Regional meteorological element forecast error evaluation method based on multi-dimensional deviation

Abstract

The invention provides a regional meteorological element forecast error evaluation method based on multi-dimensional deviation, and belongs to the field of meteorological forecast. The problems of low meteorological element forecast error evaluation efficiency and poor correction function are solved; the method specifically comprises the following steps: acquiring original information and prediction information; analyzing the original information and the prediction information, and judging whether an error in the prediction information is in a reasonable range or not; if yes, not processing; if not, analyzing the change of the meteorological factors along with the time in the prediction information, and correcting the change; if correction cannot be carried out, historical information is acquired, the change trend of the temperature, humidity and wind power of the target area is analyzed, then a rainfall or snowfall segmentation probability model of the target area is established, and the target area is predicted again; according to the method, the actual meteorological information and the predicted meteorological information of the target region are correspondingly analyzed, judged and processed, so that the part with errors in the predicted information is corrected, and the accuracy of meteorological element forecast evaluation is improved.

Description

Technical Field

[0001] The present invention relates to a method for evaluating regional meteorological element forecast errors based on multi-dimensional deviation, which belongs to the field of meteorological forecasting. Background Art

[0002] The existing methods for evaluating meteorological element forecast errors have the following deficiencies: Difficult data processing: The existing evaluation methods need to process multi-dimensional data such as space, time, and intensity simultaneously, and the data volume is too redundant. At the same time, the joint analysis of multi-dimensional deviation requires complex algorithm support, such as spatial interpolation, time series analysis, statistical modeling, etc., and the technical threshold is relatively high. Moreover, the existing evaluation methods do not specifically screen and process the data during the calculation process, resulting in a large amount of calculation.

[0003] Strong data dependence: The accuracy of the existing error evaluation of meteorological element forecasts depends on high-quality observation data. However, in some regions (such as the ocean, plateau, and remote areas), the observation data is sparse or of low quality, which will affect the evaluation results. At the same time, in order to achieve precise analysis of multi-dimensional deviation, it is necessary to efficiently assimilate the observation data and model forecast data, which places higher requirements on the performance of the data assimilation system.

[0004] Method limitations: Most of the existing partial evaluation methods use fixed threshold evaluation. When defining the deviation threshold, this evaluation method has subjectivity, and the threshold selection for different regions or different meteorological elements will also affect the final evaluation results. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technology, the purpose of the present invention is to provide a method for evaluating regional meteorological element forecast errors based on multi-dimensional deviation, aiming to solve the problems of low efficiency and poor correction function in the evaluation of meteorological element forecast errors.

[0006] To achieve the above purpose, the present invention is implemented through the following technical solutions: The method for evaluating regional meteorological element forecast errors based on multi-dimensional deviation includes: Obtain the meteorological information and satellite cloud images of the target area in the recent month as the original information; Use the method to be tested to obtain the prediction information corresponding to the original information; analyze the original information and the prediction information to determine whether the error of the method to be tested is within a reasonable range. If so, do not process it. If not, analyze the change of meteorological factors in the prediction information over time, find out the meteorological factors with errors, and correct the method to be tested. If the method to be tested cannot be corrected, obtain the meteorological information of the corresponding month in the past three years of the target area as historical information; Extract the temperature and humidity data of the target area in the past three years based on historical information, analyze the changing trends of temperature and humidity in the target area, and reconstruct a linear model based on the changing trends as the prediction part of the temperature and humidity of the method to be tested; perform vector splitting on the wind speed in the target area according to historical information, and then predict the wind force change in the target area based on trigonometric function modeling; count the rainfall and snowfall frequencies in the target area, and establish a piecewise probability model for precipitation or snowfall in the target area with temperature, humidity, air pressure and other meteorological conditions as judgment conditions to predict the weather in the target area.

[0007] Further, the specific steps for determining whether the error of the method to be tested is within a reasonable range are as follows: Prediction data representation: After using the method to be tested for prediction in the target area, the predicted temperature, predicted humidity, predicted wind direction, predicted wind speed and predicted weather are obtained; Judge whether the difference in numerical information in the meteorological information is within ε, and judge whether the difference in text information in the meteorological information is within ε; ε represents the error determination coefficient; If both the difference in numerical information and the difference in text information between the meteorological information and the predicted data are within ε, it means that the error of the method to be tested is within a reasonable range, the method to be tested is credible, and the subsequent steps are skipped; If both the difference in numerical information and the difference in text information between the meteorological information and the predicted data are not within ε, it means that the error of the method to be tested is not within a reasonable range, and the method to be tested is not credible; Take the months in the past month as the target months, obtain the meteorological information in the target months in the past three years in the target area as historical information, and do not perform the subsequent correction steps; If the difference in numerical information between the meteorological information and the predicted data is within ε or the difference in text information is within ε, it means that the error of the method to be tested is partially reasonable, analyze the meteorological factors with errors in the predicted information, and correct the method to be tested.

[0008] Further, the specific steps for correcting the method to be tested are as follows: If the difference in numerical information between the meteorological information and the predicted data is within ε, compare the magnitudes of the predicted temperature, predicted humidity and predicted wind speed in the predicted information with the temperature, humidity and wind speed in the meteorological information, find out the meteorological factors with errors in the predicted information as deviation factors; analyze the change of the deviation factors over time and correct the method to be tested; If the difference in text information between the meteorological information and the predicted data is within ε, analyze the predicted wind speed and predicted weather in the predicted information.

[0009] Further, the specific steps for analyzing the change of the deviation factors over time and correcting the method to be tested are as follows: If the deviation factor is temperature, based on the temperature in the meteorological information, analyze the change of the predicted temperature in the prediction information with respect to temperature, and correct the temperature error in the method to be tested; If the deviation factor is humidity, repeat the same steps for correcting the temperature error in the method to be tested to correct the humidity error in the method to be tested; If the deviation factor is wind speed, perform vector decomposition on the predicted wind speed in the prediction information according to the wind direction in the meteorological information to obtain the lateral predicted wind speed hfv （1） ~hfv （mn） and the longitudinal predicted wind speed zfv （1） ~zfv （mn） ; Correct the predicted lateral wind speed and predicted longitudinal wind speed of the method to be tested.

[0010] Furthermore, the specific steps for correcting the temperature error in the method to be tested are as follows: Denote the predicted temperature in the prediction information for the past month as ft （1,1） ~ft （mn,24） , and denote the temperature in the original information as pt （1,1） ~pt （mn,24） ; According to ft （1,1） ~ft （mn,24） and pt （1,1） ~pt （mn,24） Calculate the Kalman gain Kk （1） ~Kk （mn） and the observation mapping coefficient H （1） ~H （mn） ; Calculate the average value aKk of Kk （1） ~Kk （mn） , and the average value aH of H （1） ~H （mn） ; According to ft （1,1） ~ft （mn,24） and pt （1,1） ~pt （mn,24） Calculate the state transition matrix A from 1 to 2 o'clock in the past month （1，1-2） ~A （mn，1-2） ; Similarly, the state transition matrix A from 23 to 24 o'clock （1，23-24） ~A （mn，23-24） ; Calculate the mean matrix of A （1，1-2） ~A （mn，1-2） as the change matrix aA of temperature from 1 to 2 o'clock （1-2） Similarly, calculate A （1，23-24） ~A （mn，23-24）The mean matrix of the temperature change matrix aA from 23:00 to 24:00 （23-24） ; Assume the actual temperature of the target area at time ti is t （ti） , assuming that the estimated temperatures at ti and (ti+1) of the method to be tested are ft （ti） and ft （ti+1） ; Assume that the change matrix from time ti to (ti+1) is aA （ti-（ti+1）） , the estimated temperature after correction (ti+1) is amt （ti+1） ; Construct relation A1: ; The predicted temperature in the method to be tested is corrected according to the relation A1.

[0011] Furthermore, the Kk （1） and H （1） And the state transfer matrix A （1，1-2） ~A （1，23-24） The specific calculation steps are as follows: Construct the matrix Mf （1-2） : ; Construct matrix Mp （1-2） : ; Calculate the state transfer matrix A （1，1-2） : ; Calculate the matrix Mf （1-2） Transformed into matrix Mp （1-2） The process noise matrix wM （1-2） : ; Calculate the observation mapping coefficient H （1，1-2） : ; Calculate the covariance RM of the observation noise （1-2） : ; Calculate the covariance QM of the process noise （1-2） : ; Calculate the prediction error covariance Pk （1-2） : ; Calculate Kk （1，1-2） : .

[0012] Furthermore, the specific steps for correcting the predicted crosswind speed of the method to be measured are as follows: Calculate hv （1） ~hv （mn） to obtain the average value ahv; Calculate the white noise et （1） ~et （mn） for the 1st to the mnth day in the past month; Calculate the first-order difference of hfv （1） ~hfv （mn） to obtain ▽hfv （2） ~▽hfv （mn） ; Calculate the first-order difference of hv （1） ~hv （mn） to obtain ▽hv （2） ~▽hv （mn） ; Let the date be da, and the first-order difference of the crosswind speed on date da be ▽hv （da） ; The first-order difference of the predicted crosswind speed on date (da + 1) is ▽hfv （da+1） , and the first-order difference of the crosswind speed is ▽hv （da+1） ; The predicted crosswind speed hfv on date da （da） , and the crosswind speed hv （da） ; The white noise on the da-th day is et （da） , and the white noise on the (da + 1)-th day is et （da+1） ; Construct the relationship formula A2-1: ; φ represents the autoregressive coefficient, and θ represents the moving average coefficient; Based on the relationship formula A2-1, use the maximum likelihood estimation algorithm to calculate φ and θ; Construct the relationship formula A2-2: ; α and β represent the relationship coefficients of the predicted crosswind speed with respect to the crosswind speed; ηt （da+1） represents the residual term on the (da + 1)-th day: ; Based on the relationship formula A2-2, use the maximum likelihood estimation algorithm to calculate α and β; According to the relationship formula A2-1 and the relationship formula A2-2, correct the crosswind prediction speed in the method to be measured.

[0013] Furthermore, the specific steps for the predicted wind speed and predicted weather in the analysis and prediction information are as follows: If the predicted wind speed in the prediction information does not match the wind speed in the meteorological information, then record the lateral predicted wind speed as uv and the longitudinal predicted wind speed as vv; Calculate the combined wind speed wv: ; Calculate the wind direction angle jv: ; If jv is between 0 o ~90 o , then the wind direction is southwest; if jv is between 90 o ~180 o , then the wind direction is southeast; if jv is between 180 o ~270 o , then the wind direction is northeast; if jv is between 270 o ~360 o , then the wind direction is northwest; If the predicted weather in the prediction information does not match the weather in the meteorological information, then connect the multi-spectral fusion algorithm to the method to be measured, obtain the data of the infrared band, visible band, and short-wave infrared band of the corresponding cloud layer in the target area through the satellite, and use the GPM algorithm to correct the weather in the target area.

[0014] Compared with the prior art, the beneficial effects of the present invention are: Multi-dimensional analysis: In the time dimension, the present invention analyzes the errors of the forecast elements in the time series, such as the change trends of humidity and temperature; at the same time, it quantifies the numerical deviations of the meteorological elements, such as the differences between the predicted values and the actual values of temperature, humidity, and wind speed, etc.; through the joint analysis of multi-dimensional deviations, it can more comprehensively reflect the sources and characteristics of the forecast errors and avoid the limitations of single indicators.

[0015] Forecast optimization: After completing the task of evaluating the meteorological element forecast errors of the method to be measured, the present invention provides a detailed optimization method for the method to be measured; by analyzing the changes of various meteorological factors in the target area over time and using a variety of statistical regression methods, it analyzes the meteorological changes in the target area and significantly improves the prediction accuracy of the method to be measured.

[0016] Wide adaptability: All the mathematical models in the present invention are based on the existing meteorological data analysis and construction in the target area. This processing method can make the present invention have strong adaptability in dealing with complex terrain areas (such as mountainous areas and coastal areas); through multi-dimensional deviation analysis, it can reveal the systematic errors of the forecast in these areas and eliminate the forecast errors of the method to be measured; based on the error evaluation results and optimization schemes, it improves the forecast method and enhances the accuracy of regional forecasts. Description of the Drawings

[0017] Other features, objects, and advantages of the present invention will become more apparent by reading the following detailed description of non - restrictive embodiments with reference to the accompanying drawings: Figure 1 Schematic diagram of the method of the present invention; Figure 2 Schematic diagram of the data of the present invention; Figure 3 Schematic diagram of the wind direction of the present invention. Detailed implementation manners

[0018] To make the above - mentioned objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific implementation manners.

[0019] Please refer to Figure 1 and Figure 2 , the method for evaluating the forecast error of regional meteorological elements based on multi - dimensional deviation includes: Step S1: Obtain the meteorological information and satellite cloud images of the target area in the recent month as the original information; The meteorological information includes: daily temperature, (air) humidity, wind direction, wind speed, and weather; Step S2: Use the method to be tested to obtain the prediction information corresponding to the original information; analyze the original information and the prediction information to determine whether the error of the method to be tested is within a reasonable range; if so, do not process; if not, analyze the change of meteorological factors in the prediction information over time, find out the meteorological factors with errors, and correct the method to be tested; if the method to be tested cannot be corrected, obtain the meteorological information of the corresponding month in the past three years in the target area as historical information; It should be noted that in the present invention, the "method to be tested" means: a method provided by a user or relevant technical personnel and evaluated for error by the present invention (the method for evaluating the forecast error of regional meteorological elements based on multi - dimensional deviation); The specific steps of Step S2 are as follows: Step S21: The prediction data refers to: within the recent month, after the relevant department in the target area uses the method to be tested for prediction, the daily predicted temperature, predicted humidity, predicted wind direction, predicted wind speed, and predicted weather obtained; Judge whether the (numerical) difference in numerical information between the temperature, humidity, and wind speed in the meteorological information and the predicted temperature, predicted humidity, and predicted wind speed in the prediction data is within ε, and judge whether the difference in text information between the wind direction and weather in the meteorological information and the predicted wind direction and predicted weather in the prediction data is within ε (the difference in text information means: the non - matching rate of text information); Among them, ε represents the error determination coefficient, and the value of ε is 20%; the user or relevant technical personnel can adjust the value of ε according to actual needs; If the differences in numerical information and literal information between the meteorological information and the predicted data are both within ε, it indicates that the error of the method to be tested is within a reasonable range, the method to be tested is credible, and the subsequent steps S22 to S23 and the subsequent step S3 are skipped. If the differences in numerical information and literal information between the meteorological information and the predicted data are not both within ε, it indicates that the error of the method to be tested is not within a reasonable range, the method to be tested is not credible, and the subsequent steps S22 to S23 are skipped. Take the month nearly one month as the target month, obtain the meteorological information of the target area in the target month in the past three years as historical information, and enter the subsequent step S3. The above "correction step" refers to steps S22 to 23. If the difference in numerical information between the meteorological information and the predicted data is within ε or the difference in literal information is within ε, it indicates that part of the error of the method to be tested is reasonable, the method to be tested is partially credible, skip the subsequent step S3, analyze the meteorological factors with errors in the predicted information, and enter steps S22 to S23. Step S22: If the difference in numerical information between the meteorological information and the predicted data is within ε, compare the predicted temperature, predicted humidity, and predicted wind speed in the predicted information with the temperature, humidity, and wind speed in the meteorological information, find out the meteorological factors with errors in the predicted information as deviation factors; analyze the change of the deviation factors over time and correct the method to be tested. Step S221: If the deviation factor is temperature (that is, the numerical gap between the predicted temperature in the predicted information and the temperature in the meteorological information exceeds ε), take the temperature in the meteorological information as the benchmark, analyze the change of the predicted temperature in the predicted information with temperature, and correct the temperature error in the method to be tested. Step S2211: Denote the number of days in the nearly one month as mn; denote the predicted temperature in the predicted information for the nearly one month as ft （1,1） ~ft （mn,24） , denote the temperature in the original information as pt （1,1） ~pt （mn,24） ; According to ft （1,1） ~ft （mn,24） and pt （1,1） ~pt （mn,24） Calculate the Kalman gain Kk （1） ~Kk （mn） and the observation mapping coefficient H （1） ~H （mn） ; Calculate the average value aKk of Kk （1） ~Kk （mn） , and the average value aH of H （1） ~H （mn） ; According to ft （1,1） ~ft （mn,24） and pt （1,1） ~pt （mn,24） Calculate the state transition matrix A from 1:00 to 2:00 in the past month （1，1-2） ~A （mn，1-2） ; The state transition matrix A from 2:00 to 3:00 （1，2-3） ~A （mn，2-3） ; And so on, the state transition matrix A from 23:00 to 24:00 （1，23-24） ~A （mn，23-24） ; Calculate the mean matrix of A （1，1-2） ~A （mn，1-2） as the change matrix aA of the temperature from 1:00 to 2:00 （1-2） : ; Calculate the mean matrix of A （1，2-3） ~A （mn，2-3） as the change matrix aA of the temperature from 2:00 to 3:00 （2-3） : ; And so on, calculate the mean matrix of A （1，23-24） ~A （mn，23-24） as the change matrix aA of the temperature from 23:00 to 24:00 （23-24） : ; Let the actual temperature at time ti in the target area be t （ti） , and let the predicted temperatures at time ti and (ti + 1) predicted by the method to be tested be ft （ti） and ft （ti+1） ; Let the change matrix from time ti to (ti + 1) be aA （ti-（ti+1）） , and the predicted temperature at (ti + 1) after correction be amt （ti+1） ; where the value range of ti is: 1 - 23, aA （ti-（ti+1）） ∈{aA （1-2） ~aA （23-24）}; Construct the relationship A1: ; Correct the method to be tested according to the relationship A1; (Among ft （1,1） ~ft （mn,24） and pt （1,1） ~pt （mn,24） ) Extract the predicted temperature ft （1,1） of the first day~ft （1,24） and the temperature is denoted as pt （1,1） ~pt （1,24） , calculate Kk （1） and H （1） and the state transition matrix A （1，1-2） ~A （1，23-24） ; Step S2212: Calculate the Kalman gain Kk from 1:00 to 2:00 on the first day （1，1-2） , the observation mapping coefficient H （1，1-2） ; and the state transition matrix Construct matrix Mf （1-2） : ; Construct matrix Mp （1-2） : ; where ft （1,2） represents the predicted temperature at 2:00 on the first day, and pt （1,2） represents the temperature at 2:00 on the first day; Calculate the state transition matrix A （1，1-2） (matrix A （1，1-2） is a (2×2) matrix): ; where T represents the transpose of the matrix, * represents matrix multiplication, and -1 represents the inverse of the matrix; Calculate matrix Mf （1-2） to become the process noise matrix wM （1-2） of matrix Mp （1-2） (matrix wM （1-2） is a (1×2) matrix): ; Calculate the observation mapping coefficient H （1，1-2） : ; where - represents matrix subtraction; Calculate the covariance RM of the observation noise （1-2） : ; Calculate the covariance QM of the process noise （1-2） : ; Calculate the prediction error covariance Pk （1-2） : ; Calculate Kk （1，1-2） : ; Step S2213: Repeat Kk （1，1-2） and H（1，1-2） Calculation steps to calculate the Kalman gain Kk from 2:00 to 3:00, 3:00 to 4:00, until 23:00 to 24:00 on the first day （1，2-3） 、Kk （1，3-4） ~Kk （1，23-24） ,observation mapping coefficient H （1，2-3） 、H （1，3-4） ~H （1，23-24） and state transition matrix A （1，2-3） 、A （1，3-4） ~A （1，23-24） ; Calculate Kk （1，1-2） ~Kk （1，23-24） and take the average value as Kk （1） ,H （1，1-2） ~H （1，23-24） and take the average value as H （1） ; Step S2214: Repeat the calculation steps of Kk （1） and H （1） and A （1，1-2） ~A （1，23-24） to calculate the Kalman gain Kk corresponding to the 2nd to the mnth day （2） ~Kk （mn） and observation mapping coefficient H （2） ~H （mn） ; Calculate the state transition matrix A from 1:00 to 2:00, 2:00 to 3:00, until 23:00 to 24:00 corresponding to the 2nd to the mnth day （2，1-2） 、A （2，2-3） ~A （mn，23-24） ; Step S222: If the deviation factor is humidity (i.e., the numerical difference between the predicted humidity in the prediction information and the humidity in the meteorological information exceeds ε), then repeat the same steps to correct the temperature error in the method to be measured, and correct the humidity error in the method to be measured; Step S223: Refer to Figure 3 . If the deviation factor is wind speed (i.e., the numerical difference between the predicted wind speed in the prediction information and the wind speed in the meteorological information exceeds ε), decompose the predicted wind speed in the prediction information vectorially with the wind direction in the meteorological information to obtain the transverse predicted wind speed hfv from the 1st, 2nd, until the mnth day of the last month （1） 、hfv （2） ~hfv （mn） and longitudinal predicted wind speed zfv （1） 、zfv （2） ~zfv （mn） ; Decompose the wind speed in the meteorological information vectorially with the wind direction in the meteorological information to obtain the transverse wind speed hv from the 1st, 2nd, until the mnth day of the last month （1）, hv （2） ~hv （mn） and the longitudinal wind speed zv （1） , zv （2） ~zv （mn） ; According to hfv （1） ~hfv （mn） and hv （1） ~hv （mn） correct the predicted lateral wind speed of the method to be measured; Using hv （1） ~hv （mn） as the reference data, calculate the average value ahv of hv （1） ~hv （mn） ; Calculate the white noise et （1） on the first day of the last month (i.e., hv （1） ): et （1） = hv （1） −ahv; The white noise et （2） on the second day of the last month (i.e., hv （2） ): et （2） = hv （2） −ahv; And so on, the white noise et （mn） on the mn-th day of the last month (i.e., hv （mn） ): et （mn） = hv （mn） −ahv; Calculate the first-order difference corresponding to hfv （1） ~hfv （mn） to obtain ▽hfv （2） , ▽hfv （3） ~▽hfv （mn） ; Among them, the calculation formula of ▽hfv （2） is: ▽hfv （2） = hfv （2） −hfv （1） ; The calculation formula of ▽hfv （3） is: ▽hfv （3） = hfv （3） −hfv （2） ; And so on, the calculation formula of ▽hfv （mn） is: ▽hfv （mn） = hfv （mn） −hfv （mn-1） ; Among them, hfv （3）Indicates the lateral predicted wind speed on the 3rd day of the last month in the prediction information; similarly, hfv （mn-1） Indicates the lateral predicted wind speed on the (mn - 1)th day of the last month in the prediction information; Repeat the calculation steps of ▽hfv （2） ~▽hfv （mn） to calculate hv （1） ~hv （mn） to obtain the first-order difference of hv, getting ▽hv （2） ~▽hv （mn） ; Let the date be da, and the first-order difference of the lateral wind speed on date da is ▽hv （da） ; The first-order difference of the predicted lateral wind speed on date (da + 1) is ▽hfv （da+1） , and the first-order difference of the lateral wind speed is ▽hv （da+1） ; The predicted lateral wind speed hfv on date da （da） , and the lateral wind speed hv （da） ; The white noise on the da-th day is et （da） , and the white noise on the (da + 1)-th day is et （da+1） ; where the value range of da is: 2~(mn - 1); Construct relationship formula A2-1: ; where φ represents the autoregressive coefficient and θ represents the moving average coefficient; Based on relationship formula A2-1, use the maximum likelihood estimation algorithm in the statsmodels library to calculate φ and θ; Construct relationship formula A2-2: ; where α and β represent the relationship coefficients of "the predicted lateral wind speed in the prediction information" with respect to "the lateral wind speed in the meteorological information"; ηt （da+1） represents the residual term on the (da + 1)-th day, and the calculation formula of ηt （da+1） is: ; Based on relationship formula A2-2, use the maximum likelihood estimation algorithm in the statsmodels library to calculate α and β; According to relationship formula A2-1 and relationship formula A2-2, correct the lateral predicted wind speed in the method to be measured; Repeat the same steps of correcting the lateral predicted wind speed in the method to be measured. According to zfv （1） ~zfv （mn） and zv （1）~zv （mn） Modify the predicted longitudinal wind speed of the method to be measured; Step S23: If the difference between the meteorological information and the text information in the prediction data is within ε, then analyze the predicted wind speed and predicted weather in the prediction information; Step S231: If the predicted wind speed in the prediction information does not match the wind speed in the meteorological information, then Determine whether there is a numerical difference between the wind speed in the prediction information and the wind speed in the meteorological information; If there is no numerical difference, record the lateral predicted wind speed as uv (positive to the west, negative to the east), and record the longitudinal predicted wind speed as vv (positive to the south, negative to the north); Calculate the combined wind speed wv: ; Calculate the wind direction angle jv: ; If jv is between 0 o ~90 o , the wind direction is southwest; if jv is between 90 o ~180 o , the wind direction is southeast; if jv is between 180 o ~270 o , the wind direction is northeast; if jv is between 270 o ~360 o , the wind direction is northwest; Step S232: If there is a numerical difference, then based on the lateral predicted wind speed and longitudinal predicted wind speed corrected in step S223, repeat the processing steps of step S231 to correct the predicted wind direction in the prediction information; Step S233: If the predicted weather in the prediction information does not match the weather in the meteorological information, then connect the multi-spectral fusion algorithm to the method to be measured, obtain the data of the infrared band, visible band, and short-wave infrared band of the corresponding cloud layer in the target area through the satellite, and use the GPM or DPR algorithm to correct the weather in the target area; Step S3: Extract the temperature and humidity data of the target area in the past three years (in the target month) according to the historical information, and analyze the change trends of the temperature and humidity in the target area (within the target month). Reconstruct a linear model based on the change trends (using polynomial regression or time series) as the prediction part of the temperature and humidity of the method to be measured; Perform vector splitting on the wind speed of the target area according to the historical information, and then predict the wind force change in the target area based on trigonometric function modeling; The frequency of rainfall and snowfall in the target area (in the target month of the past three years) is counted, and a segmented probability model of precipitation or snowfall in the target area is established based on temperature, humidity, air pressure and other meteorological conditions (such as air convection, air lift, vertical temperature distribution of the atmosphere, etc.) as judgment conditions to predict the weather in the target area; It should be noted that the processing method of "Step S3" in the present invention is a complete reconstruction of the above-mentioned "method to be tested provided by the user or relevant technical personnel". It is to completely reconstruct the "method to be tested" under the premise that the "method to be tested" is completely unreliable and unavailable. Therefore, "Step S3" in the present invention only provides a reconstruction idea of ​​the "method to be tested", and the specific reconstruction method is determined by the user or relevant technical personnel (such as the NWP digital weather forecast system).

[0020] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technical personnel in this field according to actual conditions. For example, if there are weight coefficients and proportional coefficients, their set sizes are to quantify each parameter to obtain a specific value, which is convenient for subsequent comparison. Regarding the size of the weight coefficient and the proportional coefficient, it is sufficient as long as it does not affect the proportional relationship between the parameter and the quantized value.

[0021] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A regional meteorological element forecast error assessment method based on multidimensional deviation, characterized in that: The method comprises: Obtain the target area's meteorological information and satellite cloud images for the past month as raw information; Obtain the forecast information corresponding to the original information; analyze the original information and the forecast information to determine whether the error in the forecast information is within a reasonable range; if so, do not process it; if not, analyze the change of meteorological factors in the forecast information over time, find out the meteorological factors with errors, and correct them; if correction is not possible, obtain the meteorological information of the corresponding months in the target area in the past three years as historical information; Based on historical information, the temperature and humidity data of the target area in the past three years are extracted, and the changing trends of the temperature and humidity in the target area are analyzed. Based on the changing trends, the linear model is reconstructed to predict the temperature and humidity in the target area; the wind speed in the target area is vector-splitting based on historical information, and then the wind speed changes in the target area are predicted based on trigonometric function modeling; the frequency of rainfall and snowfall in the target area is counted, and the temperature, humidity, air pressure and other meteorological conditions are used as judgment conditions to establish a segmented probability model of precipitation or snowfall in the target area to predict the weather in the target area.

2. The regional meteorological element forecast error assessment method based on multidimensional deviation according to claim 1 is characterized in that: The specific steps of determining whether the error of the method to be tested is within a reasonable range are as follows: The forecast data represents: forecast temperature, forecast humidity, forecast wind direction and forecast wind speed, and forecast weather; Determine whether the difference of numerical information in meteorological information is within ε, and determine whether the difference of text information in meteorological information is within ε; ε represents the error determination coefficient; If the numerical information difference and text information difference between the meteorological information and the forecast data are within ε, it means that the error in the forecast information is within a reasonable range, and the subsequent steps are skipped; If the difference between the numerical information and the text information in the meteorological information and the forecast data is not within ε, it means that the forecast information is unreliable; The month of the previous month is taken as the target month, and the meteorological information of the target area in the target month in the past three years is obtained as historical information, and the subsequent correction steps are not performed; If the difference between the numerical information in the meteorological information and the forecast data is within ε or the difference between the text information is within ε, it means that the error part in the forecast information is reasonable. Analyze the meteorological factors that cause the error in the forecast information and correct the forecast information.

3. The regional meteorological element forecast error assessment method based on multidimensional deviation according to claim 2 is characterized in that: The specific steps of correcting the prediction information are as follows: If the difference between the meteorological information and the numerical information in the forecast data is within ε, then compare the predicted temperature, predicted humidity and predicted wind speed in the forecast information with the temperature, humidity and wind speed in the meteorological information, find out the meteorological factor with error in the forecast information, and use it as the deviation factor; analyze the change of the deviation factor over time and correct the forecast information; If the difference between the meteorological information and the text information in the forecast data is within ε, the forecast wind speed and forecast weather in the forecast information are analyzed.

4. The method for evaluating regional meteorological element forecast errors based on multidimensional deviation according to claim 3, characterized in that: The specific steps of analyzing the change of the deviation factor over time and correcting the prediction information are as follows: If the deviation factor is temperature, the temperature in the meteorological information is used as a reference, the predicted temperature in the forecast information is analyzed as it changes with temperature, and the temperature error in the forecast information is corrected; If the bias factor is humidity, repeat the same steps of correcting the temperature error in the forecast information to correct the humidity error in the forecast information; If the deviation factor is wind speed, the predicted wind speed in the forecast information is vector-decomposed by the wind direction in the meteorological information to obtain the horizontal predicted wind speed hfv from the 1st to the mnth day in the past month （1） ~hfv （mn） and the longitudinal predicted wind speed zfv （1） ~zfv （mn） ; Correct the predicted lateral wind speed and predicted longitudinal wind speed of the forecast information.

5. The method for evaluating regional meteorological element forecast errors based on multidimensional deviation according to claim 4, characterized in that: The specific steps of correcting the temperature error in the prediction information are as follows: The predicted temperature for the past month in the forecast information is recorded as ft （1,1） ~ft （mn,24） , the temperature in the original information is recorded as pt （1,1） ~pt （mn,24） ; According to ft （1,1） ~ft （mn,24） and pt （1,1） ~pt （mn,24） Calculate the Kalman gain Kk （1） ~Kk （mn） and the observation mapping coefficient H （1） ~H （mn） ; Calculate Kk （1） ~Kk （mn） The average value of aKk,H （1） ~H （mn） The average value aH; According to ft （1,1） ~ft （mn,24） and pt （1,1） ~pt （mn,24） Calculate the state transition matrix A from 1:00 to 2:00 in the past month （1，1-2） ~A （mn，1-2） ; Similarly, the state transition matrix A from 23:00 to 24:00 （1，23-24） ~A （mn，23-24） ; Calculate A （1，1-2） ~A （mn，1-2） The mean matrix of the temperature change matrix aA from 1 to 2 （1-2） Similarly, calculate A （1，23-24） ~A （mn，23-24） The mean matrix of the temperature change matrix aA from 23:00 to 24:00 （23-24） ; Assume the actual temperature of the target area at time ti is t （ti） , assuming that the estimated temperatures at time ti and (ti+1) are ft （ti） and ft （ti+1） ; Assume that the change matrix from time ti to (ti+1) is aA （ti-（ti+1）） , the estimated temperature after correction (ti+1) is amt （ti+1） ; Construct relation A1: ； The predicted temperature in the prediction information is corrected according to the relational expression A1.

6. The method for evaluating regional meteorological element forecast errors based on multidimensional deviation according to claim 5, characterized in that: Kk （1） and H （1） And the state transfer matrix A （1，1-2） ~A （1，23-24） The specific calculation steps are as follows: Construct the matrix Mf （1-2） : ; Construct matrix Mp （1-2） : ; Calculate the state transfer matrix A （1，1-2） : ； Calculate the matrix Mf （1-2） Transformed into matrix Mp （1-2） The process noise matrix wM （1-2） : ； Calculate the observation mapping coefficient H （1，1-2） : ； Calculate the covariance RM of the observation noise （1-2） : ； Calculate the covariance QM of the process noise （1-2） : ； Calculate the prediction error covariance Pk （1-2） : ； Calculate Kk （1，1-2） : 。 7. The method for evaluating regional meteorological element forecast errors based on multidimensional deviation according to claim 4, characterized in that: The specific steps of correcting the predicted lateral wind speed in the prediction information are as follows: Calculate hv （1） ~hv （mn） The average value of ahv; Calculate the white noise et from the 1st to the mnth day in the past month （1） ~et （mn） ; Calculate HFV （1） ~hfv （mn） The first-order difference of ▽hfv （2） ~▽hfv （mn） ; Calculate hv （1） ~hv （mn） The first-order difference of ▽hv （2） ~▽hv （mn） ; Let date be da, the first-order difference of the lateral wind speed on date da is ▽hv （da） ; The first-order difference of the predicted lateral wind speed on date (da+1) is ▽hfv （da+1） , the first-order difference of the lateral wind speed is ▽hv （da+1） ; Forecasted lateral wind speed hfv for date da （da） , transverse wind speed hv （da） ; The white noise of the first day is et （da） , the white noise of the (da+1)th day is et （da+1） ; Construct relation A2-1: ； φ represents the autoregressive coefficient, θ represents the moving average coefficient; Based on equation A2-1, use the maximum likelihood estimation algorithm to calculate φ and θ; Construct relation A2-2: ； α and β represent the relationship coefficients of the predicted lateral wind speed with respect to the lateral wind speed; ηt （da+1） Represents the residual term on the (da+1)th day: ； Based on equation A2-2, use the maximum likelihood estimation algorithm to calculate α and β; The lateral predicted wind speed in the prediction information is corrected according to the relationship A2-1 and the relationship A2-2.

8. The method for evaluating regional meteorological element forecast errors based on multidimensional deviation according to claim 3, characterized in that: The specific steps of analyzing and predicting wind speed and weather are as follows: If the predicted wind speed in the forecast information does not match the wind speed in the meteorological information, the horizontal predicted wind speed is recorded as uv and the vertical predicted wind speed is recorded as vv; Calculate the total wind speed wv: ; Calculate the wind direction angle jv: ; If jv is 0 o ~90 o If JV is between 90 and 100 degrees, the wind direction is southwest. o ~180 o If JV is between 180 and 200 degrees, the wind direction is southeast. o ~270 o If JV is between 270 and 310 degrees, the wind direction is northeast. o ~360 o Between, the wind direction is northwest; If the predicted weather in the forecast information does not match the weather in the meteorological information, the multi-spectral fusion algorithm is used to obtain the infrared band, visible band and short-wave infrared band data of the corresponding cloud layer in the target area through satellite, and the GPM algorithm is used to correct the weather in the target area.

Images