Method and system for medium and long term weather prediction based on variable grid model and medium

Abstract

The application provides a medium, a long-term weather prediction method and system based on a variable grid model, wherein the method comprises the following steps: obtaining weather products based on initial prediction data and back calculation data; performing bias correction on the weather products based on a preset correction algorithm to obtain a prediction result; and performing evaluation based on the corrected prediction result, and outputting the prediction result when the evaluation result meets a demand target. The application has the characteristics of high calculation efficiency and a large number of ensemble samples, improves the refinement degree and prediction efficiency of the forecast, and adopts a traditional statistical and artificial intelligence hybrid correction method, so that the prediction accuracy can be greatly improved, and the demand of new energy development for medium and long-term weather prediction products can be met.

Description

Technical Field

[0001] This invention relates to the field of meteorological forecasting technology, and in particular to a medium- and long-term meteorological forecasting method, system, and medium based on a variable grid model. Background Technology

[0002] my country's wind resources are concentrated and large-scale, with most wind power bases located far from load centers. This has led to the proposal to transmit surplus wind power to load centers for consumption. However, transmitting wind power across regions to load centers involves a very large volume of wind power. Implementing short-term power trading would have too great an impact on power system dispatch and operation, making it impractical. Therefore, long-term power trading is needed. Consequently, medium- to long-term wind power generation forecasts are required for the regions where large wind power bases are located to provide a basis for the effective implementation of cross-regional, long-term power trading.

[0003] For current medium- and long-term wind resource forecasting, following the approach of short-term forecasting, if we start with numerical weather prediction and then use algorithms to forecast medium- and long-term wind resources, although we can obtain meteorological forecasts for the corresponding time scale, the accuracy and precision are low. Therefore, medium- and long-term forecasting cannot simply copy the methods of short-term forecasting. For example, even using a global uniform grid model cannot achieve high-resolution (10km) long-term multi-ensemble integral forecasting due to the extremely large computational load. Furthermore, S2S (Sub-seasonal to Seasonal) seasonal forecasting based on a global uniform grid model has low dynamic forecasting skill and cannot meet the needs of actual production and daily life. Summary of the Invention

[0004] The purpose of this invention is to provide a medium- and long-term weather forecasting method, system, and medium based on a variable grid model, to solve the problem of medium- and long-term weather forecasting.

[0005] Firstly, this application provides a medium- to long-term weather forecasting method based on a variable grid model, the method comprising:

[0006] Meteorological products are obtained based on initial forecast data and back-calculation data;

[0007] The meteorological products are corrected for deviations based on a preset correction algorithm to obtain prediction results;

[0008] The prediction results are evaluated based on the corrected prediction results, and the prediction results are output when the evaluation results meet the target requirements.

[0009] In one possible implementation of this application, the acquisition of meteorological products based on initial forecast data and back-calculation data specifically includes:

[0010] Obtain the initial prediction data and the back-calculation data predicted in real time by the prediction system;

[0011] A general standardization operation for the variable mesh is performed based on the initial prediction data, and the results of the standardization operation are used as the boundary forcing field of the variable mesh model.

[0012] The initial prediction data within the boundary forcing field are downscaled based on preset weights, and the scale bias is corrected.

[0013] Based on the back-calculation data, probability forecasts and anomaly forecasts are performed, and the meteorological product is obtained by combining the initial forecast data after correcting for scale bias. The meteorological product includes at least one meteorological element.

[0014] In one possible implementation of this application, the step of correcting the meteorological product based on a preset correction algorithm to obtain the prediction result specifically includes:

[0015] The correction algorithm is obtained, wherein the correction algorithm includes at least a classification meteorological element bias correction algorithm, a Kalman filter bias correction algorithm, a similarity method bias correction algorithm, a Bayesian bias correction algorithm, an empirical orthogonal function decomposition bias correction algorithm, and a deep learning bias correction algorithm.

[0016] The meteorological products are corrected based on the classification meteorological element deviation correction algorithm, the empirical orthogonal function decomposition deviation correction algorithm, and the deep learning deviation correction algorithm.

[0017] The variable mesh model is corrected based on the Kalman filter bias correction algorithm, the similarity method bias correction algorithm, and the Bayesian bias correction algorithm.

[0018] In one possible implementation of this application, the meteorological product is corrected based on the classification meteorological element deviation correction algorithm, specifically including:

[0019] Correcting temperature forecast biases, including average temperature, daily maximum temperature, and daily minimum temperature;

[0020] Correcting precipitation forecast bias, including correcting both the forecast precipitation frequency and the observed precipitation frequency;

[0021] Correcting wind speed forecast bias, including hourly average wind speed and direction, and maximum wind speed and direction.

[0022] In one possible implementation of this application, the variable mesh model is corrected based on the Kalman filter bias correction algorithm, specifically including:

[0023] The model error is updated to adjust the bias scale, which includes estimating the model forecast bias and determining the model forecast bias after weighted averaging.

[0024] Obtain error correction coefficients and correct the variable mesh model based on the error correction coefficients.

[0025] In one possible implementation of this application, the meteorological product is corrected based on a deep learning bias correction algorithm, specifically by processing and analyzing the time series data using a long short-term memory neural network model to correct the meteorological product.

[0026] In one possible implementation of this application, the evaluation based on the corrected prediction result, and the output of the prediction result when the evaluation result meets the required objective, specifically includes:

[0027] The first weather forecast score is obtained based on the surface non-precipitation elements in the meteorological products.

[0028] A second weather forecast score is obtained based on the precipitation elements in the aforementioned meteorological products;

[0029] The evaluation result is obtained based on the first weather forecast score and the second weather forecast score, wherein when the evaluation result meets the required objective, the corresponding prediction result is output.

[0030] Secondly, this application provides a medium- to long-term weather forecasting system based on a variable grid model, the system comprising:

[0031] The acquisition module is used to acquire meteorological products based on initial forecast data and back-calculation data.

[0032] The correction module is used to correct the deviation of the meteorological products based on a preset correction algorithm to obtain the prediction results;

[0033] The evaluation module is used to evaluate the corrected prediction results, and output the prediction results when the evaluation results meet the requirements.

[0034] Thirdly, this application provides the aforementioned computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the medium- and long-term weather forecasting method based on a variable grid model.

[0035] Fourthly, this application provides the aforementioned electronic device, which includes a processor and a memory; wherein the memory is used to store a computer program, and the processor is used to load and execute the computer program so that the electronic device executes the medium- and long-term weather forecasting method based on the variable grid model.

[0036] As described above, the medium- and long-term meteorological forecasting method, system, and medium based on the variable grid model of the present invention have the characteristics of high computational efficiency and large sample size, which improves the precision and efficiency of forecasts. Furthermore, the use of a hybrid correction method combining traditional statistics and artificial intelligence can greatly improve forecast accuracy and meet the demand of new energy development for medium- and long-term meteorological forecast products. Attached Figure Description

[0037] Figure 1 The diagram shows a schematic representation of the method steps in one embodiment of the medium- and long-term weather forecasting method based on a variable grid model of the present invention.

[0038] Figure 2 The diagram shows a schematic representation of the method steps in one embodiment of the medium- and long-term weather forecasting method based on a variable grid model of the present invention.

[0039] Figure 3 The diagram shows a schematic representation of the method steps in one embodiment of the medium- and long-term weather forecasting method based on a variable grid model of the present invention.

[0040] Figure 4 The diagram shows a schematic representation of the method steps in one embodiment of the medium- and long-term weather forecasting method based on a variable grid model of the present invention.

[0041] Figure 5 The diagram shows a schematic representation of the method steps in one embodiment of the medium- and long-term weather forecasting method based on a variable grid model of the present invention.

[0042] Figure 6 The diagram shows a schematic representation of the method steps in one embodiment of the medium- and long-term weather forecasting method based on a variable grid model of the present invention.

[0043] Figure 7 The diagram shows a training process of the medium- and long-term weather forecasting method based on a variable grid model according to an embodiment of the present invention.

[0044] Figure 8 The diagram shown is a structural schematic of a medium- and long-term meteorological forecasting system based on a variable grid model according to an embodiment of the present invention.

[0045] Figure 9 The diagram shown is a structural schematic of an electronic device according to an embodiment of the present invention.

[0046] Component designation explanation

[0047] S102~S106 step S202~S208 step S302~S306 step S402~S406 step S502~S504 step S602~S606 step 80 Medium- and Long-Term Meteorological Forecasting System Based on Variable Grid Model 81 Get Module 82 Correction module 83 Evaluation module Detailed Implementation

[0048] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0049] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0050] Wind resource forecasting is a crucial step in wind energy resource development planning. Assessing and forecasting long-term wind resource meteorological parameters for a region is beneficial for determining regional wind energy reserves and provides strong reference for wind farm site selection, wind turbine selection, and scheme formulation. The results of wind energy resource forecasting and assessment are essential for national and local governments in formulating wind power development plans, evaluating the benefits of wind farms, and long-term electricity trading.

[0051] In the early stages of wind farm construction, assessing and forecasting regional wind resources is essential. Analysis shows that in the initial phase, continuous and effective meteorological information on wind resources for the selected site for 1-3 years is required. Furthermore, the site selection process necessitates the collection of basic data, numerical simulation of regional wind energy resources, and estimation of regional wind energy reserves. This forces owners to invest significant time, manpower, and resources in finding suitable wind farm locations. Therefore, medium- and long-term wind resource forecasting is particularly important in the early stages of wind farm construction. More refined medium- and long-term regional wind resource forecasts can shorten the initial investigation period, save time and costs, and provide a scientific basis for the selection of meteorological towers and the final site selection of the wind farm.

[0052] Not only in the initial stages of wind farm construction, but also after wind farms are built and put into operation, long-term wind resource forecasting plays a crucial role in long-term electricity trading. The impact of large-scale wind power grid connection on the safe and stable operation of the power system is becoming increasingly significant. The contradiction between wind power development and system safety operation is also gradually emerging. This can lead to some wind curtailment, resulting in energy waste. Therefore, solving the problem of large-scale wind power consumption is urgently needed.

[0053] Furthermore, the medium- and long-term forecasts described in the background section cannot simply adopt the methods used for short-term forecasts. The specific reasons are as follows: ① The spatial resolution of weekly, monthly, and seasonal forecasts produced by models is relatively low, mostly greater than 50 km. Although downscaling can be used to refine the forecasts, dynamic downscaling is costly, while statistical downscaling requires a long period of historical data, and several years of data is insufficient for seasonal forecasts; ② The temporal resolution is also low, unlike the hourly level of short-term weather forecasts. The results are often monthly or even seasonal averages, and the relationship between monthly average wind speed and monthly average wind power is poor, with a similar situation for photovoltaics; ③ Seasonal forecast products mostly provide 10m wind speeds, while wind power applications often require specific wind speed forecasts; ④ Medium- and long-term forecasts have large errors, and models exhibit significant "drift" phenomena. Based on these pain points and difficulties in medium- and long-term meteorological element forecasting, there is significant room for development and improvement in medium- and long-term wind resource forecasting. Therefore, to strengthen corresponding meteorological capabilities and develop core meteorological technologies, it is urgent to develop a forecasting mechanism that can be independently updated and masters core technologies to achieve self-reliance in meteorological science and technology.

[0054] The foundation of medium- and long-term wind resource forecasting stems from the current development of medium- and long-term weather forecasting. However, the current focus of medium- and long-term weather forecasting is on "medium- and long-term forecasting," a global challenge and widely recognized as a "gap" in the development of seamless weather-climate prediction systems. With rapid economic and social development, more and more fields require longer-term weather forecast information. In current weather and climate forecasting operations, weekly, monthly, and seasonal medium- and long-term weather forecasts are the most challenging aspect of "seamless forecasting." Due to incomplete theoretical research, accurate medium- and long-term weather forecasting remains difficult, but the urgent need for medium- and long-term forecasting services has made it a research hotspot internationally. Weekly, monthly, and seasonal medium- and long-term forecasts have been a challenge in international atmospheric science research for over 20 years, with many meteorological experts studying medium- and long-term forecasting methods from different perspectives. The difficulty of medium- and long-term forecasting lies in the fact that its forecast lead time exceeds the theoretical upper limit of deterministic forecasting, while the time scale of the forecast object is smaller than the time scale of climate prediction.

[0055] In 2013, the World Weather Research Programme and the World Climate Research Programme, both under the World Meteorological Organization, jointly launched the Sub-seasonal to Seasonal (S2S) forecasting program. The European Centre for Medium-Range Weather Forecasts (ECMWF), in its "Roadmap 2025," proposed increasing the lead time for high-impact weather events to two weeks and for large-scale circulation patterns and transitions to four weeks ahead within 10 years. The U.S. National Academy of Sciences (NAS) released a forecasting research plan for the next two weeks to 12 months, aiming to broaden and deepen the application of sub-seasonal to seasonal scale forecasts to the current level of weather forecasting within 10 years. The U.S. National Oceanic and Atmospheric Administration (NOAA) plans to develop a sub-seasonal to interannual scale forecasting system based on the existing North American Multimodel Ensemble Forecasting System, providing forecasts for the next 3-4 weeks. In its "Action Plan for Comprehensively Promoting Meteorological Modernization (2018-2020)" and "Action Plan for Intelligent Grid Forecasting (2018-2020)," the China Meteorological Administration explicitly identified medium- and long-term forecasts of meteorological elements and important weather processes as one of its key tasks and core technologies to be tackled.

[0056] Sub-seasonal to seasonal (S2S) forecasts have a longer lead time than weather-by-weather forecasts, with the contribution of initial atmospheric information continuously diminishing. They also have a shorter lead time than short-term climate forecasts, where the effects of underlying surface signals such as sea surface temperature, snow cover, and soil moisture are not fully realized. This results in a significant difference in predictability between medium- and long-term forecasts and those of weather and short-term climate forecasts. The predictability signals for S2S originate from both initial conditions and underlying surface factors. Errors in initial and boundary conditions are closely related to forecast lead time. Numerous studies have shown that several potential predictability sources exist for this forecast period, such as tropical low-frequency oscillations and intra-seasonal oscillations in the Northern Hemisphere summer atmosphere, tropical ocean-atmosphere interactions, stratospheric and tropospheric interactions, soil moisture, snow cover and sea ice, and tropical-tropical extratropical teleconnections, all of which provide a scientific basis for forecasts in this period.

[0057] Following the establishment of the General Circulation Model (GCM), many scholars have used this model to conduct numerous medium- and long-term lunar dynamic forecasts. With a deepening understanding of the climate system, especially of medium- and long-term predictable information, such as the "10-90 day" low-frequency signal, a foundation for the development of medium- and long-term forecasts has been laid. In the last decade or so, data assimilation techniques and model performance have greatly improved, and ensemble forecasting techniques have been widely applied, all providing favorable conditions for improving medium- and long-term forecasting skills. Therefore, the world's major numerical weather prediction centers have once again invested great enthusiasm in medium- and long-term forecasting and have gradually developed their own operational or quasi-operational dynamic medium- and long-term forecasting systems.

[0058] Although the theoretical foundation is not yet complete, there are still many usable forecasting factors for medium- and long-term forecasting. According to existing research, low-frequency signals such as ISO, MJO, and stratospheric signals are the main sources of predictability for medium- and long-term weather. With the advancement of various observation methods such as satellite remote sensing, the amount of climate observation data obtained in the past decade has increased rapidly. This large amount of scientific data reflects and characterizes complex natural phenomena and relationships, and can be regarded as new signals. Currently, the main methods for medium- and long-term forecasting are dynamical models, classical statistics, big data methods, and methods combining dynamical and statistical approaches.

[0059] Currently, medium- and long-term wind resource forecasting is mainly used in practical engineering projects for wind farm site selection and wind power prediction, with wind power prediction being the most prevalent application. Wind power forecasting is rapidly developing; theoretical research on wind power forecasting in countries such as Denmark, the United States, and Germany began as early as the 1990s, and they have developed several relatively complete forecasting software programs applicable to actual wind farms. Examples include Germany's WPMS forecasting software, the WPPT forecasting software developed by the Technical University of Denmark, and the Prediktor forecasting software developed by the Danish National Laboratory.

[0060] Germany's WPMS system can predict wind power output for individual wind farms, regional wind farms, and the entirety of Germany. The error margin for the former is 10%-15% of the installed capacity of the wind farm; for regional wind power, the error margin is 6%-7% of the installed capacity; and for the entirety of Germany, the error margin is 5%-6% of the total installed wind power capacity. Short-term and ultra-short-term predictions are even more accurate. Denmark's WPPT software was developed earliest and was already operational in Danish wind farms in 1994 and 1999. This regional prediction software accurately predicts wind power generation within a specific area by monitoring the power output of all wind turbines and forecasting wind speed, temperature, and humidity. Prediktor prediction software is another type of software that predicts wind power output based on wind speed forecasts. First, the wind speed distribution is provided by the numerical weather prediction system. Then, the WASP system performs a more accurate analysis of the geographical location of the wind turbines, such as the turbine height and rough set data, to determine a more accurate wind speed forecast. Finally, the power generation is predicted based on the wind speed forecast.

[0061] Please see Figure 1 In one embodiment of the invention, the medium- and long-term weather forecasting method based on a variable grid model of the present invention includes the following steps:

[0062] Step S102: Obtain meteorological products based on initial forecast data and back-calculation data;

[0063] Step S104: Correct the deviation of the meteorological product based on the preset correction algorithm to obtain the prediction result;

[0064] Step S106: Evaluate the corrected prediction results, and output the prediction results when the evaluation results meet the target requirements.

[0065] It should be noted that the meteorological products obtained based on initial forecast data and back-calculation data, such as... Figure 2 As shown, the specific steps include the following:

[0066] Step S202: Obtain the initial prediction data and the back-calculation data predicted in real time by the prediction system;

[0067] Step S204: Perform a general standardization operation on the variable mesh based on the initial prediction data, and use the result of the standardization operation as the boundary forcing field of the variable mesh model;

[0068] Step S206: Downscale the initial prediction data within the boundary forcing field based on preset weights and correct the scale bias.

[0069] Step S208: Based on the back-calculation data, perform probability forecasting and anomaly forecasting, and combine the initial forecast data after correcting for scale bias to obtain the meteorological product, wherein the meteorological product includes at least one meteorological element.

[0070] It should be noted that the forecasting system includes my country's CMA-CPSv3 and the US CFS forecasting system. Real-time forecast data and back-calculation data from CMA-CPSv3 and the US CFS forecasting system are collected. The forecast data is standardized using a variable grid universal forcing method. The standardized CMA-CPSv3 and US CFS forecasts are assimilated into the variable grid model as the boundary forcing field of the model. Then, based on preset weights, the initial forecast data within the boundary forcing field is downscaled and the scale bias is corrected. Based on the back-calculation data, probability forecasts and anomaly forecasts are made. The meteorological products are obtained by combining the scale bias-corrected initial forecast data.

[0071] Specifically, at the initial reporting time, the CFS / CMA model f000 time analysis field data of the day is used as the initial field of the model. A prediction set is generated using time-lag perturbation and parameter initial value perturbation methods. In one embodiment, 16 prediction sets can be generated. The prediction sets are first downscaled using CMA-CPSv3 and CFS as boundary forcing with a weight of 50%. The variable grid model assimilates the CMA-CPSv3 and CFS data into the model integral to correct large-scale biases using a large circle Newton relaxation iterative algorithm. Correspondingly, real-time prediction adopts an on-the-time prediction method synchronized with ECMWF. In one embodiment of the Fly method, while making real-time predictions, a back-calculation of nearly 20 years is performed. For example, if the reporting start date is January 1, 2023, the back-calculation from 2003 to 2023 is performed simultaneously to obtain the back-calculation data, which is used for the final probability forecast and anomaly forecast. Finally, the proportion of CMA-CPSv3 and CFS forced field is optimized to form the meteorological product with a daily rolling update of 6-10KM resolution and a daily (pentad, ten-day) multi-meteorological element set forecast for 11-60 days. Accordingly, the meteorological product includes at least one meteorological element, such as wind, rain, temperature, etc.

[0072] Furthermore, in one embodiment of the invention, as Figure 3 As shown, the deviation correction of the meteorological product based on the preset correction algorithm to obtain the prediction result specifically includes the following steps:

[0073] Step S302: Obtain the correction algorithm, wherein the correction algorithm includes at least the classification meteorological element bias correction algorithm, Kalman filter bias correction algorithm, similarity method bias correction algorithm, Bayes bias correction algorithm, empirical orthogonal function decomposition bias correction algorithm, and deep learning bias correction algorithm.

[0074] Step S304: Correct the meteorological product based on the classification meteorological element deviation correction algorithm, the empirical orthogonal function decomposition deviation correction algorithm, and the deep learning deviation correction algorithm;

[0075] Step S306: Correct the variable mesh model based on the Kalman filter bias correction algorithm, the similarity method bias correction algorithm, and the Bayesian bias correction algorithm.

[0076] It should be noted that the bias correction for the meteorological products includes methods such as classification meteorological element bias correction, Kalman filter bias correction, similarity method bias correction, Bayesian method bias correction, EOF bias correction, and deep learning correction techniques. For example, Figure 4 As shown, the meteorological product is corrected based on the aforementioned classification meteorological element deviation correction algorithm, specifically including the following steps:

[0077] Step S402: Correct the temperature forecast deviation, wherein the temperature correction objects include the average temperature, daily maximum temperature and daily minimum temperature;

[0078] Step S404: Correct precipitation forecast bias, wherein the corrected precipitation parameters include the forecast precipitation frequency and the observed precipitation frequency;

[0079] Step S406: Correct the wind speed forecast deviation, wherein the wind speed correction objects include hourly average wind speed and direction, and maximum wind speed and direction.

[0080] It should be noted that the correction of temperature forecast deviations includes the average temperature, daily maximum temperature, and daily minimum temperature, and the methods used include the decreasing average method, the univariate linear regression method, the average method, and the double-weighted average method.

[0081] The decreasing averaging method is an adaptive error correction method that reduces the error scale through lag averaging. The correction principle is as follows:

[0082] ;

[0083] Where B(t) is the lag average error, b(t) is the forecast error, b(t) = f(t) - a(t), f(t) is the forecast value, a(t) is the actual analysis value; B(t-1) is the average error of the previous day; These are the weighting coefficients.

[0084] Specifically, (1) a cold start is performed when t=1, i.e., B(t-1)=0; (2) the optimal weight coefficients for the past 60 days of training are calculated. (3) Calculate the lag mean error B(t) according to the formula; (4) Repeat step (3). After 60 days of training and iterative accumulation, the error has become stable and can characterize the system error to a certain extent; (5) Subtract the latest lag mean error B(t) from the current forecast value to obtain the corrected forecast value, and the weight coefficient The magnitude of the weighting coefficient reflects the error memory capacity of the correction model and directly affects the correction result. This method adopts a stepwise incremental scheme, with the weighting coefficients increasing in increments of 0.001 from 0 to 1. The accuracy of the temperature forecast over the past 60 days for each grid point at different forecast lead times for the next 3 days is calculated cyclically, and the weighting coefficients corresponding to the highest scores for each grid point at each forecast lead time are identified to form a weighting coefficient array.

[0085] Simple linear regression is a method for analyzing the linear correlation between two variables with only one independent variable (x and y). Continuous random variables in meteorology often follow a normal distribution, so linear regression equations are frequently used. Where a is the regression coefficient and b is a constant. When a is positive (negative), it indicates that the factor increases (decreases) linearly during the calculation period.

[0086] The averaging method corrects the forecasts for each station based on the observation increments at each station, as shown in the following formula:

[0087] ;

[0088] ;

[0089] in, To observe the temperature at 2m, The forecast temperature is interpolated to the station, where k is the station number and n is the observation time. For the revised forecast field, For the revised forecast.

[0090] Finally, the double-weighted averaging method corrects the forecast for each station based on the double-weighted average of the observed increments. This is generally similar to the average method, but it reduces the influence of outlier data on the average. The formula is as follows:

[0091] ;

[0092] in, The double-weighted average of the observation increments of the k-th station, and the weights related.

[0093] Furthermore, regarding the correction of precipitation forecast biases, the methods employed include the decreasing averaging method and the frequency or area method. The precipitation parameters used for correction include forecast precipitation commentary and observed precipitation frequency. Specifically, taking the decreasing averaging method as an example, the forecast precipitation frequency and observed precipitation frequency for the day are statistically analyzed:

[0094] ;

[0095] in, This indicates the average frequency of precipitation on that day. represents the frequency of precipitation on that day, and w represents the weighting coefficient of the frequency of precipitation on the previous day. This represents the average frequency of precipitation on the previous day.

[0096] Based on the statistical analysis of the frequency of precipitation occurrence under different threshold conditions in forecasts and previous observations (which can be spatial, temporal, or both), for a given threshold, assuming that its frequency in the forecast should be consistent with its frequency in the actual observation, the forecast precipitation should be corrected to match the actual precipitation amount. This method of maintaining consistent frequency is called the "frequency matching method." If understood spatially, the magnitude of the frequency is actually the size of the spatial range (the number of stations or grid points). Adjustments are made to the forecast rain area based on wet or dry biases. The "frequency matching method" can also be called the "area matching method," as this method considers the consistency of the area of ​​the forecast and actual rain areas.

[0097] Furthermore, regarding the correction of wind speed forecast bias, the correction targets include hourly average wind speed and direction, and maximum wind speed and direction. Methods employed include random forest algorithms and support vector machine algorithms. The random forest algorithm is a classifier containing multiple decision trees. It uses feature vectors established from various meteorological elements output by the model as input, and the wind speed corresponding to these feature vectors as the prediction result. It is fitted using training samples and used to correct the model's wind speed prediction results. The random forest modeling process is as follows: Define a wind speed prediction training set X→Y, where Y is the true value in the prediction model, and X is the feature vector In→X established from other meteorological elements. Based on the determined training set, a single regression decision tree is established. Using the feature vector In in the training samples and its corresponding true value Y, the splitting variable and splitting value are searched. The regression decision tree divides the entire vector space into m partitions {R1, R2, …, Rm}. Any partition can be mapped to a model Cm, which divides the vector space into two parts based on the value of a certain feature, expressed as:

[0098] R1(j, s) = {I | Ij ≤ s};

[0099] R2(j, s) = {I | Ij > s};

[0100] In the formula, j represents an influence factor, and s represents the value at the time of splitting. The objective function for searching the vector space splitting variables and splitting values ​​is:

[0101] ;

[0102] In the formula, z represents the minimum variance of the real wind speed, yi represents the real wind speed of the i-th sample, xi represents the value corresponding to the influence factor vector of the i-th sample, c1 represents the mean of the real wind speed values ​​in the first part, and c2 represents the mean of the real wind speed values ​​in the second part; and the entire random forest is constructed based on the construction of a single decision tree. The generated random forest is a multivariate nonlinear regression analysis model, and the predicted value is the average of the predicted values ​​of all decision trees.

[0103] Furthermore, the correction of the meteorological products also includes corrections for shortwave irradiance, sea level pressure, and relative humidity deviations, wherein bias elimination methods are used to correct the model forecast results hourly. For each time moment, , These are the average values ​​of model forecasts and observations over a period of time. , These are the model forecast value and the bias correction result for day i, respectively, as shown in the following formula: Correction is performed using Model Output Statistics (MOS), which selects model output forecast factors that are correlated with the forecasted object and mapped to the station's latitude and longitude from 06:00 to 19:00 daily. , , , ...), and establish a statistical relationship with the total radiation observation (Y). In order to eliminate the diurnal variation characteristics of radiation, a forecast equation is established for each moment and corrected hourly.

[0104] Furthermore, in one embodiment of the invention, as Figure 5 As shown, the variable mesh model is corrected based on the Kalman filter bias correction algorithm, which specifically includes the following steps:

[0105] Step S502, update the model error to adjust the bias scale, including estimating the model forecast bias and determining the model forecast bias after weighted averaging.

[0106] Step S504: Obtain error correction coefficients and correct the variable mesh model based on the error correction coefficients.

[0107] It should be noted that the Kalman filter method is simple in algorithm, fast in processing, and capable of correcting systematic errors. Systematic errors are obtained by continuously updating model forecast biases and weighted averaging the forecast bias values ​​over different time periods (Kalman RE, 1960). Because it does not require long-term statistical regularities of model errors, it is very suitable for frequent model system upgrades and constantly changing error patterns, leading to its widespread application in meteorological operations. Therefore, the Kalman filter method is used to estimate the biases of variables with Gaussian distribution characteristics and then perform corresponding first-order bias corrections. Firstly, the effectiveness of this method in error correction is analyzed and tested.

[0108] Specifically, the first-order adaptive Kalman filtering method continuously updates the model error to obtain an estimate of the error at any given time, thereby reducing the bias scale and overcoming the interference of noise inherent in the variable grid model. This includes estimating the model forecast bias, as shown in the formula: bi,j(t) = fi,j(t) – ai,j(t), where bi,j(t) is the grid forecast bias, defined as the difference ai,j(t) between the grid forecast data fi,j(t) in the test area and the corresponding analysis field data, and bi,j(t) varies with forecast lead time. It also includes determining the model forecast bias after weighted averaging. The weighted averaged model forecast bias is determined using an approximation method, as shown in the formula: Bi,j(t) = (1 – w)Bi,j(t-1) + w bi,j(t), where Bi,j(t) is defined as the weighted averaged model forecast bias estimate, obtained by weighted averaging of model forecast biases from two different time periods. Using the discrete approximation method, the model forecast bias estimate is updated continuously. Studies on this time-varying weighted forecast average error show that selecting an appropriate weighting factor w can help extract useful forecast bias information, determine the optimal system average weighted error, and greatly improve forecast accuracy. Furthermore, it allows for the correction of model forecast errors using the following formula: Fi,j(t) = fi,j(t) - Bi,j(t), where the system average weighted error Bi,j(t) is used to correct the latest model forecast fi,j(t). The model forecast error correction process is performed for all grid points across different forecast lead times.

[0109] Furthermore, the error correction coefficients are selected; specifically, for the correction equation Bt= of the adaptive Kalman filter. *(FA)+ (1- ) *Bt-1, where This determines how long a sample period in the near future will affect the forecast correction for that day; different... For different weight change curves, for different The decay rates of the curves are different. The larger the value, the faster the decay; additionally, for different... The impact of sample data from the same day on the decreasing mean bias is different. The larger the value, the greater the weight of samples closer to the correction date, and the smaller the weight of samples further away from the correction date; conversely, the smaller the value, the greater the weight. Different grid points, different forecast times, different physical quantities, and different numbers of days in the model all correspond to different decreasing average biases. Currently, in practical applications, The usual method is as follows: Take samples through experimentation. The values ​​are 0.01, 0.02, 0.03, ..., 0.99. The total forecast error for all points is compared, and the value that minimizes this error is used to determine the prediction. It is a constant value that does not change with time, forecast lead time, or location.

[0110] Furthermore, the above embodiments illustrate that when correcting the variable mesh model, it can also be corrected using the similarity method bias correction algorithm and the Bayesian bias correction algorithm. The similarity method bias correction uses the similarity of the model's prediction field to correct the systematic error of the current variable mesh model, while the Bayesian bias correction algorithm also corrects the variable mesh model itself. The specific details are common algorithms for those skilled in the art and will not be elaborated here.

[0111] Furthermore, in one embodiment of the invention, as Figure 6 As shown, the prediction results are evaluated based on the corrected prediction results. When the evaluation results meet the target requirements, the prediction results are output. The specific steps include the following:

[0112] Step S602: Obtain a first weather forecast score based on the surface non-precipitation elements in the meteorological product;

[0113] Step S604: Obtain a second weather forecast score based on the precipitation elements in the meteorological product;

[0114] Step S606: Based on the first weather forecast score and the second weather forecast score, the evaluation result is obtained, wherein when the evaluation result meets the required objective, the corresponding prediction result is output.

[0115] It should be noted that for non-precipitation factors at ground level, including meteorological variables such as "2m air temperature, 10m wind speed, and 2m humidity," the specific items for the first meteorological forecast score obtained for the next season's forecast include mean absolute error, root mean square error, anomaly correlation coefficient, and temperature forecast accuracy. For monthly and seasonal forecasts, the specific items for the score include the anomaly correlation coefficient. The corresponding calculation formula for the first meteorological forecast score (PS score) is as follows:

[0116] ;

[0117] Wherein, N0 is the number of stations with correct climate trend prediction, N1 is the number of stations with correct prediction of primary anomalies, N2 is the number of stations with correct prediction of secondary anomalies, M is the number of stations with actual temperature anomalies ≥3℃ or ≤-3℃ without a predicted secondary anomaly, and a, b and c are the weighting coefficients of the climate trend item, primary anomaly item and secondary anomaly item, respectively. In one embodiment, a=2, b=2 and c=4 can be taken.

[0118] Furthermore, for precipitation elements, PS scores also need to be calculated, thus the PS calculation formula is consistent with that of the first meteorological forecast score. However, for the second meteorological forecast scenario, the corresponding M is the number of stations where there is no forecast of secondary anomalies but the actual precipitation anomaly percentage is ≥100% or equal to -100% (also known as missed reporting stations). The second meteorological forecast score is obtained by calculating the corresponding PS scores, so that the evaluation result can be obtained based on the first meteorological forecast score and the second meteorological forecast score. When the evaluation result meets the required objective, the corresponding prediction result is output. The prediction result score comprehensively considers the PS scores of non-precipitation elements and precipitation elements, and performs a weighted average of the score results of different meteorological elements to obtain the comprehensive prediction technique of the variable grid model.

[0119] Furthermore, in one embodiment of the invention, the meteorological product correction based on the deep learning bias correction algorithm is specifically achieved by processing and analyzing the time series data using a long short-term memory neural network model to correct the meteorological product. Specifically, the long short-term memory neural network model proposed in this application is a four-layer multi-memory gate neural network, an improved version of the traditional long short-term memory neural network (LSTM). It also considers the spatial convolution of meteorological data. By introducing multiple memory gates, the long-term memory of meteorological stations, grid reanalysis data, forecast data, and satellite data can be simultaneously incorporated into the model training. The ADAM and MSE+MAE operators are selected as the loss functions for model training, with an initial learning rate of "e-4". Each LSTM layer has "192" learning channels, and the convolution kernel size is "3". The validation set and training set are distinguished by a "3:7" ratio, and the trained mmLSTM prediction model is finally obtained. By evaluating the prediction skill of the prediction model and verifying the model training degree, the neural network model is finally improved and optimized. The training flowchart is as follows. Figure 7 As shown.

[0120] It should be noted that Long Short-Term Memory (LSTM) neural networks have an inherent advantage in processing time series problems and can correct model prediction products, improving prediction accuracy. Therefore, the time series can be processed and analyzed using LSTM models to correct the meteorological products. The above embodiments illustrate that the meteorological products can also be corrected using the Empirical Orthogonal Function Decomposition (EOF) bias correction algorithm. Specifically, the Empirical Orthogonal Function (EOF) method, also known as eigenvector analysis or principal component analysis, is a method for analyzing the structural features of matrix data and extracting key data features. Therefore, in practical operations, the EOF bias correction algorithm can also be used to correct meteorological products. Correspondingly, the EOF algorithm is a common algorithm for those skilled in the art and will not be elaborated upon here.

[0121] This application also provides a medium- and long-term weather forecasting system based on a variable grid model. The medium- and long-term weather forecasting system based on a variable grid model can implement the medium- and long-term weather forecasting method based on a variable grid model described in this application. However, the implementation device of the medium- and long-term weather forecasting method based on a variable grid model described in this application includes, but is not limited to, the structure of the medium- and long-term weather forecasting system based on a variable grid model listed in this embodiment. All structural modifications and substitutions of the prior art made in accordance with the principles of this application are included within the protection scope of this application.

[0122] Please see Figure 8 In one embodiment, this embodiment provides a medium- to long-term weather forecasting system 80 based on a variable grid model, the system comprising:

[0123] Module 81 is used to acquire meteorological products based on initial forecast data and back-calculation data;

[0124] The correction module 82 is used to correct the deviation of the meteorological product based on a preset correction algorithm to obtain the prediction result;

[0125] The evaluation module 83 is used to evaluate the corrected prediction results and output the prediction results when the evaluation results meet the requirements.

[0126] Since the specific implementation of this embodiment corresponds to the aforementioned method embodiment, the same details will not be repeated here, and those skilled in the art should also understand this. Figure 8The division of the modules in the embodiments is only a logical functional division. In actual implementation, they can be fully or partially integrated into one or more physical entities. These modules can be fully implemented in software through processing element calls, fully implemented in hardware, or some modules can be implemented in software through processing element calls and some modules can be implemented in hardware.

[0127] See Figure 9 This embodiment provides an electronic device, which includes at least a memory and a processor connected via a bus. The memory stores a computer program, and the processor executes the computer program stored in the memory to perform all or part of the steps in the aforementioned method embodiment.

[0128] In summary, this invention features high computational efficiency and a large sample size, which improves the precision and efficiency of forecasts. Furthermore, by employing a hybrid correction method combining traditional statistics and artificial intelligence, it can significantly enhance forecast accuracy and meet the demand for medium- and long-term meteorological forecast products in the development of new energy.

[0129] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, or methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules / units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or units may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of apparatuses or modules or units may be electrical, mechanical, or other forms.

[0130] The modules / units described as separate components may or may not be physically separate. The components shown as modules / units may or may not be physical modules; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules / units can be selected to achieve the objectives of the embodiments of this application, depending on actual needs. For example, the functional modules / units in the various embodiments of this application may be integrated into one processing module, or each module / unit may exist physically separately, or two or more modules / units may be integrated into one module / unit.

[0131] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0132] This application also provides a computer-readable storage medium. Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing a processor. The program can be stored in a computer-readable storage medium, which is a non-transitory medium, such as random access memory, read-only memory, flash memory, hard disk, solid-state drive, magnetic tape, floppy disk, optical disk, and any combination thereof. The storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., digital video disc (DVD)), or a semiconductor medium (e.g., solid-state drive (SSD)).

[0133] This application embodiment may also provide a computer program product comprising one or more computer instructions. When the computer instructions are loaded and executed on a computing device, all or part of the processes or functions described in this application embodiment are generated. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0134] When the computer program product is executed by a computer, the computer performs the method described in the foregoing method embodiments. The computer program product can be a software installation package; when the foregoing method is required, the computer program product can be downloaded and executed on the computer.

[0135] The descriptions of the processes or structures corresponding to the above figures each have their own emphasis. For parts of a process or structure that are not described in detail, please refer to the relevant descriptions of other processes or structures.

[0136] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A method for medium and long term weather prediction based on a variable mesh model, characterized in that, The method comprises: obtaining meteorological products based on initial prediction data and back calculation data, specifically comprising: obtaining the initial prediction data and the back calculation data predicted by the prediction system in real time; performing a variable grid universal standardization operation based on the initial prediction data, and taking the standardization operation result as a boundary forcing field of the variable grid model; performing a downscaling operation on the initial prediction data in the boundary forcing field based on a preset weight, and correcting the scale bias; performing probability prediction and anomaly prediction based on the back calculation data, and obtaining the meteorological products in combination with the initial prediction data after the scale bias is corrected, wherein the meteorological products comprise at least one meteorological element; correcting the bias of the meteorological products based on a preset correction algorithm to obtain a prediction result, specifically comprising: obtaining the correction algorithm, wherein the correction algorithm at least comprises a classified meteorological element bias correction algorithm, a Kalman filter bias correction algorithm, a similarity method bias correction algorithm, a Bayesian bias correction algorithm, an empirical orthogonal function decomposition bias correction algorithm and a deep learning bias correction algorithm; correcting the meteorological products based on the classified meteorological element bias correction algorithm, the empirical orthogonal function decomposition bias correction algorithm and the deep learning bias correction algorithm; correcting the variable grid model based on the Kalman filter bias correction algorithm, the similarity method bias correction algorithm and the Bayesian bias correction algorithm; evaluating based on the corrected prediction result, and outputting the prediction result when the evaluation result meets the demand target. 2.The medium and long term weather prediction method based on the variable mesh model according to claim 1, characterized in that, correcting the meteorological products based on the classified meteorological element bias correction algorithm, specifically comprising: correcting temperature prediction bias, wherein the temperature correction object comprises average temperature, daily maximum temperature and daily minimum temperature; correcting precipitation prediction bias, wherein the corrected precipitation parameter comprises predicted precipitation frequency and observed precipitation frequency; correcting wind speed prediction bias, wherein the wind speed correction object comprises hourly average wind speed and wind direction, maximum wind speed and wind direction. 3.The medium and long term weather prediction method based on the variable mesh model according to claim 1, characterized in that, correcting the variable grid model based on the Kalman filter bias correction algorithm, specifically comprising: updating the model error to adjust the bias scale, which comprises estimating the model prediction bias and determining the model prediction bias after weight averaging; obtaining an error correction coefficient, and correcting the variable grid model based on the error correction coefficient. 4.The medium and long term weather prediction method based on the variable mesh model according to claim 1, characterized in that, The meteorological products are corrected based on the deep learning bias correction algorithm, specifically by processing and analyzing the time series through a long short-term memory neural network model to correct the meteorological products. 5.The medium and long term weather forecast method based on variable mesh model according to claim 1, characterized in that, The evaluation is performed based on the corrected prediction result, and the prediction result is output when the evaluation result meets the demand target, specifically comprising: obtaining a first meteorological prediction score based on ground non-precipitation elements in the meteorological products; obtaining a second meteorological prediction score based on precipitation elements in the meteorological products; obtaining the evaluation result based on the first meteorological prediction score and the second meteorological prediction score, wherein the corresponding prediction result is output when the evaluation result meets the demand target.

6. A medium and long term weather prediction system based on a variable mesh model, characterized in that, comprises: The acquisition module is configured to acquire the meteorological product based on the initial prediction data and the back-calculation data, specifically including: acquiring the initial prediction data and the back-calculation data predicted by the prediction system in real time; performing a variable-grid general standardization operation based on the initial prediction data, and taking the standardization operation result as a boundary forcing field of the variable-grid model; performing a downscaling operation on the initial prediction data in the boundary forcing field based on a preset weight, and correcting a scale bias; performing probability prediction and anomaly prediction based on the back-calculation data, and obtaining the meteorological product in combination with the initial prediction data after the scale bias is corrected, wherein the meteorological product includes at least one meteorological element; The correction module is configured to correct a bias of the meteorological product based on a preset correction algorithm to obtain a prediction result, specifically including: acquiring the correction algorithm, wherein the correction algorithm at least includes a classified meteorological element bias correction algorithm, a Kalman filter bias correction algorithm, a similarity method bias correction algorithm, a Bayesian bias correction algorithm, an empirical orthogonal function decomposition bias correction algorithm, and a deep learning bias correction algorithm; correcting the meteorological product based on the classified meteorological element bias correction algorithm, the empirical orthogonal function decomposition bias correction algorithm, and the deep learning bias correction algorithm; correcting the variable-grid model based on the Kalman filter bias correction algorithm, the similarity method bias correction algorithm, and the Bayesian bias correction algorithm; The evaluation module is configured to evaluate the prediction result after correction, and output the prediction result when the evaluation result meets the demand.

7. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the method for long-term meteorological prediction based on the variable-grid model in any one of claims 1 to 5.

8. An electronic device, comprising: The electronic device includes a processor and a memory, wherein the memory is configured to store a computer program, and the processor is configured to execute the computer program stored in the memory to enable the electronic device to perform the method for long-term meteorological prediction based on the variable-grid model in any one of claims 1 to 5.

Images