Machine learning correction forecasting method for wind and light elements of new energy power station in complex terrain

Abstract

The invention discloses a machine learning correction forecasting method for wind and light elements of a new energy power station in a complex terrain, which relates to the technical field of data processing and comprises the following steps: acquiring a multi-source observation data set of a target area, and performing numerical forecasting assimilation on the multi-source observation data set of the target area; obtaining an output data set of the water-wind-light power station in the target area, and determining a forecasting factor through a statistical method based on the multi-source observation data set and the output data set assimilated by numerical forecasting; establishing and training a joint prediction model according to the prediction factors; and obtaining weather forecast data of the target area in the to-be-predicted time period, and predicting the hydroelectric power generation power, the wind power generation power and the photovoltaic power generation power of the water-wind-light power station in the target area in the to-be-predicted time period through the joint prediction model based on the weather forecast data of the target area in the to-be-predicted time period. According to the method, the initial field is improved through data assimilation, so that the numerical forecasting effect is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to a machine learning correction and prediction method for wind and light elements of new energy power stations under complex terrains. Background Art

[0002] Vigorously developing clean energies such as water, wind, and light is a major strategic measure to ensure future energy security and address global warming. Wind energy and solar energy are the new energies with the most promising large-scale development prospects. However, both are easily affected by meteorological factors, and their output powers have strong random volatility and are difficult to predict. With the continuous advancement of the construction of wind and photovoltaic power generation bases, the large-scale direct grid connection of wind and photovoltaic power will bring great pressure to the peak shaving, load regulation, and stable operation of the power system. Hydropower units have the characteristics of rapid start-stop, flexible operation, large amplitude of output power change, and fast response to load changes, and are ideal peak shaving power sources. Utilizing the natural complementarity of resources and the flexibility of hydropower to aggregate various energies such as water, wind, and light to form a multi-energy complementary power generation system is an effective way to reduce the impact of new energy grid connection and improve the utilization rate of basin resources.

[0003] Current power prediction methods all separately predict the power of hydropower, wind power, and photovoltaic power generation, without considering the influence among these several energies of water, wind, and light. Therefore, how to comprehensively consider the influence among water, wind, and light and perform joint power prediction on the power generation of water, wind, and light is an issue that needs to be considered to further improve the accuracy of power prediction for hydropower, wind power, and photovoltaic power.

[0004] Therefore, it is necessary to provide a machine learning correction and prediction method for wind and light elements of new energy power stations under complex terrains to improve the accuracy of power prediction for water, wind, and light power generation. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides a machine learning correction and prediction method for wind and light elements of new energy power stations under complex terrains, which can solve the problem that traditional power prediction methods all separately predict the power of hydropower, wind power, and photovoltaic power generation without considering the influence among these several energies of water, wind, and light.

[0007] To solve the above technical problems, the present invention provides the following technical solution. A machine learning correction and prediction method for wind and light elements of new energy power stations under complex terrains includes: Obtain the multi-source observation data set of the target area, and perform numerical prediction assimilation on the multi-source observation data set of the target area; obtain the output data set of the hydropower, wind power and photovoltaic power stations in the target area, and determine the prediction factors based on the multi-source observation data set of numerical prediction assimilation and the output data set through statistical methods; establish and train a joint prediction model according to the prediction factors; obtain the meteorological forecast data of the target area in the period to be predicted, and predict the hydropower generation power, wind power generation power and photovoltaic power generation power of the hydropower, wind power and photovoltaic power stations in the target area in the period to be predicted through the joint prediction model based on the meteorological forecast data of the target area in the period to be predicted.

[0008] As a preferred solution of the machine learning correction prediction method for the wind and light elements of new energy power stations under complex terrain according to the present invention, wherein: the determination of the prediction factors includes determining the spatio-temporal correlation between the meteorological factors and the total power generation, the spatio-temporal correlation between the meteorological factors and the hydropower generation power, the spatio-temporal correlation between the meteorological factors and the wind power generation power, and the spatio-temporal correlation between the meteorological factors and the photovoltaic power generation power based on the multi-source observation data set of numerical prediction assimilation and the output data set through statistical methods; Determine the complementary correlation between the hydropower generation power and the wind power generation power, the complementary correlation between the hydropower generation power and the photovoltaic power generation power, and the complementary correlation between the wind power generation power and the photovoltaic power generation power based on the output data set through statistical methods; Determine the prediction factors based on the spatio-temporal correlation between the meteorological factors and the total power generation, the spatio-temporal correlation between the meteorological factors and the hydropower generation power, the spatio-temporal correlation between the meteorological factors and the wind power generation power, the spatio-temporal correlation between the meteorological factors and the photovoltaic power generation power, the complementary correlation between the hydropower generation power and the wind power generation power, the complementary correlation between the hydropower generation power and the photovoltaic power generation power, and the complementary correlation between the wind power generation power and the photovoltaic power generation power; According to the lag factor corresponding to each prediction factor, establish a sequence corresponding to the prediction factor, and the joint prediction model assigns independent LSTM channels to each prediction factor to extract features respectively.

[0009] As a preferred solution of the machine learning correction prediction method for the wind and light elements of new energy power stations under complex terrain according to the present invention, wherein: the spatio-temporal correlation includes, for each of the meteorological factors, calculating the maximum mutual information coefficient between the meteorological factor and the total power generation based on the multi-source observation data set of numerical prediction assimilation and the output data set as the spatio-temporal correlation between the meteorological factor and the total power generation.

[0010] For each of the meteorological factors, calculate the maximum mutual information coefficient between the meteorological factor and the hydropower generation power based on the multi-source observation data set of numerical prediction assimilation and the output data set as the spatio-temporal correlation between the meteorological factor and the hydropower generation power.

[0011] For each of the above-mentioned meteorological factors, based on the multi-source observation dataset of numerical weather prediction assimilation and the output dataset, calculate the maximum mutual information coefficient between the meteorological factor and the wind power generation, as the spatio-temporal correlation between the meteorological factor and the wind power generation.

[0012] For each of the above-mentioned meteorological factors, based on the multi-source observation dataset of numerical weather prediction assimilation and the output dataset, calculate the maximum mutual information coefficient between the meteorological factor and the photovoltaic power generation, as the spatio-temporal correlation between the meteorological factor and the photovoltaic power generation.

[0013] As a preferred embodiment of the machine learning correction prediction method for new energy power station wind-solar elements under complex terrain according to the present invention, wherein: the spatio-temporal correlation includes generating a scatter plot of the meteorological factor and the total power generation based on the multi-source observation dataset of numerical weather prediction assimilation and the output dataset by statistical methods; using a variety of grid division schemes to divide the scatter plot into grids; for each of the grid division schemes, calculate the mutual information value of each grid under the grid division scheme, determine the maximum mutual information value corresponding to the grid division scheme, and normalize the maximum mutual information value corresponding to the grid division scheme; based on the maximum mutual information value corresponding to each grid division scheme, determine the maximum mutual information coefficient between the meteorological factor and the total power generation.

[0014] As a preferred embodiment of the machine learning correction prediction method for new energy power station wind-solar elements under complex terrain according to the present invention, wherein: the maximum mutual information coefficient includes calculating the mutual information value of each grid under the grid division scheme based on the following formula: , wherein, is the mutual information value of the grid, is the total number of values of the meteorological factor x in the grid, is the total power generation is the total number of values in the grid, is the meteorological factor and the total power generation is the joint probability density function of, is the marginal probability density function of the total power generation, is the marginal probability density function of the meteorological factor x.

[0015] As a preferred embodiment of the machine learning correction prediction method for new energy power station wind-solar elements under complex terrain according to the present invention, wherein: normalize the maximum mutual information value corresponding to the grid division scheme based on the following formula: , Among them, is the maximum mutual information value after normalization, is the minimum value function, a is the number of rows of the grid, and b is the number of columns of the grid.

[0016] As a preferred solution of the machine learning correction and prediction method for wind and light elements of new energy power stations under complex terrain according to the present invention, wherein: the multi-source observation data set of the target area includes hydropower, wind power and photovoltaic power station observation data, satellite meteorological observation data, radar meteorological observation data, ground meteorological observation data, aircraft meteorological observation data and radiosonde meteorological observation data.

[0017] As a preferred solution of the machine learning correction and prediction method for wind and light elements of new energy power stations under complex terrain according to the present invention, wherein: predicting the hydraulic power generation power, wind power generation power and photovoltaic power generation power of the hydropower, wind power and photovoltaic power stations in the target area during the period to be predicted includes using the WRF regional model for prediction to obtain numerical prediction data; using a machine learning algorithm to post-correct the numerical prediction data based on the processed multi-source observation data set; performing feature extraction, feature preprocessing, feature classification construction and feature combination on the numerical prediction data; using a machine learning algorithm to post-correct the numerical prediction data based on the feature combination of the numerical prediction data and the processed multi-source observation data set.

[0018] A computer device includes a memory and a processor, the memory stores a computer program, and is characterized in that when the processor executes the computer program, the steps of the machine learning correction and prediction method for wind and light elements of new energy power stations under complex terrain are realized.

[0019] A computer-readable storage medium stores a computer program thereon, and is characterized in that when the computer program is executed by a processor, the steps of the machine learning correction and prediction method for wind and light elements of new energy power stations under complex terrain are realized.

[0020] The beneficial effects of the present invention: The method of the present invention improves the initial field through data assimilation to improve the numerical prediction effect. Considering the spatio-temporal correlation of each element in the system and the complementarity between hydropower, wind power and photovoltaic power during the regulation process, and reflecting this correlation and complementarity through statistical methods, and then determining the prediction factors based on the correlation and complementarity; according to the prediction factors, a joint prediction model is established and trained to predict the hydraulic power generation power, wind power generation power and photovoltaic power generation power, improving the accuracy of the prediction of hydropower, wind power and photovoltaic power generation. Description of the Drawings

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0022] Figure 1 Schematic flow chart of the machine learning correction and prediction method for wind and light elements of a new energy power station under complex terrain provided by an embodiment of the present invention.

[0023] Figure 2 Schematic flow chart of numerical prediction assimilation of multi-source observation data sets by the machine learning correction and prediction method for wind and light elements of a new energy power station under complex terrain provided by an embodiment of the present invention. Detailed implementation manners

[0024] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will provide a detailed description of the specific implementation manners of the present invention with reference to the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0025] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0026] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that excludes other embodiments.

[0027] The present invention is described in detail in combination with schematic diagrams. When detailing the embodiments of the present invention, for the convenience of explanation, the cross-sectional views showing the device structure will be enlarged locally in a non-general proportion, and the schematic diagrams are only examples and should not limit the protection scope of the present invention herein. In addition, in actual production, three-dimensional spatial dimensions including length, width, and depth should be included.

[0028] Meanwhile, in the description of the present invention, it should be noted that the orientation or positional relationship indicated by terms such as "upper, lower, inner, and outer" is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be construed as a limitation to the present invention. In addition, the terms "first, second, or third" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance.

[0029] Unless otherwise clearly defined and limited in the present invention, the terms "installed, connected, and coupled" shall be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may also be a mechanical connection, an electrical connection, or a direct connection, and may also be indirectly connected through an intermediate medium, or may be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0030] Example 1, referring to Figure 1 - Figure 2 , which is the first embodiment of the present invention. This embodiment provides a machine learning correction and prediction method for wind and light elements of a new energy power station in complex terrain, including: Step 110, obtaining a multi-source observation data set of the target area.

[0031] In some embodiments, the multi-source observation data set of the target area at least includes observation data of a water-wind-solar power station, satellite meteorological observation data, radar meteorological observation data, ground meteorological observation data, aircraft meteorological observation data, and radiosonde meteorological observation data.

[0032] Step 120, performing numerical weather prediction assimilation on the multi-source observation data set of the target area.

[0033] For more descriptions on performing numerical weather prediction assimilation on the multi-source observation data set of the target area, reference can be made to Figure 2 and its related descriptions, which will not be elaborated here.

[0034] Step 130, obtaining the output data set of the water-wind-solar power station in the target area.

[0035] The output data set of the water-wind-solar power station in the target area may include the total output data, hydroelectric power output data, and photovoltaic power output data of the water-wind-solar power station. The output data set of the water-wind-solar power station in the target area can be trimmed and completed.

[0036] Step 140, determining the prediction factors based on the multi-source observation data set and the output data set obtained by numerical weather prediction assimilation through statistical methods.

[0037] In some embodiments, step 140 may specifically include: Based on the multi-source observation dataset and output dataset of numerical weather prediction assimilation through statistical methods, determine the spatio-temporal correlations between various meteorological factors and the total power generation, between various meteorological factors and the hydropower generation, between various meteorological factors and the wind power generation, and between various meteorological factors and the photovoltaic power generation.

[0038] Based on the output dataset through statistical methods, determine the complementary correlations between hydropower generation and wind power generation, between hydropower generation and photovoltaic power generation, and between wind power generation and photovoltaic power generation.

[0039] Based on the spatio-temporal correlations between various meteorological factors and the total power generation, between various meteorological factors and the hydropower generation, between various meteorological factors and the wind power generation, between various meteorological factors and the photovoltaic power generation, the complementary correlations between hydropower generation and wind power generation, between hydropower generation and photovoltaic power generation, and between wind power generation and photovoltaic power generation, determine the forecasting factors.

[0040] In some embodiments, based on the multi-source observation dataset and output dataset of numerical weather prediction assimilation through statistical methods, determining the spatio-temporal correlations between various meteorological factors and the total power generation, between various meteorological factors and the hydropower generation, between various meteorological factors and the wind power generation, and between various meteorological factors and the photovoltaic power generation includes: For each meteorological factor, based on the multi-source observation dataset and output dataset of numerical weather prediction assimilation, calculate the maximum mutual information coefficient between the meteorological factor and the total power generation as the spatio-temporal correlation between the meteorological factor and the total power generation.

[0041] For each meteorological factor, based on the multi-source observation dataset and output dataset of numerical weather prediction assimilation, calculate the maximum mutual information coefficient between the meteorological factor and the hydropower generation as the spatio-temporal correlation between the meteorological factor and the hydropower generation.

[0042] For each meteorological factor, based on the multi-source observation dataset and output dataset of numerical weather prediction assimilation, calculate the maximum mutual information coefficient between the meteorological factor and the wind power generation as the spatio-temporal correlation between the meteorological factor and the wind power generation.

[0043] For each meteorological factor, based on the multi-source observation dataset and output dataset of numerical weather prediction assimilation, calculate the maximum mutual information coefficient between the meteorological factor and the photovoltaic power generation as the spatio-temporal correlation between the meteorological factor and the photovoltaic power generation.

[0044] In some embodiments, based on the multi-source observation dataset and the output dataset of numerical weather prediction assimilation, calculating the maximum mutual information coefficient between the meteorological factors and the total power generation as the spatio-temporal correlation between the meteorological factors and the total power generation, includes: Generating a scatter plot of the meteorological factors and the total power generation based on the multi-source observation dataset and the output dataset of numerical weather prediction assimilation by statistical methods; dividing the scatter plot using multiple grid division schemes; for each grid division scheme, calculating the mutual information value of each grid under the grid division scheme, determining the maximum mutual information value corresponding to the grid division scheme, and normalizing the maximum mutual information value corresponding to the grid division scheme; determining the maximum mutual information coefficient between the meteorological factors and the total power generation based on the maximum mutual information value corresponding to each grid division scheme.

[0045] In some embodiments, the mutual information value of each grid under the grid division scheme can be calculated based on the following formula: , where, is the mutual information value of the grid, is the total number of values of the meteorological factor x in the grid, is the total power generation is the total number of values in the grid, is the meteorological factor and the total power generation is the joint probability density function of, is the marginal probability density function of the total power generation, is the marginal probability density function of the meteorological factor x.

[0046] In some embodiments, the maximum mutual information value corresponding to the grid division scheme is normalized based on the following formula: , where, is the normalized maximum mutual information value, is the minimum value function, a is the number of rows of the grid, and b is the number of columns of the grid.

[0047] In some embodiments, the maximum mutual information coefficient between the meteorological factors and the total power generation can be determined based on the following formula according to the maximum mutual information value corresponding to each grid division scheme: , where, ab < B(T) is the constraint condition of the total number of grid divisions, and B(T) is the 0.6th power of the total number of period data T.

[0048] The maximal information coefficient (MIC) is an index for measuring the correlation between two variables based on mutual information, which can measure the amount of information contained in one random variable about another random variable. MIC can measure the linear and non-linear relationships between two variables, and is not easily affected by outliers in the data, featuring universality, robustness and fairness.

[0049] The method for determining the spatio-temporal correlation between multiple meteorological factors and hydropower generation power, the spatio-temporal correlation between multiple meteorological factors and wind power generation power, and the spatio-temporal correlation between multiple meteorological factors and photovoltaic power generation power is similar to the method for determining the spatio-temporal correlation between multiple meteorological factors and total power generation, which will not be elaborated here.

[0050] Only by way of example, a meteorological factor with at least one of the spatio-temporal correlations with total power generation, the spatio-temporal correlation with hydropower generation power, the spatio-temporal correlation with wind power generation power, and the spatio-temporal correlation with photovoltaic power generation power greater than a preset spatio-temporal correlation threshold can be used as a forecasting factor.

[0051] Considering the time lag effect of the forecasting factors on the forecasting of hydropower generation power, wind power generation power and photovoltaic power generation power, lag correlation analysis is used to determine the influence degree of the forecasting factors on hydropower generation power, wind power generation power and photovoltaic power generation power at different lag periods, and the lag factor corresponding to each forecasting factor is determined.

[0052] Step 150, establish and train a joint prediction model according to the forecasting factors.

[0053] The joint prediction model can be a long short-term memory (LSTM) model.

[0054] Specifically, according to the lag factor corresponding to each forecasting factor, a sequence corresponding to the forecasting factor is established. For example: wind speed sequence: [wind speedt-1, wind speedt-2, ..., wind speedt-k]; irradiance sequence: [irradiancet-0, irradiance t-1, ..., irradiance t-m]. The joint prediction model assigns independent LSTM channels to each forecasting factor to extract its features respectively. An attention layer is added before feature fusion to dynamically adjust the importance of different forecasting factors to generate the fused features, and the output layer of the joint prediction model simultaneously predicts hydropower, wind power and photovoltaic power generation according to the fused features.

[0055] In some embodiments, during the process of training the joint prediction model, four indicators, namely the Nash-Sutcliffe efficiency coefficient (NSE), Bias, root mean square error (RMSE), and mean absolute error (MAE), can be selected to evaluate the prediction effect. The calculation formulas for each indicator are as follows: , , , , where, is the actual power value at the t-th time period; is the predicted power value at the t-th time period; T is the total number of time periods, and T represents the total number of time periods.

[0056] Step 160: Obtain the meteorological forecast data of the target area for the to-be-predicted time period.

[0057] Step 170: Based on the meteorological forecast data of the target area for the to-be-predicted time period, use the joint prediction model to predict the hydraulic power generation, wind power generation, and photovoltaic power generation of the water-wind-solar power station in the target area for the to-be-predicted time period.

[0058] Figure 2 is a schematic flow chart of numerical weather prediction assimilation for a multi-source observation dataset according to some embodiments of this specification. As Figure 2 shown, in some embodiments, the numerical weather prediction assimilation of the multi-source observation dataset of the target area includes the following steps: Step 210: Perform assimilation and data processing on the initial field of the multi-source observation dataset to obtain the multi-source observation dataset after data processing.

[0059] Perform repeatability, validity, persistence, and spatio-temporal consistency checks on the multi-source observation dataset, and give the quality-controlled observation data and corresponding quality labels, specifically including: Repeatability check: mainly judge whether there are multiple groups of approximate detection data when they are relatively close in time, longitude, latitude, and altitude.

[0060] Validity check: Its purpose is to judge the availability of the current message based on the observation data.

[0061] Continuous check: Check for unchanged or minimally changed observation data caused by instrument malfunctions. It is required that the difference between adjacent message data be greater than a given criterion. If the difference between adjacent message data is less than the given criterion and occurs continuously for 3 or more times, it is considered that the instrument malfunction fails the continuous check.

[0062] Extreme value check: It is required that the observed values of the ground area station's underlying surface consistency monitoring module for the observed elements should be within a reasonable range under the corresponding latitude and altitude conditions.

[0063] Position consistency check: Mainly to exclude data quality problems caused by time acquisition and positioning errors. The front and back messages can be used to determine whether the current message maintains consistency in terms of time, longitude, latitude, and altitude changes.

[0064] Temporal-spatial consistency check: The temporal-spatial consistency check can explore the reliability of messages during the processes of temporal and spatial changes.

[0065] Suspicious data check: Suspicious data refers to message data for which the above quality control methods are not applicable. For example, there is only one or two message quantities for the same flight, or there are too many error messages in the same flight, resulting in normal quality control being impossible. Such data is marked with a suspicious identifier for further processing manually or by other means.

[0066] Based on the position of the simulated area station in the model calculated by the model observation operator, collect the underlying surface information, compare it with the actual situation, and mark those with typical differences, specifically including: Based on the position of the simulated area station in the model calculated by the model observation operator; Collect the underlying surface information and compare it with the actual situation; Mark those with typical differences (such as land-water differences).

[0067] For all observation stations, especially those in harsh environments, detect the observation errors caused by equipment being exposed to sunlight, rain, sand and dust erosion, etc., as well as the incorrect observation data caused by obstacles within the observation field of view. Detect the obvious observation data without meteorological significance and eliminate it. Also, check the persistence, background consistency, etc. of the data, and correct the bias of the observation data. Specifically including: Station extreme value check: The station extreme value refers to the maximum and minimum values of a certain element that have occurred in the history of the detected station.

[0068] Time consistency check: The purpose of the time consistency check is to test the time change rate of the observation information or observed elements and identify undesirable sudden changes.

[0069] Spatial consistency check: The spatial consistency check is a method for quality control of the observed data of a station by making full use of the relationship between the observed data of the station to be inspected and the observed data of multiple adjacent stations at the same time.

[0070] Background field consistency check: Compare the difference between the observed data and the background field (abbreviated as the observed residual) with its criterion.

[0071] Satellite data needs to be converted into the PrepBUFR or BUFR (Binary Universal Form for the Representation of meteorological data) format. In the BUFR format, in addition to the brightness temperature of each observation channel of the satellite, data such as the longitude and latitude of the data grid points, the solar altitude angle, the solar azimuth angle, the satellite azimuth angle, the brightness temperature variance, and the clear sky ratio are also required. Among them, the brightness temperature variance and the clear sky ratio need to be calculated according to the diluted number of grid points. Using the BUFR program attached to the GSI module, the FY-2F observation data can be converted into the BUFR format.

[0072] The lowest resolution of the original FY-2F satellite data is 5 km (infrared channel), which is higher than the resolution of GSI assimilation (0.06°). Through the sparsification process, the error correlation between adjacent observations can be reduced. Referring to the BUFR data information of the GOES satellite, taking 9*9 pixel points as an observation unit, the values of the brightness temperature, longitude and latitude, solar altitude angle, etc. are taken at the center point, and the brightness temperature variance and clear sky coverage rate of the observation unit are calculated. Therefore, the resolution of the BUFR data becomes 40 km - 60 km.

[0073] Under clear sky conditions, in the infrared band, only the emission and absorption of the atmosphere and the surface are considered, and effects such as scattering are not considered. The radiative transfer equation can be written as: , where I is the radiation intensity received at the top of the atmosphere, is the surface emission, τ represents the optical thickness in the vertical direction from the surface to the atmosphere, is the optical thickness in the vertical direction of the entire atmosphere, µ is the cosine of the zenith angle, and B is the radiation intensity emitted by the atmosphere. The first term of the radiative transfer equation is the part of the radiation emitted by the surface that reaches the top of the atmosphere after being absorbed by the entire atmosphere, and the second term is the contribution of the radiation emitted by the atmosphere itself. Among them, for the part of the atmosphere emission, after step-by-step integration, the entire equation can be rewritten as: , where, , which is the variation of transmittance with height and is called the weighting function; H represents the vertical height range of the entire atmosphere. It can be regarded as the weight of each layer of the atmosphere in the radiation contribution of the entire atmosphere. To some extent, the height corresponding to the peak of the weighting function is the height that the satellite can "see". Most of the contributions to satellite observation information come from the heights with larger weighting functions. The distribution of the weighting function is related not only to the wavelength band but also to the gas composition, content, and atmospheric state profile in the atmosphere.

[0074] Since the radiation information in the infrared channel is very sensitive to clouds, cloud detection is an important part of the quality control of infrared channel radiation data. The FY-2F nominal disk data non-encrypted observation times include the cloud classification information calculated by the DPC system of the Satellite Meteorological Center. To remove cloud-contaminated data as much as possible and avoid mixing model information with observation field information, three independent cloud detection algorithms that do not rely on model calculations are used, and cloud detection is carried out respectively by setting empirical thresholds. The detection method is as follows: When the i-th grid point satisfies , and , it is judged as a cloudy pixel point. Among them, represents the vertical lapse rate of atmospheric temperature, and its value is 7K / 1000m. is the standard deviation of terrain height, is the standard deviation of the IR1 brightness temperature (taking 3*3 pixel points centered on i). , respectively represent the brightness temperatures of the i-th pixel point in the IR1 and IR4 channels, is the maximum brightness temperature value of the IR1 channel among 3*3 pixel points. The empirical thresholds corresponding to land and ocean surfaces are different; for the IR1IR4T cloud detection method, since the IR4 near-infrared is easily affected by solar radiation during the day, the empirical threshold is also different between day and night. It is defined that when the solar altitude angle at a certain point is greater than 0, the empirical threshold takes , and when the solar altitude angle is less than 0, the empirical threshold takes . The empirical thresholds , , , are shown in Table 1.

[0075] Table 1 , Radar assimilation is also a crucial component of data assimilation. Radar data quality issues caused by terrain obstruction and environmental noise are inspected and controlled. Fuzzy logic and other methods are used to address ground clutter, data shortages due to other factors, and clear-air echo quality issues, resulting in high-quality radar data after quality control. Interpolation and extension of data shortages can be implemented by leveraging the continuity of meteorological data, interpolating data from the same unit point over different time periods. Outlier echo processing: Doppler radar echoes have a threshold. Under normal circumstances, the echo will not exceed this threshold. However, outlier data often exceed this threshold, providing a basis for identifying outliers. Two-dimensional median filtering is often used to address outliers. Clear-air echo removal: Doppler radars sometimes generate echoes when detecting cloudless areas in the atmosphere. Terrain obstruction processing: Most echoes generated by terrain obstruction remain unchanged. Terrain echoes are identified and processed to remove those caused by terrain obstruction from the radar data. Using the longitude, latitude, and altitude of a Cartesian grid point, the elevation, azimuth, and slant range in the spherical coordinate system are calculated. Based on the calculated elevation, azimuth, and slant range in the radar spherical coordinate system, radial and azimuth interpolation is performed using the nearest neighbor method, and radial interpolation is performed vertically using linear interpolation. A value is assigned to each grid point to obtain the analysis value at that grid point. Three-dimensional gridded reflectivity data are constructed to form the data for assimilation system analysis. Rapid loop analysis based on numerical forecast results is conducted. A GSI assimilation module is established based on a Bayesian prior estimation model and a Gaussian probability distribution model to assimilate collected conventional and unconventional observation data. Variable scale iteration techniques, direct reflectivity assimilation, weak wind constraint techniques, cloud analysis false observation and particle information constraint assimilation, and convective scale control variable techniques are used to improve the quality of the analysis field, providing a better initial field for the next forecast.

[0076] The NMC method is used to calculate the forecast samples of the background error covariance. The forecast samples are obtained by performing forecast at different start times.

[0077] By comparing the effects of different control variables on assimilation and selecting the best one, we can construct a suitable background error covariance matrix that is consistent with the climatological scale characteristics of the simulated region. At the same time, we conduct variable analysis on the final statistical background error covariance matrix to find out its variation in different model heights.

[0078] Variable conversion: Establish a computational model to convert model prediction variables into control variables.

[0079] Latitude averaging: Use filtering and other methods to average the differences in the latitude of the control variables.

[0080] Decorrelation among variables: Through dynamic and statistical balance, the correlation of background errors between variables is removed and converted into uncorrelated control variables.

[0081] Vertical transformation: According to the empirical orthogonal function (EOF) decomposition, the vertical correlation within the control variables is eliminated, and the background error eigenvalues and eigenvectors are calculated.

[0082] Horizontal transformation: The recursive filter is used to calculate the horizontal characteristic length scale of the control variables, representing the characteristics of the background error field in the horizontal direction.

[0083] After the collected conventional / non-conventional observation data pass strict quality control, they need to be converted into a specific format and assimilated into the GSI assimilation system.

[0084] Select the conventional / non-conventional observation data with strict quality control during the forecast or simulation time period, and pick out the observation data in the 1.5-hour time domain before and after the assimilation time.

[0085] The selected observation data may have various storage formats, such as ASCII, BUFR, MADIS. Write a program to process all the observation data into the LITTLE_R format.

[0086] Run the OBSPROC data preprocessing program in the GSI assimilation system to convert the observation data in the LITTLE_R format into the PREBUFR format for assimilation.

[0087] For different types of observation data, considering different weather backgrounds, establish an observation operator that can reasonably map the relationship between the physical quantities of the atmospheric state and the observed physical quantities. Taking the establishment of the near-surface observation operator for surface observation data as an example, surface observation data is greatly affected by terrain and landform, and there is generally a certain height difference between the terrain of the general model and the terrain of the actual observation station. How to effectively solve the problem of the height difference between the model terrain and the observation station terrain has become a basic problem in surface data assimilation. Then, considering establishing a surface observation operator based on the dynamic and thermodynamic constraints of the near-surface boundary layer is an effective way to solve the above problem.

[0088] Incorporate the dynamic and thermodynamic processes of the boundary layer, and consider the height difference between the model terrain and the observation station terrain to establish a new regional near-surface observation operator.

[0089] Conduct strict tests and evaluations on the established near-surface observation operator, including accuracy tests (adjoint test and gradient test) of the tangent linear mode and adjoint mode of the model. Check the test results to ensure that the new near-surface observation operator can better describe the role of near-surface observations in assimilation.

[0090] The forecast effects vary with the height difference between different model terrains and the observed terrain. Therefore, it is necessary to determine the selection of the optimal critical height difference.

[0091] Use the newly established observation operator for subsequent assimilation experiments to ensure that the three-dimensional variational analysis system under the new surface observation operator can correctly reflect the interaction relationship between the wind field and other variables.

[0092] For a certain observation data and a certain forecast object, select an integer multiple of the horizontal radius dx of the model grid as the horizontal influence radius and conduct assimilation experiments. Increase or decrease the horizontal influence radius and conduct assimilation experiments.

[0093] Assimilate several times, compare the analysis fields obtained after assimilation under different horizontal influence radii with the observed field and the background field before assimilation, and evaluate the assimilation effects of different horizontal influence radii through increment or deviation analysis. Select the experiment with the best assimilation effect through comparison and determine the selection range of the horizontal influence radius.

[0094] Conduct assimilation experiments by changing the vertical influence scale size, compare the assimilation effects, and determine the selection range of the vertical influence radius.

[0095] Based on the Bayesian prior estimation model and the Gaussian probability distribution model, establish a 3DVAR assimilation module to assimilate the collected conventional / non-conventional observation data.

[0096] Use the meteorological initial field processed by WPS or the meteorological field predicted by WRF as the background field for three-dimensional variational assimilation. Select a reasonable background error covariance matrix statistically, link the preprocessed observation data, background data, and background error covariance with specific names, run the GSI assimilation system, and obtain the analysis field after assimilation at the corresponding moment as the background field for continued forecasting at the next moment.

[0097] According to the set assimilation time window, conduct multiple cyclic assimilations to continuously improve the forecast effect.

[0098] During the assimilation analysis process, for the same observation data and different forecast objects, change the influence scale factor in different iterative cycles to fully extract the effective information of the observation data at different scales.

[0099] For a certain observation data and a certain forecast object, at the initial analysis moment, select a value within the previously determined horizontal or vertical influence radius range as the horizontal or vertical influence scale.

[0100] As the weather system develops, the influence scales of observational data at different analysis times may not be the same. The analysis field obtained after the first assimilation is used as the background field for the next forecast. Continue the forecast until the next assimilation time. Select the same horizontal or vertical influence radius as the previous assimilation for the assimilation experiment. At the same time, conduct assimilation experiments by increasing or decreasing the influence radius in a cyclic manner. Compare the assimilation effects under different influence scales, and select the influence scale corresponding to the set of experiments where the obtained analysis field is closest to the observations as the horizontal or vertical influence scale for assimilation at this time.

[0101] For each subsequent analysis time, find the optimal horizontal or vertical influence radius for assimilation at this time, so as to fully extract the information of observational data at different scales.

[0102] For variable-scale assimilation analysis, multiple iterations are required. Utilize multi-threading and parallel computing technologies to improve the computational efficiency of the algorithm and accelerate the analysis efficiency of a large amount of data. Enable the system to meet the short-term and rapid analysis requirements.

[0103] Consult the literature to find the mathematical model of structural optimization design based on system reliability.

[0104] Improve the optimization criterion method used in solving the optimization model in the past.

[0105] Through the solution process of examples, verify the good effect of the improved optimization criterion method and ensure that the iterative convergence speed is accelerated.

[0106] Directly adopt a multi-level observation operator that includes various air particle phases, without inverting the reflectivity, to avoid the errors introduced in the inversion process and also avoid the problem that the warm cloud parameterization scheme, as an observation operator, cannot correctly reflect the ice phase process. Use the background temperature as the basis for particle classification to make the hydrometeor after assimilation more accurately reflect the actual observations.

[0107] The updated thermodynamic variables after assimilating the absorption observational data include water vapor mixing ratio, perturbation pressure, perturbation potential temperature, perturbation geopotential height, etc. This does not include the rain water mixing ratio that is closely related to the basic radar reflectivity factor. Therefore, in order to directly assimilate the echo data of the Doppler weather radar, in the assimilation framework, an analysis variable of rain water mixing ratio is introduced in the calculation.

[0108] Relate the echo intensity and the rain water mixing ratio according to the relationship between the radar reflectivity Z (dBZ) and the rain water mixing ratio qr.

[0109] Relate the rain water mixing ratio and the total liquid water mixing ratio through the warm rain scheme introduced as a constraint condition in the GSI assimilation system.

[0110] Based on the above relationships, realize the direct assimilation of radar echo intensity.

[0111] In the GSI objective function, add a divergence constraint term for analyzing the flow field. This term will serve as a weak constraint term to provide a balance relationship constraint for the wind field analysis results and improve the wind field analysis results.

[0112] The principle of GSI assimilation is to solve for the analysis field corresponding to the minimum value of the cost function composed of the background field, observation field, analysis field, observation operator, background error covariance, observation error covariance, etc. Without making any changes, use the original cost function for the assimilation experiment.

[0113] Add a small term related to the divergence wind in the original objective function for the balance constraint of the wind field to conduct the assimilation experiment.

[0114] Analyze and compare the assimilation results of the original scheme and after changing the objective function to ensure that the added constraint term is beneficial to the wind field analysis results. On this basis, continuously improve the constraint term until the wind field analysis results reach the best.

[0115] Construct a cloud analysis module based on cloud physics initialization. Cloud physics initialization is to extract information about clouds (mainly cumulus and stratus clouds) from ground truth, sounding, satellite, and radar observations, etc., improve the three-dimensional structure of clouds, and thus adjust the atmospheric humidity, temperature distribution, vertical and horizontal wind fields to make the initial state of the model closer to the true situation of the atmosphere. Through an empirical function, convert the relative humidity in the initial field into the cloud cover amount on the corresponding grid to obtain the three-dimensional cloud cover amount background field. Add the cloud base height and cloud cover amount information in the mesoscale observation data to obtain a comprehensive horizontal cloud field analysis.

[0116] The cloud top is obtained from satellite infrared data, while the radar reflectivity is mainly used to obtain information about clouds in the middle troposphere, and visible satellite cloud images are used to estimate the total cloud amount and calibrate the overestimation of clouds. Add infrared satellite cloud image data to obtain cloud top information. The three-dimensional cloud cover amount obtained from the first two steps, plus the initial temperature field, are used to correct the cloud analysis to make the cloud top brightness temperature of the grid analysis consistent with the observations. Add radar echo data. First, interpolate the reflectivity factor to the model grid points. Compare the reflectivity factor values at the grid points within the radar scan range with the threshold. If the reflectivity factor is lower than the threshold, it is regarded as clear sky. If it is higher than the threshold, correct the cloud base and cloud top. Add satellite visible cloud images to estimate the total cloud amount and calibrate the overestimation of clouds, and comprehensively correct the three-dimensional cloud amount to obtain the final three-dimensional cloud amount distribution of the model.

[0117] Multiple variables can be obtained in the cloud distribution area through cloud analysis. Different from traditional cloud analysis, these variables will enter the 3DVAR assimilation as false observations. The cloud analysis information will become a constraint information for the model assimilation analysis and enter the assimilation system. Through this analysis method, the lack of cloud and rain information, as well as the heat, humidity and other information in the cloud and rain area in the model, can be better supplemented. At the same time, in 3DVAR, due to the constraints of the model background field and the background error covariance matrix, the analysis variables will be more coordinated. Through the cloud analysis module, a strictly corrected three-dimensional cloud amount distribution map of the model is obtained. Variables such as humidity, cloud water, rain water, and temperature are extracted from the cloud distribution area. These variables are assimilated into the 3DVAR assimilation system as false observations as a constraint information for the assimilation analysis, so that the lack of cloud and rain information, as well as the heat, humidity and other information in the cloud and rain area in the model The control variables and the analysis variables are unified as u, v, w, t, q, and the stream function and potential function are no longer used as control variables. Such analysis results can better display the subtle features of meso- and small-scale. For any forecast object, the environmental field and meso- and small-scale features are analyzed at each analysis time. For convective-scale weather systems, in the GSI assimilation system, the control variables and the analysis variables are uniformly set as u, v, w, t, q. The assimilation results are analyzed to see whether the analysis results can better display the subtle features of meso- and small-scale.

[0118] Starting from a cold start at 1200 UTC every day, three-dimensional variational assimilation is updated every 3 hours. The assimilation analysis is carried out simultaneously in three regions of the model. The assimilation includes multi-source observational data (including conventional surface, ship, buoy, automatic station, aircraft message, sounding, etc. observational data) within the assimilation window of 1.5 hours before and after the analysis time, as well as unconventional observational data such as satellite and radar. Then, an 18-hour forecast is made until a new cold start at 1200 UTC the next day. The startup of the numerical weather prediction model requires a driving field, and the 3-hourly forecast requires the driving field to be updated at any time. This module is responsible for monitoring the update of the driving data in real time, collecting the driving data, and using it for the startup of the numerical weather prediction model. Use shell scripts to regularly monitor the update of global forecast fields such as ECMWF, and obtain the latest driving data information. Use shell scripts to start the download software, automatically download the driving data, and store and back up the data by category and date for convenient future data retrieval. According to business requirements, determine the spatial range and model resolution of the model area. Consider whether to set nested domains for large-scale fine forecasts, and determine the appropriate projection method according to business requirements. Set the start and end times of the model simulation. If the 3-hourly startup system is used, use shell scripts to update the time settings in the model. The startup of the model requires surface information as the lower boundary condition. This module interpolates the available global terrain static data of the model to the simulation area. Read the simulation area information in the model preprocessing module, and calculate the scale factors of the longitude, latitude, and each grid point in the map. According to the interpolation methods of each variable, interpolate variables such as soil type, land use, terrain height, monthly vegetation cover, and monthly albedo to the model grid points.

[0119] The numerical weather prediction model needs to be started with specific variable data and data formats. When the output results of the global model or reanalysis data are used as the driving field of the mesoscale model, due to their special storage formats and diverse variables, they cannot directly drive the numerical model. Therefore, this module is required to decode and extract the driving field data. Select the coding table applicable to the driving field. The data released by the commonly used NCEP and ECMWF adopt the GRIB format, and both formats have their own coding tables, namely Vtable files. Before extracting the driving field, it is necessary to first select the applicable Vtable. Extract the driving variables. Through the determined coding table, determine which fields need to be extracted from the GRIB file, and write the extracted file into an intermediate file in a transitional format. Interpolate the extracted meteorological elements horizontally onto the simulation area determined in the model preprocessing module and write them in a data format that can be directly absorbed by the model. Read the TBL file. This module uses the TBL file to control how the meteorological elements are interpolated. The TBL file provides an interval for each meteorological element, within which the interpolation method of the element field is determined. After interpolation, the meteorological elements are written. Usually, the interpolated meteorological elements are written in a data format that can be directly absorbed by the model, usually in NetCDF format, which is convenient for visualization software to achieve visualization.

[0120] Through horizontal and vertical interpolation, process the global model background field data onto the grid of the regional numerical weather prediction to produce the lateral boundary conditions for the regional numerical weather prediction. Read the driving field data, prepare the soil field for the model (usually interpolated to the required height), and verify that the soil classification, land use, soil temperature, and surface temperature are consistent with each other. After verifying the soil field, interpolate the upper-air variables to the model calculation surface, that is, the set vertical levels, to generate the initial conditions and lateral boundary conditions.

[0121] Focusing on the occurrence and development of small and medium-scale high-impact weather systems, the settings of the model are the same as above. Through multiple sets of sensitivity tests, compare and obtain the parameterization scheme suitable for the simulation area. Read the initial conditions and lateral boundary conditions generated in the lateral boundary processing module, and start the model to generate the simulation results. Organize and back up the simulation results.

[0122] Step 220, use the WRF regional model for forecasting to obtain numerical weather prediction data.

[0123] Step 230, use machine learning algorithms to post-correct the numerical weather prediction data based on the processed multi-source observation data set.

[0124] In some embodiments, step 230 may specifically include: Extract features, preprocess features, construct feature classification, and combine features for the numerical weather prediction data; Use machine learning algorithms to post-correct the numerical weather prediction data based on the feature combination of the numerical weather prediction data and the processed multi-source observation data set.

[0125] Specifically, the functions of GSI and WRF not only improve the spatio-temporal resolution of the numerical weather prediction but also further improve the prediction accuracy of the model. However, on this basis, a machine learning method is adopted to more accurately predict the meteorological elements at the target site in order to lay a good foundation for the following power prediction. The process includes the following steps: Download historical observation data and perform data cleaning on the historical observation data, including screening valid data, judging the target threshold, judging the uniqueness of dates, supplementing dates, and processing missing data; Download historical NWP numerical weather prediction data and perform feature engineering on the NWP data, including feature extraction, feature preprocessing, feature classification construction, and feature combination; Input the features and the cleaned observed data into the model for learning, and different models will have different learning results. Select the best model from them and save the model parameters; Directly input the NWP data into the previously trained model to output the corrected single-station numerical weather prediction product.

[0126] In some embodiments, four machine learning models, namely Decision Tree (DT), Random Forest, GBDT (Gradient Boosting Decision Tree), and XGBoost, are used for optimal selection.

[0127] Finally, it should be understood that the embodiments described in this specification are only used to illustrate the principles of the embodiments of this specification. Other deformations may also fall within the scope of this specification. Therefore, by way of example and not limitation, alternative configurations of the embodiments of this specification can be regarded as consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly introduced and described in this specification.

[0128] Embodiment 2 The second embodiment of the present invention is different from the previous one in that: If the said function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or a part of this technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. And the aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0129] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks.

[0130] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks.

[0131] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks.

[0132] Although the preferred embodiments of the present application have been described, those skilled in the art can make additional changes and modifications once they learn the basic creative concepts. Therefore, the appended claims are intended to be construed to include the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0133] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these changes and modifications.

Claims

1. Machine learning correction and prediction method for wind and light elements of new energy power stations under complex terrain, characterized in that: including, obtaining a multi-source observation data set of a target area, and performing numerical prediction assimilation on the multi-source observation data set of the target area; obtaining an output data set of a water-wind-solar power station in the target area, and determining a prediction factor based on the multi-source observation data set obtained by numerical prediction assimilation and the output data set through a statistical method; establishing and training a joint prediction model according to the prediction factor; obtaining meteorological forecast data of the target area in a to-be-predicted time period, and predicting the hydraulic power generation power, wind power generation power, and photovoltaic power generation power of the water-wind-solar power station in the target area in the to-be-predicted time period through the joint prediction model based on the meteorological forecast data of the target area in the to-be-predicted time period.

2. The machine learning correction and prediction method for wind and light elements of a new energy power station under complex terrain according to claim 1, wherein: The determining of the prediction factor includes: determining the spatio-temporal correlation between a meteorological factor and the total power generation power, the spatio-temporal correlation between the meteorological factor and the hydraulic power generation power, the spatio-temporal correlation between the meteorological factor and the wind power generation power, and the spatio-temporal correlation between the meteorological factor and the photovoltaic power generation power based on the multi-source observation data set obtained by numerical prediction assimilation and the output data set through a statistical method; determining the complementary correlation between the hydraulic power generation power and the wind power generation power, the complementary correlation between the hydraulic power generation power and the photovoltaic power generation power, and the complementary correlation between the wind power generation power and the photovoltaic power generation power based on the output data set through a statistical method; determining the prediction factor based on the spatio-temporal correlation between the meteorological factor and the total power generation power, the spatio-temporal correlation between the meteorological factor and the hydraulic power generation power, the spatio-temporal correlation between the meteorological factor and the wind power generation power, the spatio-temporal correlation between the meteorological factor and the photovoltaic power generation power, the complementary correlation between the hydraulic power generation power and the wind power generation power, the complementary correlation between the hydraulic power generation power and the photovoltaic power generation power, and the complementary correlation between the wind power generation power and the photovoltaic power generation power; establishing a sequence corresponding to the prediction factor according to the lag factor corresponding to each prediction factor, and the joint prediction model allocates independent LSTM channels for each prediction factor to extract features respectively.

3. The machine learning correction and prediction method for wind and light elements of a new energy power station under complex terrain according to claim 2, wherein: The spatio-temporal correlation includes: for each of the meteorological factors, calculating the maximum mutual information coefficient between the meteorological factor and the total power generation power based on the multi-source observation data set obtained by numerical prediction assimilation and the output data set, and using it as the spatio-temporal correlation between the meteorological factor and the total power generation power; for each of the meteorological factors, calculating the maximum mutual information coefficient between the meteorological factor and the hydraulic power generation power based on the multi-source observation data set obtained by numerical prediction assimilation and the output data set, and using it as the spatio-temporal correlation between the meteorological factor and the hydraulic power generation power; for each of the meteorological factors, calculating the maximum mutual information coefficient between the meteorological factor and the wind power generation power based on the multi-source observation data set obtained by numerical prediction assimilation and the output data set, and using it as the spatio-temporal correlation between the meteorological factor and the wind power generation power; for each of the meteorological factors, calculating the maximum mutual information coefficient between the meteorological factor and the photovoltaic power generation power based on the multi-source observation data set obtained by numerical prediction assimilation and the output data set, and using it as the spatio-temporal correlation between the meteorological factor and the photovoltaic power generation power.

4. The machine learning correction and prediction method for wind and light elements of a new energy power station under complex terrain according to claim 3, wherein: The spatio-temporal correlation includes generating a scatter plot of the meteorological factors corresponding to the total power generation based on the multi-source observation data set assimilated by numerical prediction and the output data set through a statistical method; Using a variety of grid division schemes to divide the scatter plot into grids; For each of the grid division schemes, calculating the mutual information value of each grid under the grid division scheme, determining the maximum mutual information value corresponding to the grid division scheme, and normalizing the maximum mutual information value corresponding to the grid division scheme; Based on the maximum mutual information value corresponding to each of the grid division schemes, determining the maximum mutual information coefficient between the meteorological factors and the total power generation.

5. The machine learning correction prediction method for wind and light elements of a new energy power station under complex terrain according to claim 4, characterized in that: The maximum mutual information coefficient includes calculating the mutual information value of each grid under the grid division scheme based on the following formula: ， Among them, is the mutual information value of the grid, is the total number of values of the meteorological factor x in the grid, is the total generated power is the total number of values in the grid, is the meteorological factor and the total generated power is the joint probability density function, is the marginal probability density function of the total generated power, is the marginal probability density function of the meteorological factor x.

6. The machine learning correction and prediction method for wind and light elements of a new energy power station under complex terrain according to claim 5, characterized in that: Normalizing the maximum mutual information value corresponding to the grid division scheme based on the following formula: ， Among them, is the maximum mutual information value after normalization, is the minimum value function, a is the number of rows of the grid, and b is the number of columns of the grid.

7. The machine learning correction and prediction method for wind and light elements of a new energy power station under complex terrain according to claim 6, wherein: The multi-source observation data set of the target area includes hydropower, wind power and photovoltaic power station observation data, satellite meteorological observation data, radar meteorological observation data, ground meteorological observation data, aircraft meteorological observation data and radiosonde meteorological observation data.

8. The machine learning correction and prediction method for wind and light elements of new energy power stations under complex terrains according to claim 7, characterized in that: Predicting the hydropower generation power, wind power generation power and photovoltaic power generation power of the hydropower, wind power and photovoltaic power stations in the target area during the period to be predicted includes: Using the WRF regional model for forecasting to obtain numerical forecast data; Using a machine learning algorithm to post-correct the numerical forecast data based on the multi-source observation data set after data processing; Performing feature extraction, feature preprocessing, feature classification construction and feature combination on the numerical forecast data; Using a machine learning algorithm to post-correct the numerical forecast data based on the feature combination of the numerical forecast data and the multi-source observation data set after data processing.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1 to 8.

Images