Medium and long term forecasting method, system, equipment and medium for wind and light meteorological resources

Abstract

The invention discloses a medium-and-long-term forecasting method, system and equipment for wind-light meteorological resources and a medium. The method comprises the following steps: acquiring meteorological data at the current moment; carrying out climate modal feature extraction on the meteorological data at the current moment to obtain a feature vector; substituting the feature vector into a trained multi-modal fusion forecasting model for calculation to obtain a forecasting result of the period to be predicted; wherein the multi-modal fusion forecasting model is constructed by taking output data obtained by inputting a feature vector into a random forest model as input data of a diffusion model; according to the method, through climate modal feature extraction, the explanation capability of medium and long-term scale wind and light resource changes is improved; end-to-end prediction is realized through a multi-modal fusion prediction model, and the prediction precision and reliability are improved; through combination of wavelet analysis and band-pass filtering, enhancement or weakening of the MJO in different ENSO stages can be identified.

Description

A method, system, equipment and medium for medium- and long-term forecasting of wind, solar and meteorological resources. Technical Field

[0001] This application relates to the field of new energy technology, specifically to a method, system, equipment, and medium for medium- and long-term forecasting of wind, solar, and meteorological resources. Background Technology

[0002] Traditional medium- and long-term weather forecasts primarily rely on numerical models, such as the Weather Research and Forecasting Model (WRF) or the European Centre for Medium-Range Weather Forecasts (ECMWF) model. However, their forecast accuracy is limited by the sensitivity to initial conditions and the constraints of physical parameterization schemes, particularly at sub-seasonal to seasonal scales (10-90 days), where forecasting skill is relatively low. Furthermore, while existing methods consider some climatic factors, they fail to fully exploit the real-time phase propagation characteristics of the tropical intraseasonal oscillation (MJO) and the dynamic impact of seasonal sea surface temperature anomalies associated with El Niño-Southern Oscillation (ENSO) on wind and solar resources. This results in insufficient accuracy in forecasts at regional scales and for extreme events, failing to meet the urgent need for high-precision, probabilistic data in the renewable energy sector. For example, existing numerical forecasting models often neglect the modulation effect of the MJO phase relative to the East Asian radiation field and do not integrate generative AI models (such as diffusion models) with physical constraints, leading to inaccurate uncertainty quantification.

[0003] Therefore, the problems of short forecast lead time, insufficient accuracy, and weak uncertainty quantification in existing technologies urgently need to be solved. Summary of the Invention

[0004] To address the problems of short forecast lead time, insufficient accuracy, and weak uncertainty quantification in existing technologies, this application proposes a medium- to long-term forecasting method for wind, solar, and meteorological resources. The method includes: acquiring meteorological data at the current moment, including the tropical atmospheric intraseasonal oscillation index, the El Niño-Southern Oscillation index, and historical ERA5 data; extracting climate modal features from the current meteorological data to obtain feature vectors; and substituting the feature vectors into a trained multimodal fusion forecasting model for calculation to obtain the forecast result for the period to be predicted. The multimodal fusion forecasting model is constructed by using the output data obtained from the feature vector input to a random forest model as input data to a diffusion model.

[0005] Preferably, the construction process of the multimodal fusion forecast model includes: acquiring meteorological data at historical times; extracting climate modal features from the meteorological data at historical times to obtain a historical feature vector set; training a random forest model based on the historical feature vector set as input data and the radiation and wind speed anomalies of the period to be predicted as target variables; inputting the radiation and wind speed anomalies output by the random forest model into a UNet architecture and training it in conjunction with physical constraints to obtain a diffusion model; and fusing the random forest model and the diffusion model to construct a multimodal fusion forecast model.

[0006] Preferably, it also includes: optimizing the multimodal fusion forecast model by analyzing meteorological data from historical moments to determine the weights of seasonal adaptive factors and the rules for correcting extreme event biases.

[0007] Preferably, before the climate modal feature extraction, the method further includes: performing outlier identification, removal, and spatial noise smoothing operations based on the tropical atmospheric intraseasonal oscillation index, El Niño-Southern Oscillation index, and ERA5 historical data to obtain a clean dataset; extracting the intraseasonal oscillation signal using Lanczos bandpass filtering based on the clean dataset to obtain a filtered signal; calculating the power spectrum using Morlet wavelet analysis based on the clean dataset to obtain a power spectrum; calculating the coherence between the tropical atmospheric intraseasonal oscillation index and the El Niño-Southern Oscillation index based on the clean dataset to obtain a coherence coefficient; and performing standardization operations on the filtered signal, power spectrum, and coherence coefficient.

[0008] Preferably, the meteorological data is subjected to climate mode feature extraction to obtain feature vectors, including: based on the phase corresponding to the tropical atmospheric intraseasonal oscillation index, querying the mapping relationship database to determine the quantitative influence of the current phase on shortwave radiation and wind speed in the target area; using empirical orthogonal function analysis to extract key spatial modes of the solar radiation field and calculating the correlation between the mode and the El Niño-Southern Oscillation Index; calculating the conditional probability of each phase occurring under different El Niño-Southern Oscillation stages based on the phases corresponding to different El Niño-Southern Oscillation stages; correcting the quantitative influence law based on the conditional probability of each phase occurring under different El Niño-Southern Oscillation stages to obtain the corrected influence law; generating feature vectors based on the quantitative influence law, key spatial modes and correlations, and the corrected influence law; the mapping relationship database is obtained through composite analysis of historical tropical atmospheric intraseasonal oscillation indices and their corresponding phases.

[0009] Preferably, the calculation based on feature vectors combined with a pre-constructed multimodal fusion forecast model to obtain forecast results for a future set period includes: calculating the radiation and wind speed anomalies for the future set period using a random forest algorithm based on the feature vectors to obtain point estimates and preliminary probability distributions; and generating a high-resolution ensemble forecast field as the forecast result for the period to be predicted through a progressive denoising process based on the point estimates and preliminary probability distributions of radiation and wind speed anomalies for the future set period.

[0010] Based on the same concept, this application also proposes a medium- to long-term forecasting system for wind, solar, and meteorological resources, comprising: a meteorological data acquisition module for acquiring meteorological data at the current moment, the meteorological data including: tropical atmospheric intraseasonal oscillation index, El Niño-Southern Oscillation index, and ERA5 historical data; a climate model feature extraction module for extracting climate modal features from the meteorological data at the current moment to obtain feature vectors; and a model calculation module for substituting the feature vectors into a trained multimodal fusion forecasting model for calculation to obtain the forecast results for the period to be predicted; wherein, the multimodal fusion forecasting model is constructed based on the output data obtained by inputting the feature vectors into a random forest model, which is then used as the input data for a diffusion model.

[0011] Preferably, the system further includes a model building module; the model building module is specifically used for: acquiring meteorological data at historical moments; extracting climate modal features from the meteorological data at historical moments to obtain a historical feature vector set; training a random forest model based on the historical feature vector set as input data and the radiation and wind speed anomalies of the period to be predicted as target variables; inputting the radiation and wind speed anomalies output by the random forest model into the UNet architecture and training it in conjunction with physical constraints to obtain a diffusion model; and fusing the random forest model and the diffusion model to construct a multimodal fusion forecast model.

[0012] Preferably, the model building module further includes: optimizing the multimodal fusion forecast model by analyzing meteorological data from historical moments to determine the weights of seasonal adaptive factors and the rules for correcting extreme event deviations.

[0013] Preferably, the system further includes a data processing module; the data processing module is specifically used for: identifying, removing, and smoothing spatial noise based on the tropical atmospheric intraseasonal oscillation index, the El Niño-Southern Oscillation index, and historical ERA5 data to obtain a clean dataset; extracting the intraseasonal oscillation signal using Lanczos bandpass filtering based on the clean dataset to obtain a filtered signal; calculating the power spectrum using Morlet wavelet analysis based on the clean dataset to obtain a power spectrum; calculating the coherence between the tropical atmospheric intraseasonal oscillation index and the El Niño-Southern Oscillation index based on the clean dataset to obtain a coherence coefficient; and standardizing the filtered signal, power spectrum, and coherence coefficient.

[0014] Preferably, the climate model feature extraction module is specifically used for: querying a mapping database based on the phase corresponding to the tropical atmospheric intraseasonal oscillation index to determine the quantitative influence of the current phase on shortwave radiation and wind speed in the target area; extracting key spatial modes of the solar radiation field using empirical orthogonal function analysis and calculating the correlation between the mode and the El Niño-Southern Oscillation Index; calculating the conditional probability of each phase occurring under different El Niño-Southern Oscillation stages based on the phases corresponding to different El Niño-Southern Oscillation stages; correcting the quantitative influence law based on the conditional probability of each phase occurring under different El Niño-Southern Oscillation stages to obtain the corrected influence law; and generating a feature vector based on the quantitative influence law, key spatial modes and correlations, and the corrected influence law; the mapping database is obtained through composite analysis of historical tropical atmospheric intraseasonal oscillation indices and their corresponding phases.

[0015] Preferably, the model calculation module is specifically used for: calculating the radiation and wind speed anomalies in the future set period using the random forest algorithm based on the feature vector, and obtaining the point estimates and preliminary probability distributions of the radiation and wind speed anomalies in the future set period; and generating a high-resolution ensemble forecast field as the forecast result for the period to be predicted through a progressive denoising process based on the point estimates and preliminary probability distributions of the radiation and wind speed anomalies in the future set period.

[0016] Furthermore, this application also proposes an electronic device, comprising: at least one processor and a memory; the memory and the processor are connected via a bus; the memory is used to store one or more programs; when the one or more programs are executed by the at least one processor, a long-term forecasting method for wind, solar and meteorological resources as described above is implemented.

[0017] Furthermore, this application also proposes a readable storage medium on which an executable program is stored, which, when executed, implements a long-term forecasting method for wind, solar and meteorological resources as described above.

[0018] Compared with existing technologies, the beneficial effects of this application are as follows: a method, system, equipment, and medium for medium- and long-term forecasting of wind, solar, and meteorological resources, comprising: acquiring meteorological data at the current moment, wherein the meteorological data includes: tropical atmospheric intraseasonal oscillation index, El Niño-Southern Oscillation index, and ERA5 historical data; extracting climate modal features from the meteorological data at the current moment to obtain feature vectors; substituting the feature vectors into a trained multimodal fusion forecasting model for calculation to obtain the forecast result for the period to be predicted; wherein, the multimodal fusion forecasting model is constructed based on the output data obtained by inputting the feature vectors into a random forest model, which is then used as the input data for a diffusion model; this application improves the interpretability of medium- and long-term wind and solar resource changes through climate modal feature extraction; achieves end-to-end forecasting through a multimodal fusion forecasting model, improving forecast accuracy and reliability; and the combination of wavelet analysis and bandpass filtering in this application can identify the enhancement or weakening of the MJO at different ENSO stages. Attached Figure Description

[0019] Figure 1 is a flowchart of a medium- and long-term forecasting method for wind, solar and meteorological resources according to this application; Figure 2 is a flowchart of a medium- and long-term forecasting method for wind, solar and meteorological resources based on MJO and ENSO modes according to this application; Figure 3 is a structural diagram of a medium- and long-term forecasting system for wind, solar and meteorological resources according to this application; Figure 4 is a diagram of the electronic equipment according to this application. Detailed Implementation

[0020] This invention aims to construct a multimodal fusion model for medium- and long-term wind, solar, and meteorological resources by combining the spatiotemporal propagation characteristics of MJO and ENSO with machine learning techniques. This model addresses the problems of short forecast lead times, insufficient accuracy, and weak uncertainty quantification in existing technologies, providing highly reliable and customized support for new energy planning, power grid dispatching, and meteorological disaster early warning. To better understand this application, the following description, in conjunction with the accompanying drawings and embodiments, further illustrates the content of this application.

[0021] Example 1: A medium- to long-term forecasting method for wind, solar and meteorological resources. The specific process is shown in Figure 1, including: Step 1, acquiring meteorological data at the current moment, the meteorological data including: tropical atmospheric intraseasonal oscillation index, El Niño-Southern Oscillation index and ERA5 historical data; Step 2, extracting climate modal features from the meteorological data at the current moment to obtain feature vectors; Step 3, substituting the feature vectors into a trained multimodal fusion forecasting model for calculation to obtain the forecast results for the period to be predicted; wherein, the multimodal fusion forecasting model is constructed based on the output data obtained by inputting the feature vectors into a random forest model, which is then used as the input data for a diffusion model.

[0022] Before step 1, the process includes constructing a multimodal fusion forecasting model, as follows: Constructing the multimodal fusion forecasting framework: Random Forest Model: Input features include ocean-atmosphere driving factors (such as NINO3.4, NAO, PDO index, MJO amplitude / phase), and target variables are radiation and wind speed anomalies for the next two months. Model training uses historical data (1979-2020), employing 100 decision trees with a maximum depth of 15, and optimizing hyperparameters through grid search. Outputs are point estimates of radiation / wind speed and feature importance rankings.

[0023] The Diffusion Model (DDPM) is a generative model using the UNet architecture. Inputs include a noisy weather field (added with Gaussian noise) and conditional variables (MJO phase and ENSO index). A high-resolution weather field set is generated through progressive denoising (T=1000 steps), increasing the resolution to 0.1°×0.1°. Physical constraints such as energy conservation regularization are incorporated into the model. Training uses the Adam optimizer, and the loss function is the mean squared error plus KL divergence.

[0024] Fusion method: Random forest output is used as the initial condition of the diffusion model to achieve end-to-end prediction, generating 50-100 set members and quantifying uncertainty.

[0025] Seasonal dynamic adjustment: The factor weights are adaptively adjusted according to the season. For example, the NAO weight is increased to 0.4 in winter (based on historical correlation analysis), and the ENSO weight is increased to 0.5 in summer. Dynamic optimization is achieved using an attention mechanism.

[0026] Ensemble processing: Deterministic forecasts are calculated by ensemble averaging, and probability distributions are generated by kernel density estimation (e.g., the probability of radiation exceeding the threshold is >80%). Bias corrections are made specifically for extreme events (e.g., wind speed >15m / s).

[0027] Validation and iteration: The model is evaluated using cross-validation, with RMSE as the metric, and the weights are updated iteratively.

[0028] Step 1 involves acquiring meteorological data at the current moment. This meteorological data includes the tropical atmospheric seasonal oscillation index, the El Niño-Southern Oscillation index, and historical ERA5 data. Specifically, this includes: firstly, acquiring historical data, including the ERA5 reanalysis dataset provided by the European Centre for Medium-Range Weather Forecasts (ECMWF), which covers variables such as shortwave radiation (unit: W / m2), emitted longwave radiation (OLR, unit: W / m2), and near-surface wind field (10m wind speed, unit: m / s). The temporal resolution is hourly, the spatial resolution is 0.25°×0.25°, covering the entire globe, and the time span is from 1979 to the present.

[0029] MJO Tropical Atmospheric Seasonal Oscillation Index: Real-time Madden-Julian Oscillation (RMM) index, including amplitude and phase 1-8, obtained from NOAA or BOM.

[0030] ENSO El Niño-Southern Oscillation Index: NINO3.4 sea surface temperature anomaly index, Indian Ocean Dipole (IOD) index, etc., obtained from NOAA Climate Prediction Center.

[0031] Next, data preprocessing is performed: Quality control: Identify and remove outliers, such as radiation values ​​that exceed the physical range of 0-1366 W / m2; use median filtering to smooth noise.

[0032] Time-frequency analysis: Lanczos bandpass filtering is applied to extract intraseasonal oscillation signals; Morlet wavelet analysis is used to extract significant periods, for example, to identify the dominant MJO signal with a period of 20-90 days; wavelet power spectrum and coherence are calculated to quantify signal strength.

[0033] Standardization: All variables are Z-score standardized to ensure that the model input is dimensionless.

[0034] In step 2, climate modal features are extracted from the meteorological data at the current moment to obtain feature vectors. Specifically, this includes analyzing the influence of different phases 1-8 of the MJO on shortwave radiation and wind speed in various regions. For example, phase 3 leads to a decrease in radiation in South China, while phase 6 leads to an increase in wind speed in North China.

[0035] MJO Phase Analysis: Based on the RMM index, the MJO is divided into 8 phases. Through composite analysis, the shortwave radiation and wind speed anomalies of each phase in each region are calculated. For example, phase 3 leads to a 15%-20% reduction in radiation in South China (due to enhanced convection suppressing solar radiation), while phase 6 leads to a 10%-15% increase in wind speed in North China (due to the intensified westerly jet stream).

[0036] ENSO mode extraction: Key modes of solar radiation were extracted using empirical orthogonal function (EOF) analysis, such as the first mode (uniform variation across the entire region, explaining variance > 50%) and the second mode (inverse variation from east to west). Correlation models between these modes and sea surface temperature (SST) were constructed, and the correlation between the Nino-like index and the Indian Ocean Basin Warming (IOB) index was calculated using the Pearson correlation coefficient (r > 0.6).

[0037] Cross-impact assessment: Calculate the conditional probability distribution of the MJO phases in different ENSO phases (El Niño, La Niña, Neutral) and identify interaction patterns (such as the prolonged residence time of MJO phases 4-5 under El Niño, leading to an anomalous increase in northwest radiation).

[0038] Step 3 involves substituting the feature vector into the trained multimodal fusion forecast model for calculation to obtain the forecast result for the period to be predicted. Specifically, this includes: using the random forest algorithm based on the feature vector to calculate the point estimates and preliminary probability distributions of radiation and wind speed anomalies for the future set period; and generating a high-resolution ensemble forecast field as the forecast result for the period to be predicted through a progressive denoising process based on the point estimates and preliminary probability distributions of radiation and wind speed anomalies for the future set period.

[0039] Within a 0-10 day lead time, the shortwave radiation forecast RMSE is below 15%, and the wind speed forecast RMSE is more than 20% better than the traditional WRF model. Combining the sub-seasonal signals from the MJO and the seasonal signals from the ENSO, it covers forecasting needs from 10 to 90 days, improving the accuracy of extreme events by 30%. It can provide customized forecasting products for wind farms and photovoltaic power plants, supporting the integration of new energy sources and grid stability.

[0040] Example 2: The present invention will be further described below with reference to Figure 2 using a specific implementation method: A method for medium- and long-term forecasting of wind, solar and meteorological resources, comprising: Step 1, acquiring meteorological data at the current moment, wherein the meteorological data includes: tropical atmospheric intraseasonal oscillation index, El Niño-Southern Oscillation index and ERA5 historical data; Step 2, extracting climate modal features from the meteorological data at the current moment to obtain feature vectors; Step 3, substituting the feature vectors into a trained multimodal fusion forecasting model for calculation to obtain the forecast result for the period to be predicted; wherein, the multimodal fusion forecasting model is constructed based on the output data obtained by inputting the feature vectors into a random forest model, which is used as the input data of a diffusion model.

[0041] Before step 1, the process also includes building a multimodal fusion prediction model, as follows: Prediction Model Construction - Random Forest Part: Input feature vectors into a pre-trained random forest model (100 trees, maximum depth 15, training data 1979-2020, hyperparameters optimized by grid search: n_estimators=100, max_depth=15, min_samples_split=5); n_estimators=100. Specify the number of decision trees in the random forest, set to 100.

[0042] `max_depth=15` limits the maximum depth that each decision tree can grow to, setting it to 15.

[0043] `min_samples_split=5`. A tree node must contain at least 5 samples to be allowed to split further. If a node has fewer than 5 samples, it becomes a leaf node and will not be split further.

[0044] The model predicts monthly average anomalies of radiation and wind speed over the next 60 days.

[0045] The calculation formula is:

[0046] in, The monthly average anomaly of radiation or wind speed over the next 60 days. The input vector features (key information about the current climate state, including the phase and magnitude of the MJO, the intensity and mode of ENSO, and the interaction between the two). For sorting the tree, For the first Tree prediction.

[0047] Output point estimation: radiation anomaly -8 W / m2, wind speed anomaly +0.5 m / s; characteristic importance: MJO phase contribution 25%, ENSO 30%.

[0048] Confidence intervals were generated: using OOB samples for estimation, the 95% CI was wind speed [+0.3, +0.7] m / s.

[0049] Output: Preliminary forecast estimates and probability distributions.

[0050] Forecast Model Construction – Diffusion Model Part and Fusion: Using the output of the random forest model as the conditional input of the diffusion model.

[0051] Denoising process of the diffusion model DDPM: Using UNet (4 layers, 64-512 channels, Python torch.nn.Module), T=1000 steps, denoising at each step:

[0052] in, For the predicted mean of UNET, This refers to the noise field (such as MJO phase) at step t-1 of the denoising process. Let t be the noise field at step t (such as MJO phase, etc.). This is the number of steps (maximum 1000 steps). Let be the standard deviation of the noise in the model at step t. This is random noise sampled from a standard Gaussian distribution (mean 0, variance 1).

[0053] Add physical constraints: Loss = λ * (energy non-conservation penalty), λ = 0.1.

[0054] Generate set: Generate 100 members (resolution 0.1°×0.1°, upsampled from 0.25°), and adjust for extreme events: If the probability of peak wind speed >15 m / s >20%, adjust the bias.

[0055] Output: High-resolution ensemble forecast field, field mean plus perturbation.

[0056] It also includes seasonal adaptive adjustment: currently in summer, an attention mechanism is applied.

[0057] in, For historical correlation of factors, for example, ENSO weight 0.45, NAO weight 0.15. As weight, Seasonally adaptive tuning parameters are optimized for model learning (determined by performance on the validation set).

[0058] Ensemble processing: Calculate the average forecast (deterministic output: mean radiation 250 W / m2, wind speed 5.5 m / s); generate PDF by kernel density estimation, for example, P(radiation > 300 W / m2) = 65%, PoE = 10% for extreme wind speeds > 15 m / s.

[0059] Validation: Compare with similar historical cases (El Niño years 1979-2020), RMSE < 10%; Iteration: If the deviation > 5%, update the weights and rerun steps 3-4.

[0060] Output: Final probabilistic forecast product.

[0061] Step 1 involves acquiring current meteorological data, including the tropical atmospheric seasonal oscillation index, the El Niño-Southern Oscillation index, and historical ERA5 data. Specifically, this includes downloading the current amplitude of the MJO RMM index (1.5) and the current values ​​of the Phase 4 and ENSO NINO 3.4 indices (0.8, indicating weak El Niño) from the NOAA website. Simultaneously, historical ERA5 data (1979-2024, variables including shortwave radiation (SW), emitted longwave radiation (OLR), and near-surface wind speed u / v components at 10m, spatially limited to 70°E-140°E, 15°N-55°N) is acquired.

[0062] Quality control: Check whether the radiation value is within the range of 0-1366 W / m2, eliminate abnormalities (such as non-numbers or extreme values ​​>1500), use a 3x3 median filter to smooth spatial noise, and the filter window covers 9 grid points (approximately 0.75°×0.75°).

[0063] Time-frequency analysis: Lanczos bandpass filtering was applied to extract the seasonal MJO signal; Morlet wavelet was used to calculate the power spectrum, identify the dominant period, and calculate the coherence between MJO and ENSO. A coherence coefficient >0.7 indicates strong interaction.

[0064] Standardization: Apply Z-score standardization to all variables to generate input tensors.

[0065] Output: Preprocessed dataset, MJO phase 4 confirmed, ENSO weak El Niño, data stored in NetCDF format.

[0066] In step 2, the climate modal features of the meteorological data at the current moment are extracted to obtain the feature vector, which specifically includes: MJO phase analysis: query the impact of phase 4 from the mapping relationship library pre-built based on historical composite analysis: radiation anomaly in the southwest region -12%, wind speed anomaly +7%.

[0067] Calculation formula: Anomaly = (Current phase composite mean - Climatic mean) / Climatic standard deviation × 100%.

[0068] ENSO mode extraction: The first two modes of the radiation field were extracted using EOF analysis, and the correlation coefficient was calculated.

[0069] in, The correlation coefficient, For EOF time coefficient, For NIN O3.4, for standard deviation Let Y be the standard deviation.

[0070] Result: r=0.65, indicating a strong correlation.

[0071] Cross-impact assessment: Calculate the conditional probability P(MJO phase | ENSO = El Niño). The probability of residence in phase 4-5 increases by 20%, resulting in an 8% increase in northwest radiation anomaly. Adjust regional anomaly: The southwest radiation baseline is adjusted to -10% to -15%, and wind speed to +5% to +10%.

[0072] Output: Feature vector (dimension: 1×20, including MJO phase, ENSO index, EOF coefficient, etc.).

[0073] Step 3, which involves substituting the feature vector into the trained multimodal fusion forecast model for calculation to obtain the forecast result for the period to be predicted, specifically includes: using the random forest algorithm based on the feature vector to calculate the point estimates and preliminary probability distributions of radiation and wind speed anomalies for the future set period; and generating a high-resolution ensemble forecast field as the forecast result for the period to be predicted through a progressive denoising process based on the point estimates and preliminary probability distributions of radiation and wind speed anomalies for the future set period.

[0074] Example 3: A medium- to long-term forecasting system for wind, solar and meteorological resources, the structure of which is shown in Figure 3, includes: a meteorological data acquisition module for acquiring meteorological data at the current moment, the meteorological data including: tropical atmospheric intraseasonal oscillation index, El Niño-Southern Oscillation index and ERA5 historical data; a climate model feature extraction module for extracting climate modal features from the meteorological data at the current moment to obtain feature vectors; and a model calculation module for substituting the feature vectors into a trained multimodal fusion forecasting model for calculation to obtain the forecast results for the period to be predicted; wherein, the multimodal fusion forecasting model is constructed based on the output data obtained by inputting the feature vectors into a random forest model, which is then used as the input data for a diffusion model.

[0075] It also includes a model building module; the model building module is specifically used for: acquiring meteorological data at historical moments; extracting climate modal features from the meteorological data at historical moments to obtain a historical feature vector set; training a random forest model based on the historical feature vector set as input data and the radiation and wind speed anomalies of the period to be predicted as target variables; inputting the radiation and wind speed anomalies output by the random forest model into the UNet architecture and training it in conjunction with physical constraints to obtain a diffusion model; and fusing the random forest model and the diffusion model to construct a multimodal fusion forecast model.

[0076] The model building module also includes: optimizing the multimodal fusion forecast model by analyzing meteorological data from historical moments to determine the weights of seasonal adaptive factors and the rules for correcting extreme event deviations.

[0077] It also includes a data processing module; the data processing module is specifically used for: identifying, removing, and smoothing spatial noise based on the tropical atmospheric intraseasonal oscillation index, the El Niño-Southern Oscillation index, and ERA5 historical data to obtain a clean dataset; extracting the intraseasonal oscillation signal using Lanczos bandpass filtering based on the clean dataset to obtain a filtered signal; calculating the power spectrum using Morlet wavelet analysis based on the clean dataset to obtain a power spectrum; calculating the coherence between the tropical atmospheric intraseasonal oscillation index and the El Niño-Southern Oscillation index based on the clean dataset to obtain a coherence coefficient; and standardizing the filtered signal, power spectrum, and coherence coefficient.

[0078] The climate model feature extraction module is specifically used for: querying a mapping database based on the phase corresponding to the tropical atmospheric intraseasonal oscillation index to determine the quantitative influence of the current phase on shortwave radiation and wind speed in the target area; extracting key spatial modes of the solar radiation field using empirical orthogonal function analysis and calculating the correlation between the mode and the El Niño-Southern Oscillation Index; calculating the conditional probability of each phase occurring under different El Niño-Southern Oscillation stages based on the phases corresponding to different El Niño-Southern Oscillation stages; correcting the quantitative influence law based on the conditional probability of each phase occurring under different El Niño-Southern Oscillation stages to obtain the corrected influence law; and generating a feature vector based on the quantitative influence law, key spatial modes and correlations, and the corrected influence law. The mapping database is obtained through composite analysis of historical tropical atmospheric intraseasonal oscillation indices and their corresponding phases.

[0079] The model calculation module is specifically used to: calculate the radiation and wind speed anomalies for a future set period using the random forest algorithm based on the feature vector; and generate a high-resolution ensemble forecast field based on the point estimates and preliminary probability distributions of the radiation and wind speed anomalies for the future set period through a progressive denoising process, which serves as the forecast result for the period to be predicted.

[0080] Example 4: As shown in Figure 4, the present invention also provides an electronic device, which may be a computer device, a microcontroller device, a smart mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, processor, and transceiver component are connected via a bus; the memory can be used to store executable programs, and an exemplary executable program may include instructions; the processor is used to execute the instructions stored in the memory. The memory can also be used to store data, which can be accessed and / or modified when instructions are executed.

[0081] The processor may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and it is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the storage medium to implement the corresponding method flow or corresponding function, so as to realize the steps of the long-term forecasting method for wind, solar and meteorological resources in the above embodiments.

[0082] Example 5: Based on the same inventive concept, this invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory). This readable storage medium is a memory device within an electronic device, used to store programs and data. It is understood that the storage medium here can include both the built-in storage medium of the electronic device and extended storage media supported by the electronic device. The storage medium provides storage space, which stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more executable programs (including program code). It should be noted that the storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. Loading and executing one or more instructions stored in the storage medium by the processor can implement the steps of a long-term forecasting method for wind, solar, and meteorological resources in the above embodiments.

[0083] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0084] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more blocks of the flowchart illustrations and / or one or more blocks of the block diagrams.

[0085] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more block diagrams.

[0086] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams.

[0087] The above are merely embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application shall be included within the scope of the claims of this application pending approval.

Claims

1. A method for medium- and long-term forecasting of wind, solar, and meteorological resources, characterized in that, include: The system acquires meteorological data at the current moment, including the tropical atmospheric seasonal oscillation index, the El Niño-Southern Oscillation index, and historical ERA5 data. Climate modal features are extracted from the current meteorological data to obtain feature vectors. These feature vectors are then substituted into a trained multimodal fusion forecasting model to calculate the forecast results for the period to be predicted. The multimodal fusion forecasting model is constructed using the output data obtained from the feature vector input to the random forest model as input data to the diffusion model.

2. The method according to claim 1, characterized in that, The construction process of the multimodal fusion forecast model includes: acquiring meteorological data at historical moments; extracting climate modal features from the meteorological data at historical moments to obtain a historical feature vector set; training the random forest model based on the historical feature vector set as input data and the radiation and wind speed anomalies of the period to be predicted as target variables; inputting the radiation and wind speed anomalies output by the random forest model into the UNet architecture and training it in conjunction with physical constraints to obtain a diffusion model; and fusing the random forest model and the diffusion model to construct the multimodal fusion forecast model.

3. The method according to claim 2, characterized in that, Also includes: By analyzing historical meteorological data, the weights of seasonal adaptive factors and the rules for correcting extreme event biases are determined to optimize the multimodal fusion forecast model.

4. The method according to claim 1 or 2, characterized in that, Before extracting the climate modal features, the process includes: identifying, removing, and smoothing outliers and spatial noise based on the tropical atmospheric intraseasonal oscillation index, the El Niño-Southern Oscillation index, and historical ERA5 data to obtain a clean dataset; extracting the intraseasonal oscillation signal using Lanczos bandpass filtering based on the clean dataset to obtain the filtered signal; calculating the power spectrum using Morlet wavelet analysis based on the clean dataset to obtain the power spectrum; calculating the coherence between the tropical atmospheric intraseasonal oscillation index and the El Niño-Southern Oscillation index based on the clean dataset to obtain the coherence coefficient; and standardizing the filtered signal, power spectrum, and coherence coefficient.

5. The method according to claim 1 or 2, characterized in that, The meteorological data is subjected to climate modal feature extraction to obtain feature vectors, including: based on the phase corresponding to the tropical atmospheric intraseasonal oscillation index, querying the mapping relationship database to determine the quantitative influence of the current phase on shortwave radiation and wind speed in the target area; using empirical orthogonal function analysis to extract key spatial modes of the solar radiation field and calculating the correlation between the mode and the El Niño-Southern Oscillation Index; calculating the conditional probability of each phase occurring under different El Niño-Southern Oscillation stages based on the phases corresponding to different El Niño-Southern Oscillation stages; correcting the quantitative influence law based on the conditional probability of each phase occurring under different El Niño-Southern Oscillation stages to obtain the corrected influence law; and generating feature vectors based on the quantitative influence law, key spatial modes and correlations, and the corrected influence law; the mapping relationship database is obtained through composite analysis of historical tropical atmospheric intraseasonal oscillation indices and their corresponding phases.

6. The method according to claim 2, characterized in that, The calculation based on feature vectors combined with a pre-constructed multimodal fusion forecast model to obtain forecast results for a future specified period includes: calculating the radiation and wind speed anomalies for the future specified period using a random forest algorithm based on the feature vectors, and generating a high-resolution ensemble forecast field as the forecast result for the period to be predicted through a progressive denoising process based on the point estimates and preliminary probability distributions of radiation and wind speed anomalies for the future specified period.

7. A medium- to long-term forecasting system for wind, solar, and meteorological resources, characterized in that, include: The meteorological data acquisition module is used to acquire meteorological data at the current moment, including: tropical atmospheric seasonal oscillation index, El Niño-Southern Oscillation index, and ERA5 historical data; the climate model feature extraction module is used to extract climate modal features from the meteorological data at the current moment to obtain feature vectors; the model calculation module is used to substitute the feature vectors into a trained multimodal fusion forecast model for calculation to obtain the forecast results for the period to be predicted; wherein, the multimodal fusion forecast model is constructed based on the output data obtained by inputting the feature vectors into a random forest model, which is then used as the input data for a diffusion model.

8. The system according to claim 7, characterized in that, It also includes a model building module; the model building module is specifically used for: acquiring meteorological data at historical moments; extracting climate modal features from the meteorological data at historical moments to obtain a historical feature vector set; training a random forest model based on the historical feature vector set as input data and the radiation and wind speed anomalies of the period to be predicted as target variables; inputting the radiation and wind speed anomalies output by the random forest model into the UNet architecture and training it in conjunction with physical constraints to obtain a diffusion model; and fusing the random forest model and the diffusion model to construct a multimodal fusion forecast model.

9. The system according to claim 8, characterized in that, The model building module also includes: optimizing the multimodal fusion forecast model by analyzing meteorological data from historical moments to determine the weights of seasonal adaptive factors and the rules for correcting extreme event deviations.

10. The system according to claim 7 or 8, characterized in that, It also includes a data processing module; the data processing module is specifically used to: perform outlier identification, removal and spatial noise smoothing operations based on tropical atmospheric seasonal oscillation index, El Niño-Southern Oscillation index and ERA5 historical data to obtain a clean dataset; Based on the clean dataset, the intraseasonal oscillation signal was extracted using Lanczos bandpass filtering to obtain the filtered signal. The power spectrum was calculated using Morlet wavelet analysis based on the clean dataset. The coherence between the tropical atmospheric seasonal oscillation index and the El Niño-Southern Oscillation index was calculated based on a clean dataset, and the coherence coefficient was obtained. The filtered signal, power spectrum, and coherence coefficient were then standardized.

11. The system according to claim 7, characterized in that, The climate model feature extraction module is specifically used to: query the mapping relationship database based on the phase corresponding to the tropical atmospheric seasonal oscillation index to determine the quantitative influence of the current phase on shortwave radiation and wind speed in the target area; extract the key spatial modes of the solar radiation field using empirical orthogonal function analysis, and calculate the correlation between the mode and the El Niño-Southern Oscillation Index. Based on the various phases corresponding to different El Niño-Southern Oscillation stages and tropical atmospheric intraseasonal oscillation indices, the conditional probabilities of the occurrence of each phase under different El Niño-Southern Oscillation stages are calculated. The quantitative influence law is modified based on the conditional probability of the occurrence of each phase under different El Niño-Southern Oscillation stages to obtain the modified influence law; feature vectors are generated based on the quantitative influence law, key spatial modes and correlations, and the modified influence law. The mapping database was obtained through historical tropical atmospheric seasonal oscillation indices and their corresponding phase composite analysis.

12. The system according to claim 7, characterized in that, The model calculation module is specifically used to: calculate the radiation and wind speed anomalies for a future set period using the random forest algorithm based on the feature vector; and generate a high-resolution ensemble forecast field as the forecast result for the period to be predicted through a progressive denoising process based on the point estimates and preliminary probability distributions of the radiation and wind speed anomalies for the future set period.

13. An electronic device, characterized in that, include: At least one processor and memory; The memory and the processor are connected via a bus; the memory is used to store one or more programs. When the one or more programs are executed by the at least one processor, a long-term forecasting method for wind, solar and meteorological resources as described in any one of claims 1 to 6 is implemented.

14. A readable storage medium, characterized in that, It contains an execution program, which, when executed, implements a long-term forecasting method for wind, solar and meteorological resources as described in any one of claims 1 to 6.

Images