Correction method for measuring ground wind speed considering complex convective environmental characteristics

Abstract

The present application provides a method and device for correcting the measurement of ground wind speed considering the complex environmental characteristics of convection. The method includes: obtaining the original historical data of the target site; preprocessing the original historical data to generate standardized historical data; judging the category of the strong wind process based on the standardized historical data such as meteorological observations and forecasts, early warning information, disaster-causing events, etc. according to the thresholds of wind speed, duration, and convection parameters, and screening out ordinary cases, cold air systematic strong wind cases, and convective strong wind cases; based on the standardized historical data such as meteorological forecasts and meteorological observations, training and debugging LSTM, XGboost, and Stacking models under different categories of strong wind processes to generate a wind speed measurement classification correction model; for real-time wind speed measurement, judging the category of the strong wind process, and automatically matching the appropriate wind speed measurement correction model in the wind speed measurement classification correction model to correct the wind speed measurement.

Description

Technical Field

[0001] The present application relates to the technical field of data processing, and more specifically, to a correction method and device for calculating ground wind speed considering the characteristics of complex convective environments. Background Art

[0002] Currently, the correction of ground wind speed measurement data is usually obtained by using error statistical analysis methods or artificial intelligence models based on topographic features, meteorological observations, etc. For example: by combining high-resolution topographic elevation data and geomorphic data, using the empirical relationship between topographic features and wind speed, constructing a preliminary correction result of the wind speed at the target point or local high-resolution wind speed, and further based on the historical wind speed observations at the target point, using statistical analysis methods to calculate the wind speed measurement error at the target point, and combining the preliminary correction result of the wind speed to obtain the final wind speed correction result; or, using high-resolution topographic and geomorphic data, meteorological forecast data, and historical observation data to train an artificial intelligence model to obtain a correction model for ground wind speed measurement for correcting wind speed measurement. The above wind speed measurement correction models usually do not consider different weather processes and meteorological characteristics under different weather backgrounds. In the specific practice process, it is found that although these wind speed measurement correction models can reduce the overall error level of wind speed measurement, the wind speed correction effect is poor under the conditions of strong wind processes, especially convective strong wind processes. Therefore, it is necessary to construct a correction method for ground wind speed measurement considering the characteristics of complex convective environments. Summary of the Invention

[0003] The purpose of the embodiments of the present application is to provide a correction method for ground wind speed measurement considering the characteristics of complex convective environments, which is used to improve the problem that the error is still relatively large after the current site wind speed measurement is corrected.

[0004] The present application provides a correction method for ground wind speed measurement considering the characteristics of complex convective environments, which includes:

[0005] Obtaining the original historical data of the target site including meteorological forecasts, meteorological observations, warning information, and disaster-causing events;

[0006] Preprocessing the obtained original historical data of the target site to generate standardized historical data; based on the thresholds including wind speed, duration, and convective parameters, judging the category of strong wind processes based on the standardized historical data, and screening out ordinary cases, cold air systematic strong wind cases, and convective strong wind cases;

[0007] Based on the standardized historical data, training and debugging LSTM, XGboost, and Stacking models according to different strong wind process categories to generate a classification correction model for wind speed measurement;

[0008] For real-time wind speed measurement, in combination with thresholds including wind speed, duration, and convective parameters, the category of the strong wind process is judged, and accordingly, the appropriate wind speed measurement correction model is automatically matched in the wind speed measurement classification correction model to correct the wind speed measurement result.

[0009] Specifically, the generation of the wind speed measurement correction model A for ordinary cases is as follows:

[0010]

[0011] The generation of the wind speed measurement correction model B for cold air systematic strong wind cases is as follows:

[0012]

[0013] The generation of the wind speed measurement correction model C for convective strong wind cases is as follows:

[0014]

[0015] Among them, Xa and Xb are feature tensors constructed respectively from the standardized historical data including conventional meteorological element forecasts, the standardized historical data of meteorological observations of wind speed, and the standardized historical data of wind speed measurement errors for ordinary cases and cold air systematic strong wind cases, Xc is a feature tensor constructed from the standardized historical data including conventional meteorological element forecasts, the standardized historical data of convective parameter forecasts, the standardized historical data of meteorological observations of wind speed, and the standardized historical data of wind speed measurement errors for convective strong wind cases, Y is a label tensor constructed from the standardized historical data of wind speed measurement errors for the target forecast batch, T is the wind speed measurement error threshold, Y predict is the predicted value of the wind speed measurement error, WS orig is the standardized data of wind speed measurement, WS corr is the corrected value of the standardized data of wind speed measurement, Y' predict is the predicted value of the wind speed measurement error after threshold control.

[0016] Specifically, the specific process of the Stacking model includes:

[0017] Based on the standardized historical data of wind speed measurement and the standardized historical data of wind speed observations, calculate the standardized historical data of wind speed measurement errors;

[0018] Use the standardized historical data of meteorological forecasts including conventional meteorological elements such as wind speed, temperature, humidity, and air pressure and the standardized historical data of meteorological observations of wind speed to construct a feature tensor, and use the standardized historical data of wind speed errors to construct a label tensor; divide the feature tensor and the label tensor into a training set and a test set;

[0019] Input the training set and the test set into the artificial intelligence model for training and adjusting the model parameters to obtain multiple wind speed measurement error prediction models as the Stacking first-level learners;

[0020] Use the outputs of the multiple wind speed measurement error prediction models as new features to input into the second-level learner, train the Stacking second-level learner and adjust the parameters;

[0021] Construct a wind speed measurement error prediction model based on the Stacking first-level learner and the Stacking second-level learner.

[0022] Specifically, construct a feature tensor using the standardized historical data of meteorological forecasts and the standardized historical data of meteorological observations of wind speed, including:

[0023] Based on the standardized historical data of wind speed measurement and the standardized historical data of wind speed observation, calculate the standardized historical data of wind speed measurement error;

[0024] Based on the standardized historical data of wind speed measurement, the standardized historical data of wind speed observation, and the standardized historical data of wind speed measurement error of the same forecast batch in the past several days, construct the feature X di ; where the dimension of X di is R×C, the row R represents different forecast time steps, and the column C represents different meteorological elements;

[0025] Based on the feature X di , splice the features of the past several days to form a feature X dn with the dimension of R×(nC), where n represents the number of past days; based on the X dn , use a scaling method to obtain the input features of the machine learning model; the label feature is the error of the historical wind speed forecast in the current batch.

[0026] Adjust the number of days n, conduct a sensitivity experiment to obtain the optimal value of n, and thus construct the input feature X and the label feature Y of the machine learning model.

[0027] Specifically, generate a wind speed measurement correction model, including:

[0028] Use the wind speed measurement error prediction model to calculate the wind speed measurement error prediction value;

[0029] Use the wind speed measurement error prediction value and the standardized historical data of wind speed measurement to calculate the preliminary correction value of the standardized historical data of wind speed measurement;

[0030] Based on the standardized historical data of wind speed observations, the preliminary corrected value of the standardized historical data of wind speed measurement is tested, and the wind speed measurement error threshold is set according to the test results. This process is repeated to obtain the optimal threshold of wind speed measurement error to control the range of the predicted value of wind speed measurement error;

[0031] Based on the wind speed measurement error prediction model and the optimal threshold of wind speed measurement error, a wind speed measurement correction model is constructed.

[0032] Specifically, based on the standardized historical data including meteorological observations, early warning information, and disaster-causing events, the categories of gale processes are judged, including:

[0033] Based on the standardized historical data including early warning information and disaster-causing events, combined with national standards and industry standards, the thresholds including wind speed, gale duration, and convective parameters are established;

[0034] Based on the standardized historical data of meteorological observations, combined with the thresholds including wind speed, gale duration, and convective parameters, the categories of gale processes are judged.

[0035] Specifically, the judgment of the gale process category includes:

[0036] Based on the standardized historical data of meteorological observations, combined with the thresholds including wind speed and gale duration, ordinary cases and gale cases are determined;

[0037] Based on the standardized historical data of meteorological forecasts, within the gale cases, using the convective parameters, Pearson correlation analysis is applied to obtain highly correlated convective parameters. Based on the highly correlated convective parameters, the k-means clustering method is further used to screen out cold air systematic gale cases and convective gale cases.

[0038] Specifically, based on the highly correlated convective parameters, the k-means clustering method is used to screen out cold air systematic gale cases and convective gale cases, including:

[0039] The data including highly correlated convective parameters and historical meteorological element data are sorted out, and a training data set that meets the k-means input format is formed;

[0040] Based on the training data set that meets the k-means input format, the k-means model is trained, and the number of centers is determined through sensitivity experiments and used as the number of case types;

[0041] Based on the results of the sensitivity experiment, cold air systematic gale cases and convective gale cases are screened out.

[0042] Specifically, for real-time wind speed measurement, the category of the strong wind process is judged by combining the thresholds of wind speed, duration, and convection parameters, and accordingly, the appropriate wind speed measurement correction model is automatically matched in the wind speed measurement classification correction model to correct the wind speed measurement, including:

[0043] Obtain real-time wind speed measurement and preprocess it into standardized real-time data for weather forecasting;

[0044] Based on the thresholds of wind speed, duration, and convection parameters, judge the category of the strong wind process;

[0045] According to the category of the strong wind process, match the appropriate wind speed measurement correction model from the wind speed measurement classification correction model to correct the wind speed measurement.

[0046] The present application also provides a ground wind speed correction device considering the characteristics of the convective complex environment, which includes:

[0047] A data acquisition module for the original historical data including weather forecasting, meteorological observation, early warning information, and disaster-causing events of the target site;

[0048] A data preprocessing module for preprocessing the original historical data of the target site obtained above to generate standardized historical data;

[0049] A strong wind process classification module for judging the category of the strong wind process based on the thresholds of wind speed, duration, and convection parameters, and screening out ordinary cases, cold air systematic strong wind cases, and convective strong wind cases based on the standardized historical data including meteorological observation, early warning information, and disaster-causing events;

[0050] A wind speed measurement correction model training module for training and debugging LSTM, XGboost, and Stacking models according to different strong wind process categories based on the standardized historical data including weather forecasting and meteorological observation to generate a wind speed measurement classification correction model;

[0051] A real-time wind speed measurement correction module for judging the category of the strong wind process according to the thresholds of wind speed, duration, and convection parameters, and accordingly selecting an appropriate wind speed correction model to correct the real-time measured wind speed.

[0052] Beneficial technical effects: This method can significantly improve the accuracy of wind speed measurement. After correction, the mean absolute error of wind speed measurement is greatly reduced, which means that the average deviation degree between the measured value and the actual wind speed value is significantly reduced; this method can enhance the timeliness and adaptability of wind speed measurement. This method designs a set of efficient real-time wind speed measurement correction methods. By combining the thresholds of wind speed, duration, and convection parameters, it can quickly judge the category of the strong wind process and automatically match the most suitable model in the wind speed measurement classification correction model to immediately optimize the wind speed measurement result. Description of the Drawings

[0053] Figure 1 Flow chart of a correction method for ground wind speed measurement considering complex convective environmental characteristics;

[0054] Figure 2 Schematic diagram of a device for a correction method for ground wind speed measurement considering complex convective environmental characteristics;

[0055] Figure 3 Schematic diagram of the correction effect of wind speed measurement at a certain meteorological station in the middle of January 2023;

[0056] Figure 4 Schematic diagram of the correction effect of wind speed measurement at a certain meteorological station in the middle of July 2023. Detailed Implementation Manner

[0057] As Figure 1 shown, an embodiment of the present application provides a ground wind speed measurement correction method considering complex convective environmental characteristics, including:

[0058] Step S110: Obtain original historical data such as meteorological forecasts, meteorological observations, early warning information, and disaster-causing events of the target station.

[0059] The original historical data of meteorological forecasts includes: historical forecast data of conventional meteorological elements such as temperature, humidity, air pressure, precipitation, dew point temperature, wind speed, and wind direction, and historical forecast data of convective parameters such as convective available potential energy (CAPE), convective inhibition energy (CIN), and K index;

[0060] The original historical data of meteorological observations includes: historical observation data of meteorological elements such as gusts, wind speed, and wind direction;

[0061] The original historical data of early warning information includes: historical early warning information issued by meteorological departments at all levels and industry historical early warning information;

[0062] The original historical data of disaster-causing events includes: disaster records of government departments such as civil affairs disaster relief meteorology at all levels and the industry.

[0063] Step S120: Preprocess the obtained original historical data of the target station (quality control, filling or discarding missing values, standardization, etc.).

[0064] For the original historical data of meteorological forecasts, use spatial interpolation methods to extract forecasts of various meteorological elements at the target point, organize the forecast batches, discard the forecast batches with a large number of missing forecast times, perform time interpolation filling on the forecast batches with a small number of missing forecast times, and then generate standardized historical data of meteorological forecasts according to the prefabricated format;

[0065] Perform quality control on the original historical meteorological observation data (such as threshold value check, internal consistency check, spatial consistency check, temporal consistency check, etc.), discard the observation periods with a large number of consecutive missing observations, interpolate and fill the observation periods with a small number of consecutive missing observations, extract the required meteorological elements and generate standardized historical meteorological observation data in a prefabricated format;

[0066] For the original historical warning information data, extract the warning information related to the target stations, eliminate the duplicate warnings in the multi-level warnings at the national, provincial, city, and county levels, and generate standardized historical meteorological warning information data;

[0067] For the original historical disaster-causing event data, extract the disaster-causing information related to the target stations, eliminate the duplicate records in the multi-level disaster-causing event records at the national, provincial, city, and county levels and the industry disaster records, and generate standardized historical disaster-causing event data.

[0068] Step S130: Based on the thresholds of wind speed, duration, and convective parameters, judge the category of the gale process based on the standardized historical data such as meteorological observations, warning information, and disaster-causing events, and select ordinary cases, cold air systematic gale cases, and convective gale cases.

[0069] Based on the standardized historical data of warning information, disaster-causing event data, meteorological observation data, etc., combined with national standards, industry standards, etc., establish the thresholds of wind speed, gale duration, convective parameters, etc.;

[0070] Based on the thresholds of the wind speed, gale duration, etc. and the standardized historical meteorological observation data, judge the category of the gale process and determine ordinary cases and gale cases.

[0071] Within the gale cases, based on the standardized historical meteorological forecast data, use Pearson correlation analysis for multiple convective parameters to obtain highly correlated convective parameters;

[0072] Furthermore, based on the highly correlated convective parameters, further use the k-means clustering method to screen out cold air systematic gale cases and convective gale cases, including: organizing the highly correlated convective parameters, historical meteorological element data, etc. and forming a training data set that meets the k-means input format; training the k-means model based on the training data set that meets the k-means input format, determining the number of centers through sensitivity experiments and using it as the number of case categories; based on the results of the sensitivity experiment, screening out cold air systematic gale cases and convective gale cases.

[0073] Step S140: Based on the standardized historical data such as meteorological forecasts and meteorological observations, train and debug LSTM, XGboost, and Stacking models according to different gale process categories to generate a wind speed measurement classification correction model.

[0074] For the general cases, based on the historical standardized meteorological forecast data and historical standardized meteorological observation data of conventional meteorological elements such as wind speed, temperature, humidity, and air pressure, an LSTM model is trained to obtain a wind speed measurement correction model A;

[0075] For the cold air systematic gale cases, based on the historical standardized meteorological forecast data and historical standardized meteorological observation data of conventional meteorological elements such as wind speed, temperature, humidity, and air pressure, an LSTM is trained to obtain a wind speed measurement correction model 1, an LSTM and XGboost are trained to obtain a wind speed measurement correction model 2, and further using the Stacking method, a wind speed measurement correction model B is obtained;

[0076] For the convective gale cases, based on the historical standardized meteorological forecast data of conventional meteorological parameters such as wind speed, temperature, humidity, air pressure and highly correlated convective parameters, and historical standardized meteorological observation data, an LSTM is trained to obtain a wind speed measurement correction model 3, an LSTM and an XGboost model are trained to obtain a wind speed measurement correction model 4, and further using the Stacking method, a wind speed measurement correction model C is obtained.

[0077] The wind speed measurement classification correction model is composed of the wind speed measurement correction model A, the wind speed measurement correction model B, and the wind speed measurement correction model C.

[0078] Step S150: For real-time wind speed measurement, in combination with the wind speed, duration, highly correlated convective parameters and corresponding thresholds, judge the category of the gale process, and accordingly automatically match a suitable wind speed correction model in the wind speed measurement classification correction model to correct the wind speed measurement.

[0079] Obtain real-time wind speed measurement and process it into standardized meteorological forecast data;

[0080] Based on the wind speed, duration, highly correlated convective parameters and corresponding thresholds, judge the category of the gale process;

[0081] According to the category of the gale process, match a suitable wind speed measurement correction model from the wind speed measurement classification correction model (wind speed measurement correction model A, wind speed measurement correction model B, wind speed measurement correction model C) to correct the wind speed measurement.

[0082] In the implementation process of the above solution, by using the historical data such as meteorological forecast, meteorological observation, early warning information, and disaster-causing events of the target site to judge different gale weather processes and establishing wind speed measurement correction models respectively, the ground wind speed measurement correction model can adapt to the changes of different weather processes, effectively improving the accuracy of the ground wind speed measurement data.

[0083] As an alternative implementation of the above step S110, the original historical meteorological forecast data of the target site can be obtained from the China Meteorological Administration Mesoscale Numerical Weather Prediction Operational System (CMA-MESO); the original historical meteorological observation data of the target site can be obtained from the national surface meteorological observation station network of the China Meteorological Administration; the original historical warning information data of the target site can be obtained from the national meteorological warning information database and the industry warning information database of the China Meteorological Administration; the original historical data of disaster-causing events of the target site can be collected from government departments such as civil affairs and disaster relief meteorology at all levels and industry records.

[0084] As an alternative implementation of the above step S120, preprocessing (quality control, missing value filling or discarding, standardization, etc.) is performed on the obtained original historical data of the target site, including:

[0085] Step S121: For the original historical meteorological forecast data, use spatial interpolation methods to extract multiple meteorological element forecasts at the target location; organize the forecast batches, discard the forecast batches with a large number of missing forecast times, and perform time interpolation filling on the forecast batches with a small number of missing forecast times; then generate standardized historical meteorological forecast data according to the prefabricated format.

[0086] An example of the implementation of the above step S121: The original historical meteorological forecast data comes from numerical forecasts such as CMA-MESO and is grid data. Use spatial interpolation methods (such as nearest point interpolation, bilinear interpolation, machine learning statistical regression, etc.) to obtain multiple historical meteorological element forecast data at the target location; count the number of forecast times included in each forecast batch, discard the forecast batches with the number of forecast times lower than the preset threshold, and retain the forecast batches with the number of forecast times higher than the preset threshold; for the retained forecast batches, if some forecast times are missing, use time interpolation methods (such as linear interpolation, time series statistics, machine learning regression, etc.) to fill the multiple historical meteorological element forecast data at the target location; then generate standardized historical meteorological forecast data according to the prefabricated format.

[0087] Step S122: Perform quality control on the original historical meteorological observation data, discard the observation periods with a large number of consecutive missing observations, interpolate and fill the observation periods with a small number of consecutive missing observations, extract the required meteorological elements, and generate standardized historical meteorological observation data according to the prefabricated format.

[0088] For example, the implementation of step S122 is as follows: historical meteorological observation data is sorted through quality control methods (such as limit value checks, internal consistency checks, spatial consistency checks, time consistency checks, etc.), observation periods with a large number of consecutive missing observations are discarded, and interpolation (such as time series analysis interpolation, linear interpolation, machine learning regression interpolation, etc.) is used to fill observation periods with a small number of consecutive missing observations. Then, the required meteorological elements are extracted and standardized historical meteorological observation data is generated in a prefabricated format.

[0089] Step S123: For the original historical warning information data, extract the warning information related to the target site, eliminate duplicate warnings in multi-level warnings at the national, provincial, municipal, county levels and industry warnings, and generate standardized historical warning information data.

[0090] For example, the implementation of step S123 is as follows: important fields (such as time period, location, warning category, warning level, etc.) are extracted through a matching algorithm, and multi-level warnings at the national, provincial, municipal, county levels and industry warnings are integrated through a deduplication algorithm to obtain preliminary effective warning information data, and then standardized historical warning information data is generated in a preset format.

[0091] Step S124: For the original historical disaster-causing event data, extract the strong wind disaster-causing information related to the target site, eliminate duplicate records in multi-level disaster-causing event records at the national, provincial, municipal, county levels and industry disaster records, and generate standardized historical disaster-causing event data.

[0092] For example, the implementation of step S124 is as follows: important fields (such as time period, affected area, disaster category, disaster level, etc.) are extracted through a matching algorithm, and multi-level disaster-causing event records at the national, provincial, municipal, county levels and industry disaster records are integrated through a deduplication algorithm to obtain preliminary effective disaster-causing event data, and then standardized historical disaster-causing event data is generated in a preset format.

[0093] As an alternative implementation of step S130 above, based on thresholds such as wind speed and duration and convective parameters, judging the category of strong wind process based on standardized historical data such as meteorological observations, warning information, and disaster-causing events may include:

[0094] Step S131: Based on standardized historical data such as warning information, disaster-causing events, and meteorological observations, combined with national standards, industry standards, etc., establish thresholds such as wind speed and strong wind duration.

[0095] For example, the implementation of step S131 is as follows: for the target location, target industry, and target operation scenario, combined with national standards, industry standards, etc., obtain threshold 1 for wind speed, strong wind duration, etc.; for standardized historical data such as warning information and disaster-causing events, match standardized historical meteorological observation data to obtain threshold 2; use threshold 2 to correct threshold 1 to determine thresholds such as wind speed and strong wind duration.

[0096] Step S132: Based on the thresholds such as the wind speed and the duration of strong wind, and the standardized historical meteorological observation data, determine the category of the strong wind process, and identify ordinary cases and strong wind cases.

[0097] An implementation manner of the above step S132 is as follows: For example, use the thresholds such as the wind speed and the duration of strong wind to match the corresponding observation periods in the standardized historical meteorological observation data, and obtain the standardized historical meteorological observation data during the strong wind period and the standardized historical meteorological observation data during the ordinary period; according to the above-mentioned strong wind period and ordinary period, select the corresponding standardized historical forecast data during the strong wind period and the standardized historical forecast data during the ordinary period from the standardized historical meteorological forecast data; an ordinary case is composed of the standardized historical meteorological observation data during the ordinary period and the standardized historical meteorological forecast data during the ordinary period; a strong wind case is composed of the standardized historical meteorological observation data during the strong wind period and the standardized historical meteorological forecast data during the strong wind period.

[0098] Step S133: Within the strong wind cases, based on the standardized historical meteorological forecast data, according to the convective parameter forecasts therein, use Pearson correlation analysis to obtain highly correlated convective parameters, and further screen out cold air systematic strong wind cases and convective strong wind cases based on the highly correlated convective parameters using the k-means clustering method.

[0099] An implementation manner of the above step S133 is as follows: For example, in the strong wind cases, use the Pearson correlation analysis method to establish the correlations between various convective parameters CV such as convective available potential energy, convective inhibition energy, K index, and helicity and the wind speed U , and select the highly correlated convective parameters; for the selected highly correlated convective parameters, use the k-means clustering method to further screen the strong wind cases to obtain cold air systematic strong wind cases and convective strong wind cases, including: organizing the above-mentioned highly correlated convective parameters, historical meteorological element data, etc. to form a training data set that meets the k-means input format; training the k-means model based on the training data set that meets the k-means input format, and determining the number of centers through sensitivity experiments and taking it as the number of case types; based on the results of the sensitivity experiments, screen out cold air systematic strong wind cases and convective strong wind cases. As an optional implementation manner of the above step S140, the implementation manner of establishing the wind speed measurement classification correction model may include:

[0100] Step S141: Make a feature tensor and a label tensor based on the standardized historical meteorological forecast data and the standardized historical meteorological observation data.

[0101] The implementation of the above step S141 is as follows: Based on the standardized historical data of wind speed measurement and the standardized historical data of wind speed observation, calculate the standardized historical data of wind speed measurement error; Based on the standardized historical data of wind speed measurement, the standardized historical data of wind speed observation, and the standardized historical data of wind speed measurement error in the same forecast batch (such as the 08:00 batch, the 20:00 batch, etc.) in the past several days (-3d, -4d, -5d, etc.), construct the feature X di ; where the dimension of X di is R*C, the row R represents different forecast time steps, and the column C represents different meteorological elements (forecast, convective parameter, observation, wind speed measurement error); Based on the feature X di , splice the features of the past several days to form a feature X dn with the dimension of R×(nC), where n represents the number of past days; Based on the X dn , use scaling methods (such as normalization method, standardization method, etc.) to obtain the input features of the machine learning model; The label feature is the error of the historical wind speed forecast in the current batch; Adjust the number of days n, conduct a sensitivity experiment to obtain the optimal value of n, and thus construct the input feature X and label feature Y of the machine learning model.

[0102] Step S142: For the ordinary cases, based on the feature tensor and the label tensor, train the LSTM model to obtain the wind speed measurement correction model A.

[0103] The implementation of the above step S142 is as follows: In the ordinary cases, construct the feature tensor Xa with the standardized historical data of the conventional meteorological elements (such as wind speed, temperature, humidity, air pressure, etc.) forecast in the same forecast batch in the past several days, the standardized historical data of wind speed meteorological observation, and the standardized historical data of wind speed measurement error, and construct the label tensor Ya with the standardized historical data of wind speed measurement error in the target forecast batch; Divide (the feature tensor Xa, the label tensor Ya) into a training set and a test set; Input the training set and the test set into the LSTM long short-term memory neural network model for training and adjust the model parameters to obtain the wind speed measurement error prediction model A; Use the wind speed error prediction value and the standardized historical data of wind speed measurement to calculate the preliminary correction value of the standardized historical data of wind speed measurement; Based on the standardized historical data of wind speed observation, test the preliminary correction value of the standardized historical data of wind speed measurement, and set the wind speed measurement error threshold according to the test results, and repeat this process to obtain the best wind speed measurement error threshold to control the range of the wind speed measurement error prediction value; The above constitutes the wind speed measurement correction model A. The specific process is as follows,

[0104]

[0105] Among them, Xa is a feature tensor constructed from the standardized historical data including conventional meteorological element forecasts of ordinary cases, the standardized historical data of meteorological observations of wind speed, and the standardized historical data of wind speed measurement errors; Ya is a label tensor constructed from the standardized historical data of wind speed measurement errors of the target forecast batches of ordinary cases; Ta is the wind speed measurement error threshold of ordinary cases; Ya predict is the predicted value of the wind speed measurement error of ordinary cases; WS orig is the standardized data of wind speed measurement; WS corr is the correction value of the standardized data of wind speed measurement; Ya′ predict is the predicted value of the wind speed measurement error of ordinary cases after threshold control.

[0106] Step S143: For the cold air systematic gale cases, based on the feature tensor and the label tensor, train the LSTM to obtain the wind speed correction model 1, train the LSTM and XGboost to obtain the wind speed correction model 2, and further use the Stacking method to obtain the wind speed measurement correction model B.

[0107] For example, in the implementation of the above step S143: in the case of cold air systematic strong wind, construct the feature tensor Xb with the standardized historical data of conventional meteorological elements (wind speed, temperature, humidity, air pressure, etc.) forecast in the same forecast batch in the past several days, the standardized historical data of meteorological observations of wind speed, and the standardized historical data of wind speed measurement error. Construct the label tensor Yb with the standardized historical data of wind speed measurement error in the target forecast batch of the cold air systematic strong wind case. Divide (feature tensor Xb, label tensor Yb) into a training set and a test set; input the training set and the test set into the LSTM long short-term memory neural network model for training and adjust the model parameters to obtain the wind speed measurement error prediction model B1; input the training set and the test set into the LSTM long short-term memory neural network model, and the input result Ob_lstm of the LSTM model and the label tensor Yb are combined into new input data and connected to the extreme gradient boosting model XGBoost. Train the LSTM and XGBoost models and adjust the model parameters to obtain the wind speed measurement error prediction model B2; use the wind speed measurement error prediction model B1 and the wind speed measurement error prediction model B2 as the Stacking first-level learners, and their outputs are used as new feature inputs and train the Stacking second-level learner (linear model or other machine learning models). Obtain the wind speed measurement error prediction model B from the above Stacking first-level learner and Stacking second-level learner; use the wind speed error prediction value and the standardized historical data of wind speed measurement to calculate the preliminary correction value of the standardized historical data of wind speed measurement; based on the standardized historical data of wind speed observation, test the preliminary correction value of the standardized historical data of wind speed measurement, and set the wind speed measurement error threshold according to the test result. Repeat this process to obtain the optimal wind speed measurement error threshold to control the range of the wind speed error prediction value; the above constitutes the wind speed measurement correction model B. The specific process is as follows:

[0108]

[0109] Among them, Xb is the feature tensor constructed with the standardized historical data including conventional meteorological element forecasts, the standardized historical data of meteorological observations of wind speed, and the standardized historical data of wind speed measurement error in the case of cold air systematic strong wind. Yb is the label tensor constructed with the standardized historical data of wind speed measurement error in the target forecast batch of the cold air systematic strong wind case. Tb is the wind speed measurement error threshold in the case of cold air systematic strong wind. Yb predict is the wind speed measurement error prediction value in the case of cold air systematic strong wind. WS orig is the standardized wind speed measurement data. WS corr is the correction value of the standardized wind speed measurement data. Yb′ predict is the wind speed measurement error prediction value after threshold control in the case of cold air systematic strong wind.

[0110] Step S144: For the convective gale case, based on the feature tensor and the label tensor, train the LSTM to obtain the wind speed correction model 3, train the LSTM and the XGboost model to obtain the wind speed correction model 4, and further use the Stacking method to obtain the wind speed measurement and correction model C.

[0111] The implementation method of the above step S143 is as follows: In the convective gale case, construct the feature tensor Xc with the standardized historical data of the conventional meteorological elements (wind speed, temperature, humidity, air pressure, etc.) forecast in the same forecast batch in the past several days, the standardized historical data of the convective parameter forecast, the standardized historical data of the meteorological observation of the wind speed, and the standardized historical data of the wind speed measurement error. Construct the label tensor Yc with the standardized historical data of the wind speed measurement error in the target forecast batch; divide (the feature tensor Xc, the label tensor Yc) into a training set and a test set; input the training set and the test set into the LSTM long short-term memory neural network model for training and adjust the model parameters to obtain the wind speed measurement error prediction model C1; input the training set and the test set into the LSTM long short-term memory neural network model, and the input result Oc_lstm of the LSTM model and the label tensor Yc are combined into new input data and connected to the extreme gradient boosting model XGBoost. Train the LSTM and the XGBoost model and adjust the model parameters to obtain the wind speed measurement error prediction model C2; use the wind speed measurement error prediction model C1 and the wind speed measurement error prediction model C2 as the Stacking first-level learners, and their outputs are used as new feature inputs and train the Stacking second-level learner (linear model or other machine learning models). The wind speed measurement error prediction model C is obtained from the above Stacking first-level learner and Stacking second-level learner; use the wind speed error prediction value and the standardized historical data of the wind speed measurement to calculate the preliminary correction value of the standardized historical data of the wind speed measurement; based on the standardized historical data of the wind speed observation, check the preliminary correction value of the standardized historical data of the wind speed measurement, and set the wind speed measurement error threshold according to the inspection result. Repeat this process to obtain the optimal wind speed measurement error threshold to control the range of the wind speed error prediction value; the above constitutes the wind speed measurement and correction model C. The specific process is as follows:

[0112] Among them, Xc is the feature tensor constructed with the standardized historical data including the conventional meteorological element forecast, the standardized historical data of the convective parameter forecast, the standardized historical data of the meteorological observation of the wind speed, and the standardized historical data of the wind speed measurement error in the convective gale case. Yc is the label tensor constructed with the standardized historical data of the wind speed measurement error in the target forecast batch of the convective gale case. Tc is the wind speed measurement error threshold in the convective gale case. Yc predict is the wind speed measurement error prediction value in the convective gale case. WS orig is the standardized data of the wind speed measurement. WScorr Yc′ is the standardized data correction value for wind speed measurement. predict It is the predicted value of the wind speed measurement error after threshold control for convective gale cases.

[0113] As an alternative implementation of the above step S150, the implementation of correcting the real-time wind speed measurement may include:

[0114] Step S151: Obtain the real-time wind speed measurement and process it into standardized real-time meteorological forecast data.

[0115] For example, the implementation of the above step S151: Obtain real-time forecast data from the China Meteorological Administration Mesoscale Numerical Weather Prediction Operational System (CMA-MESO); extract the forecast data of conventional meteorological elements such as wind speed, temperature, humidity, and air pressure, and highly correlated convective parameters at the target station; generate standardized real-time meteorological forecast data according to the prefabricated format.

[0116] Step S152: Based on thresholds such as wind speed, gale duration, and convective parameters, determine the category of the gale process described in the real-time forecast.

[0117] For example, the implementation of the above step S152: Based on thresholds such as wind speed, duration, and convective parameters, compare the wind speed, gale duration, convective parameters, etc. in the standardized real-time meteorological forecast data to determine the category of the gale process described in the real-time forecast.

[0118] Step S153: According to the category of the gale process, match a suitable wind speed measurement correction model from the wind speed measurement classification correction models to correct the wind speed measurement.

[0119] For example, the implementation of the above step S153: According to the determined category of the gale process, match a suitable wind speed measurement correction model from the wind speed measurement classification correction models (wind speed measurement correction model A, wind speed measurement correction model B, wind speed measurement correction model C), establish the corresponding input feature tensor according to the selected wind speed measurement correction model, and generate the corrected wind speed measurement after inputting the feature tensor into the model.

[0120] As Figure 2 shown, the embodiment of the present application provides a ground wind speed correction device 200 considering the characteristics of the convective complex environment, including:

[0121] A data acquisition module 210, configured to acquire original historical data such as historical meteorological forecasts, historical meteorological observations, historical warning information, and historical disaster-causing events of the target station.

[0122] A data preprocessing module 220, which is used to perform preprocessing such as cleaning, quality control, filling or discarding default values, and standardization on various historical data of the target site obtained above, and generate standardized historical data.

[0123] A gale process classification module 230, which is used to judge the gale process category based on data such as wind speed and duration thresholds, historical meteorological observations, historical warning information, and historical disaster-causing events, screen out ordinary cases and gale cases, and further screen out cold air systematic gale cases and convective gale cases by using the k-means method according to convective parameters (such as CAPE) in the gale cases.

[0124] A wind speed measurement classification correction model training module 240, which is used to train and debug LSTM, XGboost, and Stacking models based on historical meteorological forecast data and historical meteorological observation data according to the gale process category, and generate a wind speed measurement classification correction model.

[0125] A real-time wind speed measurement correction module 250, which is used to judge the gale process category according to wind speed, duration and other thresholds and convective parameters, and select a suitable wind speed correction model accordingly to correct the real-time wind speed measurement.

[0126] As an optional implementation manner of the above device, the original historical data includes: original historical meteorological forecast data, original historical meteorological observation data, original historical warning information data, and original historical disaster-causing event data; the data acquisition module includes:

[0127] An original historical meteorological forecast data acquisition module, which is acquired from the China Meteorological Administration Mesoscale Numerical Weather Prediction Operational System (CMA-MESO).

[0128] An original historical meteorological observation data acquisition module, which is acquired from the national surface meteorological observation station network of the China Meteorological Administration.

[0129] An original historical warning information data acquisition module, which is acquired from the national meteorological warning information database of the China Meteorological Administration and the industry warning information database.

[0130] An original historical disaster-causing event data acquisition module, which is acquired from government departments such as civil affairs disaster relief meteorology at all levels and industry records.

[0131] As an optional implementation manner of the above device, the data preprocessing module includes:

[0132] A meteorological forecast data preprocessing module, which is used to extract the original historical forecasts of various meteorological elements at the target point, fill in the missing forecast times, and generate standardized historical meteorological forecast data according to the prefabricated format;

[0133] A meteorological observation data preprocessing module, which is used to perform quality control on the original historical meteorological observation data, discard the observation data with relatively many consecutive missing measurements, interpolate and fill the observation data with relatively few consecutive missing measurements, extract the required meteorological elements, and generate standardized historical meteorological observation data according to the prefabricated format;

[0134] An early warning information preprocessing module, which extracts the original historical data of early warning information related to the target station, eliminates the duplicate early warnings in the multi-level early warning information of the country, province, city, county and industry early warning information, and generates standardized historical meteorological early warning information;

[0135] A disaster-causing event preprocessing module, which extracts the original historical data of disaster-causing information related to the target station, eliminates the duplicate records in the multi-level disaster-causing event records of the country, province, city, county and industry disaster records, and generates standardized historical disaster-causing event data.

[0136] As an optional implementation manner of the above device, a strong wind process classification module includes:

[0137] A threshold determination module, which is used to use the standardized historical data of early warning information, disaster-causing event standardized historical data, meteorological observation standardized historical data, etc., and combine national standards, industry standards, etc., to establish thresholds such as wind speed and strong wind duration;

[0138] A preliminary classification module, which is used to use thresholds such as wind speed and strong wind duration and standardized historical meteorological observation data to judge the category of strong wind process and determine ordinary cases and strong wind cases.

[0139] A refinement classification module, which is used within the strong wind cases, uses the convective parameter data in the standardized historical meteorological forecast data, obtains highly correlated convective parameters through Pearson correlation analysis, and further screens out cold air systematic strong wind cases and convective strong wind cases based on the highly correlated convective parameters using the k-means clustering method.

[0140] As an optional implementation manner of the above device, a wind speed measurement and classification correction model training module includes:

[0141] An ordinary wind speed measurement and correction model training module, which is used for ordinary cases, and trains an LSTM model based on the standardized historical meteorological forecast data and meteorological observation standardized historical data of conventional meteorological elements such as wind speed, temperature, humidity, and air pressure to obtain a wind speed correction model A;

[0142] The module for training the systematic gale wind speed calculation and correction model is used for the case of systematic gale in cold air. Based on the standardized historical data of meteorological forecast and meteorological observation of conventional meteorological elements such as wind speed, temperature, humidity, and air pressure, LSTM is trained to obtain wind speed correction model 1. LSTM and XGboost are trained to obtain wind speed correction model 2. The stacking method is further used to obtain the final wind speed correction model B.

[0143] The convective gale wind speed estimation and correction model training module is used for convective gale cases. Based on the standardized historical data of meteorological forecasts and meteorological observations of conventional meteorological parameters such as wind speed, temperature, humidity, and air pressure and highly correlated convective parameters, the LSTM is trained to obtain the wind speed correction model 3. The LSTM and XGboost models are trained to obtain the wind speed correction model 4. The Stacking method is further used to obtain the final wind speed correction model C.

[0144] As an optional implementation of the above device, the real-time wind speed measurement and correction module includes:

[0145] Real-time data standardization module, used to obtain real-time wind speed measurement and process it into standardized real-time data for weather forecast;

[0146] Real-time high wind process classification module, used to determine the high wind process category using thresholds such as wind speed, duration, and convection parameters;

[0147] The wind speed measurement classification correction module is used to select a suitable wind speed measurement correction model from the wind speed measurement classification correction models (wind speed measurement correction model A, wind speed measurement correction model B, wind speed measurement correction model C) to correct the wind speed measurement according to the category of the strong wind process.

[0148] It should be understood that the device corresponds to the above-mentioned wind farm forecast data correction method embodiment and can execute the various steps involved in the above-mentioned method embodiment. The specific functions of the device can be found in the above description, and the detailed description is appropriately omitted here. The device includes at least one software function module that can be stored in a memory in the form of software or firmware or solidified in the operating system (OS) of the device.

[0149] The following are examples of application scenarios applicable to the method claimed in this application: energy and power construction and operation, transportation construction and operation, engineering construction, etc. Specifically, in the application scenario of energy and power construction and operation, wind farm operators can use the corrected wind farm forecast data to more accurately predict future wind speed and direction, reasonably arrange the start-stop and maintenance plans of wind turbines, and correct the power output, which helps wind power enterprises participate in the electricity market bidding and improve the wind power utilization rate and economic benefits. In the application scenario of transportation construction and operation, railway or light rail operation departments can use the corrected wind farm forecast data to issue wind speed warnings in a timely manner to improve operation efficiency and safety. In the application scenario of engineering construction, wind farm forecasting is crucial for safe construction and safe hoisting. Construction parties can use the corrected wind farm forecast data to issue wind speed warnings in advance and adjust the construction plan to ensure construction safety. In addition, meteorological departments can use refined wind farm forecast data and corrected wind farm forecast data to predict and warn in advance of natural disasters caused by strong winds on the local ground.

[0150] The correction effect was tested for multiple strong wind cases in the Beijing area from January to October 2023. The comprehensive test results of multiple meteorological stations showed that after correction by the ground wind speed measurement correction model, the mean absolute error of wind speed forecast (1.64 m / s) decreased by 0.58 m / s compared with the original forecast (2.22 m / s); the mean error (0.02 m / s) decreased by 1.39 m / s compared with the original forecast (1.41 m / s), and the optimization effect of wind speed forecast was obvious.

[0151] As Figure 3 , 4 shown, the test results of individual cases of wind speed measurement correction at a certain meteorological station in mid-January and mid-July 2023 showed that after correction by the ground wind speed measurement correction model, the average wind speed forecast (blue line) at each moment was significantly lower than the original forecast (gray line), and the time series was closer to the actual observation (gray shadow).

Claims

1. A method for calculating and correcting ground wind speed considering complex convective environment characteristics, characterized in that: include: Obtain the original historical data of the target site, including weather forecasts, weather observations, warning information, and disaster-causing events; Preprocessing the original historical data of the target site obtained above to generate standardized historical data; According to the thresholds including wind speed, duration, and convection parameters, the category of the gale process is determined based on standardized historical data, and ordinary cases, cold air systemic gale cases, and convective gale cases are screened out; Based on standardized historical data, according to different categories of strong wind processes, LSTM, XGboost, and Stacking models are trained and debugged to generate a wind speed measurement classification correction model. Among them, the generation of the wind speed measurement correction model A for ordinary cases is specifically as follows: The generation of wind speed calculation correction model B for cold air systematic gale cases is as follows: The generation of wind speed calculation correction model C for convective gale cases is as follows: Among them, Xa and Xb are feature tensors constructed with the standardized historical data of conventional meteorological element forecasts, standardized historical data of meteorological observations of wind speed, and standardized historical data of wind speed measurement errors for common cases and cold air systematic gale cases, respectively; Xc is a feature tensor constructed with the standardized historical data of conventional meteorological element forecasts, standardized historical data of convective parameter forecasts, standardized historical data of meteorological observations of wind speed, and standardized historical data of wind speed measurement errors for convective gale cases; Y is a label tensor constructed with the standardized historical data of wind speed measurement errors of the target forecast batch; T is the wind speed measurement error threshold; Y is the label tensor constructed with the standardized historical data of wind speed measurement errors of the target forecast batch; predict is the wind speed measurement error prediction value, WS orig Standardized data for wind speed measurement, WS corr Y′ is the corrected value of wind speed measurement standard data, predict is the wind speed measurement error prediction value after threshold control; For real-time wind speed measurement, the category of strong wind process is judged by combining the thresholds including wind speed, duration and convection parameters, and the appropriate wind speed measurement correction model is automatically matched in the wind speed measurement classification correction model to correct the wind speed measurement results.

2. The method according to claim 1, characterized in that For real-time wind speed measurement, the wind speed, duration, and convection parameter thresholds are combined to determine the category of the strong wind process, and the appropriate wind speed measurement correction model is automatically matched in the wind speed measurement classification correction model to correct the wind speed measurement, including: Obtain real-time wind speed measurement and preprocess it into standardized real-time data for weather forecast; Determine the category of the high wind process based on thresholds including wind speed, duration and convection parameters; According to the category of strong wind process, the wind speed measurement is corrected by matching the appropriate wind speed measurement correction model from the wind speed measurement classification correction model.

3. A surface wind speed correction device considering the complex convective environment characteristics, characterized in that: include: Data acquisition module, which is used for the original historical data of target sites, including weather forecasts, weather observations, warning information, and disaster-causing events; A data preprocessing module is used to preprocess the original historical data of the target site obtained above to generate standardized historical data; The gale process classification module is used to determine the gale process category based on thresholds including wind speed, duration, and convection parameters, and based on standardized historical data including meteorological observations, warning information, and disaster-causing events, and to screen out common cases, cold air systemic gale cases, and convective gale cases; The wind speed estimation correction model training module is used to train and debug LSTM, XGboost, and Stacking models according to the standardized historical data including weather forecasts and meteorological observations, and according to different categories of strong wind processes, to generate a wind speed estimation classification correction model; among them, the generation of the wind speed estimation correction model A for ordinary cases is specifically as follows: The generation of wind speed calculation correction model B for cold air systematic gale cases is as follows: The generation of wind speed calculation correction model C for convective gale cases is as follows: Among them, Xa and Xb are feature tensors constructed from the standardized historical data of conventional meteorological element forecasts, standardized historical data of meteorological observations of wind speed, and standardized historical data of wind speed measurement errors for common cases and cold air systematic gale cases, respectively. Xc is a feature tensor constructed from the standardized historical data of conventional meteorological element forecasts, standardized historical data of convective parameter forecasts, standardized historical data of meteorological observations of wind speed, and standardized historical data of wind speed measurement errors for convective gale cases. Y is a label tensor constructed from the standardized historical data of wind speed measurement errors of the target forecast batch. T is the wind speed measurement error threshold, and Y is the label tensor constructed from the standardized historical data of wind speed measurement errors of the target forecast batch. predict is the wind speed measurement error prediction value, WS orig Standardized data for wind speed measurement, WS corr Y′ is the corrected value of wind speed measurement standard data, predict is the wind speed measurement error prediction value after threshold control; The real-time wind speed measurement and correction module is used to judge the category of the strong wind process based on the thresholds including wind speed, duration, and convection parameters, and select the appropriate wind speed correction model to correct the real-time measured wind speed accordingly.

Images