Wind and light energy prediction method and device and electronic equipment

Abstract

The embodiment of the invention provides a wind and light energy prediction method and device and electronic equipment, and relates to the field of wind and light energy prediction. The wind and light energy prediction method comprises the following steps: acquiring multi-source observation data; performing time and spatial resolution resampling on the obtained multi-source observation data; a wind field solving module and a radiation transmission module are integrated in the numerical weather forecasting mode; a turbulence closed equation set is embedded in the wind field solving module; enabling an aerosol correction factor in the radiation transfer module; in combination with observation data in a historical period, parameter tuning is carried out on the formed coupling system; different surface thermal field disturbance schemes are set in the target area, and multiple groups of wind field and solar radiation field forecasts are generated; according to the method, the problems of insufficient coupling of the physical process and the chemical process of the atmospheric boundary layer, lack of dynamic feedback of meteorological and pollutant disturbance, insufficient high-temporal-spatial-resolution prediction and the like in the prior art are effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind and solar energy prediction, and in particular to a wind and solar energy prediction method, device and electronic equipment. Background Art

[0002] As clean and renewable energy sources, wind power and solar energy are increasingly receiving attention from the power industry and environmental protection departments. Existing wind and solar energy forecasts usually use numerical weather forecast models or machine learning algorithms to simulate and correct wind speed and radiation fields. However, due to the complexity of factors such as nonlinear turbulence in the atmospheric boundary layer, the interaction between aerosols and radiation, and random disturbances in the distribution of pollution sources, traditional methods often find it difficult to accurately describe the impact of micro-meteorological changes and atmospheric chemical processes on wind and light energy, and the prediction accuracy and temporal and spatial resolution are still insufficient.

[0003] To further improve the accuracy of wind and solar energy forecasts, it is necessary to construct a coupled online model with multi-physical parameterization and multi-chemical parameterization schemes to address problems such as insufficient coupling between atmospheric physics and chemistry, lack of dynamic feedback on aerosol and pollution source disturbances, and insufficient model integration capabilities for multi-source data. In addition, it should combine meteorological field ensemble perturbation experiments with rolling correction strategies, fully consider the impact of complex terrain, pollution source lists, and real-time observation data, and achieve high-precision prediction and flexible scheduling of the spatiotemporal distribution of wind and solar energy. Summary of the invention

[0004] According to an embodiment of the present invention, a wind and solar energy prediction method, device and electronic device are provided to solve the problems raised by the above background technology.

[0005] In a first aspect of the present invention, a method for predicting wind and solar energy is provided, comprising the following steps:

[0006] Step 1. Obtain multi-source observation data: collect wind speed, solar radiation intensity, cloud distribution and environmental pollutant content in the target area, and extract corresponding terrain and land use information;

[0007] Step 2. Resample the multi-source observation data obtained in step 1 in terms of time and space resolution, and use data assimilation techniques to correct outliers or missing values;

[0008] Step 3. Integrate the wind field solution module and the radiation transfer module in the numerical weather forecast model to simulate the boundary layer structure and solar energy arrival in the target area; embed the turbulence closed equation group in the wind field solution module to process the terrain influence and surface friction in blocks; enable the aerosol correction coefficient in the radiation transfer module to improve the simulation accuracy of atmospheric transmittance;

[0009] Step 4. Based on the preprocessing results of step 2, the wind speed and solar radiation intensity are predicted and corrected using a regression algorithm or a deep neural network model, and the corrected data are output to the multi-stage numerical simulation model constructed in step 3;

[0010] Step 5. Combine the observation data in the historical period to optimize the parameters of the coupled system formed by steps 3 and 4; dynamically adjust the key physical parameters of the wind field solution module and the radiation transmission module through the adaptive optimization algorithm, and correct the network structure or hyperparameters of the machine learning model;

[0011] Step 6. Set different surface thermal field disturbance schemes in the target area to generate multiple sets of wind field and solar radiation field forecasts; for the wind energy prediction value and light energy prediction value generated by each set of disturbance schemes, calculate the deviation from the actual observation, and select the set of results with the smallest comprehensive error as the final prediction output;

[0012] Step 7. Visualize the set of wind energy prediction values ​​and solar energy prediction values ​​selected in step 6, and output a corresponding prediction report in combination with the grid dispatching needs and environmental management requirements.

[0013] Preferably, the data assimilation technology described in step 2 adopts a combination of interpolation correction and ensemble Kalman filtering to dynamically assign weights to data from different observation sources, thereby effectively reducing data uncertainty.

[0014] Preferably, the multi-stage numerical simulation model described in step 3 also includes a cloud droplet condensation and water vapor diffusion simulation unit, which is used to evaluate the weakening effect of cloud cover on solar radiation and the influence of cloud top height on wind energy distribution.

[0015] Preferably, the machine learning algorithm fusion process described in step 4 establishes a multi-input and multi-output network structure to simultaneously perform end-to-end prediction of wind speed, wind direction, and light intensity, thereby reducing the cumulative error caused by single-factor prediction.

[0016] Preferably, the adaptive optimization algorithm described in step 5 includes at least one of a genetic algorithm, a particle swarm algorithm or a differential evolution algorithm, which is used to automatically search for the optimal parameter combination of the numerical simulation model and the machine learning model during the iteration process.

[0017] Preferably, the collective perturbation scheme described in step 6 randomly samples different surface temperature gradients, humidity distributions, and aerosol absorption coefficients in the target area, and simultaneously introduces the variation of surface roughness to enhance the diversity of the perturbation test;

[0018] The comprehensive error assessment described in step 6 includes the weighted average of wind energy prediction deviation, light energy prediction deviation and pollutant dispersion simulation deviation, and is further combined with the stability index of extreme weather scenario testing for an overall score.

[0019] Preferably, the visualization presentation described in step 7 includes a wind power generation curve, a photovoltaic power generation curve and a geographic information system overlay layer in different time periods, so as to assist the power dispatching department in allocating wind and solar energy on different time scales;

[0020] The forecast report described in step 7 also includes a rolling assessment mechanism for the deviation between the model and the measured data. As new observation data are continuously added, the numerical simulation model and machine learning algorithm are periodically calibrated to continuously improve the forecast accuracy.

[0021] In a second aspect of the present invention, a wind and solar energy prediction device is provided, comprising:

[0022] Multi-source observation data acquisition module, used to obtain wind speed, solar radiation intensity, cloud cover and environmental pollutant content in the target area, and output for subsequent processing;

[0023] A data preprocessing and assimilation module, connected to the multi-source observation data acquisition module, is used to resample the observation data in time and space resolution, and correct outliers or missing values ​​using a data assimilation algorithm;

[0024] A multi-stage numerical simulation model building module, connected to the data preprocessing and assimilation module, is used to integrate the wind field solution submodule and the radiation transfer submodule in the numerical weather forecast model to simulate the boundary layer structure and solar energy arrival of the target area;

[0025] A machine learning fusion module is connected to the multi-stage numerical simulation model construction module, and is used to predict and correct the wind speed and light intensity, and feed the correction results back to the multi-stage numerical simulation model construction module;

[0026] A mode iteration training and adaptive optimization module is connected to the machine learning fusion module, and is used to use the observation data in the historical period to perform parallel parameter adjustment on the multi-stage numerical simulation model construction module and the machine learning fusion module, and automatically search for the optimal physical parameters and network structure according to the adaptive algorithm;

[0027] An ensemble perturbation and verification module, connected to the model iterative training and adaptive optimization module, is used to perform multiple numerical simulations for different surface and meteorological perturbation scenarios, and compare the simulation results with actual observations to select the wind and light energy prediction scheme with the smallest error;

[0028] The wind and solar energy forecast result publishing module is connected to the collective disturbance and verification module, and is used to visualize the selected forecast scheme and output the wind power generation curve and photovoltaic power generation curve, as well as generate a forecast report for power grid scheduling and environmental protection management.

[0029] Preferably, a real-time data transmission channel is set between the wind and solar energy prediction result publishing module and the multi-source observation data acquisition module, which is used to periodically calibrate the multi-stage numerical simulation model construction module and the machine learning fusion module after obtaining new observation data, so as to realize rolling optimization of the wind and solar energy prediction device.

[0030] In a third aspect of the present invention, an electronic device is provided, comprising: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate via the bus, and the machine-readable instructions are executed by the processor to execute the steps of the above-mentioned wind and solar energy prediction method.

[0031] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0032] The present invention incorporates aerosol radiation correction, terrain friction refinement and pollution source disturbance scheme design into the numerical weather forecast model, and combines machine learning algorithms to perform deviation correction and real-time rolling optimization of wind speed and radiation field, effectively overcoming the problems of insufficient coupling of physical and chemical processes in the atmospheric boundary layer, lack of dynamic feedback of meteorological and pollutant disturbances, and insufficient high-temporal and spatial resolution prediction in the prior art; at the same time, the adaptability and robustness of the model are further enhanced through collective perturbation and adaptive optimization, so that wind energy and solar energy can still be predicted with high precision under complex boundary layers and extreme weather conditions, thereby providing more reliable and flexible decision support for power grid dispatching and environmental management.

[0033] It should be understood that the contents described in the summary of the invention are not intended to limit the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The above and other features, advantages and aspects of the embodiments of the present invention will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, wherein:

[0035] Figure 1 A flow chart of a wind and solar energy prediction method according to an embodiment of the present invention is shown;

[0036] Figure 2 A system block diagram of a wind-solar energy prediction device according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] In addition, the term "and / or" in this article is only a description of the association relationship between the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0039] like Figure 1 As shown, in a first aspect of the present invention, a method for predicting wind and solar energy is provided, comprising the following steps:

[0040] Step 1. Acquisition of multi-source observation data;

[0041] Deploy or call existing monitoring stations in the target area to collect real-time or scheduled data on wind speed, solar radiation intensity, cloud distribution, and environmental pollutant content. According to research needs, wind speed at 10 meters (or other appropriate standard heights), total sky radiation, equivalent direct radiation value, cloud percentage, and PM2.5, O3 and other pollutant monitoring data can be selected.

[0042] Obtain terrain data (such as digital elevation model DEM) and land use type information (such as forests, farmlands, urban areas, water bodies, etc.) of the target area from remote sensing images or geographic information system (GIS) databases.

[0043] In windy areas, install high-precision ultrasonic anemometers or LiDAR to obtain wind profile information with higher spatial resolution;

[0044] Spectral irradiance meters are deployed in photovoltaic concentrated areas or desert areas to monitor the impact of solar spectrum characteristics on photovoltaic power generation.

[0045] Step 2. Data preprocessing and assimilation;

[0046] The observation data obtained in step 1 are interpolated or aggregated according to the time resolution required by the target prediction model (for example, 15 minutes, 30 minutes or 1 hour); for spatial resolution, the data can also be gridded using methods such as bilinear interpolation or inverse distance weighting to keep consistent with the grid resolution of the numerical weather forecast model.

[0047] Perform threshold determination, box plot method or cluster screening on obvious noise or outliers in data of wind speed, radiation intensity and other elements;

[0048] Use time series interpolation (such as linear interpolation, spline interpolation) or machine learning algorithms (such as KNN filling) to reasonably estimate missing values.

[0049] Data assimilation methods such as ensemble Kalman filtering (EnKF) or three-dimensional variational (3D-Var) can be used to combine the observational data with the initial or boundary fields of numerical weather prediction (NWP) to improve the accuracy of initial conditions for subsequent numerical simulations.

[0050] By comparing the deviations between the observed values ​​and the output values ​​after model correction, the correction effect of the assimilation process on wind speed, solar radiation intensity and pollutant concentration field is evaluated.

[0051] Step 3. Construction of multi-stage numerical simulation model;

[0052] WRF (Weather Research and Forecasting), COSMO or other regional numerical weather forecast models can be selected, and the model's terrain, land use and emission source list can be regionalized.

[0053] Turbulence closed equations setting: Use appropriate boundary layer schemes (such as YSU, MYJ, etc.) to simulate turbulent transport, and select different horizontal and vertical resolutions according to the complexity of the terrain in the study area;

[0054] Terrain influence and surface friction block processing: Divide the target area into several sub-areas with similar terrain features and surface roughness, and assign values ​​to the surface heat flux and friction coefficient respectively;

[0055] Aerosol correction factor is enabled: Based on long-term statistics or online coupling, the absorption and scattering effects of aerosols in the atmosphere are corrected to improve the simulation accuracy of solar radiation transmittance and reflectivity;

[0056] Atmospheric longwave and shortwave radiation scheme: RRTMG (Rapid Radiative Transfer Model for GCMs) or other advanced radiation schemes are preferred, and corrected in combination with aerosol optical depth (AOD).

[0057] The present invention divides wind field solution and radiation transmission into two stages: initial meteorological forecast and refined radiation field simulation. The initial meteorological forecast provides a background for large-scale airflow fields or regional circulations, and the refined radiation field simulation considers microphysical processes such as aerosols and cloud droplet condensation on this basis.

[0058] Step 4. Machine learning algorithm integration;

[0059] Input: meteorological elements such as wind speed, solar radiation intensity, and pollutant concentration after assimilation in step 2;

[0060] Output: Corrected values ​​for wind speed, light intensity, etc. or intermediate results for directly predicting wind energy and light energy.

[0061] Regression algorithms (such as support vector regression SVR, random forest RF) or deep neural networks (DNN) can be used. If DNN is used, structures such as multi-layer perceptrons or convolutional neural networks can be built to capture spatiotemporal features.

[0062] Forecast bias correction: The wind speed and radiation field output by the numerical weather forecast model are used as the initial input, and their deviations are corrected through machine learning to obtain wind speed and light intensity that are closer to actual observations;

[0063] Feature extraction: If a deep network is used, the coupling relationship between local meteorological conditions and pollutant concentrations can be extracted in the hidden layer to provide more refined input for subsequent steps.

[0064] Step 5. Pattern iterative training and adaptive optimization;

[0065] The observation data within a historical period (such as the past 1 year or longer) is paired with the output of the numerical weather forecast model to form a training sample, covering different seasons, different weather types and different pollution source emission states.

[0066] Numerical model tuning: using multi-source observation data and assimilation results, adjust the boundary layer parameterization, cloud microphysics parameterization, and aerosol correlation coefficient;

[0067] Machine learning model tuning: optimize network depth, learning rate, regularization coefficient, or kernel function parameters.

[0068] Genetic algorithm, particle swarm algorithm or differential evolution algorithm can be preferably selected to automatically search for the optimal physical parameters and network structure in the iterative process by defining a comprehensive objective function (such as minimizing the prediction deviation of wind energy and light energy).

[0069] When the training error or validation error is lower than the set threshold, or the number of iterations reaches the upper limit, the adaptive optimization is stopped and the optimal model parameter set is output.

[0070] Step 6. Collective perturbation and verification;

[0071] Set up some disturbance solutions in the target area, including but not limited to:

[0072] Surface thermal field disturbance: introducing different degrees of positive and negative disturbances into the surface temperature field or surface radiation balance;

[0073] Aerosol distribution disturbance: adjust the aerosol optical depth (AOD) or pollution source emission inventory to simulate the impact of different pollution scenarios on wind and sunlight forecasts;

[0074] Boundary layer turbulence perturbations: Modify key parameters in the turbulence parameterization scheme, such as roughness length, mixing length, etc.

[0075] The input data under each disturbance scheme is sequentially sent to the optimal coupling mode obtained in step 5, and multiple sets of wind speed fields, radiation fields and corresponding wind energy and light energy calculation results are output.

[0076] By comparing with the actual observation values, the deviation and correlation coefficient of each group of results at different times and measurement points are calculated;

[0077] Based on deviation statistics (such as RMSE, MAE, BIAS) and stability indicators (such as prediction reliability under extreme weather conditions), a set of results with the smallest error and the best robustness are comprehensively selected as the final prediction output.

[0078] Step 7. Release of wind and solar energy forecast results;

[0079] Visualize selected wind and solar energy forecast results, such as:

[0080] Wind energy output curve by time period: shows the trend of wind energy output changing over time within one day or several days;

[0081] Light energy distribution heat map: showing the photovoltaic power generation potential in different geographical locations.

[0082] According to the grid dispatching needs, the wind and solar forecast results are summarized with information such as load curves and the operating status of the energy storage system to provide a basis for the power dispatching center to make adjustments. Combined with environmental protection management requirements, the pollutant concentration and the impact of emission sources are analyzed to ensure the use of clean energy while taking into account environmental quality.

[0083] The above analysis results are made into a forecast report, which includes forecast accuracy assessment, possible extreme scenarios and response strategy recommendations, and serves as a reference for decision-making by power departments or environmental protection authorities.

[0084] The present invention incorporates aerosol radiation correction, terrain friction refinement and pollution source disturbance scheme design into the numerical weather forecast model, and combines machine learning algorithms to perform deviation correction and real-time rolling optimization of wind speed and radiation field, effectively overcoming the problems of insufficient coupling of physical and chemical processes in the atmospheric boundary layer, lack of dynamic feedback of meteorological and pollutant disturbances, and insufficient high-temporal and spatial resolution prediction in the prior art; at the same time, the adaptability and robustness of the model are further enhanced through collective perturbation and adaptive optimization, so that wind energy and solar energy can still be predicted with high precision under complex boundary layers and extreme weather conditions, thereby providing more reliable and flexible decision support for power grid dispatching and environmental management.

[0085] The data assimilation technique described in step 2 uses a combination of interpolation correction and ensemble Kalman filtering;

[0086] For the temperature, humidity, wind speed, pollutant concentration and other element data collected by different observation stations in the region, the interpolation algorithm can be used to align the time and space and make preliminary corrections. For example, the inverse distance weighted method (IDW) is used to interpolate the observation data with inconsistent geographical locations. The interpolation process can be expressed as:

[0087]

[0088] Among them, V(x) represents the value of the element at the target position x after interpolation, V i is the element value of the known observation site i, d i is the distance between observation site i and target location x, and p is the distance decay exponent (usually 2).

[0089] The Ensemble Kalman Filter (EnKF) fuses the initially processed observation data with the background field of the numerical model on the basis of interpolation correction to reduce data uncertainty. The update process of the Ensemble Kalman Filter can be expressed as follows:

[0090]

[0091] in, is the prior (predicted) state vector of the ensemble Kalman filter at the kth moment, is the a posteriori (analysis) state vector, is the observation vector, H is the observation operator, is the covariance matrix of the prior state vector, R is the observation error covariance matrix, and K is the Kalman gain matrix.

[0092] When there are several ensemble members in the target area (such as 30 sets of numerical simulations under different initial conditions), by comparing the differences between actual observations and the outputs of each ensemble member, the wind field, pollutant distribution field and other elements in the numerical model are updated in real time to obtain more accurate initial / boundary conditions and reduce the deviation between the background field and the observed values.

[0093] In this embodiment, the multi-stage numerical simulation model described in step 3 further includes a cloud droplet condensation and water vapor diffusion simulation unit;

[0094] Cloud droplet condensation process, in the numerical weather forecast model (WRF, COSMO or other regional models), add cloud droplet condensation and water vapor diffusion simulation units, mainly evaluate:

[0095] The weakening effect of cloud cover on solar radiation: By adding parameters such as cloud water content and cloud ice content to the radiation transfer equation, the absorption and scattering of shortwave radiation by clouds is calculated;

[0096] The impact of cloud top height on wind energy distribution: As the cloud body develops, it causes disturbances to factors such as temperature gradient and vertical velocity, which in turn affects wind shear in the boundary layer.

[0097] The following approximate formula for cloud droplet growth rate can be used:

[0098]

[0099] Where r is the cloud droplet radius and S is the supersaturation (i.e., the ratio of actual water vapor pressure to saturated water vapor pressure). In the numerical model, piecewise integration is used for different cloud droplet scales to accurately describe the cloud microphysical process.

[0100] Water vapor diffusion simulation, in cloud microphysics or microscale models, the calculation of water vapor diffusion can be done using the finite difference method, performing the following discretization for each grid or vertical layer:

[0101]

[0102] Among them, q v is the water vapor mixing ratio, u is the wind speed vector, and D is the diffusion coefficient. This unit is used to simulate the cloud-rain process and water vapor redistribution to better reflect the actual impact of cloud cover on the distribution of light and wind speed.

[0103] In this embodiment, the machine learning algorithm fusion process described in step 4 is through establishing a multi-input multi-output network structure;

[0104] The multi-input multi-output structure design is used to simultaneously predict multiple meteorological factors such as wind speed, wind direction, and light intensity. The neural network can be configured as follows:

[0105] Input layer: includes multi-source data, such as temperature, humidity, surface roughness, pollutant concentration, cloud water content, etc. after assimilation;

[0106] Hidden layer: Several fully connected layers or convolutional layers to extract spatiotemporal features;

[0107] Output layer: defines a multi-dimensional output vector They correspond to the predicted values ​​of wind speed, wind direction, radiation intensity and other factors.

[0108] End-to-end prediction, through an end-to-end approach, the predicted values ​​of multiple meteorological quantities can be obtained in one forward propagation, avoiding the cumulative errors caused by single-factor models one by one and the coupling inconsistency problems between different models.

[0109] Furthermore, a weighted multi-objective loss function can be defined, for example:

[0110]

[0111] Among them, MSE represents mean square error, and α and β are weight coefficients of different factors to balance the attention to the accuracy of each meteorological element.

[0112] In this embodiment, the adaptive optimization algorithm described in step 5 includes at least one of a genetic algorithm, a particle swarm algorithm, or a differential evolution algorithm;

[0113] Taking the genetic algorithm (GA) as an example, a number of “individuals” or “chromosomes” can be defined in the initial population, each of which contains a numerical pattern and a set of parameters of the machine learning model: X = (p 1 ,p 2 , ..., p m ), where p m Represents a model parameter or network structure hyperparameter to be optimized.

[0114] Fitness function, which can be defined as the inverse or negative value of the prediction error, such as:

[0115] F(X)=-(γ 1 ×E wind +γ 2 ×E solar )

[0116] Among them, E wind ,E solar They represent the forecast deviations of wind energy and solar energy respectively (mean square error or absolute error can be used), γ 1 ,γ 2 is the weight coefficient.

[0117] According to the crossover and mutation operations of the genetic algorithm, the current population is iteratively updated:

[0118] Crossover: Randomly select two chromosomes and exchange some gene segments;

[0119] Mutation: fine-tune certain gene positions in the chromosome with a certain probability (for example, make small random changes to the selection of physical parameterization schemes or learning rate values). When the fitness function no longer improves significantly or reaches the preset maximum number of iterations, output the optimal solution X * , and apply it to numerical simulation models and machine learning models.

[0120] In this embodiment, the collective perturbation scheme described in step 6 performs random sampling of different surface temperature gradients, humidity distributions, and aerosol absorption coefficients within the target area.

[0121] Furthermore, the surface temperature gradient can be randomly sampled according to a normal distribution within [-2°C, +2°C];

[0122] Humidity distribution: Add ±5% random disturbance to the original background field;

[0123] The aerosol absorption coefficient can be sampled according to the historical observation range, such as 0.8, 1.20.8, 1.20.8, and 1.2 times the basic value.

[0124] In addition, the surface roughness z is introduced 0 The change in the amount of logarithmic distribution or uniform distribution in [z 0 -δ,z 0 +δ] to simulate different terrain and underlying surface conditions.

[0125] Numerical patterns and machine learning models are used to jointly predict each set of disturbance parameter combinations to form an "ensemble member" (usually no less than 10 groups) to cover more potential meteorological scenarios and extreme conditions.

[0126] The comprehensive error assessment described in step 6 includes the weighted average of wind energy prediction deviation, light energy prediction deviation and pollutant dispersion simulation deviation, and is combined with the stability index of the extreme weather scenario test for an overall score.

[0127] Let E wind represents the wind energy forecast deviation, E solar represents the light energy prediction deviation, E pollutant Represents the pollutant dispersion prediction deviation, and the weighted average error can be defined as:

[0128] E total =λ 1 E wind +λ 2 E solar +λ 3 E pollutant

[0129] Among them, λ 1 ,λ 2 ,λ 3 It is an adjustable weight coefficient, which is set according to power dispatch or environmental protection requirements.

[0130] In order to measure the prediction performance under extreme weather conditions (such as heavy rain, sandstorms, extreme high temperatures, etc.), a stability index SI can be defined to evaluate the degree of discreteness of the ensemble prediction results, such as:

[0131]

[0132] Among them, E i is the prediction error value corresponding to the ith set disturbance member, is the average error value. The smaller the value, the lower the prediction dispersion of the model in extreme scenarios, that is, the higher the stability.

[0133] E can be considered comprehensively total Assigned together with SI, define:

[0134]

[0135] Among them, α and β are adjustment coefficients, which are used to balance the weights of comprehensive error and stability. Finally, the set of results with the largest score is taken as the optimal prediction solution.

[0136] In this embodiment, the visualization presentation described in step 7 includes time-divided wind power generation curves, photovoltaic power generation curves, and geographic information system overlay layers to assist the power dispatching department in allocating wind and solar energy on different time scales.

[0137] Furthermore, by calculating the wind speed field and unit characteristic curve of the target area hourly (or shorter time intervals), a time series of wind power generation is generated and presented in the form of a line graph or a bar graph, which is convenient for the power dispatching department to grasp the daily and multi-day trends of wind power output. In view of factors such as light intensity, component conversion efficiency and temperature compensation, the photovoltaic power generation is calculated hourly or daily by the radiation field and meteorological elements output by each module of the present invention, and a visualization curve is formed. The curve can intuitively reflect the peak light period, the morning and evening attenuation trend, and the fluctuation of photovoltaic production capacity on rainy days. The data output by numerical simulation and machine learning of wind field, solar radiation distribution, aerosol concentration, etc. are used as raster layers or vector layers, and superimposed on the regional map for spatial visualization. Operators can interactively browse the distribution of wind and light energy in different geographical locations in the GIS platform, identify high potential areas or potential risk areas, and compare them with terrain features, land use types and power grid infrastructure layout.

[0138] The forecast report described in step 7 also includes a rolling assessment mechanism for the deviation between the model and the measured data. As new observation data are continuously added, the numerical simulation model and machine learning algorithm are periodically calibrated to continuously improve the forecast accuracy.

[0139] After the prediction model releases the results, the actual observed wind power, photovoltaic power generation and related meteorological data are compared with the prediction output, and the prediction deviation is recorded; according to the physical process (such as boundary layer turbulence, aerosol absorption, etc.) or data-driven process (such as under-adjustment of a layer of parameters in the machine learning network) where the deviation may come from, the symptoms are corrected; combined with the actual needs of power dispatch or monitoring cycle (such as daily, weekly or monthly), the measured data is continuously collected and updated, and the prediction model deviation is dynamically tracked to ensure the real-time evaluation results. After the new observation data is continuously added, it is added to the historical data set, and the initial / boundary conditions of the original numerical simulation model and the training samples of the machine learning algorithm are incrementally or fully updated; after a certain amount or time span of data accumulation is met, the parameters of the numerical simulation model and the machine learning algorithm are automatically triggered to be recalibrated; if a machine learning model is used, the network structure can be automatically searched or the weight can be fine-tuned regularly; the calibrated model output is compared with the measured data again. If the prediction deviation is significantly improved, the new model replaces the old model to maintain the continuous high accuracy of the prediction system.

[0140] like Figure 2 As shown, in a second aspect of the present invention, a wind and solar energy prediction device is provided, comprising:

[0141] The multi-source observation data acquisition module is used to obtain observation data such as wind speed, solar radiation intensity, cloud cover and environmental pollutant content in the target area, and transmit it to subsequent modules. Meteorological stations, high-altitude monitoring equipment and pollutant monitors can be deployed on site, or corresponding data can be obtained through the open data interface of the meteorological and environmental protection departments. The temporal and spatial consistency of data can be ensured by reasonably setting the acquisition frequency and synchronization mechanism. It is connected to the downstream data preprocessing and assimilation modules to transmit the collected raw data via the network or storage medium for subsequent multi-stage numerical simulation and machine learning processing.

[0142] The data preprocessing and assimilation module is used to receive multi-source observation data, perform operations such as outlier removal, missing value filling, interpolation resampling, and correct the deviation between the numerical background field and the observation data through the data assimilation algorithm. The interpolation algorithm is first used to spatially process the unevenly distributed observation data, and then the assimilation strategy such as ensemble Kalman filtering or three-dimensional variational method is used to merge the processed measured data with the initial / boundary conditions of the numerical model to obtain more accurate meteorological fields and pollutant fields. It is connected between the multi-source observation data acquisition module and the multi-stage numerical simulation model construction module. On the one hand, it obtains the original data from the observation end, and on the other hand, it outputs the assimilated results to provide higher quality input for subsequent numerical simulation and machine learning modules.

[0143] The multi-stage numerical simulation model construction module is connected to the data preprocessing and assimilation module, and is used to integrate the wind field solution submodule and the radiation transfer submodule in the numerical weather forecast mode to simulate the boundary layer structure and solar energy arrival of the target area. In the numerical weather forecast mode, the wind field solution submodule and the radiation transfer submodule are respectively integrated to simulate the boundary layer wind field and solar energy arrival of the target area; and it includes a cloud droplet condensation simulation unit and a surface friction calculation unit to further evaluate the weakening effect of cloud cover on radiation and the correction of complex terrain on the wind field. The wind field solution submodule sets the corresponding turbulence closure scheme according to the terrain and underlying surface characteristics; the radiation transfer submodule combines the factors of cloud droplet condensation and water vapor diffusion to perform refined calculations on short-wave and long-wave radiation. The surface friction calculation unit divides different friction parameters for mountainous areas, plains or coastal areas; the cloud droplet condensation simulation unit can combine factors such as cloud water content and temperature gradient to simulate the attenuation degree of cloud cover on radiation. It receives the high-precision initial / boundary conditions output by the data preprocessing and assimilation module, provides the basic numerical results of the wind field and radiation field for the machine learning fusion module, and can adjust the internal physical parameters after obtaining the correction signal.

[0144] The machine learning fusion module is connected to the multi-stage numerical simulation model construction module. On the one hand, it receives the numerical simulation results for correction, and on the other hand, it returns the corrected prediction results or deviation signals to the numerical model so as to correct the physical parameterization configuration in subsequent iterative training. The machine learning fusion module is used to predict and correct key meteorological elements such as wind speed and light intensity, and feed back the correction results to the multi-stage numerical simulation model construction module. Carry out prediction correction for key meteorological elements such as wind speed, wind direction, and light intensity; realize end-to-end prediction of multiple meteorological quantities through a multi-input and multi-output network structure, and feed back the prediction deviation signal to the numerical simulation model construction module. In the hidden layer, fully extract the correlation information between factors such as terrain features, aerosol concentration, cloud cover, and wind field and radiation field, and then correct the deviation of the numerical model output. Through a multi-input and multi-output method, the predicted values ​​of wind speed, wind direction, irradiance, etc. are generated at one time, reducing the error accumulation caused by the series connection of multiple models.

[0145] The mode iteration training and adaptive optimization module is connected to the machine learning fusion module. On the one hand, it obtains the prediction error information from the machine learning fusion module, and on the other hand, it obtains the simulation deviation of the physical process from the numerical simulation model; through interactive adjustment, it outputs the optimal parameter combination to the collective perturbation and verification module. The mode iteration training and adaptive optimization module is used to use the observation data in the historical period to adjust the parameters of the multi-stage numerical simulation model construction module and the machine learning fusion module in parallel, and automatically search for the optimal physical parameters and network structure according to the adaptive algorithm. On the basis of making full use of historical observation data, the mode iteration training and adaptive optimization module adjusts the parameters of the multi-stage numerical simulation model and the machine learning fusion module in parallel; and searches for the optimal physical parameters and network structure through the particle swarm algorithm, genetic algorithm or differential evolution algorithm. A comprehensive error index can be set to quantify the prediction errors of wind energy, light energy, etc. into one or more objective functions; then the adaptive optimization algorithm iteratively updates in the candidate parameter space to select the optimal solution. This process can be triggered periodically according to the addition of new observation data to maintain the high adaptability of the model in different seasons or extreme weather.

[0146] The ensemble perturbation and verification module is connected to the model iteration training and adaptive optimization module, accepts the optimal parameter set output by the model iteration training and adaptive optimization module, and generates multiple sets of perturbation schemes; after comparing the evaluation results, the optimal prediction scheme is passed to the wind and solar energy prediction result release module, and the error information can be fed back for further optimization iteration. The ensemble perturbation and verification module is used to perform multiple numerical simulations for different surface and meteorological perturbation scenarios, and compare the simulation results with actual observations to select the wind and solar energy prediction schemes with the smallest error. For the optimized numerical simulation and machine learning models, a variety of surface and meteorological perturbation scenarios (such as temperature gradient, aerosol absorption coefficient, humidity distribution, etc.) are designed, and the wind and solar energy prediction results under each perturbation scenario are compared and analyzed to select the scheme with the smallest comprehensive error and the highest stability. Under the original optimal parameter conditions, random perturbations of different amplitudes can be applied to key elements such as surface temperature, humidity, and aerosols to generate several ensemble members; then the output corresponding to each member is compared with the measured data to comprehensively evaluate the deviations of wind energy, solar energy, and pollutant dispersion simulations.

[0147] The wind and solar energy forecast result release module is connected to the collective disturbance and verification module to receive the optimal solution finally selected; it provides the graphical forecast report and error assessment results to the power dispatching department or environmental protection management department, so that they can quickly understand and make relevant decisions. The wind and solar energy forecast result release module is used to visualize the verified optimal forecast solution and output wind power and photovoltaic power generation curves; it includes a visualization submodule and an uncertainty assessment submodule to provide risk assessment recommendations in extreme weather scenarios. The visualization submodule is responsible for presenting wind power and photovoltaic power generation in different time periods in the form of curves or charts, and can be superimposed on the GIS map layer to achieve an intuitive display of high-potential areas or wind and solar resource distribution.

[0148] The uncertainty assessment submodule integrates the simulation results of multiple groups of disturbance members and gives the prediction error distribution or confidence interval, reserving response strategies and deployment measures for the power grid dispatching department and the ecological management department.

[0149] In this embodiment, a real-time data transmission channel is set between the wind and solar energy prediction result publishing module and the multi-source observation data acquisition module, which is used to periodically calibrate the multi-stage numerical simulation model construction module and the machine learning fusion module after obtaining new observation data, so as to realize rolling optimization of the wind and solar energy prediction device.

[0150] In a third aspect of the present invention, an electronic device is provided, comprising: a processor, a memory and a bus. The memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and the machine-readable instructions are executed by the processor to execute the steps of the above-mentioned wind and solar energy prediction method.

[0151] Furthermore, the processor is used to execute machine-readable instructions stored in the memory to acquire, assimilate, and integrate numerical weather forecasting and machine learning algorithms for multi-source observation data. The processor may be a CPU, GPU, or other chip with parallel computing capabilities (such as FPGA, ASIC) to accommodate the large amount of data and complex numerical calculation requirements in wind and solar energy forecasting. When executing the wind and solar energy forecasting method of the present invention, the processor schedules multiple threads or multiple processes to execute numerical simulation and machine learning modules in parallel or in blocks; the parameters are updated in real time according to the adaptive optimization strategy to ensure high-precision forecasts in different seasons or extreme weather scenarios.

[0152] The memory is used to store machine-readable instructions that can be executed by the processor, as well as intermediate data, historical observation data and model parameters generated during the prediction process. The memory may include flash memory, DRAM or other non-volatile and volatile storage media. Multi-source observation data are stored in layers through appropriate file system or database management methods; the trained machine learning model parameters, numerical weather forecast initial fields and boundary conditions, and the settings of the ensemble perturbation scheme are persistently saved to facilitate seamless recovery of the prediction process when the device is restarted or switched; a cache area can be set up in the memory as needed to improve the access speed to high-frequency read and write data (such as real-time observations, rolling calibration parameters).

[0153] The bus is used to transmit instructions and data between the processor and the memory, ensuring that the various steps of the wind and solar energy prediction method can be efficiently linked at the hardware level. High-speed parallel bus, PCIe or multi-channel memory controller and other technologies are used to reduce data transmission bottlenecks in model training and numerical simulation stages; multi-level bus architecture can be selected according to the system scale to achieve data interaction with external sensors, network ports or other peripherals.

[0154] After the electronic device is powered on or started, the processor loads machine-readable instructions related to the wind and solar energy prediction method from the memory and initializes the required data structures, including the observation data buffer, the model parameter table, and the optimization algorithm configuration.

[0155] According to the stored program logic, the processor obtains multi-source observation data through the network interface or sensor drive, and uses the assimilation module to interpolate the data, process missing values ​​and correct the background field; finally, the assimilated results are stored in a pre-allocated cache area.

[0156] The processor calls the wind field solution, radiation transfer and multi-input multi-output network structure modules in the order of machine-readable instructions to calculate the boundary layer wind field, cloud droplet condensation and solar radiation; and then combines the machine learning module to perform bias correction and end-to-end meteorological quantity prediction.

[0157] During each training cycle or when new observation data is monitored, the processor executes an adaptive optimization algorithm (such as particle swarm, genetic algorithm, etc.) to search for and update the optimal parameter combination; then the ensemble perturbation module is called to simulate different surface, aerosol and humidity scenarios, and the result deviations are comprehensively compared and screened.

[0158] Finally, the processor outputs the optimal wind and solar energy prediction plan in the form of wind power curves, photovoltaic power curves or GIS layers according to program instructions, and writes or transmits it to an external interface so that power dispatching and environmental protection management departments can obtain relevant report information.

[0159] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A wind and solar energy prediction method, characterized in that: The steps include: Step 1. Obtain multi-source observation data: collect wind speed, solar radiation intensity, cloud distribution and environmental pollutant content in the target area, and extract corresponding terrain and land use information; Step 2. Resample the multi-source observation data obtained in step 1 in terms of time and space resolution, and use data assimilation techniques to correct outliers or missing values; Step 3. Integrate the wind field solution module and the radiation transfer module in the numerical weather forecast model to simulate the boundary layer structure and solar energy arrival in the target area; embed the turbulence closed equation group in the wind field solution module to process the terrain influence and surface friction in blocks; enable the aerosol correction coefficient in the radiation transfer module to improve the simulation accuracy of atmospheric transmittance; Step 4. Based on the preprocessing results of step 2, the wind speed and solar radiation intensity are predicted and corrected using a regression algorithm or a deep neural network model, and the corrected data are output to the multi-stage numerical simulation model constructed in step 3; Step 5. Combine the observation data in the historical period to optimize the parameters of the coupled system formed by steps 3 and 4; dynamically adjust the key physical parameters of the wind field solution module and the radiation transmission module through the adaptive optimization algorithm, and correct the network structure or hyperparameters of the machine learning model; Step 6. Set different surface thermal field disturbance schemes in the target area to generate multiple sets of wind field and solar radiation field forecasts; for the wind energy prediction value and light energy prediction value generated by each set of disturbance schemes, calculate the deviation from the actual observation, and select the set of results with the smallest comprehensive error as the final prediction output; Step 7. Visualize the set of wind energy prediction values ​​and solar energy prediction values ​​selected in step 6, and output a corresponding prediction report in combination with grid dispatching needs and environmental management requirements.

2. The wind and solar energy prediction method according to claim 1, characterized in that: The data assimilation technology described in step 2 uses a combination of interpolation correction and ensemble Kalman filtering to dynamically assign weights to data from different observation sources, thereby effectively reducing data uncertainty.

3. The wind and solar energy prediction method according to claim 2, characterized in that: The multi-stage numerical simulation model described in step 3 also includes a cloud droplet condensation and water vapor diffusion simulation unit, which is used to evaluate the weakening effect of cloud cover on solar radiation and the influence of cloud top height on wind energy distribution.

4. The wind and solar energy prediction method according to claim 1, characterized in that: Step 4 establishes a multi-input and multi-output network structure to simultaneously perform end-to-end prediction of wind speed, wind direction, and light intensity, thereby reducing the cumulative error caused by single-factor prediction.

5. The wind and solar energy prediction method according to claim 1, characterized in that: The adaptive optimization algorithm described in step 5 includes at least one of a genetic algorithm, a particle swarm algorithm or a differential evolution algorithm, which is used to automatically search for the optimal parameter combination of the numerical simulation model and the machine learning model during the iteration process.

6. The wind and solar energy prediction method according to claim 1, characterized in that: Step 6: Randomly sample different surface temperature gradients, humidity distributions, and aerosol absorption coefficients in the target area, and simultaneously introduce the variation of surface roughness to enhance the diversity of the disturbance test; Step 6 includes the weighted average of wind energy prediction deviation, light energy prediction deviation and pollutant dispersion simulation deviation, and further combines the stability index of extreme weather scenario test to make an overall score.

7. The wind and solar energy prediction method according to claim 1, characterized in that: The visualization presentation described in step 7 includes time-divided wind power generation curves, photovoltaic power generation curves, and geographic information system overlay layers, which are used to assist the power dispatching department in allocating wind and solar energy on different time scales; The forecast report described in step 7 also includes a rolling assessment mechanism for the deviation between the model and the measured data. As new observation data are continuously added, the numerical simulation model and machine learning algorithm are periodically calibrated to continuously improve the forecast accuracy.

8. A wind and solar energy prediction device, characterized in that: include: Multi-source observation data acquisition module, used to obtain wind speed, solar radiation intensity, cloud cover and environmental pollutant content in the target area, and output for subsequent processing; A data preprocessing and assimilation module, connected to the multi-source observation data acquisition module, is used to resample the observation data in time and space resolution, and correct outliers or missing values ​​using a data assimilation algorithm; A multi-stage numerical simulation model building module, connected to the data preprocessing and assimilation module, is used to integrate the wind field solution submodule and the radiation transfer submodule in the numerical weather forecast model to simulate the boundary layer structure and solar energy arrival of the target area; A machine learning fusion module, connected to the multi-stage numerical simulation model construction module, is used to predict and correct the wind speed and light intensity, and feed the correction results back to the multi-stage numerical simulation model construction module; A mode iteration training and adaptive optimization module is connected to the machine learning fusion module, and is used to use the observation data in the historical period to perform parallel parameter adjustment on the multi-stage numerical simulation model construction module and the machine learning fusion module, and automatically search for the optimal physical parameters and network structure according to the adaptive algorithm; An ensemble perturbation and verification module, connected to the model iterative training and adaptive optimization module, is used to perform multiple numerical simulations for different surface and meteorological perturbation scenarios, and compare the simulation results with actual observations to select the wind and light energy prediction scheme with the smallest error; The wind and solar energy forecast result publishing module is connected to the collective disturbance and verification module, and is used to visualize the selected forecast scheme and output the wind power generation curve and photovoltaic power generation curve, as well as generate a forecast report for power grid scheduling and environmental protection management.

9. The wind and solar energy prediction device according to claim 8, characterized in that: A real-time data transmission channel is set between the wind and solar energy prediction result publishing module and the multi-source observation data acquisition module, which is used to periodically calibrate the multi-stage numerical simulation model construction module and the machine learning fusion module after obtaining new observation data, so as to realize rolling optimization of the wind and solar energy prediction device.

10. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and the machine-readable instructions are executed by the processor to execute the steps of the wind and solar energy prediction method as described in any one of claims 1 to 7.

Images