New energy power prediction method integrating typhoon meteorological information and micrometeorological forecast results

Abstract

The present invention relates to the field of renewable energy power generation prediction technology, and in particular to a renewable energy power prediction method that integrates typhoon meteorological information and micrometeorological forecast results. The method comprises the following steps: collecting multi-source observations such as satellite scatterometers, radar wind profiles, and laser wind measurement and unifying their coordinates; constructing a mesoscale typhoon field using four-dimensional variational assimilation, inferring the micrometeorological field using a spectral embedding diffusion network, and generating a multiscale meteorological field using frequency domain phase consistency fusion; sampling probabilistic wind speeds in a conditional reversible flow model that satisfies mass and momentum conservation, inputting an energy conservation graph neural network and coupling the wake effect to obtain unit power quantiles; feeding the power probabilities into a risk-weighted model, adjusting the weights in real time according to peak load, spare capacity, and ramp rate risks, outputting a dispatching power curve and uncertainty, and periodically updating the weights and the last layer parameters of the reversible model using the grid-connected power error. The present invention significantly improves the safe acceptance margin of the power grid for renewable energy.

Description

Technical Field

[0001] The present invention relates to the technical field of renewable energy power generation prediction, and in particular to a renewable energy power prediction method integrating typhoon meteorological information and micro-meteorological prediction results. Background Art

[0002] With the integration of high-penetration wind power in coastal and mountainous areas, strong wind shear, gusts, and wake coupling caused by the rapid passage of typhoons have become the primary cause of dramatic power fluctuations. Relying solely on conventional numerical weather prediction (NWP) data, forecasting models struggle to capture microscopic processes such as typhoon eyewall replacement and spiral rainbands on a timescale of tens of minutes, thereby weakening dispatch safety margins. Therefore, power forecasting methods that integrate typhoon macrocirculation with unit-scale micrometeorology are crucial for improving power dispatchability and reducing wind curtailment and reserve costs in extreme weather conditions.

[0003] Existing technology utilizes a "wind speed deviation identification + dual TCN power model" approach (Chinese invention patent, publication number: CN119419801B): A directed acyclic graph (DAG) network is first used to determine whether the NWP wind speed is "positively" or "negatively" offset. Two TCNs are then invoked separately, with directional penalties embedded in the loss function for correction. However, the existing technology suffers from the following drawbacks: it only processes NWP grid wind speed, ignoring the three-dimensional structure of the typhoon and micrometeorological disturbances, making it difficult to characterize wind power spikes during the rapid intensification phase of a typhoon. The dual model only provides a single power value, failing to quantify tail risk and resulting in low scheduling margin. Model weights are fixed once trained and cannot adapt to real-time grid connection errors. Accuracy plummets in typhoon path deviation scenarios. The relationship between inter-unit shading and energy conservation is not explicitly modeled, resulting in significant bias in power array effect estimation. Summary of the Invention

[0004] In response to the many problems existing in the above-mentioned existing technologies, the present invention provides a new energy power prediction method that integrates typhoon meteorological information and micrometeorological forecast results. The present invention first uses multi-source observation and four-dimensional variational assimilation to reconstruct the typhoon mesoscale field, and obtains the micrometeorological resolution wind field through spectral embedding diffusion network; then, based on the fusion field, the probabilistic wind speed that meets the conservation constraints is sampled in the reversible flow network, and the unit power is inferred by coupling the wake through the energy conservation graph neural network; finally, the probabilistic power is converted into a grid-side dispatchable curve using a risk-weighted model and adapted online with the grid connection error, significantly reducing the power peak error and scheduling risk during typhoons.

[0005] A new energy power prediction method integrating typhoon meteorological information and micro-meteorological forecast results, comprising:

[0006] Obtain wind farm satellite observations, radar wind profiles, ground wind measurements, and turbine operation data, and generate multi-source meteorological data through time correction, coordinate unification, quality control, and normalization.

[0007] Performing numerical assimilation on the multi-source meteorological data to generate a mesoscale typhoon field, inferring the mesoscale typhoon field into a micrometeorological field through super-resolution, and fusing the data using a frequency domain phase consistency method to output a multi-scale fused meteorological field;

[0008] Based on the multi-scale fusion meteorological field, probabilistic wind field data are sampled in a reversible generative model based on a streaming reversible network that satisfies mass conservation and momentum conservation constraints, and power prediction data is calculated by combining wake interaction with an energy conservation graph model based on a graph neural network.

[0009] The power forecast data of the unit is input into the risk-weighted decision model, and the dispatch power curve and power uncertainty are generated according to the peak load condition risk value, the standby capacity condition risk value and the ramp rate condition risk value. The risk-weighted decision model parameters are updated online within a preset period according to the grid-connected power feedback, and the grid-side dispatchable power forecast data is output.

[0010] Preferably, the multi-source meteorological data consists of satellite scatterometer observation data, airborne sounding profile data, three-dimensional Doppler radar wind profile data, sonic wind profiler observation data, laser wind radar observation data, drone microlaminar observation data and unit operation monitoring data, and enters the subsequent processing flow after constructing a unified data structure under a unified time base and a unified geographic coordinate system.

[0011] Preferably, the quality control first uses statistical threshold filtering to eliminate observations that exceed the abnormal judgment boundary, and then uses smooth interpolation to complete short-term continuous missing data, and uses geostatistical interpolation to complete long-term continuous missing data. After completion, a quality identification code is set for each observation record.

[0012] Preferably, the numerical assimilation adopts a four-dimensional variational assimilation algorithm, takes wind speed, temperature, humidity and pressure as control vectors, and generates a mesoscale typhoon field by minimizing a weighted error function between the background field and the observation field.

[0013] Preferably, the resolution enhancement adopts a spectrum embedding diffusion network with a frequency domain convolution architecture, and converts the mesoscale typhoon field into a micro-meteorological resolution meteorological field through forward noise perturbation and reverse denoising reconstruction sequence mapping.

[0014] Preferably, the frequency domain phase consistency method performs wavelet transform on the mesoscale typhoon field and the micro-meteorological resolution meteorological field, performs inverse transform after performing phase linear combination according to set weights within the corresponding frequency band, and generates a multi-scale fusion meteorological field.

[0015] Preferably, the reversible generation model adopts a conditionally reversible network with a residual coupling structure, records the logarithmic Jacobian in the forward mapping, and simultaneously minimizes the mass conservation residual term and the momentum conservation residual term in the reverse sampling process to obtain probabilistic wind field data.

[0016] Preferably, the energy conservation graph model uses wind turbines and meteorological towers as nodes and wake interactions as edges, calculates node latent vectors through multi-layer attention message passing, applies power conservation projection at the output layer, and maps wind speed information into unit power prediction data.

[0017] Preferably, the risk-weighted decision model calculates the risk weight coefficient based on the peak load condition risk value, the spare capacity condition risk value and the ramp rate condition risk value, applies the risk weight coefficient to the corresponding risk components in the power forecast data and performs weighted synthesis to generate a scheduling power curve and synthesize the power uncertainty according to the variance of each risk component.

[0018] Preferably, the grid-connected power feedback triggers an online update process at preset time intervals, in which the risk-weighted decision model parameters and the final stage parameters of the reversible generation model are synchronously adjusted according to the error gradient between the predicted power and the grid-connected power.

[0019] Compared with the prior art, the advantages and beneficial effects of the present invention are:

[0020] This paper uses four-dimensional variational assimilation + spectral embedding diffusion super-resolution technology to achieve multi-scale seamless fusion of typhoon macro-field and micro-meteorological field, capturing the rapidly evolving local wind speed gradient;

[0021] The present invention uses the residual coupling condition reversible flow model technology to achieve probabilistic wind field sampling that simultaneously satisfies mass-momentum conservation and outputs a credibility interval rather than a single point value.

[0022] The present invention realizes array-level power mapping and unit coupling loss compensation through energy conservation graph neural network + wake interaction constraint technology.

[0023] The present invention achieves the coordinated minimization of three-dimensional risks of peak load, spare capacity and ramp rate through risk-weighted decision-making + gradient closed-loop online update technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Schematic diagram of the process of the present invention;

[0025] Figure 2 This is a data structure association diagram of the new energy power prediction method of the present invention;

[0026] Figure 3 Schematic diagram of the conservative reversible network structure and residual feedback mechanism of the present invention. DETAILED DESCRIPTION

[0027] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0028] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise," "include," etc. used herein indicate the presence of the features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0029] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0030] like Figure 1 As shown, a new energy power prediction method integrating typhoon meteorological information and micro-meteorological forecast results includes:

[0031] Obtain wind farm satellite observations, radar wind profiles, ground wind measurements, and turbine operation data, and generate multi-source meteorological data through time correction, coordinate unification, quality control, and normalization.

[0032] This implementation describes the technical principles, key algorithms, and application results of generating multi-source meteorological data by acquiring wind farm satellite observations, radar wind profilers, ground-based wind measurements, and turbine operation data. The overall process sequentially completes spatiotemporal unification, data correction, quality control, and normalization, providing highly consistent, directly accessible input for subsequent typhoon-micrometeorological fusion and power forecasting modules.

[0033] First, along-track satellite scatterometer observations, airborne sounding profiles, 3D Doppler radar voxels, acoustic wind profiler echoes, laser wind radar volume scans, drone micro-laminar flow profiles, and crew operation records are transmitted to the data receiving node via dedicated communication links. Each observation source carries an independent timestamp and local coordinates and must be synchronized under a unified clock and coordinate architecture.

[0034] The system uses the global positioning system timing signal to construct an absolute time reference, and the original time Convert to Coordinated Universal Time :

[0035]

[0036] in Indicates the deviation between the satellite timing pulse and the internal clock of the device. The corrected timestamp error is kept at the microsecond level, which can meet the requirements of splicing high-frequency unit data and radar volume scan data. Spatial unification is achieved through a strict conversion of ellipsoidal geodetic coordinates to projected plane coordinates: the longitude and latitude triplet is converted to the Map to horizontal axis , vertical axis and senior professor After the coordinates are unified, the system aggregates multi-source observations at the same time for the same grid unit to form a multi-dimensional vector Quality control is performed in two levels. First, the sliding median-absolute deviation test is used in the time dimension to remove outliers: Calculate the sliding window median The median absolute deviation (MAD) is the same as the median absolute deviation (MAD), if:

[0037]

[0038] Then Marked as an exception, parameter According to the instrument stability setting, the spatial missing data are supplemented by three-dimensional kriging interpolation, and its semivariogram function Fitting based on neighborhood sample empirical statistics, interpolation weights Obtained by solving the Kriging linear equations.

[0039] All observed quantities are uniformly converted into dimensionless normalized form. Taking wind speed as an example, the normalization formula is:

[0040]

[0041] in is the sliding mean, Normalization maps data of different dimensions to a zero-mean unit variance space, which facilitates the subsequent reversible generation of flow models to process multiple meteorological elements simultaneously.

[0042] To verify the feasibility of this process, a typhoon-affected event was used as an example: the system installed two laser wind radars and three acoustic wind profilers at the wind farm, and enabled satellite scatterometer track coverage. After the observation data was processed through the above steps, a multi-source meteorological data matrix containing more than 50 dimensions of wind speed, wind direction, temperature, humidity, and unit power was generated. This matrix is ​​used as a unified input in the subsequent assimilation-diffusion-phase locking module. The measured splicing error is lower than the calibration accuracy of the wind tower, meeting the requirements of microstructure analysis under typhoon strong wind shear field.

[0043] Through the above technical solution, under the complex wind field conditions of a rapidly approaching typhoon, multi-source meteorological data with consistent time and space and controllable quality can be stably output, laying a solid foundation for subsequent probabilistic wind field sampling and power mapping.

[0044] Preferably, Figure 2 As shown, the multi-source meteorological data consists of satellite scatterometer observation data, airborne sounding profile data, three-dimensional Doppler radar wind profile data, sonic wind profiler observation data, laser wind radar observation data, UAV micro-laminar observation data and unit operation monitoring data, and enters the subsequent processing flow after constructing a unified data structure under a unified time base and a unified geographic coordinate system.

[0045] This implementation method describes the construction principles, key processing algorithms, and application effects of multi-source meteorological data in a new energy power prediction scenario that integrates typhoon meteorological information and micro-meteorological forecast results. The multi-source meteorological data consists of satellite scatterometer observation data, airborne sounding profile data, three-dimensional Doppler radar wind profile data, sonic wind profiler observation data, laser wind radar observation data, drone micro-laminar flow observation data, and unit operation monitoring data. By constructing a unified data structure with a unified time base and a unified geographic coordinate system, the spatial dislocation and temporal drift caused by source differences can be eliminated in the subsequent assimilation-diffusion-phase locking link, thereby improving the power prediction accuracy under typhoon extreme wind fields.

[0046] There is a deviation between the built-in clock of each data source and the timing signal of the global navigation satellite system. The time is unified by the following formula:

[0047]

[0048] in Represents the original timestamp; Indicates the difference between the timing pulse and the device clock; Represents the unified Coordinated Universal Time. After correction, all observations are aligned within a microsecond time error, ensuring synchronization between the radar volume scan voxels and the unit's 10-Hz power curve.

[0049] Satellite and radar products are given in geocentric ellipsoidal coordinates and projected to plane coordinates using the following formula:

[0050]

[0051]

[0052] in is the first radius of curvature; is the meridian curvature radius; and The height is uniformly referenced to the geoid, and the relative height of the acoustic wind profile and the laser wind radar return is converted to the orthometric height. All observation points are then mapped to a unified 3D grid index .

[0053] For each grid time series Perform two levels of quality inspection:

[0054] Level 1 test: sliding median and median absolute deviation (MAD):

[0055]

[0056] in is a constant factor.

[0057] Secondary inspection: the abnormal marked points are screened again according to the neighborhood spatial gradient, and the missing segments are filled by linear interpolation or Kriging interpolation, and Kriging weights are used. , by solving:

[0058]

[0059] represents the empirical semivariogram function; It is the Lagrange multiplier. After completion, it generates the quality identification code QF as the basis for subsequent weighting.

[0060] For each meteorological quantity Implement a zero-mean unit-variance mapping:

[0061]

[0062] represents the sliding mean; Represents the sliding standard deviation; construct a unified vector after normalization:

[0063]

[0064] in is the characteristic dimension. This vector is directly input into the mass conservation-momentum conservation reversible generation flow model without further dimensional transformation.

[0065] If the Kriging interpolation residual exceeds the threshold , the system immediately triggers redundancy switching and uses another acoustic wind profiler or laser wind radar at the same height as the compensation source. , the historical statistics field is applied to perform replay filling to ensure the integrity of the input tensor.

[0066] Embedding this implementation in the case of the typhoon "Bohai Rim," the mean square error of wind speed in the fused grid was significantly reduced compared to traditional site interpolation schemes. The probabilistic power quantile curve remained continuous near the unit cut-out wind speed, supporting the energy conservation graphical model to accurately describe wake coupling. Experimental verification showed that the RMS error of the wind vector output by the assimilation-diffusion-phase locking module near the center of the extreme typhoon was significantly reduced, and the absolute error curve of the power forecast also decreased.

[0067] In this embodiment, two laser wind radars, an airborne sounding system, and an acoustic wind profiler array are deployed in a coastal wind farm. The system aggregates the observation sources at a time granularity of one second and executes the above-mentioned spatiotemporal unification and quality control process to generate vectors. The reversible generated flow was then applied to sample the wind field and map the power curve. The results showed that the active power forecast deviation six hours before the typhoon made landfall remained within 10% of the unit's rated power, validating the effective support of multi-source meteorological data for renewable energy power forecasting scenarios.

[0068] Preferably, the quality control first uses statistical threshold filtering to eliminate observations that exceed the abnormal judgment boundary, and then uses smooth interpolation to complete short-term continuous missing data, and uses geostatistical interpolation to complete long-term continuous missing data. After completion, a quality identification code is set for each observation record.

[0069] To ensure that the subsequent typhoon-micrometeorological fusion model obtains reliable input, this implementation method constructs a quality control chain for multi-source meteorological data. The core consists of three types of operations: anomaly rejection, missing measurement completion, and quality identification. This not only makes full use of high-frequency observations, but also avoids the amplification of abnormal noise during the probabilistic wind field sampling stage.

[0070] First, the system is in a unified space-time grid For each observation Create a time series Calculate the sliding median of the series Median absolute deviation If there is an observation that satisfies:

[0071]

[0072] It is judged to be beyond the abnormal judgment boundary. Indicates the multiplication factor set for sensor stability, represents the median, Indicates the median of absolute deviation. This statistical threshold filtering can be completed within 30 milliseconds of data access and output a one-dimensional Boolean mask. After the abnormality is eliminated, the system detects the length of the continuous missing segment. .like:

[0073]

[0074] in If the threshold is preset, smooth interpolation is used to complete the function. Smooth interpolation is performed by the cubic spline function Description, its natural boundary condition is Ensure the curvature is smooth. The reconstructed observation value after interpolation is recorded as .like , then call geostatistical interpolation. The system is based on the surrounding spatial samples Empirically construct the semivariogram function:

[0075]

[0076] in represents the sample spacing, Indicates the number of sample pairs on this interval. By fitting Get the theoretical range With abutment , and then solve the Kriging equations:

[0077]

[0078] Obtain weight , and use this to calculate the estimated value of the missing grid:

[0079]

[0080] in represents the weight coefficient, Represents the distance between the point to be estimated and the known sample points, represents the Lagrange multiplier.

[0081] After interpolation, the system assigns a quality identifier, QF, to each observation record. QF=0 indicates the original observation passed the threshold test; QF=1 indicates it was interpolated using smooth interpolation; QF=2 indicates it was interpolated using geostatistical interpolation; and QF=3 indicates it was masked due to severe omissions but still occupies a placeholder. The quality identifier is weighted in the loss function of the reversible generative flow model. The contribution of low-quality samples is automatically reduced, ensuring that the conservation constraint training process remains robust to the error distribution.

[0082] Example: During a typhoon's rapid landing, the acoustic wind profiler experienced signal attenuation in a high humidity environment and was absent for two minutes. System Detection If the value is greater than the threshold, Kriging interpolation is initiated. Spatial samples are constructed using lidar and airborne sounding profiles at the same altitude. The wind speed field is reconstructed through semivariogram fitting and weighted solution. The interpolation residual variance is less than 20% of the mean squared error of the original wind tower. This interpolation result is assigned a QF of 2, which gives it a lower weight in the subsequent probabilistic wind field sampling phase, reducing the spread of data uncertainty in the anomalous segments.

[0083] Through the above-mentioned multi-level quality control, the scheme maintains the integrity of the observation vector under extreme typhoon wind speed shear, and provides noise-controlled input for the subsequent typhoon-micrometeorological fusion algorithm, ultimately significantly reducing the mean square error of the unit power prediction compared to the baseline without quality control.

[0084] Performing numerical assimilation on the multi-source meteorological data to generate a mesoscale typhoon field, inferring the mesoscale typhoon field into a micrometeorological field through super-resolution, and fusing the data using a frequency domain phase consistency method to output a multi-scale fused meteorological field;

[0085] The original multi-source meteorological data is used as observation vectors after being unified into a grid Assimilate the three-dimensional state vector of the system in a finite difference non-hydrostatic mode To control the variables, the four-dimensional variational objective function is minimized:

[0086]

[0087] in represents the background field, represents the background error covariance matrix, express The time observation operator, represents the observation error covariance matrix.

[0088] Solution Then we get the mesoscale typhoon field with a horizontal resolution of 3,000 meters. , capturing the maximum radial wind shear of the typhoon eyewall and providing a typhoon-scale forcing boundary for wind power forecasting.

[0089] Will The slices are fed into the spectrum embedding diffusion network. The network gradually injects Gaussian noise into the state tensor in the forward process and performs frequency domain convolution denoising in the reverse process to achieve reconstruction with a horizontal resolution of 30 meters and a vertical resolution of 5 meters, and output the micro-meteorological field. The process is shown to maintain the relative phase structure of the typhoon vortex and avoid the small-scale energy attenuation introduced by traditional interpolation.

[0090] In order to eliminate the phase mismatch between the meso-micro scale boundary, the system first and Perform three-dimensional discrete wavelet transform to obtain the corresponding frequency band phase and Construct the fused phase within each wavelet band:

[0091]

[0092] Represents the weight coefficient. Keep the amplitude as the micrometeorological energy spectrum amplitude ,and After combining, perform inverse wavelet transform and output multi-scale fusion meteorological field This fusion maintains both mesoscale energy and microscale details, and phase continuity ensures the diffusibility of the wind velocity field gradient, meeting the conservation constraint requirements of the subsequent reversible generation flow model. It should be noted that in the present invention, Represents a multi-scale fused meteorological field (i.e., a three-dimensional meteorological tensor that has completed typhoon-micrometeorological phase fusion and has a spatial resolution of the unit scale). If a local window needs to be extracted from the tensor or a low-dimensional conditional vector is obtained through linear embedding, it can be written as ,in Indicates the fixed processing steps of "cropping + downsampling + fully connected embedding". That is, the multi-scale fusion meteorological field itself is used Indicates; all expressions of derived conditional vectors are expressed using .

[0093] The experiment was conducted on the case of the landing typhoon "SA Qiansen": the wind vector mean square error of the traditional pure assimilation-interpolation scheme is , after introducing super-resolution-phase fusion, the error is reduced to The 95th percentile forecast error of the mapped wind farm power was reduced by more than 30%, and the unit's over-rated wind speed cut-out judgment was issued 20 minutes in advance, effectively supporting grid-connected scheduling.

[0094] In this example, two laser wind radars, a coastal radar array, and an airborne sounding system are deployed in a coastal 100MW wind farm. Two hours after the typhoon's outer circulation enters the observation area, the system completes the assimilation-inference-fusion integrated calculation and generates The data is then pushed to the reversible generation flow module. Monitoring showed that during the passage of the typhoon's eyewall, the fusion field captured gust pulses with a delay of less than 15 seconds, ultimately achieving a smooth downward adjustment of the power dispatch curve and avoiding frequent unit trips.

[0095] Preferably, the numerical assimilation adopts a four-dimensional variational assimilation algorithm, takes wind speed, temperature, humidity and pressure as control vectors, and generates a mesoscale typhoon field by minimizing a weighted error function between the background field and the observation field.

[0096] This method uses a four-dimensional variational assimilation algorithm to invert mesoscale typhoon fields from multi-source observations, laying the foundation for subsequent micrometeorological super-resolution inference and renewable energy power forecasting. Its core principle is to treat wind speed, temperature, humidity, and static pressure as state variables. Within the differentiable framework of a finite-difference nonhydrostatic numerical model, it simultaneously approximates the background and observation fields by minimizing a weighted error function.

[0097] set up represents the three-dimensional state vector to be estimated, represents the extrapolated background field of the model, Indicates the first The observation vector at time t, is the corresponding observation operator, then the objective function is:

[0098]

[0099] in represents the background error covariance matrix, which is used to measure the spatial correlation of background deviation; Represents the observation error covariance matrix, which is used to characterize the uncertainty of each observation source.

[0100] The algorithm calculates the adjoint mode , using conjugate gradient or quasi-Newton iterative search . This is a mesoscale three-dimensional wind field encompassing the typhoon's circulation, eyewall, and outer rainband structure. The horizontal grid resolution for this wind field is much smaller than that of conventional global reanalysis products, enabling resolution of radial wind shear at wind turbine blade tip height. The vertical resolution is consistent with that of the sounding profile, enabling synchronous constraint of temperature and humidity gradients and momentum coupling.

[0101] In order to enable the grid dispatcher to obtain a reliable power reduction plan before the typhoon arrives, the present invention selects the assimilation window in the forecast period before the rapid evolution of the wind field; if there is a missing satellite scatterometer band, the observation operator automatically removes the corresponding observation channel and adjusts To maintain the stability of the condition number. The abnormal wind speed observation has been assigned the quality identification code QF in the above quality control step; The weights of the corresponding diagonal elements are reduced to ensure that the iterative convergence direction is not constrained by local anomalies.

[0102] Example: During a moderate-to-strong typhoon landfall, airborne sounding, coastal-based three-dimensional radar, and two drone profile data were incorporated into the assimilation window. The resulting maximum near-surface radial wind speed for the mesoscale typhoon field deviated from the measured instantaneous wind speed at the turbine rotor center by less than half the baseline model output. Subsequent micrometeorological super-resolution inference inherited the phase of this field to generate a multi-scale fused meteorological field. After power mapping, the wind speed cut-out time for the turbine was accurately determined three hours in advance, providing a basis for the station to implement active power reduction and reduce blade root load.

[0103] Preferably, the resolution enhancement adopts a spectrum embedding diffusion network with a frequency domain convolution architecture, and converts the mesoscale typhoon field into a micro-meteorological resolution meteorological field through forward noise perturbation and reverse denoising reconstruction sequence mapping.

[0104] This method uses a spectral embedding diffusion network with a frequency-domain convolutional architecture to refine a mesoscale typhoon wind field with a horizontal grid of approximately 3 km into a micrometeorological wind field with a horizontal grid of approximately 30 m and a vertical grid of approximately 5 m, providing unit-scale wind vectors for renewable energy power forecasting. The core process and mathematical expression are as follows.

[0105] Assume that the assimilated mesoscale three-dimensional wind field tensor is The forward diffusion process Inject noise in discrete steps and get:

[0106]

[0107] in For the Step retention factor, .noise First, the three-dimensional fast Fourier transform is performed, and the wave number Dimensions are weighted by observed energy spectrum After redistribution, the inverse transform is injected into the above formula to ensure that the noise energy matches the actual energy spectrum of the typhoon.

[0108] Frequency domain convolution reverse denoising, the reverse network uses multi-level UNet, but all convolution kernels are implemented in the frequency domain: fast Fourier transform is performed on the input tensor, and the convolution kernel spectrum is compared. After multiplication, the inverse transformation is performed back to the spatiotemporal domain. Frequency domain convolution expands the receptive field and has periodic boundaries, effectively eliminating eye wall edge artifacts.

[0109] The method of spectrum embedding phase correction to maintain the typhoon phase continuity is to introduce the phase of the same layer of the encoder in the decoding stage ,according to:

[0110]

[0111] Generate fusion phase ,in is the current decoder phase, is a fixed weight. The amplitude is taken from the decoder amplitude spectrum The denoised tensor is obtained by inverse wavelet transform .

[0112] According to the highly correlated mean and standard deviation Perform denormalization:

[0113]

[0114] The symbol " " represents element-wise multiplication, The three-dimensional wind field with micro-meteorological resolution can be directly input into the subsequent conservation reversible generation flow model and power mapping model. and output spectrum The relative error is:

[0115]

[0116] The present invention is about 10 times slower than the traditional bicubic interpolation scheme. The 95th percentile power forecast deviation has been significantly reduced, enabling the dispatcher to issue power reduction instructions ten minutes in advance, reducing the risk of unit overload and tripping.

[0117] In this embodiment, dual laser wind radars, coastal multi-axis radars and airborne sounding systems are deployed at the coastal megawatt wind farm. When the outer circulation of the typhoon enters the observation area, the system uses the diffusion network to generate The measured results show that the model's response delay to gust pulses during the passage of the typhoon eyewall was less than 15 seconds, and the peak load of the unit was controlled below 90% of the rated load, verifying the practical value of the resolution enhancement link in extreme wind conditions.

[0118] Preferably, the frequency domain phase consistency method performs wavelet transform on the mesoscale typhoon field and the micro-meteorological resolution meteorological field, performs inverse transform after performing phase linear combination according to set weights within the corresponding frequency band, and generates a multi-scale fusion meteorological field.

[0119] In order to ensure that typhoon-scale dynamic information and unit-scale details are continuously presented in the same wind field, the present invention introduces the frequency domain phase consistency method after the resolution is improved to convert the mesoscale typhoon field into and micrometeorological fields Fusion into multi-scale meteorological fields This method preserves the mesoscale energy and embeds the microscale texture through linear combination of wavelet phase. The specific principle and implementation steps are as follows.

[0120] Wavelet transform and band mapping, and Perform three-dimensional discrete wavelet transform respectively to obtain the scale-frequency index The complex coefficients of:

[0121]

[0122]

[0123] 、 represents the magnitude spectrum; 、 represents the phase spectrum; is the directional subband, For the scale level.

[0124] Phase linear combination, let the phase weight coefficient be , in the same subband Linear interpolation is performed inside:

[0125]

[0126] , with increasing scale-microscale level , completely adopting the micro-meteorological phase-mesoscale main energy layer , keep the typhoon phase amplitude selection and inverse transformation amplitude to take the micro-meteorological amplitude To preserve the high-resolution spectrum:

[0127]

[0128] right Performing the inverse wavelet transform yields:

[0129]

[0130] Variable Description: —Mesoscale typhoon wind field tensor; —Micrometeorological wind field tensor obtained through spectral embedding diffusion network; —Output multi-scale fusion meteorological wind field; —Phase weight coefficients at different scale levels.

[0131] After phase consistency processing, This also maintains the typhoon eyewall circulation phase and turbine horizontal vortex details. Compared to the unfused solution, the root mean square error of wind speed in radar-unobstructed areas is reduced by approximately 35%. In the power forecasting chain, the 95th percentile power deviation is reduced and lead time is increased, enabling the dispatch system to issue power reduction commands earlier and reducing the number of turbine trips during the passage of the typhoon center.

[0132] Example: In the case of Typhoon Haiou, the three-dimensional wavelet Daubechies-4 basis is used to set the phase weight The fusion result is verified by the laser wind radar, and the average wind direction error at a height of 200 meters is down to Based on this, the station issued a cluster load reduction command ten minutes in advance, and the maximum torque of the unit was reduced by 11% compared with the unfused baseline, demonstrating the practical value of this frequency-domain phase consistency method in typhoon-driven renewable energy power forecasting.

[0133] like Figure 3As shown, with the multi-scale fusion meteorological field as a condition, probabilistic wind field data is sampled in a reversible generation model based on a streaming reversible network that satisfies mass conservation and momentum conservation constraints, and power prediction data of the group is calculated by combining wake interaction with an energy conservation graph model based on a graph neural network;

[0134] For Starting from the multi-scale fusion meteorological field, the physically credible unit power probability distribution is obtained. The present invention continuously executes two-level models in the prediction link:

[0135] Conservation condition reversible network : Generate probabilistic wind speed fields that satisfy mass-momentum conservation; Energy conservation graph neural network : Mapping probabilistic wind speed to power quantile curve considering wake interaction.

[0136] The conservation condition reversible network generates a probabilistic wind field, and the conditional vector is constructed:

[0137]

[0138] stands for "cropping + convolution downsampling + fully connected embedding"; Represents the local meteorological condition vector.

[0139] Reversible mapping and log-Jacobian:

[0140]

[0141] represents a latent variable; represents the three-dimensional wind speed tensor.

[0142] Each residual coupling block transforms only half of the channel, so:

[0143]

[0144] Indicates the Layer scale function.

[0145] Joint loss function:

[0146]

[0147] represents the mass conservation weight; represents the momentum conservation weight; represents static pressure disturbance; The second and third residuals feed back the gradient of the conservation constraint into the latent space, making the sampling results naturally conform to the fluid equation.

[0148] Sampling and statistics, extraction Group latent variables And map:

[0149]

[0150] calculate 、 、 、 and other indicators to form a probabilistic wind field set .

[0151] Energy conservation graph neural network calculates power, graph construction: suppose the wind farm contains a set of unit nodes , when the node Fall into the node Add edge inside the wake cone . Get the undirected graph .

[0152] Node and edge features, node features: probabilistic wind speed mean and variance ; Edge feature: wind direction angle Distance from center .

[0153] Message Passing and Power Conservation:

[0154]

[0155]

[0156] represents the original node vector; represents the attention weight; Represents the wind direction-similarity function. The output layer is projected to power and energy conservation is applied:

[0157]

[0158] Indicates the air density; represents the swept area; Represents the power coefficient, which does not exceed the Betz limit after Sigmoid normalization.

[0159] Probability power output, Each run , get the power sample Statistical expectation and 5% / 95% percentile , output power prediction set .

[0160] In one embodiment, in the "Meihua" typhoon test: the mean of mass conservation residuals The 95th percentile power deviation decreased from the baseline of 14% to 6%. The dispatch reserve capacity lock-in time was advanced by 11 minutes. The unit experienced no blade root overload, validating the "conservation reversible sampling + energy conservation diagram inference" approach to improving power reliability in extreme typhoon scenarios.

[0161] Preferably, the reversible generation model adopts a conditionally reversible network with a residual coupling structure, records the logarithmic Jacobian in the forward mapping, and simultaneously minimizes the mass conservation residual term and the momentum conservation residual term in the reverse sampling process to obtain probabilistic wind field data.

[0162] In multi-scale fusion of meteorological fields Under the premise that the residual coupling structure is obtained, the present invention uses the conditional reversible network (ConditionalResidual-CouplingFlow, denoted as ) performs probabilistic sampling of the typhoon's local wind field; by explicitly incorporating mass and momentum conservation penalties into the latent variable space, the generated samples are ensured to be both statistically consistent and adhere to the fundamental equations of fluid mechanics. The following section describes the network mechanism, physical constraints, sampling process, and technical results.

[0163] Network mechanism, conditional vector construction:

[0164]

[0165] represents the local meteorological condition vector; For fixed "cropping Convolutional downsampling Fully connected embedding "mapping.

[0166] Reversible mapping relationship:

[0167]

[0168] represents the three-dimensional wind speed tensor; represents a latent variable; It consists of several affine coupled blocks, are all trainable parameters.

[0169] Log-Jacobian accumulation: Each coupling block transforms only half of the channel, so the overall log-Jacobian determinant can be accumulated layer by layer:

[0170]

[0171] For the Layer scale function output; It is the channel sub-tensor of the layer that has not been affine transformed.

[0172] The physical-statistical joint loss network is trained to minimize the following:

[0173]

[0174] It is given by the latent variable density and log-Jacobian determinant to ensure statistical consistency.

[0175] represents the divergence; the two-norm term penalizes deviations from mass conservation.

[0176] represents the residual of the incompressible Navier–Stokes equations; the second-norm term penalizes deviations from momentum conservation.

[0177] represents static pressure disturbance; represents the kinematic viscosity coefficient; and are the mass conservation weight and momentum conservation weight respectively.

[0178] The conservative residual returns the gradient in a soft-constrained manner. When the sampled wind speed deviates from the conservative manifold, the latent variable gradient "pulls" the sample back to the physically feasible domain.

[0179] Probability sampling and power mapping process, sampling: fixed , extract Group latent variables ; Generate wind field .

[0180] Statistics: Pairs Calculating expectations ,variance , upper / lower quantiles, forming a probabilistic wind field .

[0181] Power inference: Feed each sample into the energy conservation graph neural network :

[0182] Nodes: turbines and wind towers; Edges: wake cone coverage relationship; Message passing: multi-head attention calculation of wind speed attenuation; Output power:

[0183] , Indicates the air density; represents the impeller swept area; Sigmoid normalization is limited to not exceed the Betz limit .

[0184] Output: Summarized expected power and 5% / 95% percentile power , forming a power prediction interval.

[0185] Example: In the "Meihua" typhoon landing test, set :

[0186] Mean divergence residual The mean square of the momentum conservation residual decreased by about 60% compared with the baseline without conservation. The 95th percentile deviation of the unit power dropped from 14% to 6%, the standby capacity locking time was advanced by 11 minutes, and no unit tripping due to sudden wind speed jumps occurred on site.

[0187] The results show that the conservation condition reversible network + energy conservation graph model can provide reliable power quantile curves under extreme typhoon wind fields, providing high-confidence support for power grid scheduling and safe operation of units.

[0188] Preferably, the energy conservation graph model uses wind turbines and meteorological towers as nodes and wake interactions as edges, calculates node latent vectors through multi-layer attention message passing, applies power conservation projection at the output layer, and maps wind speed information into unit power prediction data.

[0189] Energy conservation diagram model of the present invention The wind turbine-meteorological tower scenario is abstracted into a graph structure with physical meaning. The wake attenuation is characterized through multi-layer attention message passing. Power conservation projection is then applied at the output, mapping the probabilistic wind speed into a unit power prediction quantile curve. The following describes the graph construction, message passing mechanism, power projection formula, and technical effects, with unified annotations for all symbols. Graph Construction and Feature Definition:

[0190]

[0191] Represents a node set consisting of wind turbine nodes Node with weather tower composition; Represents an undirected edge set, if the node Located at the node Impeller wake cone inner edge . Node initial vector:

[0192]

[0193] in Represents the mean value of wind speed samples at the center of the impeller; Indicates the standard deviation of the wind speed sample;

[0194] Represents the normalized coordinates of the node height.

[0195] Edge features:

[0196]

[0197] in, Indicates the wind direction angle; Indicates the horizontal distance between units; Indicates the wake influence radius.

[0198] Attentional Message Passing, Layer update formula:

[0199]

[0200]

[0201] 、 Represents the linear transformation matrix of the self-node and neighboring nodes; represents the activation function; is the wake attention scoring function, which is as follows .

[0202] Power conservation projection, the output layer first gives the unconstrained predicted power:

[0203] Then the impeller power rating and energy conservation double projection are applied:

[0204]

[0205] in represents the final unit power forecast; Indicates the rated power of the unit; Indicates the set of units affected by wake vortex in a certain row; Represents the upstream set of turbines without wake. Projection ensures that the total power of downstream turbines does not exceed the upstream kinetic energy flux.

[0206] Symbol Notes: Indicates the mean wind speed at the impeller center; represents the standard deviation of wind speed; Indicates the wind direction angle; Indicates horizontal distance; represents the wake radius; Indicates the air density; represents the swept area; Indicates the power factor; Indicates the Layer node hidden vector; represents the attention weight.

[0207] Example: In the "Meihua" typhoon landing condition, 256 groups of probabilistic wind fields of conservative reversible sampling are input into the graph model:

[0208] Average relative error of unit level: expected power 3.8%, 95% percentile power 6.0%;

[0209] After satisfying the energy conservation projection, the upstream and downstream power difference is reduced by 42% compared with that without projection;

[0210] The dispatching system issued a power reduction command 11 minutes in advance based on the 95% curve, and no unit tripped due to misjudgment of the wake turbulence.

[0211] The results verify that multi-layer attention can express directional wake attenuation; power conservation projection prevents energy "increase", significantly improving the credibility of power prediction in extreme typhoon scenarios.

[0212] The power forecast data of the unit is input into the risk-weighted decision model, and the dispatch power curve and power uncertainty are generated according to the peak load condition risk value, the standby capacity condition risk value and the ramp rate condition risk value. The risk-weighted decision model parameters are updated online within a preset period according to the grid-connected power feedback, and the grid-side dispatchable power forecast data is output.

[0213] In the obtained unit power probability set Finally, the present invention maps random power into a grid-side dispatchable power curve through a risk-weighted decision model. and its uncertainty The model simultaneously considers three types of operational risks: peak load, spare capacity, and ramp rate. During the operation period, the weights are updated online using the grid-connected measured power closed loop to achieve dynamic scheduling optimization under extreme typhoon scenarios.

[0214] Risk indicator and weighted objective function, note:

[0215]

[0216] are the risk values ​​of peak load, spare capacity, and ramp rate conditions, respectively;

[0217]

[0218] Corresponding risk weight . The power prediction quantile vector of:

[0219]

[0220] As input, construct the weighted target:

[0221]

[0222] Indicates peak load limit; represents the lower bound of the minimum spare capacity; represents the derivative of the ninety-fifth percentile power with respect to time; Indicates the maximum ramp rate allowed by the power grid. Minimization is the criterion, Intercept the dispatchable power within the interval:

[0223]

[0224] Online weight update, assuming the grid-connected measured power is ,error:

[0225]

[0226] In the preset period Inner cumulative square error:

[0227]

[0228] Update the weights using constrained stochastic gradient descent:

[0229]

[0230] in, represents the learning rate; Represents projection onto a simplex After updating, solve again according to the above method. , realizing rolling adaptation of decision model.

[0231] Explanation of symbols: Represents the multi-scale fusion meteorological field; Indicated by Embed the obtained conditional vector; 、 Indicates unit power and quantile; Indicates peak load limit; represents the lower bound of spare capacity; Indicates the upper limit of the allowed climbing rate; , , represents the risk weight; represents the dispatch power curve; represents the scheduling power uncertainty; represents the scheduling error; Represents the weight update learning rate.

[0232] In one embodiment, during the operation before Typhoon Dusurui landed, a 30-minute cycle online update was used: the spare capacity utilization rate during off-peak hours increased by 18%; the maximum ramp-overlimit events were reduced from 9 to 1; and the unit trip rate was 0, verifying the dynamic robustness of the risk-weighted decision-making model combined with measured feedback.

[0233] Preferably, the risk-weighted decision model calculates the risk weight coefficient based on the peak load condition risk value, the spare capacity condition risk value and the ramp rate condition risk value, applies the risk weight coefficient to the corresponding risk components in the power forecast data and performs weighted synthesis to generate a scheduling power curve and synthesize the power uncertainty according to the variance of each risk component.

[0234] To predict the unit power probability The present invention constructs a risk-weighted decision model by converting it into a dispatchable power curve on both sides. The model first calculates the weight coefficient according to the three types of operation risks, and then uses the weight to weight the power components to synthesize the dispatch curve. The uncertainty is given by variance superposition .

[0235] Risk Quantification:

[0236]

[0237]

[0238]

[0239] in Indicates taking the non-negative part. Indicates peak load limit; represents the lower bound of spare capacity; Indicates the maximum allowed ramp rate.

[0240] Risk weight factor:

[0241]

[0242]

[0243] If the denominator is zero (no risk scenario), set The weights are calculated by sliding over time, which can reflect the changes in risk ratio caused by typhoon evolution.

[0244] Dispatch power curve and uncertainty:

[0245]

[0246]

[0247] 、 、 are the estimated variances of the 5% quantile, mean, and 95% quantile (calculated from the probability wind field sample set). If each risk component is considered independent, the above formula is the superposition of the variances of power uncertainty.

[0248] Online update under grid-connected feedback, assuming the measured grid-connected power is , define the error:

[0249]

[0250] In the preset period Inner cumulative square error:

[0251]

[0252] Update using projected gradient:

[0253]

[0254] Projection onto a simplex , is the learning rate. Online updates make the weights adaptive to the real-time operation status of the power grid.

[0255] Explanation of symbols: Indicates 5% quantile power; Indicates power expectation; represents the 95% quantile power; 、 They represent the peak load limit and the lower bound of the reserve capacity respectively; It represents the derivative of the 95% quantile power with respect to time; Indicates the maximum ramp rate allowed by the power grid; 、 、 Indicates three types of risk values; 、 、 Indicates the corresponding risk weight; represents the dispatch power curve; represents the power uncertainty; Indicates the measured grid-connected power; Represents the weight update learning rate.

[0256] In the embodiment, during the "Seagull" Typhoon landing drill, the risk-weighted decision-making model reduced the dispatching power by 8% during the peak load exceeding period, while maintaining sufficient spare capacity; the maximum ramp exceeding event throughout the day was reduced from 7 times to 1 time, and the unit did not experience overload or tripping, proving that the model effectively balances the three types of operating risks under extreme wind conditions and outputs an executable grid-side dispatching curve.

[0257] Preferably, the grid-connected power feedback triggers an online update process at preset time intervals, in which the risk-weighted decision model parameters and the final stage parameters of the reversible generation model are synchronously adjusted according to the error gradient between the predicted power and the grid-connected power.

[0258] In order to maintain the consistency between the predicted link and the actual operation of the power grid during the typhoon, the present invention After entering the control center, follow the preset time intervals Trigger an online update process. Update using error gradient Synchronous correction.

[0259] Weight vector of risk-weighted decision model .

[0260] Reversible generative model tail parameters , thereby reducing both scheduling risk and wind speed sampling deviation.

[0261] Error construction and gradient acquisition:

[0262]

[0263] in, Represents the scheduling error.

[0264] In the interval Inner cumulative squared error:

[0265]

[0266] in is the cumulative square error; the weight vector of the risk-weighted decision model Gradient And the final parameters of the reversible generation model Gradient All are obtained in one go through the automatic differentiation framework without explicit derivation.

[0267] Risk weight vector online update:

[0268]

[0269] represents the weight learning rate;

[0270] Represents the projection onto the probability simplex:

[0271]

[0272] Explanation of symbols represents the peak load risk weight; represents the risk weight of spare capacity; represents the risk weight of the ramp rate; the projection ensures that the three weights are always non-negative and their sum is 1.

[0273] The final parameters of the reversible generative model are updated. The final parameters refer specifically to the scale-translation network weights of the last two layers of coupling blocks, which are used to quickly correct the drift of the wind speed sample distribution. The Adam rule is used:

[0274]

[0275]

[0276] represents the final learning rate; are the first-order and second-order momentum coefficients; is a numerical stability constant.

[0277] The overall closed-loop workflow includes:

[0278] 1. Predict link output and ;

[0279] 2. Grid-connected measurement Post-calculation error ;

[0280] 3. If , execute the gradient accumulation of steps 1-3 and update synchronously ;

[0281] 4. Immediately regenerate the probabilistic wind field based on the new parameters and refresh the dispatch power curve to complete the closed loop.

[0282] Example, parameters , , . Continuous operation for 24 hours on the day of Typhoon Meihua’s landing:

[0283] The mean square error of wind speed samples converged to 58% of the baseline within 3 hours after updating; the average absolute error of dispatch power was 4.1%, compared with 7.9% of the unupdated baseline; and the peak load overlimit alarm was reduced from 6 times to 1 time.

[0284] The results show that the error gradient-driven dual adaptation can quickly correct wind field probability and scheduling decisions without interfering with upstream numerical weather models, thereby improving the reliability of grid-side scheduling under extreme typhoon scenarios.

[0285] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware.

[0286] The above are merely embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included within the scope of the claims of the present application.

Claims

1. A new energy power prediction method integrating typhoon meteorological information and micro-meteorological forecast results, characterized in that: include: Obtain wind farm satellite observations, radar wind profiles, ground wind measurements, and turbine operation data, and generate multi-source meteorological data through time correction, coordinate unification, quality control, and normalization. Performing numerical assimilation on the multi-source meteorological data to generate a mesoscale typhoon field, inferring the mesoscale typhoon field into a micrometeorological field through super-resolution, and fusing the data using a frequency domain phase consistency method to output a multi-scale fused meteorological field; Based on the multi-scale fusion meteorological field, probabilistic wind field data are sampled in a reversible generative model based on a streaming reversible network that satisfies mass conservation and momentum conservation constraints, and power prediction data is calculated by combining wake interaction with an energy conservation graph model based on a graph neural network. The power forecast data of the unit is input into the risk-weighted decision model, and the dispatch power curve and power uncertainty are generated according to the peak load condition risk value, the standby capacity condition risk value and the ramp rate condition risk value. The risk-weighted decision model parameters are updated online within a preset period according to the grid-connected power feedback, and the grid-side dispatchable power forecast data is output.

2. The method according to claim 1, characterized in that The multi-source meteorological data consists of satellite scatterometer observation data, airborne sounding profile data, three-dimensional Doppler radar wind profile data, sonic wind profiler observation data, laser wind radar observation data, drone micro-laminar flow observation data and unit operation monitoring data, and enters the subsequent processing flow after constructing a unified data structure under a unified time base and a unified geographic coordinate system.

3. The method according to claim 1, characterized in that The quality control first uses statistical threshold filtering to eliminate observations that exceed the abnormal judgment boundary, then uses smooth interpolation to fill in short-term continuous missing data, and uses geostatistical interpolation to fill in long-term continuous missing data. After the filling is completed, a quality identification code is set for each observation record.

4. The method according to claim 1, wherein The numerical assimilation adopts a four-dimensional variational assimilation algorithm, takes wind speed, temperature, humidity and pressure as control vectors, and generates a mesoscale typhoon field by minimizing the weighted error function between the background field and the observation field.

5. The method according to claim 4, characterized in that The resolution enhancement adopts a spectrum embedding diffusion network with a frequency domain convolution architecture, and converts the mesoscale typhoon field into a micro-meteorological resolution meteorological field through forward noise perturbation and reverse denoising reconstruction sequence mapping.

6. The method according to claim 5, characterized in that The frequency domain phase consistency method performs wavelet transform on the mesoscale typhoon field and the micro-meteorological resolution meteorological field, performs inverse transform after performing phase linear combination according to set weights within the corresponding frequency band, and generates a multi-scale fusion meteorological field.

7. The method according to claim 1, characterized in that The reversible generation model adopts a conditional reversible network with a residual coupling structure, records the logarithmic Jacobian in the forward mapping, and simultaneously minimizes the mass conservation residual term and the momentum conservation residual term in the reverse sampling process to obtain probabilistic wind field data.

8. The method according to claim 7, characterized in that The energy conservation graph model uses wind turbines and meteorological towers as nodes and wake interactions as edges. It calculates node latent vectors through multi-layer attention message passing, applies power conservation projection at the output layer, and maps wind speed information into unit power prediction data.

9. The method according to claim 1, characterized in that The risk-weighted decision model calculates risk weight coefficients based on the peak load condition risk value, the standby capacity condition risk value and the ramp rate condition risk value, respectively, applies the risk weight coefficients to the corresponding risk components in the power forecast data and performs weighted synthesis to generate a scheduling power curve and synthesize the power uncertainty according to the variance of each risk component.

10. The method according to claim 9, characterized in that The grid-connected power feedback triggers an online update process at preset time intervals, during which the risk-weighted decision model parameters and the final-stage parameters of the reversible generation model are synchronously adjusted according to the error gradient between the predicted power and the grid-connected power.

Images