Clear sky mixed layer height estimation method based on meteorological UAV and lidar

Abstract

The present invention is a clear sky mixed layer height estimation method based on a meteorological UAV and a laser radar, which belongs to the field of radio, and includes: collecting and processing data; calculating the spatial distance difference based on the processed data, and filtering the data through time-space matching to form meteorological profile data after time-space matching; constructing a correction fusion model to calculate the RI index of each altitude layer; obtaining the average variance profile data of a vertical wind field within the analysis time window according to the collected data combined with the DBS mode vertical wind field solution model, and constructing a profile optimization threshold; generating the average variance optimized profile data of the vertical wind field; combining the constructed deep random forest model to obtain the RI index data of each altitude layer, and obtaining the estimated value of the mixed layer height according to the mixed layer height identification condition. The present invention realizes the accurate correction of meteorological data detected by meteorological UAVs and the accurate fitting and segmented optimization of multi-peak profiles.

Description

Technical Field

[0001] The present invention relates to the field of radar detection, and in particular to a clear-air mixed layer height estimation method based on a meteorological unmanned aerial vehicle and a laser radar. Background Art

[0002] The detection of the clear-air mixing layer height is of great significance in meteorology and atmospheric science. Its detection can help understand the vertical distribution of wind speed, temperature, and humidity in the boundary layer, and then reveal its dynamic change process; through real-time monitoring of the mixing layer, it can help the meteorological department accurately predict the occurrence of storms and local thunderstorms; in addition, the height of the mixing layer and the air flow in the mixing layer directly affect the diffusion of air pollutants, and its detection can help us predict and prevent the occurrence of air pollution events, especially in large cities or heavily polluted areas. In short, the detection of the clear-air mixing layer is of great significance for in-depth understanding of the structure of the atmosphere, optimizing weather forecasts, improving air quality monitoring, and supporting wind energy and agricultural management.

[0003] At present, there are several conventional methods for obtaining the height of the clear sky mixing layer: 1. Radiosonde, which can accurately obtain the real-time change profile of the atmospheric parameters in the mixing layer, and the height of the mixing layer estimated based on these meteorological profiles has the highest accuracy. However, the time resolution and spatial resolution of this detection method are poor, and the cost of each detection is not low, so it is not suitable for high temporal and spatial resolution detection of the mixing layer; 2. Inverting the height of the mixing layer based on the data detected by lidar. This detection method has high temporal and spatial resolution, but the reliability of its inversion results is seriously dependent on the accuracy of the lidar detection data, and the results brought by different inversion methods are often quite different; 3. Observation instruments installed on meteorological towers. This detection method can detect the changing characteristics of meteorological parameters in the near-ground layer in real time and with high spatial resolution, and can accurately estimate the height of the mixing layer when it is low. However Due to the limitation of the height of the tower itself, this observation method is not feasible for the mixed layer height reaching or exceeding several hundred meters, and this observation method can only be limited to the place where the tower is deployed, and cannot achieve detection in a wider range; 4. Meteorological UAV detection. This detection method can achieve high temporal and spatial resolution detection of meteorological parameters within a few kilometers above the ground. However, the accuracy of the detection data of the meteorological observation instruments carried on the detection platform is often affected by the operation of the UAV rotor, especially the accuracy of wind speed and wind direction is most affected. Therefore, the accuracy of the mixed layer height inverted by the detection data obtained by this detection method needs to be corrected. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a clear sky mixed layer height estimation method based on meteorological drones and lidar, thereby solving the deficiencies of the prior art.

[0005] The object of the present invention is achieved by the following technical solution: a clear sky mixed layer height estimation method based on meteorological drone and laser radar, the method comprising:

[0006] S1. Collect profile data detected by meteorological drones, meteorological data and observation parameters detected by ground-based lidar, and profile data detected by radiosonde under clear sky conditions, and process the collected data;

[0007] S2, based on the data processed in step S1, the meteorological data and longitude and latitude detected by the radiosonde are used as the spatial information of the target, the spatial distance difference between the spatial position of the data point detected by the meteorological UAV and the target point is calculated, and all the data detected by the meteorological UAV are screened by the time and space matching method to form the meteorological profile data detected by the meteorological UAV and the radiosonde after time and space matching;

[0008] S3, constructing a correction fusion model of all meteorological profile data detected by the meteorological UAV according to the meteorological profile data generated in step S2, and obtaining all corrected meteorological profile data according to the correction fusion model, and inputting them into the RI index classical calculation model to calculate the RI (z) index of each altitude layer z, recorded as Data1;

[0009] S4, according to the data collected in step S1, combined with the vertical wind field solution model of the DBS mode, obtain the average variance profile data of a vertical wind field within the analysis time window, and construct the profile optimization threshold through the obtained average variance profile data of the vertical wind field;

[0010] S5, generating average variance optimized profile data of the vertical wind field according to the average variance profile data of the vertical wind field obtained in step S4 and the profile optimization threshold, recorded as Data2;

[0011] S6. Based on the obtained Data1 and Data2, the time-space matching method is used to match Data1 and Data2 in time and space, and the RI(z) index data of each altitude layer z of the meteorological UAV with high spatial resolution is obtained by combining the constructed deep random forest model, and the estimated value of the mixed layer height is obtained according to the mixed layer height identification condition.

[0012] The profile data detected by the meteorological drone includes atmospheric temperature, humidity, pressure, horizontal wind speed, horizontal wind direction, altitude, and longitude and latitude data with high temporal and spatial resolution;

[0013] The meteorological data detected by the ground-based laser radar include echo signal-to-noise ratio, radial velocity and spectrum width data of different altitude layers in DBS scanning mode, where DBS represents wind profile beam swing;

[0014] The observation parameters of the ground-based laser radar include laser wavelength, antenna elevation, azimuth and interval azimuth during beam scanning, the time period required for the radar to complete a DBS scanning mode, radial range library length, noise level, observation time, instrument height, and longitude and latitude data;

[0015] The profile data detected by the radiosonde include atmospheric temperature, humidity, pressure, horizontal wind speed, horizontal wind direction, altitude and longitude and latitude data measured by an air balloon.

[0016] The processing of the collected data in S1 includes:

[0017] A1. Calculate the average value of the atmospheric temperature profile data T(z) at different altitudes z detected by the meteorological drone. and standard deviation ;

[0018] A2. Anomalies with a score greater than the set threshold All of them are removed, and the data retained after removing the outliers is recorded as ;

[0019] A3. Confirm The position of the null value in the window is used as the center of the window, and the window size is set to width1. Based on the non-null values ​​in the window, the estimated value at the null value position is generated using the spline interpolation method, and the null value is filled with the estimated value. The data after filling the null value is recorded as ;

[0020] A4. A data point in the window is the center of the window, the window size is set to width2, the average value of the data in the window is calculated, and the average value replaces the window center value. The final processed data is recorded as ;

[0021] A5. Repeat steps A1-A4 to process other meteorological profile data detected by meteorological UAV and radiosonde respectively.

[0022] The S2 specifically includes the following contents:

[0023] B1. Based on the obtained altitude and longitude and latitude data detected by the meteorological UAV and radiosonde, as well as the processed meteorological data detected by the meteorological UAV and radiosonde, the altitude and longitude and latitude of the meteorological data detected by the radiosonde are used as the spatial information of the target point, and the spatial distance difference between the spatial position of the meteorological UAV detection data point and the target point is calculated using the Euclidean distance formula. ;

[0024] B2. Less than the spatial distance difference threshold As the screening condition, the spatial points adjacent to the target point are screened out, and then the difference between the observation time of the screened adjacent spatial points and the observation time of the target point is calculated, and the adjacent spatial point with the smallest time difference is taken as the matching point of the meteorological UAV detection data to the radiosonde target point;

[0025] B3. All data detected by the meteorological UAV are screened according to step B2, and finally the meteorological profile data detected by the meteorological UAV and radiosonde are formed after time and space matching.

[0026] The correction fusion model of all meteorological profile data detected by the meteorological UAV is constructed according to the meteorological profile data generated in step S2, including the following contents:

[0027] C1, based on the meteorological profile data detected by the meteorological UAV and radiosonde after the generated time-space matching, the temperature data detected by the meteorological UAV is used as the input data set, and the temperature data detected by the radiosonde is used as the output data set, which are input into the least squares model MOD1, the elastic network model MOD2 and the support vector machine model MOD3 respectively;

[0028] C2, using the gradient descent iteration method to optimize the performance of MOD1 to minimize the sum of squared residuals between the least squares model output and the true output;

[0029] C3. For the regularization parameters α and λ in MOD2, traverse different parameters to generate multiple combinations of α and λ, and preset α and λ for each combination in the elastic network model, run the model to obtain the output results, calculate the root mean square error RMSE between the output results of the elastic network model and the actual output data set, and select the best regularization parameters α and λ with the minimum RMSE as the screening criterion;

[0030] C4. Select the radial basis kernel function in MOD3, randomly select the combination of regularization parameter C and kernel function parameter G, and calculate the goodness of fit R between the output of the support vector machine model and the actual output data set 2 , and adjust the combination of C and G with a fixed step size to R 2 The maximum is the screening criterion, and the best combination of C and G is selected;

[0031] C5. Input the same input data set into the three optimized fitting models respectively, calculate the fitting errors between the outputs of the three fitting models and the true output data set respectively, denoted as E1, E2, and E3, and calculate the weight values ​​of the three models based on the exponential decay model, denoted as W1, W2, and W3;

[0032] C6. Construct the correction fusion model MOD_T = W1 for the temperature data detected by the meteorological drone MOD1+W2 MOD2+W3 MOD3;

[0033] C7. Use steps C1-C6 to obtain the correction fusion models of other meteorological profile data detected by the meteorological UAV.

[0034] The corrected meteorological profile data obtained according to the correction fusion model are input into the RI index classical calculation model to calculate the RI (z) index of each altitude layer z, which is recorded as Data1 and specifically includes the following contents:

[0035] According to the obtained correction fusion model of atmospheric temperature, humidity, pressure, horizontal wind speed and horizontal wind direction profile data, the atmospheric temperature, humidity, pressure, horizontal wind speed and horizontal wind direction profile data with high temporal and spatial resolution detected by the meteorological UAV are respectively input into the corresponding correction fusion model to obtain the corrected atmospheric temperature, humidity, pressure, horizontal wind speed and horizontal wind direction profile data;

[0036] Input the corrected atmospheric temperature, humidity, pressure, horizontal wind speed and horizontal wind direction profile data into In the exponential classical calculation model, the z of each altitude layer is calculated. Index, denoted as Data1.

[0037] The data collected according to step S1 is combined with the vertical wind field solution model of the DBS mode to obtain the average variance profile data of a vertical wind field within the analysis time window, which specifically includes the following contents:

[0038] According to the radial velocity, antenna elevation angle and azimuth data of different altitude layers of the ground-based laser radar in DBS scanning mode, the vertical wind field calculation model of the DBS mode is input to obtain the vertical wind field profile data of the DBS scanning mode at that moment;

[0039] According to the time period Tim_DBS required to complete a DBS scanning mode, the analysis time window is set to Tim_w, and the number of DBS scanning cycles contained in the analysis time window is calculated as Num_DBS=floor(Tim_w / Tim_DBS), where floor() means rounding down;

[0040] The Num_DBS vertical wind field profile data are combined into a two-dimensional matrix MATRIX, where the number of rows represents the number of height layers of the profile, and the number of columns represents the number of profiles;

[0041] Sort the data in the first row of MATRIX, and then count the first quartile and the third quartile of the row. Take out all the data in the row that are greater than the first quartile and less than the third quartile and calculate the average variance. Process all rows of data in this way to obtain the average variance profile data of a vertical wind field within the analysis time window.

[0042] The construction of the profile optimization threshold by using the average variance profile data of the vertical wind field obtained specifically includes the following contents:

[0043] According to the generated average variance profile data of the vertical wind field, the average value is first obtained, and then the difference between each data in the profile data and the average value is calculated, so as to generate an absolute deviation sequence;

[0044] The absolute deviation sequence is divided into M intervals, and then the number of absolute deviations in each interval is counted to generate a probability distribution curve of the absolute deviation sequence;

[0045] The probability distribution curve of the absolute deviation sequence is input into the Gaussian mixture model fitter. When the number of Gaussian distribution components N_g output by the Gaussian mixture model fitter is 1, the absolute deviation mean value output by the Gaussian mixture model fitter is calculated. and variance , construct the profile optimization threshold ;

[0046] When the number of Gaussian distribution components N_g output by the Gaussian mixture model fitter is greater than 1, the Gaussian mixture model fitter will output the mean absolute deviation ,variance and the weights of each Gaussian distribution component , i=1,2,…,N_g, thus determining the maximum value, and use the mean of absolute deviation as the axis, perform forward and backward Gaussian distribution component expansion search to find the maximum The sum of the weights of the Gaussian distribution component corresponding to the value and the adjacent k-1 Gaussian distribution components is greater than the threshold value. All Gaussian distribution components of the contour optimization threshold are constructed .

[0047] The S5 specifically includes the following contents:

[0048] D1, based on the average variance profile data of the generated vertical wind field and the calculated , set the first two data points of the profile data as the starting and ending points, draw a straight line, and calculate the vertical distances of the remaining data points from the straight line;

[0049] D2, for vertical distance less than For data points with a vertical distance greater than The data point is retained, the second data point is used as the starting point, and the nearest retained point of the second data point is used as the other point of the line segment, thereby forming a new line segment;

[0050] D3. Repeat steps D1 and D2 until the last point of the profile data, and finally generate the average variance optimized profile data of the vertical wind field, which is recorded as Data2.

[0051] The S6 specifically includes the following contents:

[0052] Generated altitude layers detected by meteorological drones The index data Data1 and the average variance optimized profile data Data2 of the generated vertical wind field are temporally and spatially matched according to the obtained meteorological drone observation altitude, longitude and latitude data and the observation parameters of the ground-based lidar, using the temporal and spatial matching method;

[0053] A deep random forest model was constructed, and the spatiotemporal matched Data2 and the corresponding height profile data were used as input data sets, and the spatiotemporal matched Data1 data were used as output data sets to train the deep random forest model. After the deep random forest model was trained, the processed average variance optimized profile data of the vertical wind field was input again to obtain the high spatial resolution of each altitude layer of the meteorological drone. Index data;

[0054] Based on the high spatial resolution meteorological UAV altitude layers Index data, with To identify the mixed layer height, find the condition that satisfies The maximum altitude layer corresponding to the time, that is, this altitude layer is the estimated value of the mixing layer altitude.

[0055] The present invention has the following advantages: a clear sky mixed layer height estimation method based on meteorological UAV and laser radar, introduces a variety of nonlinear fitting models according to the meteorological parameter profile data detected by meteorological UAV and radiosonde, makes full use of the difference in the error after fitting of each model, constructs a meteorological parameter profile correction fusion model based on exponential decay characteristics, and realizes accurate correction of meteorological data detected by meteorological UAV; based on The index calculation model and the corrected meteorological parameter profile data of the meteorological drone have achieved different altitude levels. The calculation of the index, and The exponential profile has the advantage of high temporal and spatial resolution. Based on the high temporal and spatial resolution radial velocity data in the lidar DBS scanning mode, the vertical wind field profile data with high temporal and spatial resolution is calculated, and the vertical wind field average variance profile within the set time analysis window is constructed. The probability distribution of the absolute deviation of the square variance of the vertical wind field is statistically analyzed, and a threshold calculation method and optimization strategy for the absolute deviation curve optimization in single-peak and multi-peak cases based on the Gaussian mixture model are designed, which realizes the accurate fitting and segmented optimization of multi-peak profiles. Based on the lidar vertical wind field average variance profile optimization curve and different height layers Exponential profile, a clear sky mixed layer height inversion model based on the deep random forest model was constructed, which achieved high spatial resolution and accurate estimation of the clear sky mixed layer height. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic diagram of the process of the present invention;

[0057] Figure 2 This is a flow chart for realizing the correction and fusion of meteorological profile data detected by the meteorological UAV of the present invention;

[0058] Figure 3 It is a flow chart for realizing the RI index profile of the present invention;

[0059] Figure 4 It is a flowchart for realizing the threshold value determination in the laser radar vertical wind field average variance profile optimization scheme of the present invention;

[0060] Figure 5 It is a flowchart for realizing the optimization of the average variance profile of the vertical wind field of the laser radar of the present invention;

[0061] Figure 6 This is a flowchart for realizing the clear sky mixing layer height estimation of the present invention. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of the present application provided below in conjunction with the drawings is not intended to limit the scope of protection of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application. The present invention is further described below in conjunction with the drawings.

[0063] like Figure 1 As shown, the present invention specifically relates to a clear sky mixed layer height estimation method based on a meteorological drone and a laser radar, which specifically includes the following contents:

[0064] Step 1: Under clear sky conditions, collect profile data detected by meteorological drones; collect meteorological data and observation parameters detected by ground-based lidar; collect profile data detected by radiosonde.

[0065] Among them, the profile data detected by meteorological drones mainly refers to the profile data of atmospheric temperature, humidity, pressure, horizontal wind speed, horizontal wind direction, altitude, longitude and latitude with high temporal and spatial resolution;

[0066] The meteorological data detected by ground-based lidar mainly refers to the echo signal-to-noise ratio, radial velocity, spectrum width and other data at different altitudes in the wind profile beam swing (DBS) scanning mode;

[0067] The observation parameters of the ground-based lidar mainly refer to the laser wavelength of 532 nm, the antenna elevation angle of 75 degrees, the azimuth and interval azimuth during beam scanning, the time required for the radar to complete a DBS scanning mode of 15 seconds, the radial range library length of 30 meters, the noise level, the observation time, the altitude and longitude and latitude of the instrument, etc.

[0068] The profile data collected by radiosonde detection mainly refers to the profile data of atmospheric temperature, humidity, pressure, horizontal wind speed, horizontal wind direction, altitude, longitude and latitude measured by sounding balloons;

[0069] Step 2: Based on the atmospheric temperature profile data of different altitude layers z detected by the meteorological drone in step 1 , calculate the average value of the data respectively and standard deviation ;Will The number of anomalies greater than the set threshold is 3 All of them are removed, and the data retained after removing the outliers is recorded as ;Sure The position of the null value in the window is used as the center of the window, and the window size is set to 5. Based on the non-null values ​​in the window, the estimated value at the null value position is generated using the spline interpolation method, and the null value is filled with the estimated value. The data after filling the null value is recorded as ;by A data point in the window is the center of the window, the window size is set to 4, the average value of the data in the window is calculated, and the average value is replaced by the window center value. The final processed data is recorded as The same method as above is used to process other meteorological profile data detected by meteorological UAV and radiosonde respectively.

[0070] Step 3: Based on the altitude and longitude and latitude data detected by the meteorological UAV and radiosonde obtained in step 1, and the meteorological data detected by the meteorological UAV and radiosonde processed in step 2, the altitude and longitude and latitude of the meteorological data detected by the radiosonde are used as the spatial information of the target point ( , , ), , , Respectively represent the longitude, latitude and altitude of the target, and use the Euclidean distance formula to calculate the spatial position of the meteorological UAV detection data point ( , , ) The spatial distance difference from the target point , , , Respectively represent the longitude, latitude and altitude of the meteorological drone detection data point. The screening condition is less than the spatial distance difference threshold of 0.05 km, and the spatial points adjacent to the target point are screened out. Then the difference between the observation time of the screened adjacent spatial points and the observation time of the target point is calculated, and the adjacent spatial point with the smallest time difference is taken as the matching point of the meteorological UAV detection data to the radiosonde target point. According to the above time-space matching method, all the data detected by the meteorological UAV are screened, and finally the meteorological profile data detected by the meteorological UAV and the radiosonde after time-space matching are formed.

[0071] Among them, the Euclidean distance formula is:

[0072]

[0073] Step 4: Figure 2As shown in the figure, based on the meteorological data detected by the meteorological drone and the radiosonde after time-space matching generated in step 3, the temperature data detected by the meteorological drone is used as the input data set, and the temperature data detected by the radiosonde is used as the output data set, which are respectively input into three fitting models, namely the least squares model MOD1, the elastic network model MOD2 and the support vector machine model MOD3; in the performance optimization of MOD1, the gradient descent iteration method is used to minimize the residual sum of squares between the model output result and the true output result; for the regularization parameters α and λ in MOD2, different parameters are traversed to generate a variety of combinations of α and λ, and α and λ of each combination are preset in the model, the model is run to obtain the output result, and the root mean square error RMSE between the model output result and the true output data set is calculated. The optimal regularization parameters α and λ are selected with the minimum RMSE as the screening criterion; in MOD3, the radial basis kernel function is selected, and the combination of the regularization parameter C and the kernel function parameter G is randomly selected, and the goodness of fit R between the model output result and the true output data set is calculated. 2 , and adjust the combination of C and G with a fixed step size to R 2 The maximum is the screening criterion, and the best combination of C and G is selected. After that, the same input set is input into the three optimized fitting models respectively, and the fitting errors between the outputs of the three models and the real output data sets are calculated respectively, which are recorded as E1, E2, and E3. Based on the exponential decay method model, the weight values ​​of the three models are calculated, which are recorded as W1, W2, and W3. Finally, the correction fusion model MOD_T = W1 for the temperature data detected by the meteorological drone is constructed. MOD1+W2 MOD2+W3 MOD3. The same method as above can be used to obtain the correction fusion model of other meteorological profile data detected by meteorological UAV.

[0074] The exponential decay model is: , i=1,2,3.

[0075] Step 5: Figure 3 As shown, according to the correction fusion model of atmospheric temperature, humidity, pressure, horizontal wind speed and horizontal wind direction and other profile data obtained in step 4, the atmospheric temperature, humidity, pressure, horizontal wind speed and horizontal wind direction and other profile data detected by the meteorological UAV with high temporal and spatial resolution are input into the corresponding correction fusion mode respectively, and the corrected atmospheric temperature, humidity, pressure, horizontal wind speed and horizontal wind direction and other profile data can be obtained; the above corrected atmospheric temperature, humidity, pressure, horizontal wind speed and horizontal wind direction and other profile data are input into In the exponential classical calculation model, the exponential of each altitude layer z can be calculated. Index, denoted as Data1.

[0076] Among them, the classic calculation model of RI index is:

[0077]

[0078] Where g is the acceleration of gravity, z0 is the height near the ground, z1 is the detection height, θ(z) and θ(z0) are the potential temperatures at the altitude layer z and the height near the ground z0, u(z) and u(z0) are one component of the horizontal wind speed at the altitude layer z and the height near the ground z0, respectively, and v(z) and v(z0) are the other component of the horizontal wind speed at the altitude layer z and the height near the ground z0, respectively.

[0079] Step 6: Figure 4 As shown, according to the radial velocity of different altitude layers of the ground-based laser radar in the DBS scanning mode in step 1, the antenna elevation angle is 75 degrees, and the azimuths corresponding to each beam are 0 degrees, 90 degrees, 180 degrees, and 270 degrees, which are input into the DBS mode vertical wind field solution model, the vertical wind field profile data at the DBS scanning mode moment can be obtained; according to the time cycle of 15 seconds required to complete a DBS scanning mode in step 1, the analysis time window is set to 15 minutes, and the number of DBS scanning cycles contained in the analysis time window is calculated to be 60. The 60 vertical wind field profile data are formed into a two-dimensional matrix MATRIX, the number of rows represents the number of height layers of the profile, and the number of columns represents the number of profiles; the data in the first row of MATRIX are sorted, and the first quartile and the third quartile of the row are counted, and all the data in the row greater than the first quartile and less than the third quartile are taken out and the average variance is calculated. All row data are processed in this way to obtain the average variance profile data of a vertical wind field in the analysis time window.

[0080] Step 7: According to the average variance profile data of the vertical wind field generated in step 6, first calculate its average value, then calculate the difference between each data in the profile data and the average value, so as to generate an absolute deviation sequence; divide the absolute deviation sequence into 10 intervals, and then count the number of absolute deviations in each interval to generate a probability distribution curve of the absolute deviation sequence; input the probability distribution curve of the absolute deviation sequence into the Gaussian mixture model fitter, and when the number of Gaussian distribution components output by the fitter N_g=1, according to the absolute deviation mean output by the fitter and variance , construct the profile optimization threshold ; When the number of Gaussian distribution components N_g output by the fitter is greater than 1, the fitter will output the mean absolute deviation ,variance and the weights of each Gaussian distribution component (i=1,2,…,N_g), thereby determining the maximum value, and use the mean of absolute deviation as the axis, perform forward and backward Gaussian distribution component expansion search to find the maximum The sum of the weights of the Gaussian distribution component corresponding to the value and the neighboring k-1 Gaussian distribution components is greater than all Gaussian distribution components with a threshold of 85%, and the contour optimization threshold is constructed. .

[0081] Step 8: Figure 5 As shown, according to the average variance profile data of the vertical wind field generated in step 6 and calculated in step 7 , set the first two data points of the profile data as the starting and ending points, draw a straight line, and calculate the vertical distances of the remaining data points from the straight line; for the vertical distance less than For data points with a vertical distance greater than The data point is retained, the second data point is used as the starting point, and the nearest retained point of the second data point is used as the other point of the line segment, thus forming a new line segment. The algorithm is repeated until the last point of the profile data, and finally the average variance optimized profile data of the vertical wind field is generated, which is recorded as Data2.

[0082] Step 9: Figure 6 As shown, the altitude layers detected by the meteorological drone generated in step 5 The index data Data1 and the average variance optimized profile data Data2 of the vertical wind field generated in step 8 are spatiotemporally matched according to the observation height, longitude and latitude data of the meteorological UAV obtained in step 1 and the observation parameters of the ground-based lidar, according to the spatiotemporal matching method in step 3. A deep random forest model is constructed, and the spatiotemporally matched Data2 and the corresponding height profile data are used as the input data set, and the spatiotemporally matched Data1 data is used as the output data set to train the model. After the model training is completed, the average variance optimized profile data of the vertical wind field processed according to step 8 is input again to obtain the meteorological UAV data at each height layer with high spatial resolution. Index data.

[0083] Step 10: Based on the high spatial resolution meteorological drone altitude layers obtained in step 9 Index data, with To identify the mixed layer height, find the condition that satisfies The maximum altitude layer corresponding to the time, that is, this altitude layer is the estimated value of the mixing layer altitude.

[0084] The above is only a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, and should not be regarded as excluding other embodiments, but can be used for various other combinations, modifications and improvements, and can be modified within the scope of the concept described herein through the above teachings or the technology or knowledge of the relevant field. The changes and modifications made by those skilled in the art do not deviate from the spirit and scope of the present invention, and should be within the scope of protection of the claims attached to the present invention.

Claims

1. A clear-air mixed layer height estimation method based on meteorological drones and lidar, characterized by: The method comprises: S1. Collect profile data detected by meteorological drones, meteorological data and observation parameters detected by ground-based lidar, and profile data detected by radiosonde under clear sky conditions, and process the collected data; S2, based on the data processed in step S1, the meteorological data and longitude and latitude detected by the radiosonde are used as the spatial information of the target, the spatial distance difference between the spatial position of the data point detected by the meteorological UAV and the target point is calculated, and all the data detected by the meteorological UAV are screened by the time and space matching method to form the meteorological profile data detected by the meteorological UAV and the radiosonde after time and space matching; S3, constructing a correction fusion model of all meteorological profile data detected by the meteorological UAV according to the meteorological profile data generated in step S2, and obtaining all corrected meteorological profile data according to the correction fusion model, and inputting them into the RI index classical calculation model to calculate the RI (z) index of each altitude layer z, recorded as Data1; S4, according to the meteorological data and observation parameter data detected by the ground-based laser radar collected in step S1, combined with the DBS mode vertical wind field solution model, obtain the average variance profile data of a vertical wind field within the analysis time window, and construct the profile optimization threshold through the obtained average variance profile data of the vertical wind field; S5, generating average variance optimized profile data of the vertical wind field according to the average variance profile data of the vertical wind field obtained in step S4 and the profile optimization threshold, recorded as Data2; S6. Based on the obtained Data1 and Data2, the time-space matching method is used to match Data1 and Data2 in time and space, and the RI(z) index data of each altitude layer z of the meteorological UAV with high spatial resolution is obtained by combining the constructed deep random forest model, and the estimated value of the mixed layer height is obtained according to the mixed layer height identification condition.

2. The clear-air mixed layer height estimation method based on meteorological drone and laser radar according to claim 1 is characterized in that: The profile data detected by the meteorological drone includes atmospheric temperature, humidity, pressure, horizontal wind speed, horizontal wind direction, altitude, and longitude and latitude data with high temporal and spatial resolution; The meteorological data detected by the ground-based laser radar include echo signal-to-noise ratio, radial velocity and spectrum width data of different altitude layers in DBS scanning mode, where DBS represents wind profile beam swing; The observation parameters of the ground-based laser radar include laser wavelength, antenna elevation, azimuth and interval azimuth during beam scanning, the time period required for the radar to complete a DBS scanning mode, radial range library length, noise level, observation time, instrument height, and longitude and latitude data; The profile data detected by the radiosonde include atmospheric temperature, humidity, pressure, horizontal wind speed, horizontal wind direction, altitude and longitude and latitude data measured by an air balloon.

3. The clear-air mixed layer height estimation method based on meteorological drone and laser radar according to claim 2 is characterized by: The processing of the collected data in S1 includes: A1. Calculate the average value of the atmospheric temperature profile data T(z) at different altitudes z detected by the meteorological drone. and standard deviation ; A2. The outliers T(z) greater than the set threshold score are removed, and the data retained after removing the outliers is recorded as ; A3. Confirm The position of the null value in the window is used as the center of the window, and the window size is set to width1. Based on the non-null values ​​in the window, the estimated value at the null value position is generated using the spline interpolation method, and the null value is filled with the estimated value. The data after filling the null value is recorded as ; A4. A data point in the window is the center of the window, the window size is set to width2, the average value of the data in the window is calculated, and the average value replaces the window center value. The final processed data is recorded as ; A5. Repeat steps A1-A4 to process other meteorological profile data detected by meteorological UAV and radiosonde respectively.

4. The clear sky mixed layer height estimation method based on meteorological drone and laser radar according to claim 3 is characterized by: The S2 specifically includes the following contents: B1. Based on the obtained altitude and longitude and latitude data detected by the meteorological UAV and radiosonde, as well as the processed meteorological data detected by the meteorological UAV and radiosonde, the altitude and longitude and latitude of the meteorological data detected by the radiosonde are used as the spatial information of the target point, and the spatial distance difference between the spatial position of the meteorological UAV detection data point and the target point is calculated using the Euclidean distance formula. ; B2. Less than the spatial distance difference threshold As the screening condition, the spatial points adjacent to the target point are screened out, and then the difference between the observation time of the screened adjacent spatial points and the observation time of the target point is calculated, and the adjacent spatial point with the smallest time difference is taken as the matching point of the meteorological UAV detection data to the radiosonde target point; B3. All data detected by the meteorological UAV are screened according to step B2, and finally the meteorological profile data detected by the meteorological UAV and radiosonde are formed after time and space matching.

5. The clear sky mixed layer height estimation method based on meteorological drone and laser radar according to claim 4 is characterized by: The correction fusion model of all meteorological profile data detected by the meteorological UAV is constructed according to the meteorological profile data generated in step S2, including the following contents: C1, based on the meteorological profile data detected by the meteorological UAV and radiosonde after the generated time-space matching, the temperature data detected by the meteorological UAV is used as the input data set, and the temperature data detected by the radiosonde is used as the output data set, which are input into the least squares model MOD1, the elastic network model MOD2 and the support vector machine model MOD3 respectively; C2, using the gradient descent iteration method to optimize the performance of MOD1 to minimize the sum of squared residuals between the least squares model output and the true output; C3. For the regularization parameters α and λ in MOD2, traverse different parameters to generate multiple combinations of α and λ, and preset α and λ for each combination in the elastic network model, run the model to obtain the output results, calculate the root mean square error RMSE between the output results of the elastic network model and the actual output data set, and select the best regularization parameters α and λ with the minimum RMSE as the screening criterion; C4. Select the radial basis kernel function in MOD3, randomly select the combination of regularization parameter C and kernel function parameter G, and calculate the goodness of fit R between the output of the support vector machine model and the actual output data set 2 , and adjust the combination of C and G with a fixed step size to R 2 The maximum is the screening criterion, and the best combination of C and G is selected; C5. Input the same input data set into the three optimized fitting models respectively, calculate the fitting errors between the outputs of the three fitting models and the true output data set respectively, denoted as E1, E2, and E3, and calculate the weight values ​​of the three models based on the exponential decay model, denoted as W1, W2, and W3; C6. Construct a correction fusion model for temperature data detected by meteorological drones ; C7. Use steps C1-C6 to obtain the correction fusion models of other meteorological profile data detected by the meteorological UAV.

6. The clear-air mixed layer height estimation method based on meteorological drone and laser radar according to claim 5 is characterized by: The corrected meteorological profile data obtained according to the correction fusion model are input into the RI index classical calculation model to calculate the RI (z) index of each altitude layer z, which is recorded as Data1 and specifically includes the following contents: According to the obtained correction fusion model of atmospheric temperature, humidity, pressure, horizontal wind speed and horizontal wind direction profile data, the atmospheric temperature, humidity, pressure, horizontal wind speed and horizontal wind direction profile data with high temporal and spatial resolution detected by the meteorological UAV are respectively input into the corresponding correction fusion model to obtain the corrected atmospheric temperature, humidity, pressure, horizontal wind speed and horizontal wind direction profile data; The corrected atmospheric temperature, humidity, pressure, horizontal wind speed and horizontal wind direction profile data are input into the RI index classic calculation model to calculate the RI (z) index of each altitude layer z, which is recorded as Data1.

7. The clear-air mixed layer height estimation method based on meteorological drone and laser radar according to claim 1 is characterized by: The meteorological data and observation parameter data detected by the ground-based laser radar collected in step S1 are combined with the DBS mode vertical wind field solution model to obtain the average variance profile data of a vertical wind field within the analysis time window, which specifically includes the following contents: According to the radial velocity, antenna elevation angle and azimuth data of different altitude layers of the ground-based laser radar in DBS scanning mode, the vertical wind field calculation model of the DBS mode is input to obtain the vertical wind field profile data of the DBS scanning mode at that moment; According to the time period Tim_DBS required to complete a DBS scanning mode, the analysis time window is set to Tim_w, and the number of DBS scanning cycles contained in the analysis time window is calculated as Num_DBS=floor(Tim_w / Tim_DBS), where floor() means rounding down; The Num_DBS vertical wind field profile data are combined into a two-dimensional matrix MATRIX, where the number of rows represents the number of height layers of the profile, and the number of columns represents the number of profiles; Sort the data in the first row of MATRIX, and then count the first quartile and the third quartile of the row. Take out all the data in the row that are greater than the first quartile and less than the third quartile and calculate the average variance. Process all rows of data in this way to obtain the average variance profile data of a vertical wind field within the analysis time window.

8. The clear-air mixed layer height estimation method based on meteorological drone and laser radar according to claim 7 is characterized by: The construction of the profile optimization threshold by using the average variance profile data of the vertical wind field obtained specifically includes the following contents: According to the generated average variance profile data of the vertical wind field, the average value is first obtained, and then the difference between each data in the profile data and the average value is calculated, thereby generating an absolute deviation sequence; The absolute deviation sequence is divided into M intervals, and then the number of absolute deviations in each interval is counted to generate a probability distribution curve of the absolute deviation sequence; The probability distribution curve of the absolute deviation sequence is input into the Gaussian mixture model fitter. When the number of Gaussian distribution components N_g output by the Gaussian mixture model fitter is 1, the absolute deviation mean μ and variance output by the Gaussian mixture model fitter are calculated. , construct the profile optimization threshold ; When the number of Gaussian distribution components N_g output by the Gaussian mixture model fitter is greater than 1, the Gaussian mixture model fitter will output the mean absolute deviation ,variance and the weights of each Gaussian distribution component , i=1,2,…,N_g, thus determining the maximum value, and use the mean of absolute deviation as the axis, perform forward and backward Gaussian distribution component expansion search to find the maximum The sum of the weights of the Gaussian distribution component corresponding to the value and the adjacent k-1 Gaussian distribution components is greater than the threshold value. All Gaussian distribution components of the contour optimization threshold are constructed .

9. The clear-air mixed layer height estimation method based on meteorological drone and laser radar according to claim 8 is characterized by: The S5 specifically includes the following contents: D1, based on the average variance profile data of the generated vertical wind field and the calculated , set the first two data points of the profile data as the starting and ending points, draw a straight line, and calculate the vertical distances of the remaining data points from the straight line; D2, for vertical distance less than For data points with a vertical distance greater than The data point is retained, the second data point is used as the starting point, and the nearest retained point of the second data point is used as the other point of the line segment, thereby forming a new line segment; D3. Repeat steps D1 and D2 until the last point of the profile data, and finally generate the average variance optimized profile data of the vertical wind field, which is recorded as Data2.

10. The clear-air mixed layer height estimation method based on meteorological drone and laser radar according to claim 9 is characterized in that: The S6 specifically includes the following contents: Based on the generated RI(z) index data Data1 of each altitude layer detected by the meteorological UAV and the generated average variance optimized profile data Data2 of the vertical wind field, Data1 and Data2 are temporally and spatially matched according to the obtained meteorological UAV observation altitude, longitude and latitude data and the observation parameters of the ground-based laser radar according to the temporal and spatial matching method; A deep random forest model was constructed, and the spatiotemporally matched Data2 and the corresponding height profile data were used as input data sets, and the spatiotemporally matched Data1 data were used as output data sets. The deep random forest model was trained. After the deep random forest model was trained, the processed average variance optimized profile data of the vertical wind field was input again to obtain the RI(z) index data of each altitude layer of the meteorological UAV with high spatial resolution. Based on the RI(z) index data of each altitude layer of the high spatial resolution meteorological UAV, RI(z)<0.25 is taken as the condition for identifying the mixed layer altitude, and the maximum altitude layer corresponding to RI(z)<0.25 is found, that is, this altitude layer is the estimated value of the mixed layer altitude.

Images