New energy output scene generation method and system based on meteorological factor guidance
By constructing a feature-guided pre-trained denoising diffusion probability model and a ControlNet control branch, combined with the LightGBM model, the problem of insufficient probability distribution and temporal correlation in the generation of new energy output scenarios was solved, and high-quality, stable and controllable generation of multi-day new energy output scenarios was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-11
- Publication Date
- 2026-03-27
AI Technical Summary
Existing methods for generating new energy output scenarios lack guidance from probability distribution and temporal correlation knowledge, resulting in low generation quality and insufficient generation stability and controllability, especially making it difficult to achieve controllable generation of multi-day scenarios under extreme weather conditions.
A pre-trained denoising diffusion probability model based on Hilbert curve projection distance and autocorrelation coefficient features is constructed. By controlling the branches through ControlNet and combining the LightGBM gradient boosting tree model, the conversion from weather scenarios to new energy output is realized, generating multi-day new energy output scenarios that meet the conditions for control.
It improves the probabilistic consistency and temporal correlation of the generated new energy power output scenarios, enhances the stability and controllability of the generated scenarios, reduces training costs and accelerates model convergence, and provides high-quality multi-day new energy power output scenario data.
Smart Images
Figure CN121749119A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of scene generation technology, and in particular to a method and system for generating new energy output scenes based on meteorological factors. Background Technology
[0002] To address the global energy crisis, the penetration rate of new energy sources, such as wind and solar power, in the power system continues to rise. Meteorological factors such as wind speed, temperature, and irradiance are key influences on the output of these new energy sources. With the high proportion of new energy sources integrated into the power system, meteorological factors are gradually becoming new variables affecting the stable operation of the power system. Against the backdrop of global warming, extreme weather events are becoming more frequent, and are increasingly characterized by longer durations, greater severity, and a wider variety of event types.
[0003] The strong intermittency and high-frequency fluctuations in renewable energy output under extreme weather conditions pose a severe challenge to the safe and stable operation of the power system. To analyze the impact of renewable energy output on the power system, scenario generation technology is needed to simulate various potential renewable energy output scenarios under different weather events. High-quality and diverse renewable energy output scenario data are crucial for conducting power dispatch decisions, proactive risk assessments, and planning and design.
[0004] Current methods for generating new energy power output scenarios have the following shortcomings: 1) Deep learning models lack knowledge of probability distribution and temporal correlation to guide the learning direction when generating scenes, making it difficult for the models to generate physically reasonable and diverse temporal scenes, thus reducing the generation quality of new energy output scenes.
[0005] 2) Generative models based on conditional generative adversarial networks and their variants are at risk of mode collapse during training, resulting in insufficient stability and conditional controllability of scene generation.
[0006] 3) Existing methods for generating new energy output scenarios are mostly geared towards intraday scenario generation, lacking controllable methods for generating multi-day scenarios that take into account the influence of meteorological factors, which restricts the interpretability of the generated scenarios at the weather level. Summary of the Invention
[0007] The purpose of this invention is to provide a method and system for generating new energy output scenarios based on meteorological factors. To address the problems of insufficient controllability and target orientation of multi-day new energy output scenarios, this invention constructs a feature-guided pre-trained denoising and diffusion probability model, and uses ControlNet to control branches and convert weather into new energy output, thereby achieving controllable and stable generation of multi-day new energy output scenarios covering extreme weather.
[0008] To achieve the above objectives, the present invention provides the following technical solution: On the one hand, this invention provides a method for generating new energy output scenarios based on meteorological factors, including the following steps: S1. Construct a pre-trained denoising diffusion probability model based on Hilbert curve projection distance and autocorrelation coefficient features to achieve rolling generation of weather scenes over multiple consecutive days; S2. Freeze the pre-trained denoising diffusion probability model in S1 and add it to the ControlNet control branch. Input the weather samples with conditional labels into the ControlNet control branch to train the weather scene generation model. S3. Using a weather-to-power conversion model, the generated weather scenario is converted into the output of new energy sources, resulting in a continuous multi-day new energy output scenario that meets the condition control requirements.
[0009] In the feature-guided pre-trained denoising and diffusion probability model construction stage, this invention guides the pre-trained denoising and diffusion probability model to learn the probability distribution and correlation of the real scene for the next day by adding probability distribution features and correlation feature regularization terms. By inputting the weather scene of the previous day, the model can output the weather scene of the next day. The output of the next day's scene can then be input into the model again to output the weather scene of the following day, thereby enabling the pre-trained model to continuously generate multi-day scenes.
[0010] In the controllable ControlNet scene generation model construction phase, by freezing the parameters of the pre-trained denoising and diffusion probability model and adding a conditional control branch, a weather scene generation model that meets the conditional control requirements is trained. By adjusting the output of the pre-trained denoising and diffusion probability model, weather scene generation that meets different mean conditions for the target month is achieved.
[0011] In the power output scenario conversion stage guided by meteorological factors, the wind speed scenario can be converted into a multi-day wind farm power output scenario through the mechanism formula, and the wind speed, temperature and irradiance scenario can be converted into a multi-day photovoltaic system power output scenario through the LightGBM gradient boosting tree model.
[0012] On the other hand, the present invention provides a new energy output scenario generation system based on meteorological factors, which uses the above method and includes: Model building module: Constructs a denoising diffusion probability model based on Hilbert curve projection distance and autocorrelation coefficient features as a pre-trained model to realize the rolling generation of weather scenes over multiple consecutive days; Training module: Freeze the pre-trained model in the model building module and add the ControlNet control branch. Input the weather samples with conditional labels into the ControlNet control branch to train the weather scene generation model. The conversion output module adopts a weather-to-power conversion model to convert the generated weather scene into the output of new energy, thereby obtaining a continuous multi-day new energy output scene that meets the condition control requirements.
[0013] Compared with the prior art, the present invention has the following beneficial effects: This invention improves the probability consistency and temporal correlation between the generated next day's scene and the real next day's scene by adding Hilbert curve projection distance and correlation coefficient training objectives to the pre-trained denoising diffusion probability model and inputting the previous day's weather scene.
[0014] The ControlNet architecture freezing and zero-convolution initialization mechanism of this invention reduces the training parameters of the conditional scene generation model and improves its convergence performance. By switching between weather and new energy power output scenarios, multi-day new energy power output scenarios that take into account the influence of weather factors can be constructed. When control conditions change, the weather scene generation model only needs to fine-tune the control branch without retraining the model, which helps to reduce training costs and accelerate model convergence. Attached Figure Description
[0015] Figure 1 This is a schematic diagram of the overall process of Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the weather scene generation training model based on ControlNet, which is an embodiment of the present invention. Figure 3 This is a schematic diagram of the architecture of Embodiment 1 of the present invention; Figure 4 This is a schematic diagram of the overall structure of the present invention 2. Detailed Implementation
[0016] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0017] Example 1: Please see Figures 1-3 A method for generating new energy output scenarios based on meteorological factors includes the following steps: S1. Construct a denoising diffusion probability model based on the characteristics of Hilbert curve projection distance and autocorrelation coefficient as a pre-trained model to realize the rolling generation of weather scenes for multiple consecutive days.
[0018] To achieve rolling generation of multi-day weather scenes, the pre-trained Denoising Diffusion Probabilistic Model (DDPM) primarily functions to generate the weather scene for the following day based on the previous day's scene. By incorporating a regularization term based on feature knowledge, the model is guided to generate high-quality scenes that are strongly correlated with and consistent with the probability distribution of the real-world scene for the following day.
[0019] Weather-guided scene generation aims to acquire new scenes that maintain the same probability distribution and are relevant to the original real samples. Existing noise-supervised diffusion models mainly focus on the correct direction of inverse Gaussian diffusion, lacking guidance on the correlation and probability distribution characteristics between scenes, resulting in limited scene generation quality. The denoising diffusion probability model is a generative model based on Markov chains, with generation mainly consisting of two core processes: forward diffusion and reverse denoising. The forward diffusion process gradually injects normally distributed Gaussian noise into the weather sequence, gradually disrupting the real data distribution into noise. The reverse denoising process is the inverse of learning forward diffusion. By inputting conditional feature parameters, the model is prompted to gradually reconstruct the data distribution, ultimately approximating the conditional probability distribution of the target weather sequence. This invention adjusts the learning direction of the denoising diffusion probability model by adding Hilbert curve projection Wasserstein distance and autocorrelation coefficient (ACF) regularization terms to the loss function of the denoising diffusion probability model, thereby improving the consistency of the probability distribution and temporal relevance between the generated weather scenes and the real scenes.
[0020] In a specific implementation, firstly, to measure the consistency of probability distributions between scenes, it is necessary to calculate the probability distribution distance between the generated and actual weather samples for the next day. The distance representation based on Hilbert curve projection distance prevents gradient vanishing during training, thus improving the stability of model training. To minimize the distance between the posterior and prior distributions, ensuring strong probability distribution consistency between the generated next day's scene and the actual next day's scene given a given day's scene, a Hilbert curve projection distance regularization term is added to the loss function of the pre-trained denoising diffusion probability model. for: ; In the formula, This is the squared term of the Hilbert curve projection distance between the real and generated samples; The set of joint distributions of marginal distributions; For real samples The probability distribution; To generate samples The probability distribution; These are real samples and generated samples, respectively. For in set Find the infimum in the middle; for One-dimensional numerical value after Hilbert curve projection; for One-dimensional numerical value after Hilbert curve projection; Let be the probability differential element of the joint distribution; For samples in one-dimensional space after Hilbert curve projection and The Euclidean distance.
[0021] The smaller the value, the closer the two distributions are; a value of 0 indicates that the two distributions are completely identical. (For ease of calculation...) It can be transformed into a discrete form of the Hilbert curve projection distance regularization term. for:
[0022] In the formula, , The number of real samples and the number of generated samples; For the first A real weather sample; For the first One generated weather sample; For the first dimension in the projected one-dimensional space The first real sample and the first Euclidean distance between generated samples; These are the elements of the optimal transportation matrix; This is the Hilbert curve projection function.
[0023] Secondly, the correlation between the generated scene and the actual weather scene for the next day is another important goal of scene generation. Loss functions based on mean squared error struggle to learn the correlation between multiple lag time series within a daily cycle, making it difficult for the generated weather scene to accurately depict the intensity and decay rate of the real scene. Therefore, this invention utilizes a correlation regularization term to constrain the shape consistency between the generated and real weather scenes within a daily cycle. An ACF regularization term is added to the loss function of the pre-trained denoising diffusion probability model. for: ; in, ; In the formula, To generate the maximum number of hours in a multi-day scene; The time lag is the time delay. Separate real and generated scenes at intervals At time t, the autocorrelation coefficient between time series; Calculate the expected value; , These are time series sequences for real and generated scenarios, respectively. for The mean; To and interval Time series of moments; for The mean; To and interval A time series of moments.
[0024] Finally, the loss function of the feature-guided pre-trained denoising diffusion probability model is constructed. for: ; In the formula, Calculate the expected value; To predict noise; This is real noise; This is the noise prediction loss term; The weight hyperparameters of the probability distribution; The weight hyperparameters for correlation are denoted as .
[0025] Guided by the Hilbert curve projection distance and the autocorrelation coefficient regularization term, the objective of the denoising diffusion probability model expands from minimizing the root mean square error between the real and generated weather scenes to maintaining consistency in probability distribution and temporal correlation. By guiding the learning direction of the denoising diffusion probability model, it enables the model to generate diverse and physically plausible weather scenes for the next day. It should be noted that the task of the feature-guided pre-trained denoising diffusion probability model constructed in this invention is not to generate weather scenes that maintain consistency in probability distribution and temporal correlation with the previous day's input over a long period, but rather to maintain consistency in probability distribution and temporal correlation with the real scene of the following day. After generating the scene for the next day, it can be used as input to continue generating weather scenes for the following day, thus achieving the rolling generation of weather scenes over multiple consecutive days.
[0026] S2. Freeze the pre-trained model in S1 and add it to the ControlNet control branch. Input the weather samples with conditional labels into the ControlNet control branch to train the weather scene generation model.
[0027] For multi-day scenarios across different months, the average daily wind speed, temperature, and irradiance vary significantly. Therefore, when generating multi-day scenarios, it is necessary to incorporate daily average weather and monthly condition information to control the daily weather trends and the severity of the scenarios. To this end, this invention uses daily average wind speed, average temperature, average irradiance, and the month to which the scenario belongs as conditional labels. By training the ControlNet control branch, the output of the pre-trained denoising diffusion probability model is further fine-tuned to meet the conditional control requirements of the scenarios.
[0028] The ControlNet architecture learns conditional labels without disrupting existing pre-trained denoising diffusion probabilistic models, enabling more accurate conditional scene generation. Traditional neural network models have specific input-output structures. To incorporate conditional control, ControlNet adds an extra trainable conditional control branch to the pre-trained diffusion model to learn the mapping relationship between input conditions and the target generated scene. Through parameter freezing and zero-convolution initialization, the model only needs to train the control branch when incorporating different conditional information. ControlNet's control branch typically contains only a few perceptron and convolutional modules, with a much smaller number of parameters than the pre-trained denoising diffusion probabilistic model. This significantly reduces the training parameters of the scene generation model and improves model convergence performance.
[0029] like Figure 2 As shown, in order to achieve fine-grained conditional control, the ControlNet architecture freezes the parameters of the pre-trained denoising diffusion probability model. Furthermore, a trainable branch, ControlNet, is added to the original model. The output of the weather scene with conditional control can then be denoted as: ; In the formula, The control module initialized for zero convolution; For ControlNet control branch parameters; These are the parameters for the pre-trained denoising diffusion probability model; This is the output of the pre-trained denoising diffusion probability model; Input samples for weather data; To incorporate conditional control into the weather scene output, the wind speed scene... Temperature scenarios and irradiation intensity scene composition; This is a conditional control quantity for the average daily wind speed; This refers to the conditional control quantity for the average daily temperature; This is a conditional control quantity for the average daily irradiance. This refers to the conditional control quantities for the month to which the weather scene belongs. Among them, and The calculation method is as follows: ; In the formula, For a day Wind speed at any given moment; For a day Temperature at any moment; For a day The intensity of radiation at any given moment.
[0030] In the ControlNet architecture, input weather data serves as the input to the control branch, preserving the pre-trained model's knowledge-learning capabilities and providing a foundation for subsequent conditional learning networks. Furthermore, to stabilize the model training process, all convolutional layers in ControlNet employ zero-convolution initialization. Therefore, the weather scene output with conditional control added in the early stages of training equals the pre-trained model's output. This prevents random noise from affecting the hidden states of the neural network layers in the initial training phase. The zero-convolution initialization mechanism ensures that the model's initial output matches the original pre-trained model, thus preserving the modeling capabilities learned by the pre-trained model. As training progresses, the model progressively learns the influence of weather conditions, thereby achieving stable conditional generation capabilities. When control conditions change, the scene generation model only needs to fine-tune the control branch without retraining the model, which helps reduce training costs and accelerate model convergence.
[0031] The ControlNet weather condition control branch includes a conditional projection layer, a temporal embedding layer, downsampling blocks, intermediate blocks, upsampling blocks, and a zero-convolutional layer. Two downsampling blocks, each consisting of a fully connected layer, ReLU, and a downsampling projection layer, downsample the input data after concatenating control conditions, increasing the data dimension from 64 to 128 and then to 256. Two intermediate blocks further fuse and extract features with a dimension of 256. Two upsampling blocks, each consisting of a fully connected layer, ReLU, and an upsampling projection layer, progressively reduce the data dimension from 256 to 128 and then to 64. The weights and biases of the zero-convolutional layer are initialized to 0, allowing the ControlNet output to gradually influence the final result during the early stages of training, achieving stable adjustment of the output of the pre-trained denoising diffusion probability model.
[0032] Finally, by using the aforementioned ControlNet-based weather scene condition generation method, wind speed scenes for multiple days can be obtained. Temperature scenarios and irradiation intensity scene .
[0033] S3. Using a weather-to-power conversion model, the generated wind speed scene... Temperature scenarios and irradiation intensity scene Converting to wind / solar power output scenarios yields continuous multi-day renewable energy output scenarios that meet the control requirements.
[0034] The power output of a wind turbine is affected by factors such as wind speed, blade radius, air density, and wind energy utilization factor (FE). The FE is a design parameter derived from simulation results. Air density and blade radius are typically fixed parameters. Therefore, for a specific wind turbine, the manufacturer usually provides the conversion relationship between the turbine's output power and wind speed. Once the wind speed in the area where the turbine is located is known, the active power output of the turbine can be calculated. According to Betz's law, the power of a wind turbine is generally proportional to the cube of the wind speed, and the turbine's output can be calculated using the following formula: ; In the formula, This refers to the active power output of the wind turbine. air density; The radius of the wind turbine blades; The wind energy utilization coefficient; This refers to the hub height of the wind turbine; The coefficient of friction is the wind shear coefficient. For measuring wind speed in wind turbines in a wind farm, from the wind speed scenario Extracted from the data; due to differences in wind speed at different altitudes, the measured wind speed of the wind turbine needs to be converted to a value based on its height above the ground when calculating the turbine power. wind speed at the location .
[0035] Photovoltaic (PV) power generation systems convert solar energy received by photovoltaic (PV) panels into electrical energy using inverters. For a specific PV system, its output is influenced by the interaction of multiple factors. Among weather elements, irradiance is the dominant factor affecting PV output. Higher irradiance results in higher PV output, and vice versa. Temperature and wind speed have relatively lower impacts on PV output, primarily affecting the energy conversion efficiency of the PV panels. Compared to wind turbines, the influence of meteorological information on their output is difficult to accurately represent directly using mechanistic formulas. Therefore, this invention employs a data-driven approach to construct a conversion model between meteorological elements and PV output.
[0036] The power output of a photovoltaic (PV) system under the influence of weather factors can be characterized as a multivariate nonlinear regression problem. The LightGBM gradient boosting tree model is a machine learning algorithm suitable for regression problems. This algorithm progressively corrects the model mapping error by combining multiple decision trees and outputs the PV system's power output mapping model as follows: ; In the formula, This is the predicted value of photovoltaic power; This represents the initial output value of the photovoltaic power. This represents the number of iterations. The total number of integrated regression trees, i.e., the optimal number of iterations; For data in the wind speed sequence, from wind speed scene Extract from; For data in the temperature series, from the irradiance scene Extract from; For data in the irradiance sequence, from irradiance scene Extract from; This is the output of the optimal regression tree.
[0037] In each round of training, the model first calculates the residual for each sample, and then trains the next sample. An optimal regression tree is used. The gradient boosting tree model can accurately calculate photovoltaic power output under the influence of various meteorological factors. In this embodiment, irradiance, temperature, wind speed, and photovoltaic system output are used as sample inputs, and the weather scenario is converted into photovoltaic system output using the LightGBM gradient boosting tree model.
[0038] like Figure 3 As shown, the meteorological factor-guided ControlNet new energy output scenario generation architecture includes three stages: feature-guided pre-trained denoising and diffusion probability model construction, condition-controllable ControlNet scenario generation model construction, and meteorological factor-guided output scenario conversion.
[0039] Example 2 like Figure 4 As shown, a new energy output scenario generation system based on meteorological factors uses the above method, including: Model building module: Constructs a denoising diffusion probability model guided by Hilbert curve projection distance and autocorrelation coefficient features as a pre-trained model to realize the rolling generation of weather scenes over multiple consecutive days; Training module: Freeze the pre-trained model in the model building module and add the ControlNet control branch. Input the weather samples with conditional labels into the ControlNet control branch to train the weather scene generation model. The conversion output module adopts a weather-to-power conversion model to convert the generated weather scene into the output of new energy, thereby obtaining a continuous multi-day new energy output scene that meets the condition control requirements.
[0040] This invention discloses a meteorological factor-guided new energy output scenario generation system that can be installed in a computer device. The computer device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a meteorological factor-guided new energy output scenario generation program. The memory includes at least one type of readable storage medium, including flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic storage, disk, optical disk, etc. The processor is the control core of the electronic device, connecting various components of the computer device via various interfaces and lines. It executes programs or modules stored in the memory and calls data stored in the memory to perform various functions and process data.
[0041] The module described in this invention refers to a series of computer program segments that can be executed by the processor of a computer device and can perform a fixed function, and which are stored in the memory of the computer device.
[0042] The system provided in this embodiment of the invention has the same implementation principle and technical effects as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment.
[0043] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.
[0044] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall fall within the scope of the present invention.
Claims
1. A method for generating new energy output scenarios based on meteorological factors, characterized in that, Includes the following steps: S1. Construct a pre-trained denoising diffusion probability model based on Hilbert curve projection distance and autocorrelation coefficient features to achieve rolling generation of weather scenes over multiple consecutive days; S2. Freeze the pre-trained denoising diffusion probability model in S1 and add it to the ControlNet control branch. Input the weather samples with conditional labels into the ControlNet control branch to train the weather scene generation model. S3. Using a weather-to-power conversion model, the generated weather scenario is converted into the output of new energy sources, resulting in a continuous multi-day new energy output scenario that meets the condition control requirements.
2. The method for generating new energy output scenarios based on meteorological factors according to claim 1, characterized in that, S1 includes the following steps: S11. In order to maintain a strong probability distribution consistency between the generated next day scene and the real next day scene, a Hilbert curve projection distance regularization term is added to the loss function of the denoising diffusion probability model. S12. In order to constrain the consistency of the generated shape with the real weather scene in the daily cycle, an autocorrelation coefficient regularization term is added to the loss function of the denoising diffusion probability model.
3. The method for generating new energy output scenarios based on meteorological factors according to claim 2, characterized in that, The loss function of the denoising diffusion probability model guided by the Hilbert curve projection distance and autocorrelation coefficient features is: ； In the formula, Calculate the expected value; To predict noise; This is real noise; This is the noise prediction loss term; The discrete form of the regularization term of the denoising diffusion probability model is embedded into the probability distribution knowledge, namely, the Hilbert curve projection distance regularization term; The weight hyperparameters of the probability distribution; The weight hyperparameters for the correlation; The regularization term of the denoising diffusion probability model is embedded into the correlation knowledge, namely the autocorrelation coefficient regularization term.
4. The method for generating new energy output scenarios based on meteorological factors according to claim 2, characterized in that, The Hilbert curve projection distance regularization term is: ； In the formula, This is the squared term of the Hilbert curve projection distance between the real and generated samples; The set of joint distributions of marginal distributions; For real samples The probability distribution; To generate samples The probability distribution; These are real samples and generated samples, respectively. For in set Find the infimum in the middle; for One-dimensional numerical value after Hilbert curve projection; for One-dimensional numerical value after Hilbert curve projection; Let be the probability differential element of the joint distribution; For samples in one-dimensional space after Hilbert curve projection and The Euclidean distance.
5. The method for generating new energy output scenarios based on meteorological factors according to claim 2, characterized in that, The autocorrelation coefficient regularization term is: ； in, ； In the formula, To generate the maximum number of hours in a multi-day scene; The time lag is the time delay. Separate real and generated scenes at intervals At time t, the autocorrelation coefficient between time series; Calculate the expected value; , These are time series sequences for real and generated scenarios, respectively. for The mean; To and interval Time series of moments; for The mean; To and interval A time series of moments.
6. The method for generating new energy output scenarios based on meteorological factors according to claim 1, characterized in that, In S2, the condition labels include the daily average wind speed, average temperature, average light intensity, and the month to which the scene belongs.
7. The method for generating new energy output scenarios based on meteorological factors according to claim 1, characterized in that, The ControlNet control branch learns the mapping relationship between input conditions and target generated scenes through parameter freezing and zero convolution initialization, and adjusts the output of the pre-trained denoising diffusion probability model to obtain wind speed scenes covering multiple days. Temperature scenarios and irradiation intensity scene Weather scene output results: ； In the formula, This is a conditional control quantity for the average daily wind speed; This refers to the conditional control quantity for the average daily temperature; This is a conditional control quantity for the average daily irradiance. This refers to the conditional control quantity for the month to which the weather scene belongs; The control module initialized for zero convolution; For ControlNet control branch parameters; These are the parameters for the pre-trained denoising diffusion probability model; This is the output of the pre-trained denoising diffusion probability model; Input samples for weather data.
8. The method for generating new energy output scenarios based on meteorological factors according to claim 1, characterized in that, The weather-to-power conversion model includes a wind turbine output model and a photovoltaic system output mapping model; A power output model for the fan is established by utilizing the proportional relationship between the fan's power and wind speed. The LightGBM gradient boosting tree model is used to gradually correct the model mapping error by combining multiple decision trees and output the power output mapping model of the photovoltaic system.
9. The method for generating new energy output scenarios based on meteorological factors according to claim 8, characterized in that, The power output model of the wind turbine is as follows: ； The output mapping model of the photovoltaic system: ； In the formula, This refers to the active power output of the wind turbine. air density; The radius of the wind turbine blades; The wind energy utilization coefficient; This refers to the hub height of the wind turbine; The coefficient of friction is the wind shear coefficient. For measuring wind speed in wind turbines in a wind farm, from the wind speed scenario Extract from; The height above the ground is wind speed; This is the predicted value of photovoltaic power; This represents the initial output value of the photovoltaic power. This represents the number of iterations. The total number of integrated regression trees, i.e., the optimal number of iterations; For data in the temperature series, from the irradiance scene Extract from; For data in the irradiance sequence, from irradiance scene Extract from; This is the output of the i-th optimal regression tree; For data in the wind speed sequence, from wind speed scene Extract from.
10. A new energy output scenario generation system based on meteorological factors, characterized in that, The method described using any one of claims 1-9 includes: Model building module: Constructs a denoising diffusion probability model guided by Hilbert curve projection distance and autocorrelation coefficient features as a pre-trained model to realize the rolling generation of weather scenes over multiple consecutive days; Training module: Freeze the pre-trained model in S1 and add the ControlNet control branch. Input the weather samples with conditional labels into the ControlNet control branch to train the weather scene generation model. The conversion output module adopts a weather-to-power conversion model to convert the generated weather scene into the output of new energy, thereby obtaining a continuous multi-day new energy output scene that meets the condition control requirements.