New energy probability scenario generation method and device, electronic equipment and storage medium

Abstract

The application discloses a new energy probability scene generation method and device, electronic equipment and storage medium, which are used for solving the problem that current related technologies cannot directly output a limited number of typical wind and light resource scenes with clear occurrence probability, error accumulation leads to low credibility, and long-time span and multi-scene parallel simulation are not conducive. A continuous latent trajectory of future weather is constructed according to a multi-source heterogeneous data construction condition input tensor; a conditional whitening transformation based on conditional mean-covariance joint estimation is performed on the continuous latent trajectory based on the conditional input tensor, a whitening residual trajectory is obtained, and a plurality of whitening latent scenes and scene probabilities are generated through conditional whitening diffusion; the future weather scene under the scene probability is obtained by performing decoding on the whitening latent scene based on inverse whitening; and the multi-scene numerical simulation result of the new energy system is generated based on scene selection and probability distribution according to all future weather scenes.

Description

Technical Field

[0001] This invention relates to the field of numerical simulation technology for new energy, and in particular to a method, apparatus, electronic device, and storage medium for generating probabilistic scenarios for new energy. Background Technology

[0002] As the penetration rate of new energy sources such as wind power and photovoltaics in the power system continues to increase, the random volatility, intermittency, and spatiotemporal coupling of wind and solar resources are becoming increasingly pronounced. Numerical simulation of new energy sources has gradually shifted from traditional deterministic output estimation to probabilistic scenario simulation oriented towards dispatching operations, risk assessment, and multi-scenario decision-making. Especially under the background of high proportion of new energy integration, rapid changes in wind speed, irradiance, cloud cover, temperature, and local weather systems often cause significant deviations in the trajectory of wind and solar resources in short and continuous periods, making simulation methods that rely solely on single-path outputs insufficient to meet engineering requirements such as grid dispatching, reserve capacity allocation, spot market bidding, and energy storage coordinated control.

[0003] Currently, numerical simulation methods for wind and solar resources mainly include physical mechanism-based numerical weather-driven methods, statistical learning-based time series modeling methods, and deep learning-based generative simulation methods. While these methods can achieve a certain level of accuracy in estimating wind and solar resource trajectories in practical applications, they generally suffer from the following shortcomings.

[0004] First, traditional methods primarily output data in the form of single-trajectory simulation, point value prediction, or quantile interval estimation. These methods struggle to directly provide a finite set of wind and solar resource scenarios with clearly defined probabilities of occurrence. Second, wind and solar resource time series exhibit significant non-stationarity, heteroscedasticity, and intervariate correlation. Current methods do not adequately model these statistical structures. Furthermore, in longer prediction time domains or multi-time-step rolling simulations, current autoregressive weather or resource generation methods are prone to error accumulation. These methods typically generate future states gradually over smaller time steps, using the output of the previous step as subsequent input. Therefore, once a deviation occurs in the early stages, the error propagates and amplifies in subsequent periods. When wind and solar resource simulation tasks need to cover multiple consecutive time periods from the ultra-short term to the day-ahead, this error accumulation significantly weakens the consistency and reliability of the scenario trajectories.

[0005] Furthermore, current technologies for generating new energy scenarios largely focus on a "large-scale sampling followed by clustering or reducing scenarios from the samples" approach. While this approach can form a set of typical scenarios to some extent, it still relies on sample approximation. On the one hand, the large-scale sample generation and post-processing steps increase the computational burden. On the other hand, the secondary compression process from samples to scenarios can easily cause probability distortion, leading to deviations between scenario weights and the true distribution. Meanwhile, in numerical simulation tasks for wind and solar resources, weather fields themselves exhibit significant spatial continuity and temporal correlation. Directly generating models from the original high-dimensional weather field or power sequence often results in high dimensionality, difficulty in compression, and significant structural redundancy, which is unfavorable for long-term, parallel simulations across multiple scenarios. Summary of the Invention

[0006] This invention provides a method, apparatus, electronic device, and storage medium for generating new energy probability scenarios, which solves or partially solves the technical problem that current related technologies cannot directly output a limited number of typical wind and solar resource scenarios with clear occurrence probabilities, and the accumulation of errors leads to low reliability, which is not conducive to long-term and multi-scenario parallel simulation.

[0007] This invention provides a method for generating new energy probability scenarios, the method comprising: Acquire multi-source heterogeneous data from the new energy system, and construct a continuous latent trajectory of conditional input tensor and future weather based on the multi-source heterogeneous data; Based on the conditional input tensor, a conditional whitening transformation based on conditional mean-covariance joint estimation is performed on the continuous latent trajectory to obtain the whitened residual trajectory. Based on the whitening residual trajectory, multiple whitening latent scenes are generated through conditional whitening diffusion, along with the scene probability of each whitening latent scene; For each of the whitened latent scenes, the future weather scene under the scene probability is obtained by decoding the whitened latent scene based on dewhitening. Based on all the described future weather scenarios, multi-scenario numerical simulation results of the new energy system are generated based on scenario selection and probability allocation.

[0008] Optionally, the multi-source heterogeneous data includes historical meteorological field sequences, historical station operating status sequences, future numerical weather forecast information, station static parameters, and future weather fields; the step of constructing a continuous latent trajectory of conditional input tensor and future weather based on the multi-source heterogeneous data includes: Based on the historical meteorological field sequence, the historical station operation status sequence, the future numerical weather forecast information, and the station static parameters, a conditional input tensor is constructed through data preprocessing; wherein, the data preprocessing process includes time alignment, spatial registration, anomaly handling, missing data completion, and normalization; The future weather field is encoded into a continuous latent trajectory using a pre-built continuous encoder.

[0009] Optionally, the step of performing a conditional whitening transformation on the continuous latent trajectory based on the conditional input tensor and joint estimation of conditional mean-covariance to obtain a whitened residual trajectory includes: The conditional input tensor is mapped to a conditional representation using a pre-built conditional encoder; The conditional input tensor and the conditional representation are input into a pre-constructed joint conditional estimator, and the conditional mean estimate and lower triangular matrix of the conditional input tensor are determined through joint estimation. Construct the conditional covariance matrix based on the lower triangular matrix; Based on the conditional mean estimate and the conditional covariance matrix, a conditional whitening transformation is performed on the continuous latent trajectory to obtain the whitened residual trajectory.

[0010] Optionally, the step of generating multiple latent whitening scenes through conditional whitening diffusion based on the whitening residual trajectory, and the scene probability of each latent whitening scene, includes: The whitened residual trajectory is randomly masked based on a randomly generated mask vector to obtain a whitened residual sequence. Based on the mask vector, the conditional input tensor and the whitening residual sequence, construct a future latent basis representation based on the conditional Transformer backbone network; The future latent basis representation is iteratively solved by combining forward noise diffusion and reverse noise diffusion. When the preset early stopping control condition is met, the converged basis representation is output. Based on the converged basis representation, multiple scene offset terms are generated through a pre-constructed scene head network, and then the whitened latent scene corresponding to each scene offset term is generated one by one. The pre-built probabilistic head network outputs the scene probability score for each whitened latent scene, and then each scene probability score is subjected to softmax normalization to obtain the scene probability of each whitened latent scene.

[0011] Optionally, obtaining the future weather scene with the scene probability by decoding the whitened latent scene based on dewhitening includes: The whitened latent scene is dewhitened to obtain the complete weather latent trajectory of the whitened latent scene; The complete latent weather trajectory is decoded by a pre-built decoder to obtain the future weather scenario of the whitened latent scene under the scenario probability.

[0012] Optionally, the step of generating multi-scenario numerical simulation results for the new energy system based on scenario selection and probability allocation according to all the future weather scenarios includes: For each of the future weather scenarios, extract key meteorological drivers from the future weather scenarios; Based on the aforementioned key meteorological driving factors, wind power trajectories and photovoltaic power trajectories are constructed. Obtain the energy storage state of charge and charge / discharge efficiency corresponding to the future weather scenario; construct an energy storage state evolution model based on the energy storage state of charge and the charge / discharge efficiency; and obtain the energy storage charge / discharge power trajectory by solving the energy storage state evolution model. Based on the wind power trajectory, the photovoltaic power trajectory, and the energy storage charging and discharging power trajectory, and taking into account the power reduction trajectory, a net power transmission trajectory is constructed. Based on the net power transmission trajectory, construct the wind, solar and energy storage power trajectory scenario of the new energy system under the future weather scenario; Based on all the wind, solar, and energy storage power trajectory scenarios, scenario selection and probability allocation are performed, and multi-scenario numerical simulation results of the new energy system are output. The multi-scenario numerical simulation results include multiple wind, solar, and energy storage power trajectory scenarios, as well as the occurrence probability and scenario label of each wind, solar, and energy storage power trajectory scenario.

[0013] Optionally, the key meteorological driving factors include wind speed, irradiance, and module temperature; the construction of wind power trajectories and photovoltaic power trajectories based on the key meteorological driving factors includes: Based on the wind speed, construct a wind turbine power curve model; Based on the aforementioned wind turbine power curve model, and considering both the rated power of wind power and the wind power curtailment factor, a wind power trajectory is constructed. Based on the irradiance and the module temperature, and taking into account the photovoltaic power limiting factor, inverter efficiency, module efficiency, module area, temperature correction factor and module reference temperature, a photovoltaic power trajectory is constructed.

[0014] The present invention also provides a new energy probability scenario generation device, the device comprising: A conditional input construction unit is used to acquire multi-source heterogeneous data from a new energy system and construct a continuous latent trajectory of the conditional input tensor and future weather based on the multi-source heterogeneous data. The conditional whitening transformation unit is used to perform a conditional whitening transformation on the continuous latent trajectory based on the conditional input tensor and the joint estimation of conditional mean-covariance to obtain the whitened residual trajectory. A conditional whitening diffusion unit is used to generate multiple whitening latent scenes and a scene probability for each whitening latent scene by conditional whitening diffusion based on the whitening residual trajectory. The whitening scene decoding unit is used to obtain the future weather scene under the scene probability by performing anti-whitening decoding on the whitening latent scene for each whitening latent scene. The result generation unit is used to generate multi-scenario numerical simulation results of the new energy system based on scenario selection and probability allocation, according to all the future weather scenarios.

[0015] The present invention also provides an electronic device, the device comprising a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the new energy probability scenario generation method as described above, according to the instructions in the program code.

[0016] The present invention also provides a computer-readable storage medium for storing program code for executing the new energy probability scenario generation method as described in any of the preceding claims.

[0017] As can be seen from the above technical solutions, the present invention has the following advantages: This paper presents a method for generating probabilistic scenarios for new energy sources. The first step involves acquiring multi-source heterogeneous data from a new energy system and constructing a conditional input tensor and a continuous latent trajectory of future weather based on this data. By utilizing multi-source heterogeneous data to construct multi-source conditional inputs, a conditional representation of the future weather field is built. Then, based on continuous encoding, the high-dimensional weather field is compressed into a low-dimensional continuous latent trajectory, thereby reducing the complexity of subsequent generation and modeling while preserving key spatiotemporal evolution features. The second step involves performing a conditional whitening transformation on the continuous latent trajectory based on joint estimation of conditional mean and covariance, obtaining whitened residual trajectories. By performing this conditional whitening transformation, the conditional mean and conditional covariance of the future latent trajectory are estimated from the conditional inputs. The original latent trajectory is then transformed into a whitened residual trajectory, allowing the subsequent conditional diffusion model to primarily learn the high-order residual structure and spatiotemporal coupling relationship. The third step involves generating multiple whitened latent scenarios and their probabilities through conditional whitening diffusion based on the whitened residual trajectories. Building upon the preceding steps, a finite number of representative latent scenarios and their corresponding probabilities are generated in the whitened residual space through conditional whitening diffusion. This achieves the direct output of a finite number of "scenario-probability" pairs in the whitened residual space, rather than relying on a large number of random samples followed by scenario clustering and probability matching. The fourth step involves decoding each whitened latent scenario based on dewhitening to obtain the future weather scenario with the given scenario probability. The fifth step generates multi-scenario numerical simulation results for the new energy system based on scenario selection and probability allocation, based on all future weather scenarios and scenario selection and probability allocation. By dewhitening and decoding each whitened latent scenario into future weather scenarios, each future weather scenario is mapped to a power trajectory scenario. A finite number of new energy numerical simulation results in the form of "scenario-probability" pairs with clear probabilities are directly output based on scenario selection and probability allocation. This achieves a closed-loop output from weather-driven information to wind and solar resource time-series scenarios, and then to multi-scenario numerical simulation results for new energy. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating the steps involved in generating a new energy probability scenario. Figure 2A schematic diagram of the overall process of a new energy probability scenario generation method; Figure 3 This is a structural block diagram of a new energy probability scenario generation device. Detailed Implementation

[0020] This invention provides a method, apparatus, electronic device, and storage medium for generating new energy probability scenarios. These methods address or partially address the technical problem that current related technologies struggle to directly output a limited number of typical wind and solar resource scenarios with clear occurrence probabilities, leading to low reliability due to error accumulation and hindering long-term and multi-scenario parallel simulations.

[0021] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0022] As an example, current numerical simulation methods for wind and solar resources mainly include physical mechanism-based numerical weather-driven methods, statistical learning-based time series modeling methods, and deep learning-based generative simulation methods. While these methods can achieve a certain level of accuracy in estimating wind and solar resource trajectories in practical applications, they generally suffer from the following shortcomings.

[0023] First, traditional methods primarily output data in the form of single-trajectory simulation, point value prediction, or quantile interval estimation. These methods struggle to directly provide a finite set of wind and solar resource scenarios with clearly defined probabilities of occurrence. Second, wind and solar resource time series exhibit significant non-stationarity, heteroscedasticity, and intervariate correlation. Current methods do not adequately model these statistical structures. Furthermore, in longer prediction time domains or multi-time-step rolling simulations, current autoregressive weather or resource generation methods are prone to error accumulation. These methods typically generate future states gradually over smaller time steps, using the output of the previous step as subsequent input. Therefore, once a deviation occurs in the early stages, the error propagates and amplifies in subsequent periods. When wind and solar resource simulation tasks need to cover multiple consecutive time periods from the ultra-short term to the day-ahead, this error accumulation significantly weakens the consistency and reliability of the scenario trajectories.

[0024] Furthermore, current technologies for generating new energy scenarios largely focus on a "large-scale sampling followed by clustering or reducing scenarios from the samples" approach. While this approach can form a set of typical scenarios to some extent, it still relies on sample approximation. On the one hand, the large-scale sample generation and post-processing steps increase the computational burden. On the other hand, the secondary compression process from samples to scenarios can easily cause probability distortion, leading to deviations between scenario weights and the true distribution. Meanwhile, in numerical simulation tasks for wind and solar resources, weather fields themselves exhibit significant spatial continuity and temporal correlation. Directly generating models from the original high-dimensional weather field or power sequence often results in high dimensionality, difficulty in compression, and significant structural redundancy, which is unfavorable for long-term, parallel simulations across multiple scenarios.

[0025] Further analysis reveals the following shortcomings in current numerical simulation technologies for new energy: First, current methods primarily rely on single-path numerical simulation, point value prediction, or quantile interval estimation, making it difficult to directly output a limited number of typical wind and solar resource scenarios with clearly defined probabilities of occurrence. Second, probabilistic simulation methods based on Monte Carlo sampling typically require a large number of samples to approximate future distributions, resulting in unintuitive probability expressions, insufficient coverage of tail scenarios, and high inference costs. Third, the evolution of wind and solar resources exhibits significant non-stationarity, heteroscedasticity, intervariate coupling, and spatiotemporal correlation, making it difficult for traditional methods to uniformly characterize the conditional mean, fluctuation structure, and higher-order residual correlations. Fourth, long-term rolling simulations are prone to error accumulation, leading to a decline in the temporal continuity, physical consistency, and engineering usability of the simulation trajectory.

[0026] In summary, traditional techniques currently lack a unified method for numerical simulation of wind and solar resources. This method should simultaneously possess the following capabilities: First, it should be able to learn the conditional statistical structure of wind and solar resources from historical meteorological data, future weather forecasts, and facility operation information. Second, it should be able to effectively model non-stationarity, heteroscedasticity, and intervariate correlations. Third, it should be able to avoid error accumulation during long-term rolling generation. Fourth, it should be able to directly output a finite number of typical wind and solar resource scenarios with clearly defined probabilities, without relying on extensive Monte Carlo sampling.

[0027] Based on this, it is necessary to propose a new multi-scenario numerical simulation method for new energy to achieve the integrated generation of wind and solar resource trajectories, typical operating scenarios and their occurrence probabilities, so as to better serve the analysis, scheduling decision and risk control of new energy grid-connected operation.

[0028] Therefore, one of the core inventive points of this invention is to propose a multi-scenario new energy probabilistic scenario generation method based on conditional whitening diffusion and probabilistic scenario output. By constructing multi-source conditional inputs such as historical meteorological observations, numerical weather predictions, station operating status, and equipment parameters, the future weather evolution trajectory is first characterized in a continuous latent space. Then, conditional statistical priors of the future weather latent trajectory are extracted using conditional mean-covariance joint estimation and conditional whitening is performed. In the whitening space, a masked conditional diffusion model is used to learn the residual distribution and its spatiotemporal correlation structure. Subsequently, a finite number of "scenario-probability" pairs are directly generated through the probabilistic scenario output head. Each scenario is inversely mapped to the power trajectories of wind power, photovoltaic, and energy storage, realizing a closed-loop output from weather-driven information to wind and solar resource time-series scenarios, and then to the numerical simulation results of multiple new energy scenarios.

[0029] Reference Figure 1 The diagram illustrates a flowchart of a method for generating a new energy probability scenario according to an embodiment of the present invention, which may specifically include the following steps: Step 101: Obtain multi-source heterogeneous data from the new energy system, and construct a continuous latent trajectory of conditional input tensor and future weather based on the multi-source heterogeneous data. This step primarily involves constructing multi-source conditional inputs for numerical simulation of wind and solar resources and establishing a continuous latent representation of the future weather field. The latent trajectory is used to describe hidden, potential paths or patterns within the data or system.

[0030] To more clearly illustrate the technical solution provided by this invention, the following variables and symbols are defined: Let the length of the historical time window be... The prediction time window length is Weather spatial grid size is The dimension of the meteorological variables is The dimension of the continuous latent dimension is The number of scenes is The combined output dimension of wind power, photovoltaics, and energy storage is .

[0031] Define the historical condition input as: ; in, Represents a historical meteorological field sequence; This represents a sequence of historical station operating statuses; This represents future numerical weather forecasts, satellite cloud image features, calendar features, and prior scheduling information; This indicates the static parameters of the power station, including geographical location, installed capacity, unit parameters, component parameters, energy storage parameters, and grid connection boundary parameters.

[0032] Define the future real weather field as Its continuous latent characterization is ,in: ; In the formula, This is a weather encoder with the following parameters: .

[0033] Define the weather decoder as ,satisfy: ; in, This represents the future weather field reconstructed from the latent trajectory.

[0034] In practical applications, this step uses a multi-source condition fusion and continuous latent representation method to handle the high-dimensional weather-driven problem in the numerical simulation of wind and solar resources. The core idea is to use multi-source condition inputs to construct a conditional representation of the future weather field, and then use a continuous encoder to compress the high-dimensional weather field into a low-dimensional continuous latent trajectory, thereby reducing the complexity of subsequent generation and modeling while retaining the main spatiotemporal evolution features.

[0035] The first step is to construct a multi-source conditional input. Historical meteorological fields, historical operational status, future numerical weather predictions, and station static parameters are time-aligned, spatially resampled, outlier corrected, and normalized to obtain a unified conditional input tensor. Among them, historical operating status This may include historical wind power output, historical photovoltaic power output, energy storage status of charge, charging and discharging power, power limitation markers, fault markers, and available capacity ratio.

[0036] Secondly, it involves establishing continuous latent characterizations for future weather fields. Encode the data to obtain the latent trajectory of future weather. ,Right now: ; in, Indicates time The weather latent vector.

[0037] To ensure that the latent trajectory can fully retain the main information of the original weather field, the encoder-decoder is trained using the following reconstruction loss: ; In a preferred embodiment, to enhance the continuity and stability of the latent space, the following regularization term is introduced: ; in, Indicates that it is made by the encoder The posterior distribution of the future weather latent trajectory was obtained; This represents the continuous latent trajectory corresponding to the future weather field; Indicates the future weather field; This represents the standard normal prior distribution; Represents distribution Relative to distribution Kullback-Leibler divergence (KL divergence); symbol It represents the direction relationship of divergence between two probability distributions.

[0038] The overall objective function for constructing the related steps of multi-source conditional input and continuous latent representation of future weather fields can be expressed as: ; in, This is the regularization weighting coefficient.

[0039] Finally, the output is the latent trajectory for subsequent conditional statistical modeling. After training, the main spatiotemporal variations of the future weather field are compressed into continuous latent trajectories. This serves as the foundational input for subsequent conditional mean-covariance estimation and diffusion modeling. Through the above processing, the original high-dimensional weather field generation problem can be transformed into a conditional probability modeling problem in latent space.

[0040] Based on the preceding discussion, in the specific implementation, multi-source heterogeneous data includes historical meteorological field sequences, historical station operational status sequences, future numerical weather prediction information, station static parameters, and future weather fields. The implementation process for constructing a conditional input tensor and a continuous latent trajectory of future weather based on multi-source heterogeneous data can include: constructing a conditional input tensor through data preprocessing based on historical meteorological field sequences, historical station operational status sequences, future numerical weather prediction information, and station static parameters; the data preprocessing process includes time alignment, spatial registration, anomaly handling, missing data completion, and normalization; and encoding the future weather field into a continuous latent trajectory using a pre-constructed continuous encoder.

[0041] Based on the content described in the preceding embodiments, step 101 can actually be divided into two main implementation steps. The first is data preprocessing and condition information construction, and the second is condition coding and future weather latent representation construction.

[0042] For data preprocessing and conditional information construction, specifically, historical meteorological field sequences, historical station operation status sequences, future numerical weather prediction information, and station static parameters can be collected, and time alignment, spatial registration, anomaly handling, missing data completion, and normalization can be performed to form a conditional input tensor. .

[0043] The aforementioned historical meteorological field sequences mainly include one or more of the following: wind speed, wind direction, irradiance, temperature, humidity, cloud cover, and air pressure. The historical power station operating status sequences mainly include one or more of the following: wind power output, photovoltaic power output, energy storage status of charge, energy storage charging and discharging power, available capacity ratio, power limitation markers, and fault markers. The power station static parameters mainly include one or more of the following: geographical location, installed capacity, wind turbine parameters, photovoltaic module parameters, energy storage parameters, and grid connection boundary parameters.

[0044] The above conditions are input tensors Operators constructed from conditions It is composed of historical meteorological field sequences, historical station operation status sequences, future numerical weather forecast information, and station static parameters.

[0045] More specifically, in the data preparation phase, it is necessary to complete the unified organization and conditional representation of multi-source heterogeneous data. First, historical meteorological field sequences are extracted from historical meteorological database reanalysis data. .in, This indicates the length of the historical time window, and the spatial grid size of the meteorological field is [value missing]. The dimension of the meteorological variables is .therefore, The meteorological variables include, but are not limited to, near-ground wind speed, wind direction, total irradiance, diffuse irradiance, ambient temperature, cloud cover, humidity, and air pressure.

[0046] At the same time, historical operating status sequences are extracted from the station's SCADA system. .in, This indicates the dimension of the operating state variables. The operating state variables include historical wind power output, photovoltaic power output, energy storage state of charge, energy storage charging and discharging power, available capacity ratio, power limitation flags, and fault flags, etc.

[0047] Furthermore, exogenous driving information for future forecast periods can be obtained from numerical weather prediction systems. .in, This indicates the length of the future forecast time window. Static parameters are read from the site's static database. The static parameters include geographical location, rated capacity of the unit, wind turbine cut-in wind speed, cut-out wind speed, rated wind speed, photovoltaic module area, inverter efficiency, upper limit of energy storage capacity, grid connection boundary parameters, etc.

[0048] After performing uniform time-scale resampling, missing value completion, out-of-limit truncation, and dimension normalization on the above data, a conditional input tensor is constructed. : ; This step transforms the raw, multi-source, heterogeneous data into a unified conditional input that can be used for subsequent modeling.

[0049] For conditional coding and the construction of latent representations of future weather, the conditional input tensor is... Mapping to conditional representation and the future weather field Encoded as a continuous latent trajectory This study aims to constrain the distribution of conditions for future weather evolution and characterize the driving role of exogenous factors in changes in wind and solar resources.

[0050] Among them, the above conditions characterize Conditional coding network Input a tensor based on the given conditions. Extracted latent weather trajectory Weather encoder Forecast of future weather It is obtained by performing a continuous mapping.

[0051] In addition, this step also involves building a weather decoder. To make future weather latent trajectory Reconfigurable into a future weather field The conditional encoding and latent representation construction process employs reconstruction error constraints and can further introduce continuous distribution regularization terms to ensure the latent representation's ability to preserve the main spatiotemporal structure of the future weather field.

[0052] In the conditional coding phase, the conditional coding network is first constructed. Input the conditions into the tensor The mapping is represented by the following compact conditional representation: ; in, It represents a conditional coding vector that comprehensively reflects historical weather evolution, historical station responses, future weather forecast information, and static boundary parameters.

[0053] Then, regarding future weather patterns Constructing a continuous encoder With decoder This compresses the future weather field into a continuous latent trajectory: ; in, Indicates time The weather latent vector; This represents the latent feature dimension.

[0054] Correspondingly, the decoder satisfies: ; in, This represents the weather field reconstructed from the latent trajectory.

[0055] To ensure that the latent representation both compresses high-dimensional weather information and preserves the main spatiotemporal structure, the following reconstruction error constraints are adopted during the training phase: ; If necessary, a continuous distribution regularization term can be further introduced as shown below: ; This forms the overall objective function for the encoding stage: ; in, This represents the regularization weight coefficient.

[0056] Through the above steps, the problem of numerical simulation of high-dimensional future weather fields can be transformed into a conditional trajectory modeling problem in continuous latent space, creating conditions for subsequent conditional whitening diffusion modeling.

[0057] Step 102: Based on the conditional input tensor, perform a conditional whitening transformation on the continuous latent trajectory based on the joint estimation of conditional mean and covariance to obtain the whitened residual trajectory. Based on the aforementioned embodiments, this step mainly involves performing conditional mean-covariance joint estimation on the continuous latent trajectory of future weather using the conditional input tensor, and then performing conditional whitening transformation to obtain the whitened residual trajectory.

[0058] Among them, the conditional mean-covariance joint estimation and conditional whitening method are used to deal with the non-stationarity, heteroscedasticity and correlation problems in the future weather latent trajectory. The core idea is to first estimate the conditional mean and conditional covariance of the future latent trajectory from the conditional input, and then transform the original latent trajectory into a whitened residual trajectory, so that the subsequent conditional diffusion model mainly learns the high-order residual structure and spatiotemporal coupling relationship.

[0059] Specifically, this invention first constructs a conditional mean-covariance joint estimator using the following formula: ; in, This represents the set of conditional mean estimates for each time point within the future forecast period; This represents the set of lower triangular matrices at each point in time within the future prediction period; and They are respectively the first of the Conditional mean estimate and lower triangular matrix at each time step; This represents the global conditional representation extracted from the conditional input; These are the parameters of the joint estimator.

[0060] The conditional covariance matrix is ​​further defined as follows: ; in, These are positive definite correction coefficients; It is an identity matrix.

[0061] Perform a conditional whitening transformation, that is, use the conditional mean estimate and conditional covariance to transform the time step... latent trajectory Transformed into the whitened residual latent vector at the corresponding time step : ; Correspondingly, its anti-whitening relationship can be expressed as: ; in, Represents the conditional covariance matrix The inverse square root matrix is ​​used to perform conditional whitening; Represents the conditional covariance matrix The square root matrix is ​​used to perform dewhitening recovery.

[0062] After conditional whitening, the whitening residual trajectory This indicates that removing the conditional mean estimate and linear covariance structure from the residual latent variable can significantly mitigate the adverse effects of trend terms, fluctuation scale differences, and nonlinear correlation terms in the original trajectory on the training of the generative model.

[0063] Establish a joint estimation loss function. To make... and It also possesses high estimation accuracy, and the mean fitting loss is defined as: ; Define the empirical covariance of the sliding window as The covariance fitting loss includes the Frobenius norm term. With the nuclear standard number of items : ; ; To avoid To address issues such as excessively small eigenvalues, ill-conditioned matrices, or rank degeneration, the eigenvalue constraint term is defined as follows: ; in, The minimum feature value threshold; for The Each feature value.

[0064] The total loss function for the conditional whitening transformation steps based on the joint estimation of conditional mean and covariance is shown below: ; in, The weights are constrained by the eigenvalues.

[0065] Based on the above modeling, this step outputs the conditionally whitened residual trajectory. Conditional mean estimation and conditional covariance This is used for subsequent diffusion generation and scene output.

[0066] Based on the preceding discussion, the implementation process of performing a conditional whitening transformation on a continuous latent trajectory based on joint estimation of conditional mean and covariance, to obtain a whitened residual trajectory, can include: mapping the conditional input tensor to a conditional representation using a pre-built conditional encoder; inputting the conditional input tensor and conditional representation to a pre-built conditional joint estimator to determine the conditional mean estimate and lower triangular matrix of the conditional input tensor through joint estimation; constructing a conditional covariance matrix based on the lower triangular matrix; and performing a conditional whitening transformation on the continuous latent trajectory based on the conditional mean estimate and conditional covariance matrix to obtain a whitened residual trajectory.

[0067] Based on the content described in the previous embodiments, step 102 mainly involves conditional mean-covariance estimation and conditional whitening modeling. Specifically, it utilizes the conditional input tensor... and conditional representation Latent trajectory of future weather The conditional mean and conditional covariance are jointly estimated, and the original latent trajectory is transformed into a conditionally whitened residual trajectory. .

[0068] In this step, a joint mean-covariance estimator is constructed. Output the future weather latent trajectory based on the conditional input tensor. Conditional mean estimation Lower triangular matrix sequence and conditional statistical representation Construct the conditional covariance matrix from the lower triangular matrix sequence. And based on conditional mean Conditional covariance Perform a conditional whitening transformation on the future weather latent trajectory to obtain the whitened residual trajectory. .

[0069] Conditional covariance matrix The conditional covariance matrix is ​​constructed using a lower triangular matrix product plus a positive definite correction term to ensure that the conditional covariance matrix is ​​a positive definite matrix. The joint mean-covariance estimation process can also introduce a mean fitting term, a covariance Frobenius norm constraint term, a covariance kernel norm constraint term, and a minimum eigenvalue penalty term to improve the estimation stability of the conditional mean and conditional covariance.

[0070] In the conditional statistical modeling stage, a joint mean-covariance estimator is constructed. The conditional input is mapped to a conditional mean estimate, a lower triangular factor matrix, and a global conditional representation, denoted as: ; in, Indicates time Conditional mean estimate; Indicates time A lower triangular matrix; Indicates conditional statistical characteristics.

[0071] The conditional covariance matrix is ​​further defined as follows: ; in, These are positive definite correction coefficients; It is an identity matrix.

[0072] Based on the conditional mean and conditional covariance, a conditional whitening transformation is performed on the future weather latent trajectory: ; in, Indicates time The whitened residual latent vector.

[0073] Correspondingly, its anti-whitening relationship is: ; To improve the estimation accuracy of the mean and covariance, and to suppress the ill-conditioned matrix problem caused by excessively small eigenvalues, the following loss function is constructed during the training phase: The mean fitting term is: ; The covariance Frobenius norm term is: ; The covariance norm terms are: ; And the eigenvalue penalty term is: ; in, Represents the empirical covariance matrix of the sliding window; The minimum feature value threshold; for The Each feature value.

[0074] This leads to a joint estimated loss: ; in, The weights are constrained by the eigenvalues.

[0075] Through the above steps, the conditional trends, fluctuation scale differences, and linear correlation structures in the original latent trajectory are explicitly stripped away, providing a foundation for the subsequent diffusion model to focus on learning higher-order residual distributions and complex spatiotemporal coupling structures.

[0076] Step 103: Based on the whitening residual trajectory, generate multiple whitening latent scenes through conditional whitening diffusion, and the scene probability of each whitening latent scene; Based on the aforementioned embodiments, this step mainly utilizes a masked conditional diffusion model within the conditional whitening space to recover the basic features of the future latent trajectory, generating multi-scenario weather latent scenarios and the scenario probabilities corresponding to each latent scenario.

[0077] Among them, the mask-based conditional whitening diffusion method is used to deal with the problem of generating multiple scenarios of future weather latent trajectories. Its core idea is to learn the conditional distribution of future trajectories in the whitening residual space through random masking and diffusion denoising, and then generate a finite number of representative latent scenarios through a multi-scenario query mechanism.

[0078] Specifically, the masked conditional input and diffusion forward process are first established. The residual trajectory after conditional whitening is then analyzed. Apply random mask This yielded partially visible and partially missing whitened sequences. .set up Representing timing Whether it is concealed or not, then: ; Construct the conditional Transformer backbone network as shown below: ; in, Indicates the conditional Transformer backbone network; This represents the trainable parameters of the backbone network; Represents a conditional input tensor; Indicates the trajectory of the whitened residual The mask sequence obtained after random masking; Represents a random mask matrix or mask vector; This represents the future latent basis characterization.

[0079] During the diffusion training phase, for forward noisy diffusion, the whitened residual variable is subjected to forward noisy addition as shown below: ; in, For the true whitening residual trajectory, For the diffusion step; Indicates from step 1 to step 2. The cumulative noise scheduling coefficient of the step; This represents Gaussian noise that follows a standard normal distribution. ; Represents the identity matrix; Indicates the first The noisy whitening residual variable obtained from each diffusion step; Indicates the first The noise scheduling coefficient for each diffusion step.

[0080] For inverse denoising diffusion, the main approach is to establish diffusion denoising loss and near-end deterministic constraints. Among these, the diffusion network... Based on noise samples diffusion step and substrate characterization The loss function for predicting the noise term is defined as follows: ; in, Represents a noise prediction network; Indicates the first The diffusion step is the first The noisy whitened residual variable at each time step; Representing the future latent basis characterization In the The weight of a moment; Indicates the first The true Gaussian noise at each moment; Indicates the first This moment was concealed and used as a target for diffusion recovery.

[0081] To enhance the accuracy of resource simulation in the near term, the preceding period was... Add deterministic constraints to each prediction time: ; in, For decay weights; This is the attenuation coefficient.

[0082] The basic training loss for the above-mentioned whitening diffusion-related steps is: ; in, These are deterministic constraint weights.

[0083] During the inference phase, multiple latent scenarios (i.e., multiple whitened latent scenarios) are generated, and an early stopping control mechanism is set. Specifically, during the inference phase, future trajectories are initialized with a fully masked state, and the latent basis representation is recovered through iterative revelation. The iteration terminates when the following early stopping criterion is met: ; in, Indicates the first The future latent basis characterization revealed after the next iteration; Indicates the first The future latent basis characterization revealed after the next iteration; This indicates the pre-set early stop threshold. Indicates the number of times the early stopping criterion needs to be met consecutively; This represents the Frobenius norm.

[0084] If the above conditions are met consecutively If the second convergence is achieved, then the basis representation is considered to have converged.

[0085] In obtaining convergent basis characterization Then, set Query vectors for each scenario The scene offset term is generated by the scene head network: ; Thus, we can obtain the first A whitened latent scene : ; At the same time, the probability head network outputs the first... Unnormalized probability score for each scenario The corresponding scene probabilities are obtained by normalizing using softmax. : ; ; in, Indicates the first The probability logit of a whitened latent scene; Indicates the first The probability of occurrence of a whitened latent scene (i.e., scene probability); This is the probability temperature coefficient.

[0086] Through the above processing, this step outputs... A whitened latent scene and its probability: ; Based on the preceding discussion, the specific implementation process for generating multiple whitened latent scenes through conditional whitening diffusion based on the whitening residual trajectory, and the implementation of the scene probability of each whitened latent scene, can include: performing random masking processing on the whitening residual trajectory based on a randomly generated mask vector to obtain a whitening residual sequence; constructing a future latent basis representation based on the mask vector, conditional input tensor, and whitening residual sequence using a conditional Transformer backbone network; iteratively solving the future latent basis representation by combining forward noise diffusion and backward noise denoising diffusion, and outputting the converged basis representation when a preset early stopping control condition is met; generating multiple scene offset terms through a pre-constructed scene head network based on the converged basis representation, and then generating the whitened latent scene corresponding to each scene offset term one by one; outputting the scene probability score of each whitened latent scene through a pre-constructed probability head network, and then performing softmax normalization processing on each scene probability score one by one to obtain the scene probability of each whitened latent scene.

[0087] Based on the content described in the preceding embodiments, step 103 mainly involves conditional whitening diffusion modeling and multi-scene latent scene generation. Specifically, let the future prediction time window length be... Define a random vector of whitened residual trajectory. A conditional diffusion forward perturbation model and a reverse denoising generation model are established in the whitened residual space to obtain the whitened residual trajectory in the conditional input tensor. The conditional distribution is obtained by considering the following conditions. The mathematical description is provided, and a finite number of scenarios and their probabilities are output through a multi-scenario query header.

[0088] Among them, the whitening residual trajectory Apply random mask The mask sequence is obtained. and through the conditional backbone network Constructing future latent basis representations The conditional diffusion model establishes a forward denoising process and a backward denoising process within the whitened residual space to obtain the whitened residual trajectory with respect to the conditional input tensor. The conditional distribution.

[0089] Furthermore, the forward diffusion process in this step is constructed by progressively adding Gaussian noise to the whitened residual trajectory, while the back diffusion process employs a noise prediction network. Predict the noise term for each diffusion step and establish a conditional inverse distribution. Introducing variance modulation coefficients during the reverse reasoning stage. It is used to adjust the sampling variance of each diffusion step in order to control scene diversity and generation stability.

[0090] In this step, the conditional whitening diffusion modeling also introduces deterministic constraints for the near-term period, which affect the future prediction window. Weighted reconstruction constraints are applied to the latent trajectory at each time point to improve the accuracy of numerical simulations in the near-terminal period.

[0091] In this step, the inference phase employs an iterative revelation mechanism to recover the future latent basis representation. Furthermore, an early stopping control criterion is established. Specifically, when the difference between the latent basis representations obtained from two adjacent iterations is lower than a preset threshold, and this threshold is met for a preset number of consecutive iterations... In such cases, the reverse inference process can be terminated early to reduce online computational overhead.

[0092] This step sets Query vectors for each scenario Scene-generated head network Generate the corresponding scene offset item And by the probability output head network Output the first Unnormalized probability score for each scenario The corresponding scene probabilities are obtained by normalizing using softmax. This results in a finite number of whitened latent scenes and a discrete probability scene distribution of their corresponding probabilities.

[0093] More specifically, in the diffusion modeling stage, a random mask is first applied to the residual trajectory after conditional whitening. The mask sequence is obtained. And construct future latent basis representations through conditional backbone networks: ; in, Represents the future latent basis characterization; This represents the parameters of the backbone network.

[0094] In the forward diffusion process, the first... The noise variables for each diffusion step are: ; in, Represents the true whitening residual trajectory; , This represents the diffusion scheduling coefficient.

[0095] The conditional forward distribution can be obtained as follows: ; During backward reasoning, a noisy prediction network is used. To approximate the noise term, establish a conditional inverse distribution: ; in, Indicates the first The conditional mean function of the step; Indicates the first The sampling variance of the step.

[0096] Furthermore, to control the scene diversity and stability during the inference phase, a variance modulation factor is introduced. Backsampling can be written as: ; in, For the first The variance modulation coefficient of the step. When Taking a larger value is beneficial for enhancing scene dispersion; when Taking a smaller value is beneficial for improving the focus and stability of the scene.

[0097] During the training phase, diffused noise is used to predict the loss. ; And a deterministic constraint is introduced for the near-term period: ; in, Indicates the number of proximal constraint steps; Indicates the decay weight. This represents the attenuation coefficient.

[0098] The inference phase employs an iterative revelation mechanism to recover the future basis. And set the following early termination criteria: ; When the above conditions are met consecutively a preset number of times The inference can be terminated early to reduce online computational overhead.

[0099] Based on this, set Query vectors for each scenario The scene offset is generated through the scene head network: ; Thus, the first A whitened latent scene: ; At the same time, the probability head network outputs the first... Unnormalized probability score for each scenario The corresponding scene probabilities are obtained by normalizing using softmax. : ; ; in, This represents the probability temperature parameter.

[0100] This yields the discrete probability scenario distribution: ; in, It represents the Dirac metric.

[0101] Through the above steps, a finite number of "scene-probability" pairs in the whitening residual space are directly output, instead of relying on a large number of random samples and then performing scene clustering and probability matching.

[0102] Step 104: For each whitened latent scene, obtain the future weather scene with the scene probability by performing anti-whitening decoding on the whitened latent scene; Building upon the aforementioned embodiments, this step primarily involves dewhitening and decoding each whitened latent scene into a future weather scenario. Based on this, it can be mapped to wind power, photovoltaic, and energy storage power trajectories. By performing probability allocation, scenario selection, and risk quantification on each power trajectory scenario, a finite number of "scenario-probability" pairs of new energy numerical simulation results are output.

[0103] Among them, the anti-whitening, weather decoding and wind-solar-storage power mapping methods are used to deal with the engineering implementation problem of transforming whitened latent scenarios into new energy resource trajectories. The core idea is to first restore the whitened latent scenario to the weather latent trajectory, then decode it into the future weather scenario, and finally use the wind power, photovoltaic and energy storage mechanism models to map it into the multi-scenario power trajectory that can be used in actual operation.

[0104] First, dewhitening is performed on each whitened latent scene. For the... For a whitened latent scene, the anti-whitening formula is: ; Based on this, we can obtain the first... Complete weather latent trajectory for a whitened latent scene .

[0105] Next, future weather scenarios can be obtained through weather decoding. Specifically, by inputting the latent trajectories of various weather events into the decoder, future weather scenarios can be obtained: ; Based on the preceding discussion, in a specific implementation, the process of obtaining the future weather scene under scene probability by decoding the whitened latent scene based on dewhitening can include: performing dewhitening processing on the whitened latent scene to obtain the complete weather latent trajectory of the whitened latent scene; and decoding the complete weather latent trajectory through a pre-built decoder to obtain the future weather scene of the whitened latent scene under scene probability.

[0106] Step 105: Based on all the future weather scenarios, generate multi-scenario numerical simulation results of the new energy system based on scenario selection and probability allocation.

[0107] This step mainly maps the future weather scenarios obtained in the previous steps into power trajectories for wind power, photovoltaic power, and energy storage. By performing probability allocation, scenario selection, and risk quantification on each power trajectory scenario, a finite number of new energy numerical simulation results in the form of "scenario-probability" pairs are output.

[0108] Specifically, wind speed can be extracted from weather scenes. Irradiance Component temperature Cloud cover Key meteorological drivers are used as inputs for wind and solar power resource mapping. Then, the power mapping relationship between wind power, solar power, and energy storage is constructed.

[0109] Among them, the wind power scenario power (wind power trajectory) is defined as: ; in, This refers to the rated power of the wind power. This refers to the wind power availability ratio or the wind power limitation factor. The wind turbine power curve function satisfies: ; The photovoltaic scenario power (photovoltaic power trajectory) is defined as: ; in, This refers to the photovoltaic availability ratio or the power generation limitation factor. For inverter efficiency; For component efficiency; For component area; This is a temperature correction factor; For component temperature; Indicates the reference temperature.

[0110] Energy storage state evolution is defined as: ; And satisfy the following constraints: ; ; ; In summary, the net power transmission trajectory can be obtained as follows: ; in, Indicates the first A future weather scenario at any time Net external power of new energy; Indicates wind power output; Indicates photovoltaic power; Indicates the energy storage discharge power; Indicates the energy storage charging power; This indicates the power reduction caused by power rationing, wind curtailment, or solar curtailment.

[0111] Therefore, the first The power trajectories for wind, solar, and energy storage corresponding to each future weather scenario are as follows: ; Furthermore, the core idea of ​​the engineering decision expression problem based on probabilistic scenario output and risk quantification method for processing multi-scenario results is: by jointly optimizing scenario accuracy, scenario probability and distribution consistency, the model can directly output a finite number of new energy scenarios with clear probabilities, and further form reserve, deviation and tail risk indicators.

[0112] Specifically, we first establish the scenario selection and reconstruction loss. Let the actual net power transmission trajectory be... Then, the scene index that is closest to the actual trajectory is selected based on the following formula: ; Define the scene reconstruction loss as: ; Next, we establish the scene probability loss and entropy constraints. To ensure that correct scenes have a higher probability allocation, we define the probability loss as: ; To prevent excessive probability compression to individual scenarios or excessive averaging leading to insufficient scenario diversity, scenario entropy is defined as: ; Simultaneously, the entropy constraint loss is set as follows: ; in, and These represent the lower and upper bounds of the scene probability entropy, respectively.

[0113] Then, a distributional consistency loss is established and "scenario-probability" pairs are output.

[0114] To ensure consistency between the entire scene set and the real distribution, the weighted CRPS loss is defined as: ; In summary, the total loss of the relevant steps in processing multi-scenario results based on probability scenario output and risk quantification methods is: ; in, , , and All are loss weights.

[0115] Finally, the overall training objective function of this invention can be expressed as: ; in, This represents the overall weighting coefficient.

[0116] After training is completed, this invention outputs numerical simulation results for multiple new energy scenarios in the following form: ; in, For the first The trajectory scenarios of wind, solar and energy storage power corresponding to a future weather scenario; This represents the probability of this scenario occurring. For scene tags.

[0117] The above scenario tags include, but are not limited to, scenarios with frequent sunny weather, scenarios with localized cloud cover, scenarios with rising gusts of wind, scenarios with sudden power restrictions, scenarios with capacity reduction due to faults, and scenarios with energy storage compensation linkage.

[0118] In a preferred embodiment, new energy operation risk indicators can be further calculated based on the probability of the scenario occurring. The confidence level is set to... Then the lower quantile reserve requirement can be expressed as: ; ; in, Indicates time At confidence level The lower quantile boundary of net power exported from new energy sources; Indicates the candidate power threshold; Indicates the first The scene at any moment Net power output; Indicates the first The probability of each scenario occurring; Indicates at time The net power transmitted to the outside does not exceed the threshold. A set of scene numbers.

[0119] Based on this, backup configuration boundaries, deviation penalty risk boundaries, or energy storage pre-scheduling boundaries can be generated.

[0120] Based on the preceding discussion, in the specific implementation, after obtaining multiple future weather scenarios, key meteorological driving quantities can be extracted from each future weather scenario. Then, based on these key meteorological driving quantities, wind power trajectories and photovoltaic power trajectories are constructed. The energy storage state of charge and charge / discharge efficiency corresponding to the future weather scenarios are obtained. Based on the energy storage state of charge and charge / discharge efficiency, an energy storage state evolution model is constructed, and by solving the energy storage state evolution model, the energy storage charge / discharge power trajectory is obtained. Then, based on the wind power trajectory, photovoltaic power trajectory, and energy storage charge / discharge power trajectory, while also considering the power reduction trajectory, a net power transmission trajectory is constructed. Based on the net power transmission trajectory, wind, solar, and energy storage power trajectory scenarios for the new energy system under the future weather scenarios are constructed. Scenario selection and probability allocation are performed based on all wind, solar, and energy storage power trajectory scenarios, and the multi-scenario numerical simulation results of the new energy system are output. The multi-scenario numerical simulation results include multiple wind, solar, and energy storage power trajectory scenarios, as well as the occurrence probability and scenario label of each wind, solar, and energy storage power trajectory scenario.

[0121] Furthermore, key meteorological drivers can include wind speed, irradiance, and module temperature. Based on these key meteorological drivers, the implementation process for constructing wind power trajectories and photovoltaic power trajectories can specifically include: constructing a wind turbine power curve model based on wind speed; constructing a wind power trajectory based on the wind turbine power curve model, considering both rated wind power and the wind curtailment factor; and constructing a photovoltaic power trajectory based on irradiance and module temperature, considering the photovoltaic curtailment factor, inverter efficiency, module efficiency, module area, temperature correction factor, and module reference temperature.

[0122] Based on the content described in the preceding embodiments, steps 104 to 105 mainly involve weather decoding, wind, solar, and energy storage power mapping, and probabilistic scenario output. Specifically, the future weather scenario is obtained by dewhitening and weather decoding based on the whitened latent scenario, and then mapped to wind power, solar power, and energy storage power trajectory scenarios. A finite number of new energy numerical simulation results in the form of "scenario-probability" pairs are output, completing scenario probability optimization and risk quantification.

[0123] Among them, by the first A whitened latent scene Perform the dewhitening operation to obtain the future weather latent scene. and through the weather decoder Get the corresponding future weather scenario .

[0124] From future weather scenarios Wind speed, irradiance, module temperature, and other meteorological driving factors are extracted. Based on the wind turbine power curve model, the photovoltaic irradiance-temperature correction model, and the energy storage charge-discharge equation, the wind power trajectory, photovoltaic power trajectory, and energy storage charge-discharge trajectory are calculated respectively. Furthermore, the net power export trajectory and the wind-solar-storage power trajectory are constructed. .

[0125] The wind power trajectory is determined by the rated power of the wind turbine, the wind turbine power curve function, and the available wind power ratio or curtailment factor. The photovoltaic power trajectory is determined by irradiance, module temperature, inverter efficiency, module efficiency, module area, and the available photovoltaic ratio or curtailment factor. The energy storage charge / discharge trajectory is determined by the energy storage state of charge, charge / discharge power constraints, energy storage capacity constraints, and charge / discharge efficiency.

[0126] In this step, we assume the actual renewable energy power trajectory is... Based on the actual power trajectory of new energy sources and the power trajectory of various scenarios The distance between them determines the optimal matching scenario index. Furthermore, a scene reconstruction loss term and a probability loss term are constructed to improve the consistency between the output scene and the real trajectory, as well as the probability calibration capability of the corresponding scene.

[0127] This step further constructs a weighted distribution consistency loss term. and scene entropy constraint This is to suppress excessive collapse or averaging of scene probabilities. The scene entropy constraint term sets a lower bound on the probability entropy. and the upper bound of probability entropy This is used to control the degree of concentration and dispersion in the distribution of scene probabilities.

[0128] The final output of this step is the numerical simulation results of new energy in multiple scenarios. .in, Indicates the first One new energy power trajectory scenario (i.e., wind, solar and energy storage power trajectory scenario). This indicates the probability of the scenario occurring; This indicates a scenario label. Scenario labels include one or more of the following: frequent sunny weather scenarios, localized cloud cover scenarios, gust wind scenarios, sudden power rationing scenarios, fault-induced capacity reduction scenarios, and energy storage compensation linkage scenarios.

[0129] The limited number of "scenario-probability" pairs output in this step can be used to further calculate one or more of the following: reserve demand boundary, tail risk boundary, power deviation risk index, energy storage pre-dispatch boundary, or new energy absorption capacity index, in order to serve new energy power plant operation analysis, grid dispatch, market transaction auxiliary decision-making, and risk control.

[0130] In the result generation stage, the first step is to... Perform dewhitening operation on a whitened latent scene: ; Then, the weather decoder obtains the first A future weather scenario: ; Extract wind speed from the above future weather scenarios Irradiance Component temperature Key meteorological driving factors are then mapped to new energy power trajectories.

[0131] Wind power generation is mapped using a mechanism: ; in, This refers to the rated power of the wind power. This refers to the wind power availability ratio or the wind power limitation factor. This is a function representing the power curve of the wind turbine.

[0132] Photovoltaic power is achieved using an irradiance-temperature correction model: ; in, Indicates the first The scene at any moment Photovoltaic power; This refers to the photovoltaic availability ratio or the power generation limitation factor. For inverter efficiency; For component efficiency; For component area; This is a temperature correction factor; For component temperature; Indicates the reference temperature.

[0133] The energy storage state adopts the following state equation: ; And satisfy: ; ; ; in, Indicates the first The scene at any moment The energy storage state of charge; and These represent charging power and discharging power, respectively. and These represent charging efficiency and discharging efficiency, respectively.

[0134] Furthermore, construct the net power transmission trajectory: ; No. The renewable energy power trajectory (wind, solar, and energy storage power trajectory scenario) for each scenario is as follows: ; To ensure that the scene probability matches the true distribution, let the true power trajectory be... Define the optimal matching scenario index: ; And construct the scene reconstruction loss: ; Probabilistic loss: ; And weighted distribution consistency loss: ; Meanwhile, to avoid over-scaling or over-averaging of scene probabilities, scene entropy is defined as follows: ; And construct the entropy constraint term: ; in, and These represent the lower and upper bounds of the scene probability entropy, respectively.

[0135] This results in the total loss at the output stage: ; Finally, this invention outputs numerical simulation results for multiple scenarios of new energy: ; in, For the first The trajectory scenarios of wind, solar and energy storage power corresponding to a future weather scenario; This represents the probability of this scenario occurring. These are scenario tags. The scenario tags include, but are not limited to, scenarios with frequent sunny weather, scenarios with localized cloud cover, scenarios with rising gusts of wind, scenarios with sudden power rationing, scenarios with capacity reduction due to faults, and scenarios with energy storage compensation linkage.

[0136] Based on the probability of the aforementioned scenarios, the lower quantile reserve demand, tail risk boundary, and energy storage pre-scheduling indicators can be further calculated, thereby directly serving the operation analysis of new energy power plants, grid dispatch, and risk control.

[0137] In some embodiments, the present invention also provides a dynamic parameter tuning method for adapting the foregoing embodiments to special scenarios. Specifically, in scenarios with sudden wind speed changes, the variance modulation coefficient is increased. The range of values ​​for which the scene probability entropy is determined, and the upper bound of the scene probability entropy is increased. To enhance scenario diversity, in scenarios with persistent cloud cover or sudden irradiance drops, the temperature correction coefficient and irradiance weight in the morning mapping of the photovoltaic field are adjusted, and the deterministic constraint weight in the near-term period is increased to enhance the accuracy of short-term scenarios. In scenarios with power curtailment or faults, the constraint sensitivity corresponding to the wind power availability ratio coefficient and the photovoltaic availability ratio coefficient is increased, and the scenario reconstruction loss weight is increased to improve the identification and fitting ability of abnormal operation scenarios. In scenarios with changes in grid boundaries or changes in energy storage dispatch mode, the static parameters of the power plant are updated. The energy storage state constraint parameters are adjusted and the corresponding scenario probability distribution is recalculated to ensure the model's adaptability to special operating conditions.

[0138] In summary, the overall flowchart of the new energy probability scenario generation method provided in this embodiment of the invention is shown in the figure below. Figure 2 As shown.

[0139] In this embodiment of the invention, a method for generating multi-scenario probabilistic new energy scenarios based on conditional whitening diffusion and probabilistic scenario output is proposed. First, by constructing multi-source conditional inputs including historical meteorological observations, numerical weather predictions, station operating status, and equipment parameters, the future weather evolution trajectory is characterized in a continuous latent space. Then, conditional statistical priors of the future weather latent trajectory are extracted using conditional mean-covariance joint estimation, and conditional whitening is performed to generate whitened residual trajectories. Second, a masked conditional diffusion model is used in the whitened space to learn the residual distribution and its spatiotemporal correlation structure. Subsequently, a finite number of "scenario-probability" pairs are directly generated through the probabilistic scenario output head. Finally, each scenario is inversely mapped to wind power, photovoltaic, and energy storage power trajectories, achieving a closed-loop output from weather-driven information to wind and solar resource time-series scenarios, and then to the numerical simulation results of multiple new energy scenarios. By implementing the technical solution provided by this invention, the non-stationarity, heteroscedasticity, and spatiotemporal correlation structure in the evolution of wind and solar resources can be explicitly characterized, improving the coverage of tail risk scenarios and the stability of multi-timescale numerical simulations. It is applicable to the probabilistic scenario generation, operation analysis, and scheduling decision-making of wind farms, photovoltaic power plants, wind-solar-storage integrated systems, and regional new energy clusters.

[0140] Compared with traditional technologies, the technical solution provided by this invention has at least the following beneficial effects: (1) The ability to output probabilistic scenarios is significantly enhanced. This invention transforms the traditional approach of approximating distribution through a single path or a large number of random samples into directly outputting a finite number of typical wind and light resource scenarios with clear probabilities of occurrence. In a preferred embodiment, the number of scenarios N can be set to 4 to 12, thereby reducing the complexity of post-processing of scenarios and improving the efficiency of direct access to results by the scheduling side while maintaining the ability to express the main distribution characteristics.

[0141] (2) The ability to characterize extreme disturbances and tail risks is significantly improved. By combining conditional whitening diffusion with probabilistic scenario output, this invention can simultaneously retain high-probability conventional scenarios and low-probability high-impact scenarios. It has a stronger ability to express abnormal situations such as local cloud cover, gust surge, sudden drop in irradiance, sudden power restriction and equipment failure derating, thereby enhancing the risk characterization ability of new energy numerical simulation under extreme weather and abnormal operating conditions.

[0142] (3) Spatiotemporal correlation modeling capability and simulation stability are improved simultaneously. This invention first explicitly addresses the non-stationarity, heteroscedasticity and linear correlation structure in the evolution of wind and solar resources through conditional mean-covariance joint estimation and conditional whitening mechanism, thereby improving the stability and accuracy of multivariate numerical simulation of wind and solar resources. Then, it uses a masked diffusion model to learn the high-order residual structure and complex spatiotemporal coupling relationship, and reduces the error accumulation in long-term rolling inference, thereby improving the temporal consistency and spatial coordination between scene trajectories.

[0143] (4) Enhanced engineering applicability and decision support capabilities. This invention achieves closed-loop modeling of wind power, photovoltaic, and energy storage power trajectory scenarios, encompassing historical meteorological information, future numerical weather forecasts, and power station status parameters. The output results not only include scenario trajectories but also occurrence probabilities and interpretable labels, directly serving standby configuration, spot market pricing, energy storage linkage, and new energy consumption analysis. It is more suitable for deployment in real-time auxiliary decision-making systems on the new energy power station side, dispatch side, and virtual power plant side.

[0144] Specifically, this invention is applicable to the construction of probabilistic scenarios and multi-timescale numerical simulations of wind and solar resources in wind farms, photovoltaic power plants, wind-solar-storage integrated systems, and regional-level new energy clusters. It is particularly suitable for numerical simulations of wind and solar resources on timescales ranging from 15 minutes to several days, probabilistic scenario generation, output boundary analysis, and risk verification. Simultaneously, this invention can serve multi-scenario operation analysis in power grid dispatching, reserve capacity configuration, spot market scenario assessment, energy storage collaborative control, virtual power plant aggregation optimization and new energy absorption capacity research, wind and solar resource fluctuation propagation analysis under extreme weather disturbances, and multi-scenario operation safety assessment.

[0145] Furthermore, the data foundation involved in this invention includes, but is not limited to, historical meteorological observation data, numerical weather prediction products, satellite cloud images or reanalysis meteorological fields, SCADA operation data of wind turbines and photovoltaic equipment, power generation restriction records, and multi-source heterogeneous data such as the geographical location, capacity parameters and grid connection boundary information of wind turbines and photovoltaic equipment.

[0146] Reference Figure 3 The diagram illustrates a structural block diagram of a new energy probability scenario generation device provided in an embodiment of the present invention, which may specifically include: The conditional input construction unit 301 is used to acquire multi-source heterogeneous data from the new energy system and construct a continuous latent trajectory of the conditional input tensor and future weather based on the multi-source heterogeneous data. The conditional whitening transformation unit 302 is used to perform a conditional whitening transformation on the continuous latent trajectory based on the conditional input tensor and the joint estimation of conditional mean-covariance to obtain the whitened residual trajectory. Conditional whitening diffusion unit 303 is used to generate multiple whitening latent scenes and the scene probability of each whitening latent scene by conditional whitening diffusion according to the whitening residual trajectory. Whitening scene decoding unit 304 is used to obtain the future weather scene under the scene probability by performing anti-whitening decoding on the whitening latent scene for each whitening latent scene. The result generation unit 305 is used to generate multi-scenario numerical simulation results of the new energy system based on scenario selection and probability allocation according to all the future weather scenarios.

[0147] In one optional embodiment, the multi-source heterogeneous data includes historical meteorological field sequences, historical station operation status sequences, future numerical weather forecast information, station static parameters, and future weather fields; the conditional input construction unit 301 includes: The conditional input tensor construction unit is used to construct a conditional input tensor based on the historical meteorological field sequence, the historical station operation status sequence, the future numerical weather forecast information, and the station static parameters through data preprocessing; wherein, the data preprocessing process includes time alignment, spatial registration, anomaly handling, missing data completion, and normalization; A continuous latent trajectory encoding unit is used to encode the future weather field into a continuous latent trajectory using a pre-built continuous encoder.

[0148] In one alternative embodiment, the conditional whitening transformation unit 302 includes: A conditional representation mapping unit is used to map the conditional input tensor into a conditional representation through a pre-built conditional encoder; The joint estimation unit is used to input the conditional input tensor and the conditional representation into a pre-constructed conditional joint estimator, and determine the conditional mean estimate and lower triangular matrix of the conditional input tensor through joint estimation. A conditional covariance matrix construction unit is used to construct a conditional covariance matrix based on the lower triangular matrix. The conditional whitening transformation execution unit is used to perform conditional whitening transformation on the continuous latent trajectory based on the conditional mean estimate and the conditional covariance matrix to obtain the whitened residual trajectory.

[0149] In one alternative embodiment, the conditional whitening diffusion unit 303 includes: A random masking unit is used to perform random masking processing on the whitened residual trajectory based on a randomly generated masking vector to obtain a whitened residual sequence. The future latent basis representation construction unit is used to construct the future latent basis representation based on the conditional Transformer backbone network according to the mask vector, the conditional input tensor and the whitening residual sequence. The converged basis representation generation unit is used to iteratively solve the future latent basis representation by combining forward noise diffusion and reverse noise denoising diffusion. When the preset early stopping control condition is met, the converged basis representation is output. The whitened latent scene generation unit is used to generate multiple scene offset terms through a pre-constructed scene head network based on the converged basis representation, and then generate the whitened latent scene corresponding to each scene offset term one by one. The scene probability generation unit is used to output the scene probability score of each whitened latent scene through a pre-built probability head network, and then perform softmax normalization on each scene probability score to obtain the scene probability of each whitened latent scene.

[0150] In one optional embodiment, the whitening scene decoding unit 304 includes: The anti-whitening processing unit is used to perform anti-whitening processing on the whitened latent scene to obtain the complete weather latent trajectory of the whitened latent scene. The decoding unit is used to decode the complete weather latent trajectory through a pre-built decoder to obtain the future weather scene of the whitened latent scene under the scene probability.

[0151] In one optional embodiment, the result generation unit 305 includes: A key meteorological driving force extraction unit is used to extract key meteorological driving forces from each of the future weather scenarios. The wind power and photovoltaic power trajectory construction unit is used to construct wind power trajectories and photovoltaic power trajectories based on the key meteorological driving quantities. The energy storage charging and discharging power trajectory construction unit is used to obtain the energy storage state of charge and charging and discharging efficiency corresponding to the future weather scenario, construct an energy storage state evolution model based on the energy storage state of charge and the charging and discharging efficiency, and obtain the energy storage charging and discharging power trajectory by solving the energy storage state evolution model. The net power transmission trajectory construction unit is used to construct the net power transmission trajectory based on the wind power trajectory, the photovoltaic power trajectory and the energy storage charging and discharging power trajectory, while also considering the power reduction trajectory. The wind, solar and energy storage power trajectory scenario construction unit is used to construct the wind, solar and energy storage power trajectory scenario of the new energy system under the future weather scenario based on the net power transmission trajectory. The multi-scenario numerical simulation result output unit is used to select and assign probabilities based on all the wind, solar and energy storage power trajectory scenarios, and output the multi-scenario numerical simulation results of the new energy system. The multi-scenario numerical simulation results include multiple wind, solar and energy storage power trajectory scenarios, as well as the occurrence probability and scenario label of each wind, solar and energy storage power trajectory scenario.

[0152] In one optional embodiment, the key meteorological driving factors include wind speed, irradiance, and component temperature; the wind power and photovoltaic power trajectory construction unit includes: The wind turbine power curve model construction unit is used to construct a wind turbine power curve model based on the wind speed. The wind power trajectory construction unit is used to construct the wind power trajectory based on the wind turbine power curve model, taking into account both the wind power rated power and the wind power curtailment coefficient. A photovoltaic power trajectory construction unit is used to construct a photovoltaic power trajectory based on the irradiance and the module temperature, while also considering the photovoltaic power limiting factor, inverter efficiency, module efficiency, module area, temperature correction factor and module reference temperature.

[0153] As the device embodiment is basically similar to the method embodiment, it is described in a relatively simple way. For relevant details, please refer to the description of the method embodiment above.

[0154] This invention also provides an electronic device, which includes a processor and a memory: The memory is used to store program code and transfer the program code to the processor; The processor is used to execute the new energy probability scenario generation method of any embodiment of the present invention according to the instructions in the program code.

[0155] This invention also provides a computer-readable storage medium for storing program code, which is used to execute the new energy probability scenario generation method of any embodiment of this invention.

[0156] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0157] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this invention are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0158] In the embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units through some interfaces, and may be electrical, mechanical, or other forms.

[0159] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0160] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0161] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0162] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating new energy probability scenarios, characterized in that, include: Acquire multi-source heterogeneous data from the new energy system, and construct a continuous latent trajectory of conditional input tensor and future weather based on the multi-source heterogeneous data; Based on the conditional input tensor, a conditional whitening transformation based on conditional mean-covariance joint estimation is performed on the continuous latent trajectory to obtain the whitened residual trajectory. Based on the whitening residual trajectory, multiple whitening latent scenes are generated through conditional whitening diffusion, along with the scene probability of each whitening latent scene; For each of the whitened latent scenes, the future weather scene under the scene probability is obtained by decoding the whitened latent scene based on dewhitening. Based on all the described future weather scenarios, multi-scenario numerical simulation results of the new energy system are generated based on scenario selection and probability allocation.

2. The method for generating new energy probability scenarios according to claim 1, characterized in that, The multi-source heterogeneous data includes historical meteorological field sequences, historical station operation status sequences, future numerical weather forecast information, station static parameters, and future weather fields; the construction of a continuous latent trajectory of conditional input tensor and future weather based on the multi-source heterogeneous data includes: Based on the historical meteorological field sequence, the historical station operation status sequence, the future numerical weather forecast information, and the station static parameters, a conditional input tensor is constructed through data preprocessing; wherein, the data preprocessing process includes time alignment, spatial registration, anomaly handling, missing data completion, and normalization; The future weather field is encoded into a continuous latent trajectory using a pre-built continuous encoder.

3. The method for generating new energy probability scenarios according to claim 1, characterized in that, The conditional whitening transformation based on the conditional input tensor and joint estimation of conditional mean-covariance on the continuous latent trajectory to obtain the whitened residual trajectory includes: The conditional input tensor is mapped to a conditional representation using a pre-built conditional encoder; The conditional input tensor and the conditional representation are input into a pre-constructed joint conditional estimator, and the conditional mean estimate and lower triangular matrix of the conditional input tensor are determined through joint estimation. Construct the conditional covariance matrix based on the lower triangular matrix; Based on the conditional mean estimate and the conditional covariance matrix, a conditional whitening transformation is performed on the continuous latent trajectory to obtain the whitened residual trajectory.

4. The method for generating new energy probability scenarios according to claim 1, characterized in that, The step of generating multiple latent whitening scenes through conditional whitening diffusion based on the whitening residual trajectory, and the scene probability of each latent whitening scene, includes: The whitened residual trajectory is randomly masked based on a randomly generated mask vector to obtain a whitened residual sequence. Based on the mask vector, the conditional input tensor and the whitening residual sequence, construct a future latent basis representation based on the conditional Transformer backbone network; The future latent basis representation is iteratively solved by combining forward noise diffusion and reverse noise diffusion. When the preset early stopping control condition is met, the converged basis representation is output. Based on the converged basis representation, multiple scene offset terms are generated through a pre-constructed scene head network, and then the whitened latent scene corresponding to each scene offset term is generated one by one. The pre-built probabilistic head network outputs the scene probability score for each whitened latent scene, and then each scene probability score is subjected to softmax normalization to obtain the scene probability of each whitened latent scene.

5. The method for generating new energy probability scenarios according to claim 1, characterized in that, The step of obtaining the future weather scene with the scene probability by decoding the whitened latent scene based on dewhitening includes: The whitened latent scene is dewhitened to obtain the complete weather latent trajectory of the whitened latent scene; The complete latent weather trajectory is decoded by a pre-built decoder to obtain the future weather scenario of the whitened latent scene under the scenario probability.

6. The method for generating new energy probability scenarios according to any one of claims 1 to 5, characterized in that, The process of generating multi-scenario numerical simulation results for the new energy system based on all the future weather scenarios, and on the basis of scenario selection and probability allocation, includes: For each of the future weather scenarios, extract key meteorological drivers from the future weather scenarios; Based on the aforementioned key meteorological driving factors, wind power trajectories and photovoltaic power trajectories are constructed. Obtain the energy storage state of charge and charge / discharge efficiency corresponding to the future weather scenario; construct an energy storage state evolution model based on the energy storage state of charge and the charge / discharge efficiency; and obtain the energy storage charge / discharge power trajectory by solving the energy storage state evolution model. Based on the wind power trajectory, the photovoltaic power trajectory, and the energy storage charging and discharging power trajectory, and taking into account the power reduction trajectory, a net power transmission trajectory is constructed. Based on the net power transmission trajectory, construct the wind, solar and energy storage power trajectory scenario of the new energy system under the future weather scenario; Based on all the wind, solar, and energy storage power trajectory scenarios, scenario selection and probability allocation are performed, and the multi-scenario numerical simulation results of the new energy system are output. The multi-scenario numerical simulation results include multiple wind, solar, and energy storage power trajectory scenarios, as well as the occurrence probability and scenario label of each wind, solar, and energy storage power trajectory scenario.

7. The method for generating new energy probability scenarios according to claim 6, characterized in that, The key meteorological drivers include wind speed, irradiance, and module temperature; the construction of wind power trajectories and photovoltaic power trajectories based on the key meteorological drivers includes: Based on the wind speed, construct a wind turbine power curve model; Based on the aforementioned wind turbine power curve model, and considering both the rated power of wind power and the wind power curtailment factor, a wind power trajectory is constructed. Based on the irradiance and the module temperature, and taking into account the photovoltaic power limiting factor, inverter efficiency, module efficiency, module area, temperature correction factor and module reference temperature, a photovoltaic power trajectory is constructed.

8. A new energy probability scenario generation device, characterized in that, include: A conditional input construction unit is used to acquire multi-source heterogeneous data from a new energy system and construct a continuous latent trajectory of the conditional input tensor and future weather based on the multi-source heterogeneous data. The conditional whitening transformation unit is used to perform a conditional whitening transformation on the continuous latent trajectory based on the conditional input tensor and the joint estimation of conditional mean-covariance to obtain the whitened residual trajectory. A conditional whitening diffusion unit is used to generate multiple whitening latent scenes and a scene probability for each whitening latent scene by conditional whitening diffusion based on the whitening residual trajectory. The whitening scene decoding unit is used to obtain the future weather scene under the scene probability by performing anti-whitening decoding on the whitening latent scene for each whitening latent scene. The result generation unit is used to generate multi-scenario numerical simulation results of the new energy system based on scenario selection and probability allocation, according to all the future weather scenarios.

9. An electronic device, characterized in that, The device includes a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the new energy probability scenario generation method according to any one of claims 1-7 according to the instructions in the program code.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store program code, which is used to execute the new energy probability scenario generation method according to any one of claims 1-7.

Images