a relatively mature approach to characterizing the uncertainty of renewable energy output. However, current methods for dividing historical wind and solar data samples often fail to adequately consider the varying lengths of time intervals and the lack of regularity in their distribution. Current research primarily focuses on weekly, monthly, and quarterly time interval divisions, with few methods based on solar terms. Furthermore, data-driven methods require extensive historical power data for training and supervision; when data is scarce or samples are limited, the model's learning ability is constrained. Newly built large-scale wind and solar power bases are often located in desert and Gobi areas where historical power data is scarce, limiting training conditions. Existing scene generation research rarely considers the problem of insufficient actual power samples, thus limiting the performance of scene generation.

[0004] In summary, there is a lack of methods for generating seasonal scenes that take into account the differences in the distribution of data subsets and the insufficient historical samples, in order to address the uncertainty of medium- and long-term wind and solar power in the Shagohuang region. Summary of the Invention

[0005] To address the technical problems existing in the background art, this invention aims to provide a method and system for generating seasonal landscape scenes in the Shagohuang region based on a conditional diffusion model. Based on the output characteristics of historical power data, the 24 solar terms are used as a time measurement and division system. This interval division method has a high degree of adaptability to meteorological types, accurately capturing the performance of wind and solar power output in different seasons and revealing the system's changing patterns with seasonal transitions, providing data support for subsequent medium- and long-term landscape scene generation. From the perspective of data expansion, a sample expansion strategy based on a conditional diffusion probability model is proposed to improve the performance of wind and solar simulation. Solar term labels and power data of the 24 solar terms are input into a generative network. Through small-sample data generation technology based on generative networks, the wind and solar power data of the 24 solar terms are jointly modeled, fitting a conditional joint probability distribution, allowing historical samples to share parameters and efficiently generating landscape scenes.

[0006] To solve the technical problem, the technical solution of the present invention is as follows:

[0007] A method for generating seasonal scenes of desert landscape based on a conditional diffusion model, the method comprising:

[0008] The twenty-four solar terms are used as the dividing intervals for the historical time series. The solar term samples are divided and each solar term sample is labeled.

[0009] A feature extraction network is used to map the solar term labels to the latent space, and the generated solar term feature vector c is used as a conditional vector.

[0010] Based on generative networks using few-shot learning, a solar term-based wind and solar power scene generation network was constructed using the Conditional Denoising Diffusion Probability Model (C-DDPM). The prior distribution of individual solar term feature vector mappings was transformed into a joint distribution of the twenty-four solar terms. Through iterative processes of forward noise addition and backward denoising, the uncertainty and temporal sequence of forces were captured. Noise and solar term feature vector c were input into C-DDPM to generate wind and solar power scenes under specific solar term conditions. Finally, through collaborative modeling of the twenty-four solar term data, wind and solar power scenes that conform to the characteristics of each solar term were generated.

[0011] Furthermore, the historical wind and solar power data are preprocessed, taking into account the differences in the number of days in leap years and common years, as well as the different number of days in each solar term. Based on the actual date and time of each solar term, the historical samples of each year are divided into twenty-four solar term samples with an equal number of days to construct a training set. Considering the periodicity of solar terms within the year, a two-dimensional sine and cosine vector is used to represent the solar term category, that is, the i-th solar term is encoded as c = [sin(2πi / 24), cos(2πi / 24)], and the solar term features are encoded as values ​​in the interval [-1,1] as the solar term label corresponding to each solar term sample.

[0012] Furthermore, the solar term labels are mapped to the latent space through the solar term factor feature extraction network, and the solar term factors are mapped to the feature vector set of the latent space. This effectively preserves the solar term regularity characteristics of wind and solar power output. The generated solar term feature vector is used as a condition vector, and the solar term feature vector c is used as the input condition of the conditional denoising diffusion probability model.

[0013] Furthermore, the solar term scenery scene generation network based on the conditional denoising diffusion probability model C-DDPM consists of two parts: a forward process and a backward process. In the forward process, prior noise is added iteratively to convert the scenery sequence x0 into Gaussian noise z, so that it converges to the prior distribution. In the backward process, the noise of x0 time steps is removed from the prior distribution by embedding conditional labels c to generate the scenery sequence.

[0014] The forward process is defined by a Markov chain, gradually adding Gaussian noise to the original solar term sequence x0 until it becomes pure noise; at any time during this process, x... n Only with x n-1 Therefore, the C-DDPM forward process is defined as follows:

[0015]

[0016] Where q(x) n |x n-1 The distribution is Gaussian, and can be represented as:

[0017]

[0018] {β1, …, β N} is an increasing variance time series, where β n ∈(0,1) represents the forward step α n =1-β n The noise level;

[0019] Let α n =1-β n , By employing reparameterization techniques, we can obtain x samples with arbitrary noise levels n. n for:

[0020]

[0021] In the formula: As sampling noise;

[0022] The reverse process, by embedding the solar term condition c, modifies the noise-added x. N Denoising is performed to generate x0, which also follows a Markov chain and originates from a learnable Gaussian distribution. Initially, it was defined as:

[0023]

[0024] The distribution between two adjacent latent variables is represented as follows:

[0025]

[0026] In the formula: θ is a shared parameter, which is p θ (x n-1 |x n c) Parameterization:

[0027]

[0028]

[0029] In the formula: ∈ represents the actual noise, ∈ θ For the predicted value of ∈, It is a trainable denoising function that determines how much noise should be removed in the current denoising step.

[0030] Furthermore, the training objective of the C-DDPM is to minimize the error between the actual noise and the predicted noise under different diffusion step sizes. The objective function of the training process is defined as:

[0031]

[0032] Furthermore, the input to the denoising network architecture consists of noisy samples. The output consists of the diffusion step number n and the solar term feature c, and is a denoised noise ∈ θ ;

[0033] The number of channels in the input sequence x is increased to C using a one-dimensional convolutional layer, and sigmoid linear units are used for activation.

[0034] SiLu(x)=x*Sigmoid(x) (19)

[0035] Similar to the position embedding method proposed in Transformer, a sinusoidal coding formula is applied to embed the noise level n:

[0036]

[0037] Then, the encoded noise level is projected onto a C-dimensional vector using a fully connected FC layer with shared SiLu activation and a residual-specific FC layer, and added as a bias term to the x of the convolution; the seasonal label condition c is transformed to the temporal resolution of the wind and light through the FCN U-Net network, and then the channel is transformed by a one-dimensional convolution and projected into the bias terms of n and x.

[0038] The main body of the model consists of N consistent structural layers connected in a residual manner, so that input information can be directly passed to the final layer. In each residual layer, a bidirectional dilated convolution Bi-DilConv is used to process the input, causing it to grow exponentially in the receptive field. In the i-th residual layer, the spacing between kernel points is set to 2i modulo-gated units (GUs) to activate the summation of the processed noise signal and the conditional signal. Then, the convolution vector is split into two parts and passed as the residual output and the skip output, respectively. Finally, all skip connections are summed, and two convolutional layers are used to obtain a noise vector output with the same shape as x.

[0039] A system for generating seasonal landscape scenes of Shagohuang based on a conditional diffusion model, wherein the system is applied to any of the methods described above, and the system includes:

[0040] Solar Term Sample Segmentation Module: This module uses the 24 solar terms as the dividing intervals for the historical time series, divides the solar term samples, and assigns a label to each solar term sample.

[0041] Feature extraction module: used to map the solar term labels to the latent space using the feature extraction network, and the generated solar term feature vector c is used as the condition vector;

[0042] Based on the Conditional Denoising Diffusion Probability Model Module: This module is used for generative networks based on few-shot learning. It constructs a solar term-based wind and light scene generation network based on the Conditional Denoising Diffusion Probability Model C-DDPM. It transforms the prior distribution of a single solar term feature vector mapping into a joint distribution of the twenty-four solar terms. Through the iterative process of forward noise addition and backward noise reduction, it captures the uncertainty and temporal sequence of forces. Noise and solar term feature vector c are input into C-DDPM to generate wind and light power scenes under specific solar term conditions.

[0043] Collaborative Modeling Module: This module generates wind and solar power scenarios that conform to the characteristics of each solar term by collaboratively modeling the data of the 24 solar terms.

[0044] A computer device includes 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 method for generating seasonal scenes of desert landscape based on the conditional diffusion model described above.

[0045] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for generating seasonal landscape scenes of Shagohuang based on the conditional diffusion model described above.

[0046] Compared with the prior art, the advantages of the present invention are as follows:

[0047] 1) By dividing historical samples into solar term samples of the same time interval and giving them labels, a feature extraction network is used to map the solar term labels to the feature vector set in the latent space. This not only effectively preserves the original solar term characteristics of wind and solar output, but also enhances the interpretability of the model.

[0048] 2) By using the solar term labels as conditional inputs to C-DDPM, the historical power output data of the twenty-four solar terms are jointly trained and the joint probability distribution is fitted, so that the model can share parameters, thereby effectively solving the training difficulties caused by the scarcity of historical samples of long-term simulated landscape in the desert area.

[0049] 3) The proposed C-DDPM model-generated scene generator can effectively capture the variability and uncertainty of wind and solar power output during different solar seasons. This method can also be applied to the generation of extreme wind and solar power scenarios, which is of great significance to the power balance problem faced by the power system and provides a new approach for the generation of renewable energy scenarios. Attached Figure Description

[0050] Figure 1 Flowchart for generating solar term scenarios;

[0051] Figure 2 Basic structure diagram of C-DDPM;

[0052] Figure 3 1. Denoising network structure design diagram; Detailed Implementation

[0053] The specific implementation of the present invention is described below with reference to embodiments:

[0054] It should be noted that the structures, proportions, sizes, etc. shown in this specification are only used to complement the content disclosed in the specification for those skilled in the art to understand and read, and are not intended to limit the conditions under which the present invention can be implemented. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0055] Furthermore, the terms such as "upper," "lower," "left," "right," "middle," and "one" used in this specification are merely for clarity of description and are not intended to limit the scope of the invention. Any changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention.

[0056] Conditional diffusion model: A generative model that adds conditional information to the diffusion model, ensuring that the generated samples meet specific conditions or features. It uses conditional probability to guide the data generation process, gradually transforming noisy data into target data and progressively denoising it during generation. Its core idea is to achieve precise data generation and control by simulating the diffusion process from noise to an ordered state. This model can generate high-quality data that meets specific conditions in various fields, such as image generation and text generation.

[0057] Solar term scenario generation: Solar term scenario generation involves probabilistically modeling the uncertainties within a specific solar term period in a power system and discretizing these probabilistic models into finite solar term time-series scenarios. Compared to directly using probabilistic models to describe uncertainties, solar term scenario generation is an extension of the application of probabilistic models, enabling more effective application of theoretical innovations in probabilistic models to uncertain optimization problems in practical power systems. Based on the time scale of the optimization decision-making problem being studied, solar term scenario generation is mainly used for generating medium- and long-term scenarios for power systems.

[0058] Example 1:

[0059] This invention includes key steps such as data preprocessing, feature extraction, model building, collaborative modeling, and generation. The following is an overview of this logic:

[0060] The 24 solar terms are used to divide the data and label the samples: The historical time series data is divided according to the 24 solar terms, and each solar term sample corresponds to a specific label. This step segments the data based on the characteristics of the solar terms (such as date and time).

[0061] The feature extraction network generates conditional vectors: The feature extraction network maps the solar term labels to the latent space, and the resulting latent space feature vector *c* is used as the conditional vector input into the subsequent generative model. This process preserves the feature information of the solar terms.

[0062] Conditional Probability Network Structure: A C-DDPM model is constructed using a few-shot learning method to fit the conditional probability distribution. The C-DDPM model introduces a solar term feature vector c, and through an iterative process of forward noise addition and backward noise removal, it achieves the generation of wind and solar power scenes under specific solar term conditions.

[0063] Furthermore, the Conditional Denoising Diffusion Probability Model (C-DDPM) consists of two processes: forward noise addition and inverse denoising. In the forward process, noise is iteratively added to gradually transform the original wind and solar power sequence into Gaussian noise. In the inverse process, by introducing seasonal label conditions, the noise is denoised, gradually reconstructing the wind and solar power output curves. The forward process is a Markov chain that progressively adds noise to the sequence, while the inverse process reconstructs the power sequence by learning the denoising steps.

[0064] Denoising networks capture uncertainty and temporal sequence: Denoising networks play an important role in the reverse denoising process. They can capture the fluctuation (uncertainty) and the trend of change over time (temporal sequence) of wind and solar power output, so that the generated power sequence is more consistent with the actual situation.

[0065] Furthermore, the denoising network employs residual connections and bidirectional dilated convolutional structures to process the input signal and gradually remove noise. The solar term conditions are transformed using a U-Net network and then embedded into the model for joint training. After convolution processing, the output is split into two parts for residual propagation, and finally summed through skip connections to generate a noise vector output with the same shape as the input.

[0066] Collaborative modeling to generate solar term scenarios: By collaboratively modeling all the data of the twenty-four solar terms, C-DDPM can learn the correlation and change patterns between different solar terms, and finally generate wind and solar power scenarios that conform to the characteristics of each solar term.

[0067] Furthermore, by jointly training on wind and solar power data from the desert region, this method can capture the spatiotemporal correlation and fluctuation characteristics of multimodal wind and solar variables. After training, inputting random noise and solar term labels can generate wind and solar power output scenarios related to specific solar terms.

[0068] This invention utilizes the C-DDPM model, through collaborative modeling and denoising processes, to fit the joint probability distribution law of wind and solar power output under the conditions of the 24 solar terms. It can effectively generate scenes for the 24 solar terms, and is particularly suitable for wind and solar power prediction and expansion in areas with limited sample sizes. This overall logic accurately describes the process of generating solar term wind and solar scenes based on C-DDPM, highlighting the model's capture of solar term characteristics and its handling of data uncertainty and temporal variations. It demonstrates the entire process from data partitioning, feature extraction, joint modeling, and model generation.

[0069] Example 2:

[0070] This embodiment applies to Embodiment 1. This embodiment divides the solar term samples and uses a feature extraction network to map the solar term labels to the latent space, generating solar term feature vectors as conditional vectors. A conditional diffusion model suitable for sample expansion is designed and trained through forward noise addition and reverse denoising processes to capture the correlation changes and fluctuation characteristics of wind and solar power output. By collaboratively modeling the twenty-four solar term data, its conditional joint probability distribution law is fitted, thereby efficiently generating twenty-four solar term scenes.

[0071] The specific process for generating solar term scenes is as follows: Figure 1 As shown.

[0072] First, the historical wind and solar power data are preprocessed. Considering the differences in the number of days in leap years and common years, as well as the varying number of days in each solar term, the historical samples for each year are divided into twenty-four solar term samples with an equal number of days, based on the actual date and time of each solar term, to construct a training set. Since the solar terms are periodic within the year, a two-dimensional sine and cosine vector is used to represent the solar term category. That is, the i-th solar term is encoded as c = [sin(2πi / 24), cos(2πi / 24)]. The solar term features are encoded as values ​​in the interval [-1, 1], serving as the solar term label for each solar term sample.

[0073] By using a solar term factor feature extraction network to map the solar term labels into the latent space, the solar term factors are mapped into a set of feature vectors in the latent space, effectively preserving the solar term regularity characteristics of wind and solar power output. The obtained solar term feature vectors are used as condition vectors, and the proposed conditional probability model uses the solar term feature c as a condition.

[0074] Conditional denoising diffusion probabilistic models (C-DDPM), as generative models, do not randomly generate a set of power sequences. Instead, they generate power output curves following a certain distribution based on the characteristics of the solar terms. C-DDPM establishes a mapping relationship between the conditional input and the latent space vector, as follows: Figure 2 .

[0075] C-DDPM consists of two parts: a forward process and a backward process. In the forward process, prior noise is added iteratively to transform the landscape sequence x0 into Gaussian noise z, causing it to converge to the prior distribution. The backward process removes noise from the prior distribution over x0 time steps by embedding conditional labels c to generate the landscape sequence.

[0076] The forward process is defined by a Markov chain, gradually adding Gaussian noise to the original solar term sequence x0 until it becomes pure noise. At any given time in this process, x... n Only with x n-1It is related to the solar term label c, but not to the solar term label c. Therefore, the C-DDPM forward process is defined as:

[0077]

[0078] Where q(xx|x) n-1 The distribution is Gaussian, and can be represented as:

[0079]

[0080] {β1, …, β N} is an increasing variance time series, where β n ∈(0,1) represents the forward step α n =1-β n The noise level.

[0081] Let α n =1-β n , By employing reparameterization techniques, we can obtain x samples with arbitrary noise levels n. n for:

[0082]

[0083] In the formula: As sampling noise.

[0084] The reverse process, by embedding the solar term condition c, modifies the noise-added x. N Denoising is performed to generate x0, which also follows a Markov chain and originates from a learnable Gaussian distribution. Initially, it was defined as:

[0085]

[0086] The distribution between two adjacent latent variables is represented as follows:

[0087]

[0088] In the formula: θ is a shared parameter, which is p θ (x n-1 |x n c) Parameterization:

[0089]

[0090]

[0091] In the formula: ∈ represents the actual noise, ∈ θ For the predicted value of ∈, It is a trainable denoising function that determines how much noise should be removed in the current denoising step.

[0092] The training objective of C-DDPM is to minimize the error between the actual noise and the predicted noise at different diffusion step sizes. Therefore, the objective function of the training process is defined as:

[0093]

[0094] The input to the denoising network consists of noisy samples. The output consists of the diffusion step number n and the solar term feature c, and is a denoised noise ∈ θ Its structural diagram is as follows Figure 3 As shown.

[0095] like Figure 3 As shown, a one-dimensional convolutional layer is used to increase the number of channels of the input sequence x to C, and Sigmoid linear units (SiLU) are used for activation:

[0096] SiLu(x)=x*Sigmoid(x) (29)

[0097] Similar to the position embedding method proposed in Transformer, a sinusoidal coding formula is applied to embed the noise level n:

[0098]

[0099] Then, a fully connected (FC) layer with shared SiLu activation and a residual-specific FC layer are used to project the encoded noise level onto a C-dimensional vector, which is then added as a bias term to the x-coefficient of convolution. The seasonal label condition c is transformed to the temporal resolution of the wind and light through the FCN U-Net network, and then the channel is transformed by a one-dimensional convolution and projected into the bias terms of n and x.

[0100] The main body of the model consists of N consistent structural layers connected in a residual manner, enabling direct transmission of input information to the final layer. Within each residual layer, bidirectional dilated convolutions (Bi-DilConvolution) are used. v The input is processed exponentially within the receptive field, and in the i-th residual layer, the spacing between kernel points is set to 2i. Modular gated units (GUs) are used to sum the processed noise signal and the conditional signal. The convolution vector is then split into two parts and passed as the residual output and the skip output, respectively. Finally, all skip connections are summed, and two convolutional layers are used to obtain a noise vector output with the same shape as x.

[0101] By constructing a solar term scene generation network based on C-DDPM, and utilizing an iterative process of forward noise addition and inverse denoising, the network is jointly trained and expanded to include samples with limited data in the desert region. This process enables the learning of the spatiotemporal correlations and fluctuation characteristics between wind and light multimodal variables. After the model training is complete, inputting random noise and solar term labels generates corresponding wind and light solar term scenes.

[0102] Example 3:

[0103] This invention provides a system for generating seasonal landscape scenes of the Gobi Desert based on a conditional diffusion model. This system can be used to implement the aforementioned method for generating seasonal landscape scenes of the Gobi Desert based on a conditional diffusion model. The system includes:

[0104] Solar Term Sample Segmentation Module: This module uses the 24 solar terms as the dividing intervals for the historical time series, divides the solar term samples, and assigns a label to each solar term sample.

[0105] Feature extraction module: used to map the solar term labels to the latent space using the feature extraction network, and the generated solar term feature vector c is used as the condition vector;

[0106] Based on the Conditional Denoising Diffusion Probability Model Module: This module is used for generative networks based on few-shot learning. It constructs a solar term-based wind and light scene generation network based on the Conditional Denoising Diffusion Probability Model C-DDPM. It transforms the prior distribution of a single solar term feature vector mapping into a joint distribution of the twenty-four solar terms. Through the iterative process of forward noise addition and backward noise reduction, it captures the uncertainty and temporal sequence of forces. Noise and solar term feature vector c are input into C-DDPM to generate wind and light power scenes under specific solar term conditions.

[0107] Collaborative Modeling Module: This module generates wind and solar power scenarios that conform to the characteristics of each solar term by collaboratively modeling the data of the 24 solar terms.

[0108] Example 4:

[0109] This embodiment provides a terminal device, which includes 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 corresponding method flows or corresponding functions. The processor described in this embodiment can be used in the operation of a method for generating seasonal scenes of desert landscapes based on a conditional diffusion model, including the following steps:

[0110] The twenty-four solar terms are used as the dividing intervals for the historical time series. The solar term samples are divided and each solar term sample is labeled.

[0111] A feature extraction network is used to map the solar term labels to the latent space, and the generated solar term feature vector c is used as a conditional vector.

[0112] Based on generative networks using few-shot learning, a solar term-based wind and solar power scene generation network was constructed using the Conditional Denoising Diffusion Probability Model (C-DDPM). The prior distribution of individual solar term feature vector mappings was transformed into a joint distribution of the twenty-four solar terms. Through iterative processes of forward noise addition and backward denoising, the uncertainty and temporal sequence of forces were captured. Noise and solar term feature vector c were input into C-DDPM to generate wind and solar power scenes under specific solar term conditions. Finally, through collaborative modeling of the twenty-four solar term data, wind and solar power scenes that conform to the characteristics of each solar term were generated.

[0113] Example 5:

[0114] This embodiment provides a storage medium, specifically a computer-readable storage medium (Memory), which 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.

[0115] 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 seasonal landscape scenes of desert based on a conditional diffusion model in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by the processor in the following steps:

[0116] The twenty-four solar terms are used as the dividing intervals for the historical time series. The solar term samples are divided and each solar term sample is labeled.

[0117] A feature extraction network is used to map the solar term labels to the latent space, and the generated solar term feature vector c is used as a conditional vector.

[0118] Based on generative networks using few-shot learning, a solar term-based wind and solar power scene generation network was constructed using the Conditional Denoising Diffusion Probability Model (C-DDPM). The prior distribution of individual solar term feature vector mappings was transformed into a joint distribution of the twenty-four solar terms. Through iterative processes of forward noise addition and backward denoising, the uncertainty and temporal sequence of forces were captured. Noise and solar term feature vector c were input into C-DDPM to generate wind and solar power scenes under specific solar term conditions. Finally, through collaborative modeling of the twenty-four solar term data, wind and solar power scenes that conform to the characteristics of each solar term were generated.

[0119] 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.

[0120] 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.

[0121] 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.

[0122] 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.

Claims

1. A method for generating seasonal scenes of desert landscape based on a conditional diffusion model, characterized in that, The method includes: The twenty-four solar terms are used as the dividing intervals for the historical time series. The solar term samples are divided and each solar term sample is labeled. A feature extraction network is used to map the solar term labels to the latent space, and the generated solar term feature vector c is used as a conditional vector. Based on generative networks using few-shot learning, a solar term-based wind and light scene generation network was constructed using the Conditional Denoising Diffusion Probability Model (C-DDPM). The prior distribution of the feature vector mapping of a single solar term is transformed into the joint distribution of the twenty-four solar terms. Through the iterative process of forward noise addition and backward noise reduction, the uncertainty and temporal sequence of forces are captured. Noise and solar term feature vectors are input into C-DDPM to generate wind and light power scenes under specific solar term conditions. Finally, through collaborative modeling of the twenty-four solar term data, wind and light power scenes that conform to the characteristics of each solar term are generated. The solar term scenery scene generation network based on the Conditional Denoising Diffusion Probability Model (C-DDPM) consists of two parts: a forward process and a backward process. In the forward process, prior noise is added iteratively to generate the scenery sequence. Converting to Gaussian noise z, causing it to converge to the prior distribution, and the reverse process involves embedding the solar term condition c*. The wind and light sequence is generated by removing noise from the prior distribution at a time step; The forward process is defined by a Markov chain, gradually adding Gaussian noise to the wind and light sequence. In the middle, until it becomes pure noise; at any time during the forward process Only with Therefore, the C-DDPM forward process is defined as follows: ； in, It follows a Gaussian distribution, which can be represented as: ； It is a time series with increasing variance, where Indicates the forward step The noise level; make , Arbitrary noise levels can be obtained through reparameterization techniques. downsampling for: ； In the formula: As sampling noise; The reverse process is achieved by embedding solar term conditions. After adding noise Denoising generation It also follows a Markov chain, derived from a learnable Gaussian distribution. Initially, it was defined as: ； The distribution between two adjacent latent variables is represented as follows: ； In the formula: To share parameters, Parameterization: ； ； In the formula: For real noise, for The predicted value, It is a trainable denoising function that determines how much noise should be removed in the current denoising step; The training objective of C-DDPM is to minimize the error between the actual noise and the predicted noise at different diffusion step sizes. The objective function of the training process is defined as follows: 。 2. The method for generating seasonal scenes of desert landscape based on a conditional diffusion model according to claim 1, characterized in that, Historical wind and solar power data are preprocessed, taking into account the differences in the number of days in leap years and common years, as well as the varying number of days for each solar term. Based on the actual date and time of each solar term, the historical samples for each year are divided into twenty-four solar term samples with an equal number of days to construct a training set. Considering the periodicity of solar terms within the year, a two-dimensional sine and cosine vector is used to represent the solar term category, i.e., the feature vector of the i-th solar term is... The solar term feature vector is defined as the value in the interval [−1,1], which is used as the solar term label for each solar term sample.

3. The method for generating seasonal scenes of desert landscape based on a conditional diffusion model according to claim 1, characterized in that, By using a solar term factor feature extraction network to perform latent space mapping on solar term labels, the solar term factors are mapped into a set of feature vectors in the latent space, effectively preserving the solar term regularity characteristics of wind and solar output. The generated solar term feature vectors are used as condition vectors, and the solar term feature vectors are used as input conditions for the conditional denoising diffusion probability model.

4. A system for generating seasonal scenes of desert landscapes based on a conditional diffusion model, characterized in that, The system is applied to the method according to any one of claims 1-3, and the system comprises: Solar Term Sample Segmentation Module: This module uses the 24 solar terms as the dividing intervals for the historical time series, divides the solar term samples, and assigns a label to each solar term sample. Feature extraction module: used to map the solar term labels to the latent space using the feature extraction network, and the generated solar term feature vector c is used as the condition vector; Based on the Conditional Denoising Diffusion Probability Model Module: This module is used for generative networks based on few-shot learning. It constructs a solar term-based wind and light scene generation network based on the Conditional Denoising Diffusion Probability Model (C-DDPM). It transforms the prior distribution of a single solar term feature vector mapping into a joint distribution of the twenty-four solar terms. Through the iterative process of forward noise addition and backward noise removal, it captures the uncertainty and temporal sequence of forces. Noise and solar term feature vectors are input into C-DDPM to generate wind and light power scenes under specific solar term conditions. Collaborative Modeling Module: This module generates wind and solar power scenarios that conform to the characteristics of each solar term by collaboratively modeling the data of the 24 solar terms.

5. A computer device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the method for generating seasonal scenes of desert landscape based on the conditional diffusion model as described in any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method for generating seasonal landscape scenes of Shagohuang based on the conditional diffusion model as described in any one of claims 1 to 3.

Images