Meteorological element forecasting method based on REGOCF custom height

Abstract

The invention provides a meteorological element forecasting method based on REGOCF custom height, and the method comprises the steps: collecting training data and simulation data, carrying out the processing of the training data and the simulation data according to a concern site in an operation region, and finally constructing REGOCF models of two levels, namely primary learning and secondary learning, the primary learning level employs a plurality of basic learning models, and the secondary learning level employs a plurality of basic learning models; the basic learning models are trained by using the same feature set, all prediction results obtained by the primary learning level are summarized into a new feature matrix, the new feature matrix is trained by using an OCF method to obtain a final prediction result, and an efficient and accurate meteorological element prediction effect is realized.

Description

Technical Field

[0001] The invention relates to the field of weather forecasting, and in particular to a weather element forecasting method based on REGOCF user-defined altitude. Background Art

[0002] The low-altitude economy and energy economy have put forward the demand for three-dimensional meteorological element forecasts for meteorological services. For example, fields such as aviation, agriculture, transportation, and energy need to forecast meteorological elements with defined altitudes according to industry application requirements. However, due to the complexity and nonlinear characteristics of the spatial distribution and temporal changes of low-altitude meteorological elements, traditional statistical and physical models often cannot provide accurate forecast results. Therefore, a method for forecasting meteorological elements with defined altitudes is urgently needed to solve the above problems. Summary of the invention

[0003] The purpose of the present invention is to provide a meteorological element forecasting method based on REGOCF custom height to solve the problems existing in the above-mentioned prior art.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] The present invention provides a meteorological element forecasting method based on REGOCF user-defined altitude, comprising the following steps:

[0006] S1. Collect training data and simulation data;

[0007] S2. Processing the training data and simulation data according to the locations of interest in the operation area;

[0008] S3. Construct a REGOCF model with two levels: primary learning and secondary learning. The primary learning level adopts multiple base learning models, and multiple base learning models are trained using the same feature set. All prediction results obtained at the primary learning level are summarized into a new feature matrix, and the new feature matrix is ​​trained using the OCF method to obtain the final prediction results.

[0009] Preferably, in step S1,

[0010] The training data includes:

[0011] a. Collect numerical model data, including: CMA-MESO, ECMWF Thin, National Smart Grid and Inner Mongolia Autonomous Region Ruitu regional model forecast data;

[0012] b. Collect observation data of meteorological elements in the operation area, including data from ground meteorological stations, wind towers, and laser wind radar, as well as dates of precipitation, snowfall, strong winds, and sandstorms;

[0013] c. Collect 30m resolution digital elevation model data within the operation area;

[0014] d. Collect data on land surface classification within the operation area; including cultivated land, forest land, grassland, shrub land, wetland, water bodies, tundra, artificial surface, bare land, glacier and permanent snow;

[0015] The simulated data includes:

[0016] e. Collect ECMWF ERA5 atmospheric reanalysis data, including single-layer and isobaric layer potential height, relative humidity, absolute humidity, temperature, u wind component, and v wind component meteorological element data.

[0017] Preferably, the period of all data is 2 years.

[0018] Preferably, the digital elevation model data in the operating area adopts digital elevation data within 30 meters to 500 meters, which is the 30-meter digital elevation model data of the Copernicus program of the European Space Agency.

[0019] Preferably, the surface classification data in the operating area adopts surface coverage data within 30 meters to 500 meters, which is 30-meter spatial resolution data of GlobeLand30.

[0020] Preferably, step S2 comprises:

[0021] S21. extracting temperature, wind field, humidity, and precipitation data from the numerical model data according to the latitude and longitude locations of the locations of interest and the meteorological stations in the operation area in the order of corresponding locations and time;

[0022] S22. extracting the dates of rainfall, snowfall, gale, and sandstorm weather according to the statistics of the stations from the meteorological element observation data in the operation area according to the latitude and longitude positions of the concerned locations and meteorological stations in the operation area;

[0023] S23. Extract elevation, slope, aspect data and land type data from 30-meter resolution digital elevation model data and surface classification data in the operating area according to the latitude and longitude locations of the locations of interest and the meteorological stations in the operating area;

[0024] S24. The single-layer and isobaric layer data of ECMWF ERA5 reanalysis data are used to drive the mesoscale numerical model WRF to obtain the simulated data of 40 levels of meteorological element fields from the ground to below the 500hPa isobaric surface. According to the places of interest, custom altitudes above the ground and forecast elements in the operation area, the meteorological elements corresponding to the custom altitude values ​​are extracted and sorted by time.

[0025] Preferably, in step S3, the method for constructing the REGOCF model includes:

[0026] S31. Data preparation;

[0027] S32. Feature engineering;

[0028] S33. Model training.

[0029] Preferably, step S31 includes:

[0030] S311. Missing value processing: missing values ​​in the data are supplemented by linear interpolation or mean filling method;

[0031] S312. Outlier detection, using statistical methods such as box plots to identify outliers in the data and process them;

[0032] S313. Standardization processing: standardize data of different dimensions and unify them to the same scale.

[0033] Preferably, step S32 includes:

[0034] S321. Extract time features, extract year, month, day, and day of the week features from date information;

[0035] S322. Extract spatial features, combine terrain data and underlying surface type data, and generate spatial features;

[0036] S323. Extract historical trend characteristics, calculate the average value of meteorological elements in the same period of history, the highest and lowest values ​​of meteorological elements, and form historical trend characteristics;

[0037] S324. Extract special weather characteristics, and extract the characteristics of meteorological element change values ​​when precipitation, snowfall, strong winds, and sandstorms occur.

[0038] Preferably, step S33 includes:

[0039] S331. Data segmentation: divide the cleaned data set into a training set, a validation set, and a test set, with the allocation ratios being 70%, 15%, and 15% respectively. The training set is used for the model training process, the validation set is used for the adjustment of hyperparameters, and the test set is used to evaluate the overall performance of the model.

[0040] S332. Primary learning layer training: using the training set data to train the model of each primary layer, and during the training process, selecting the optimal hyperparameter combination through the cross-validation method;

[0041] S333. Secondary learning layer training, collecting all primary model prediction results and combining them with actual observations to form a new training data set, which is then used to train the secondary model until the model performance reaches the expected effect;

[0042] S334. Parameter tuning:

[0043] For each primary model, a grid search technique is used to traverse all possible hyperparameter combinations and select the best performing hyperparameter combination;

[0044] Based on the performance on the validation set, the optimal combination of primary layer models is selected and finally evaluated on the test set.

[0045] Compared with the prior art, the present invention has achieved the following beneficial technical effects:

[0046] The present invention provides a meteorological element forecasting method based on REGOCF custom height. The method collects training data and simulation data, processes the training data and simulation data according to the focus locations in the operation area, and finally constructs a REGOCF model of two levels, primary learning and secondary learning. The primary learning level adopts multiple base learning models, and the multiple base learning models are all trained using the same feature set. All prediction results obtained at the primary learning level are summarized into a new feature matrix, and the new feature matrix is ​​trained using the OCF method to obtain the final prediction result, thereby achieving efficient and accurate meteorological element prediction effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0048] Figure 1 A schematic diagram of extracting surrounding meteorological stations according to a place of interest in a meteorological element forecasting method based on REGOCF custom height provided by the present invention;

[0049] Figure 2 A REGOCF model framework diagram in a meteorological element forecasting method based on REGOCF custom altitude provided by the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] The purpose of the present invention is to provide a meteorological element forecasting method based on REGOCF user-defined altitude to solve the problems existing in the prior art.

[0052] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0053] Embodiment 1:

[0054] This embodiment provides a meteorological element forecasting method based on REGOCF custom altitude, comprising the following steps:

[0055] S1. Collect training data and simulation data;

[0056] First, a region is fixed according to the latitude and longitude range as the operation area. Considering the limitation of computing resources, the region should not be too large and is generally based on provinces and cities.

[0057] (a) Collect numerical model data including forecast data from CMA-MESO, ECMWF Thin, National Smart Grid, and Inner Mongolia Autonomous Region Ruitu Regional Model. Prepare two years of data.

[0058] (b) Collect observation data of meteorological elements such as ground meteorological stations, wind towers, and laser wind radars in the operating area, including the dates of precipitation, snowfall, gale, and sandstorms. Prepare two years of data.

[0059] (c) The digital elevation model data within the operating area can use digital elevation data within 30 meters to 500 meters, such as the 30-meter digital elevation model data of the Copernicus program of the European Space Agency.

[0060] (d) To collect surface classification data within the operating area, surface coverage data within 30 to 500 meters can be used, such as GlobeLand30 30-meter spatial resolution data, which was updated by the Ministry of Natural Resources in 2017 to form the 2020 version. GlobeLand30 includes 10 categories, namely cultivated land, forest land, grassland, shrub land, wetland, water body, tundra, artificial surface, bare land, glacier and permanent snow.

[0061] (e) Collect ECMWF ERA5 atmospheric reanalysis data: single layer and isobaric layer potential height, relative humidity, absolute humidity, temperature, u wind component, v wind component and other meteorological element data. Prepare two years of data.

[0062] S2. Process the training data and simulation data according to the operation area content and focus location; Figure 1 As shown, the central red dot is the location of interest, and the surrounding green ones are meteorological stations within 20km of the location of interest (this range should not be too large, and the maximum is within 50km).

[0063] (1) Figure 1 ,According to the latitude and longitude positions of the places of interest and meteorological stations in the operation area, the temperature, wind field, humidity, precipitation and other data are extracted from the a data in the order of corresponding locations and time.

[0064] (2) Figure 1 ,According to the latitude and longitude positions of the places of interest and meteorological stations in the operation area, the dates of rainfall, snowfall, strong winds, and dust weather occurring by station statistics are extracted from the b data.

[0065] (3) Extract elevation, slope, aspect data, and land type data from the C and D data based on the latitude and longitude locations of the locations of interest and meteorological stations within the operating area.

[0066] (4) The single-layer and isobaric layer data of ECMWF ERA5 reanalysis data are used to drive the mesoscale numerical model WRF to obtain the simulated data of 40 levels of meteorological elements from the ground to below the 500hPa isobaric surface. According to the locations of interest in the operation area, the custom altitude above the ground (unit: m, which can be an integer altitude below 3000m, such as 10m, 20m, ..., 1000m, ...) and forecast elements (such as temperature, wind field, humidity, etc.), the meteorological elements corresponding to the custom altitude values ​​are extracted, and these values ​​are sorted by time.

[0067] S3. Construct a REGOCF model with two levels: primary learning and secondary learning. The primary learning level uses multiple base learning models, which are trained with the same feature set. All prediction results obtained at the primary learning level are aggregated into a new feature matrix, which is trained using the OCF method to obtain the final prediction results. The main body of the REGOCF model is as follows: Figure 2 As shown in the figure, DATA1, DATA1, ... these data are the data (1), (2), (3), (4) described in 1; among them, data (4) is the label data used for LEVEL1 and LEVEL2 (OCF) training; REGOCF model consists of two layers: primary learning and secondary learning. The specific structure is as follows:

[0068] Primary learning layer: Multiple base learning models, such as random forest, support vector machine, gradient boosting and other machine learning algorithms, are trained using the same feature set, and different types of algorithms are selected to ensure model diversity.

[0069] Secondary learning layer: All prediction results obtained in the primary learning stage are aggregated into a new feature matrix. Subsequently, this new feature matrix is ​​trained using the OCF method to obtain the final prediction results.

[0070] The model building method is:

[0071] Data preparation

[0072] To ensure data quality and consistency, the following cleaning and organization steps were performed on the original data (data described in (1), (2), (3), and (4)):

[0073] Missing value processing: For missing values ​​in the data, linear interpolation or mean filling method is used to supplement them to ensure the integrity of the data;

[0074] Outlier detection: Use statistical methods such as box plots to identify outliers in the data and process them to reduce the potential impact of outliers on model training;

[0075] Standardization: Standardize data of different dimensions to the same scale to facilitate model training.

[0076] Feature Engineering

[0077] Feature engineering is one of the key steps to improve model performance. In this study, the following feature engineering methods are used (referring to feature engineering of (1), (2), (3), and (4) data):

[0078] Time features - extract features such as year, month, day, and day of the week from date information to capture seasonal and cyclical changes;

[0079] Spatial characteristics - Combine terrain data and underlying surface type data to generate spatial characteristics to reflect the impact of geographical location on temperature;

[0080] Historical trend characteristics - calculate the average value of meteorological elements in the same period of history, the highest and lowest values ​​of meteorological elements, form historical trend characteristics, and help the model better understand the long-term change law;

[0081] Special weather characteristics - extract the changing value characteristics of meteorological elements when precipitation, snowfall, strong winds and sandstorms occur.

[0082] Model Training

[0083] Data segmentation: The cleaned dataset is divided into training set, validation set and test set, with the allocation ratios of 70%, 15% and 15% respectively. The training set is used for model training, the validation set is used for hyperparameter adjustment, and the test set is used to evaluate the overall performance of the model.

[0084] Primary learning layer training: Use the training set data to train the model of each primary layer. During the training process, the optimal hyperparameter combination is selected through cross-validation to reduce the risk of overfitting.

[0085] Secondary learning layer training: Collect all primary model predictions and combine them with actual observations to form a new training dataset. Then, use this dataset to train the secondary model until the model performance reaches the expected effect.

[0086] (4) Parameter tuning:

[0087] Grid Search — For each primary model, a grid search technique is used to iterate over all possible hyperparameter combinations and select the best performing hyperparameter combination.

[0088] Bayesian Optimization - For the secondary model, the Bayesian optimization algorithm is used to further optimize the hyperparameter configuration to improve the generalization ability of the model.

[0089] Best combination - Select the best combination of primary layer models based on the performance on the validation set. Finally, evaluate it on the test set to ensure it has good generalization performance.

[0090] The present invention uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only used to help understand the method and core ideas of the present invention. At the same time, for those skilled in the art, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A meteorological element forecasting method based on REGOCF custom height, characterized by: The following steps are involved: S1. Collect training data and simulation data; S2. Processing the training data and simulation data according to the locations of interest in the operation area; S3. Construct a REGOCF model with two levels: primary learning and secondary learning. The primary learning level adopts multiple base learning models, and multiple base learning models are trained using the same feature set. All prediction results obtained at the primary learning level are summarized into a new feature matrix, and the new feature matrix is ​​trained using the OCF method to obtain the final prediction results.

2. The meteorological element forecasting method based on REGOCF custom altitude according to claim 1, characterized in that: In step S1, The training data includes: a. Collect numerical model data, including: CMA-MESO, ECMWF Thin, National Smart Grid and Inner Mongolia Autonomous Region Ruitu regional model forecast data; b. Collect observation data of meteorological elements in the operation area, including data from ground meteorological stations, wind towers, and laser wind radar, as well as dates of precipitation, snowfall, gale, and sandstorms; c. Collect 30m resolution digital elevation model data within the operation area; d. Collect data on land surface classification within the operation area; including cultivated land, forest land, grassland, shrub land, wetland, water bodies, tundra, artificial surface, bare land, glacier and permanent snow; The simulated data includes: e. Collect ECMWF ERA5 atmospheric reanalysis data, including single-layer and isobaric layer potential height, relative humidity, absolute humidity, temperature, u wind component, and v wind component meteorological element data.

3. The meteorological element forecasting method based on REGOCF custom altitude according to claim 2, characterized in that: All data have a period of 2 years.

4. The meteorological element forecasting method based on REGOCF custom altitude according to claim 2, characterized in that: The digital elevation model data in the operating area uses digital elevation data within 30 meters to 500 meters, which is the 30-meter digital elevation model data of the Copernicus program of the European Space Agency.

5. The meteorological element forecasting method based on REGOCF user-defined altitude according to claim 2, characterized in that: The surface classification data in the operating area uses surface coverage data within 30 meters to 500 meters, which is the 30-meter spatial resolution data of GlobeLand30.

6. The method for forecasting meteorological elements based on REGOCF user-defined altitude according to claim 2, characterized in that: Step S2 includes: S21. extracting temperature, wind field, humidity, and precipitation data from the numerical model data according to the latitude and longitude locations of the locations of interest and the meteorological stations in the operation area in the order of corresponding locations and time; S22. extracting the dates of rainfall, snowfall, gale, and sandstorm weather according to the statistics of the stations from the meteorological element observation data in the operation area according to the latitude and longitude positions of the concerned locations and meteorological stations in the operation area; S23. Extract elevation, slope, aspect data and land type data from 30-meter resolution digital elevation model data and surface classification data in the operating area according to the latitude and longitude locations of the locations of interest and the meteorological stations in the operating area; S24. The single-layer and isobaric layer data of ECMWF ERA5 reanalysis data are used to drive the mesoscale numerical model WRF to obtain the simulated data of 40 levels of meteorological element fields from the ground to below the 500hPa isobaric surface. According to the places of interest, custom altitudes above the ground and forecast elements in the operation area, the meteorological elements corresponding to the custom altitude values ​​are extracted and sorted by time.

7. The meteorological element forecasting method based on REGOCF user-defined altitude according to claim 6, characterized in that: In step S3, the method for constructing the REGOCF model includes: S31. Data preparation; S32. Feature engineering; S33. Model training.

8. The method for forecasting meteorological elements based on REGOCF user-defined altitude according to claim 7, characterized in that: Step S31 includes: S311. Missing value processing: missing values ​​in the data are supplemented by linear interpolation or mean filling method; S312. Outlier detection, using statistical methods such as box plots to identify outliers in the data and process them; S313. Standardization processing: standardize data of different dimensions and unify them to the same scale.

9. The meteorological element forecasting method based on REGOCF user-defined altitude according to claim 7, characterized in that: Step S32 includes: S321. Extract time features, extract year, month, day, and day of the week features from date information; S322. Extract spatial features, combine terrain data and underlying surface type data, and generate spatial features; S323. Extract historical trend characteristics, calculate the average value of meteorological elements in the same period of history, the highest and lowest values ​​of meteorological elements, and form historical trend characteristics; S324. Extract special weather characteristics, and extract the characteristics of meteorological element change values ​​when precipitation, snowfall, strong winds, and sandstorms occur.

10. The meteorological element forecasting method based on REGOCF user-defined altitude according to claim 7, characterized in that: Step S33 includes: S331. Data segmentation: divide the cleaned data set into a training set, a validation set, and a test set, with the allocation ratios being 70%, 15%, and 15% respectively. The training set is used for the model training process, the validation set is used for the adjustment of hyperparameters, and the test set is used to evaluate the overall performance of the model. S332. Primary learning layer training: using the training set data to train the model of each primary layer, and during the training process, selecting the optimal hyperparameter combination through the cross-validation method; S333. Secondary learning layer training, collecting all primary model prediction results and combining them with actual observations to form a new training data set, which is then used to train the secondary model until the model performance reaches the expected effect; S334. Parameter tuning: For each primary model, a grid search technique is used to traverse all possible hyperparameter combinations and select the best performing hyperparameter combination; Based on the performance on the validation set, the optimal combination of primary layer models is selected and finally evaluated on the test set.

Images