Particle cloud macroscopic parameter quantification method, device and equipment based on machine learning

Abstract

The invention provides a particle cloud macroscopic parameter quantification method, device and equipment based on machine learning. The method comprises the following steps: acquiring foundation data, satellite data and meteorological data; performing space-time matching processing on the foundation data and the satellite data based on the meteorological data to obtain a matched foundation data set and a matched satellite data set; performing abnormal data rejection processing on the matched foundation data set, the matched satellite data set and the meteorological data to obtain rejected data; performing variable screening on the rejected data to obtain a meteorological variable, a foundation cloud macroscopic parameter variable and a satellite cloud macroscopic parameter variable; training the particle cloud macroscopic parameter initial model by adopting a meteorological variable, a foundation cloud macroscopic parameter variable and a satellite cloud macroscopic parameter variable to obtain a trained particle cloud macroscopic parameter model; and obtaining to-be-processed meteorological data, and generating a particle cloud macroscopic parameter quantification result based on the to-be-processed meteorological data and the trained particle cloud macroscopic parameter model.

Description

Technical Field

[0001] The present invention relates to the technical field of atmospheric cloud parameter quantification, and in particular to a particle cloud macroscopic parameter quantification method, device and equipment based on machine learning. Background Art

[0002] A particle cloud is an aggregate composed of particles of water vapor condensation (cloud droplets, raindrops, ice crystals, snow flakes, etc.) in different phases. It occupies a certain space and grows and migrates with the movement of the atmosphere. The macroscopic parameters of particle clouds have an important impact on the operation and safety of aircraft. For example, the size of the cloud directly affects the visibility and navigation of the aircraft. Especially when the cloud cover is high, the pilot's field of vision may be limited, increasing the need to rely on instrument flight. Secondly, the cloud base height determines the time and altitude of the aircraft entering or crossing the cloud layer, which has a direct impact on the flight path and take-off and landing operations during flight. If the cloud base height is too low, it may affect the normal take-off and landing of flights, especially near airports. In addition, the cloud top height determines whether the aircraft needs to cross the cloud layer when flying at high altitude. The higher the cloud top height, the more likely the aircraft will encounter strong convection or turbulent weather, affecting flight stability. In summary, accurate calculation of the macroscopic parameters of particle clouds is crucial for planning flight routes, ensuring flight safety, and improving flight efficiency.

[0003] Existing methods for quantifying particle cloud macro parameters include observation methods, diagnostic model methods, and statistical model methods. Observation methods mainly rely on ground observations, sounding observations, aircraft detection, and satellite remote sensing. Although ground observations can provide detailed information in the vertical direction and have the advantage of temporal continuity, their limitation is that they can only obtain cloud macro parameters at a single site, and the spatial coverage is small. Sounding observations and aircraft detection can obtain detailed data at different altitudes and provide high-precision information on particle cloud macro parameters, but their high cost limits their widespread application. In contrast, polar-orbiting satellites can obtain global observation data, which makes up for the shortcomings of ground and local observations, but due to orbital characteristics, continuous time monitoring is not possible. Geostationary satellites have the advantage of continuous observation in time, but their grasp of the three-dimensional structure of the cloud layer is not as good as other observation methods. The diagnostic model is a simplification of the complex process of cloud formation and contains a large number of empirical formulas in the past. Due to the lack of observation and detection data in the past, these empirical formulas cannot correctly describe the complex interactions between particle clouds and various atmospheric parameters. Therefore, the diagnostic model cannot effectively capture the complex nonlinear dynamic processes in particle clouds, especially when dealing with highly variable meteorological conditions or extreme weather, the performance of the model will be significantly reduced. The statistical model method relies on a large amount of historical observation data and estimates parameters by establishing statistical relationships between the data. This method has limited adaptability to different meteorological conditions, and the generalization ability of the statistical model is weak, and it often cannot accurately obtain or predict the macroscopic parameters of particle clouds. However, the diagnostic model and the statistical model can flexibly adapt to the conditions in different regions, provide quantitative results of the macroscopic parameters of particle clouds on a global scale, and still have certain predictive capabilities in the absence of real-time observations. They also have faster calculation speeds and are more widely used.

[0004] Therefore, the existing particle cloud macro-parameter quantification methods have problems such as poor adaptability to highly changing meteorological conditions and limited model generalization ability, which leads to low accuracy in quantifying cloud particle macro-parameters and makes it difficult to meet the precise quantification requirements under variable atmospheric conditions. Summary of the invention

[0005] In order to solve the above problems existing in the prior art, the present invention provides a particle cloud macroscopic parameter quantification method, device and equipment based on machine learning.

[0006] The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0007] In a first aspect, the present invention provides a particle cloud macroscopic parameter quantification method based on machine learning, comprising:

[0008] Obtain ground-based data, satellite data and meteorological data; both ground-based data and satellite data are particle cloud data;

[0009] Based on the meteorological data, the ground-based data and the satellite data are processed in time and space to obtain a matching ground-based data set and a matching satellite data set;

[0010] Abnormal data are eliminated from the matched ground-based data set, the matched satellite data set and the meteorological data to obtain eliminated data; the eliminated data includes: the eliminated ground-based data set, the eliminated satellite data set and the eliminated meteorological data;

[0011] The eliminated data are screened for variables to obtain meteorological variables, ground-based cloud macro-parameter variables, and satellite cloud macro-parameter variables;

[0012] The particle cloud macro-parameter initial model is trained using meteorological variables, ground-based cloud macro-parameter variables and satellite cloud macro-parameter variables to obtain a trained particle cloud macro-parameter model.

[0013] Obtain the meteorological data to be processed, and generate the particle cloud macro-parameter quantification results based on the meteorological data to be processed and the trained particle cloud macro-parameter model.

[0014] Optionally, performing spatiotemporal matching processing on the ground-based data and the satellite data based on the meteorological data to obtain a matched ground-based data set and a matched satellite data set includes:

[0015] The meteorological data and ground-based data are matched in time and space by the principle of temporal proximity and spatial proximity to obtain a matching ground-based data set.

[0016] The meteorological data and satellite data are temporally and spatially matched based on the temporal proximity principle and the spatial proximity principle to obtain a matching satellite data set.

[0017] Optionally, abnormal data removal is performed on the matched ground-based data set, the matched satellite data set, and the meteorological data to obtain data after removal, including:

[0018] The preset elimination threshold or isolation forest algorithm is used to eliminate abnormal data from the matched ground-based data set, the matched satellite data set and the meteorological data to obtain the eliminated ground-based data set, the eliminated satellite data set and the eliminated meteorological data.

[0019] Optionally, the matched ground-based dataset and the matched satellite dataset both include: cloud amount, cloud base height, and cloud top height;

[0020] Meteorological data include: temperature, vertical speed, specific humidity, relative humidity, liquid water content, rain water content, snow water content, ice water content, longitude U wind component, latitude V wind component, month, hour, dew point temperature difference and altitude.

[0021] Optionally, the eliminated data are subjected to variable screening to obtain meteorological variables, ground-based cloud macro-parameter variables, and satellite cloud macro-parameter variables, including:

[0022] Calculate the correlation coefficient and information entropy between the eliminated meteorological data and the eliminated ground-based data set, and obtain the ground-based correlation coefficient and ground-based information entropy accordingly;

[0023] Calculate the correlation coefficient and information entropy between the eliminated meteorological data and the eliminated satellite data set, and obtain the satellite correlation coefficient and satellite information entropy accordingly;

[0024] The ground-based correlation coefficient, ground-based information entropy, satellite correlation coefficient and satellite information entropy are used to screen the variables of the eliminated data to obtain meteorological variables, ground-based cloud macro-parameter variables and satellite cloud macro-parameter variables.

[0025] Optionally, the eliminated data are screened by using the ground-based correlation coefficient, the ground-based information entropy, the satellite correlation coefficient and the satellite information entropy to obtain meteorological variables, ground-based cloud macro-parameter variables and satellite cloud macro-parameter variables, including:

[0026] Sort the ground-based information entropy and the satellite information entropy by size to obtain the information entropy sorting result;

[0027] The data with ground-based correlation coefficient and satellite correlation coefficient greater than 0.3 and the first 75% of the information entropy sorting results were screened out to obtain meteorological variables, ground-based cloud macro-parameter variables and satellite cloud macro-parameter variables.

[0028] Optionally, the particle cloud macro-parameter initial model includes: a particle cloud macro-parameter cloud height initial model and a particle cloud macro-parameter cloud amount initial model;

[0029] The initial model of particle cloud macroscopic parameter cloud height adopts UNet++ architecture; the initial model of particle cloud macroscopic parameter cloud amount adopts LGBM architecture, and the classifier of LGBM architecture is GBDT;

[0030] The particle cloud macro parameter initial model is trained using meteorological variables, ground-based cloud macro parameter variables and satellite cloud macro parameter variables to obtain a trained particle cloud macro parameter model, including:

[0031] Taking meteorological variables as input values ​​and cloud heights of ground-based cloud macro-parameter variables and satellite cloud macro-parameter variables as true values, the particle cloud macro-parameter cloud height initial model is trained, and the particle cloud macro-parameter cloud height initial model that meets the preset convergence conditions is used as the trained particle cloud macro-parameter cloud height model;

[0032] Taking meteorological variables as input values ​​and cloud amounts of ground-based cloud macro-parameter variables and satellite cloud macro-parameter variables as true values, the particle cloud macro-parameter cloud amount initial model is trained, and the particle cloud macro-parameter cloud amount initial model that meets the preset convergence conditions is used as the trained particle cloud macro-parameter cloud amount model;

[0033] The trained particle cloud macro-parameter cloud height model and the trained particle cloud macro-parameter cloud amount model are used together as the trained particle cloud macro-parameter model;

[0034] Among them, the initial model of particle cloud macroscopic parameter cloud height adopts Dice loss; the initial model of particle cloud macroscopic parameter cloud amount adopts L2 norm loss.

[0035] Optionally, the preset convergence condition includes: the number of iterations is greater than a preset iteration threshold, or the value of the loss function is continuously less than a preset loss threshold.

[0036] In a second aspect, the present invention provides a particle cloud macro parameter quantification device based on machine learning, and the particle cloud macro parameter quantification device based on machine learning includes: an acquisition unit, a matching unit, a data elimination unit, a variable screening unit, a training unit, and a forecasting unit;

[0037] The acquisition unit is used to: acquire ground-based data, satellite data and meteorological data; both the ground-based data and satellite data are particle cloud data;

[0038] The matching unit is used to: perform spatiotemporal matching processing on the ground-based data and the satellite data based on the meteorological data to obtain a matched ground-based data set and a matched satellite data set;

[0039] The data elimination unit is used to eliminate abnormal data from the matched ground-based data set, the matched satellite data set and the meteorological data to obtain eliminated data; the eliminated data includes: the eliminated ground-based data set, the eliminated satellite data set and the eliminated meteorological data;

[0040] The variable screening unit is used to: screen the eliminated data to obtain meteorological variables, ground-based cloud macro-parameter variables and satellite cloud macro-parameter variables;

[0041] The training unit is used to: train the initial model of particle cloud macro parameters by using meteorological variables, ground-based cloud macro parameter variables and satellite cloud macro parameter variables, so as to obtain a trained particle cloud macro parameter model;

[0042] The forecasting unit is used to obtain the meteorological data to be processed, and generate the particle cloud macro-parameter quantification results based on the meteorological data to be processed and the trained particle cloud macro-parameter model.

[0043] In a third aspect, the present invention provides a particle cloud macro-parameter quantification device based on machine learning, comprising: a processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor. When the particle cloud macro-parameter quantification device based on machine learning is running, the processor and the storage medium communicate via the bus, and the processor executes the machine-readable instructions to perform the steps of the particle cloud macro-parameter quantification method based on machine learning as described in the first aspect above.

[0044] The present invention provides a particle cloud macro parameter quantification method, device and equipment based on machine learning. The particle cloud macro parameter quantification method based on machine learning includes: obtaining ground-based data, satellite data and meteorological data; the ground-based data and satellite data are both particle cloud data; performing spatiotemporal matching processing on the ground-based data and satellite data based on meteorological data to obtain a matched ground-based data set and a matched satellite data set; performing abnormal data elimination processing on the matched ground-based data set, the matched satellite data set and meteorological data to obtain eliminated data; the eliminated data includes: eliminated ground-based data set, eliminated satellite data set and eliminated meteorological data; performing variable screening on the eliminated data to obtain meteorological variables, ground-based cloud macro parameter variables and satellite cloud macro parameter variables; using meteorological variables, ground-based cloud macro parameter variables and satellite cloud macro parameter variables to train the particle cloud macro parameter initial model to obtain a trained particle cloud macro parameter model; obtaining the meteorological data to be processed, and generating the particle cloud macro parameter quantification result based on the meteorological data to be processed and the trained particle cloud macro parameter model. In the present invention, the initial model of cloud macro parameters is trained at multiple scales based on ground-based data, satellite data and meteorological data, and the influence of abnormal data on the initial model of particle cloud macro parameters is avoided based on time-space matching and abnormal data elimination processing, thereby improving the comprehensiveness and accuracy of particle cloud macro parameter quantification. Since the method of the present invention is a model processing based on overall level data, it avoids the prior art of calculating cloud amount based on a single relative variable, thereby improving the universality of the particle cloud macro parameter quantification method and the accuracy of cloud particle environmental parameter prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A schematic diagram of a process flow of a particle cloud macroscopic parameter quantification method based on machine learning provided in an embodiment of the present invention;

[0046] Figure 2 A schematic diagram of cloud recognition time series results obtained based on the method of the present invention is shown exemplarily;

[0047] Figure 3 A schematic diagram of cloud height results obtained based on the method of the present invention is shown exemplarily;

[0048] Figure 4A schematic diagram of cloud cover results obtained based on the method of the present invention is shown exemplarily;

[0049] Figure 5 A schematic diagram of the structure of a particle cloud macroscopic parameter quantification device based on machine learning provided in an embodiment of the present invention;

[0050] Figure 6 A schematic diagram of the structure of a particle cloud macroscopic parameter quantification device based on machine learning provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The present invention is further described in detail below with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.

[0052] In order to improve the comprehensiveness and accuracy of the quantification of particle cloud macroscopic parameters, an embodiment of the present invention provides a particle cloud macroscopic parameter quantification method based on machine learning. Figure 1 A schematic diagram of a process flow of a particle cloud macro parameter quantification method based on machine learning provided by an embodiment of the present invention. Figure 1 As shown, the method includes:

[0053] S101. Acquire ground-based data, satellite data, and meteorological data.

[0054] Among them, the ground-based data refers to the ground-based particle cloud data obtained by ground observation equipment, and the satellite data refers to the satellite particle cloud data observed by satellites.

[0055] The meteorological data may be the obtained reanalysis data set ERA5, and the meteorological data includes: temperature, vertical velocity, specific humidity, relative humidity, liquid water content, rain water content, snow water content, ice water content, longitude U wind component, latitude V wind component, month, hour, dew point temperature difference and altitude data. Furthermore, the ground-based data, satellite data and meteorological data are data in the same area and the same time interval.

[0056] S102: Performing spatiotemporal matching processing on the ground-based data and the satellite data based on the meteorological data to obtain a matched ground-based data set and a matched satellite data set.

[0057] Optionally, in one implementation, S102 may specifically include:

[0058] The meteorological data and ground-based data are matched in time and space by the principle of temporal proximity and spatial proximity to obtain a matching ground-based data set.

[0059] The meteorological data and satellite data are temporally and spatially matched based on the temporal proximity principle and the spatial proximity principle to obtain a matching satellite data set.

[0060] In another embodiment, in order to improve the matching accuracy, the ground-based data and satellite data may be resampled according to the spatial resolution of the meteorological data before the spatiotemporal matching, and the spatiotemporal matching process is then performed based on the resampled ground-based data and satellite data.

[0061] It can be understood that through the above-mentioned time-space matching and resampling operations, the matched ground-based dataset and the matched satellite dataset can be made consistent with the time and space of the meteorological data, thereby improving the accuracy of subsequent training data and the accuracy of model parameters.

[0062] Optionally, the matched ground-based dataset and the matched satellite dataset both include: cloud amount, cloud base height, and cloud top height.

[0063] It should be noted that cloud cover is specifically a measure of the amount of clouds when there are clouds in an area. Cloud base height is the bottom height information of the cloud obtained based on cloud cover, and cloud top height is the top height information of the cloud obtained based on cloud cover.

[0064] S103, performing abnormal data elimination processing on the matched ground-based data set, the matched satellite data set and the meteorological data to obtain eliminated data, wherein the eliminated data includes: the eliminated ground-based data set, the eliminated satellite data set and the eliminated meteorological data.

[0065] Optionally, S103 may specifically include:

[0066] The preset elimination threshold or isolation forest algorithm is used to eliminate abnormal data from the matched ground-based data set, the matched satellite data set and the meteorological data to obtain the eliminated ground-based data set, the eliminated satellite data set and the eliminated meteorological data.

[0067] A data rejection method with preset rejection thresholds can be used, for example, to reject data with relative humidity, cloud water, and cloud ice content less than 0 in meteorological data, or to delete data that does not conform to physical common sense based on prior information.

[0068] Further, in the embodiment of the present invention, when deleting abnormal data in meteorological data, data corresponding to the abnormal data in meteorological data in the matched ground-based data set and the matched satellite data set may also be deleted. Accordingly, similar processing may also be performed on the matched ground-based data set and the matched satellite data set according to the above principle, and the embodiment of the present invention will not be described in detail.

[0069] S104, performing variable screening on the eliminated data to obtain meteorological variables, ground-based cloud macro-parameter variables, and satellite cloud macro-parameter variables.

[0070] Optionally, S104 may specifically include:

[0071] Calculate the correlation coefficient and information entropy between the eliminated meteorological data and the eliminated ground-based data set, and obtain the ground-based correlation coefficient and ground-based information entropy accordingly;

[0072] Calculate the correlation coefficient and information entropy between the eliminated meteorological data and the eliminated satellite data set, and obtain the satellite correlation coefficient and satellite information entropy accordingly;

[0073] The ground-based correlation coefficient, ground-based information entropy, satellite correlation coefficient and satellite information entropy are used to screen the variables of the eliminated data to obtain meteorological variables, ground-based cloud macro-parameter variables and satellite cloud macro-parameter variables.

[0074] Optionally, the eliminated data are screened by using the ground-based correlation coefficient, the ground-based information entropy, the satellite correlation coefficient and the satellite information entropy to obtain meteorological variables, ground-based cloud macro-parameter variables and satellite cloud macro-parameter variables, including:

[0075] Sort the ground-based information entropy and the satellite information entropy by size to obtain the information entropy sorting result;

[0076] The data with ground-based correlation coefficient and satellite correlation coefficient greater than 0.3 and the first 75% of the information entropy sorting results were screened out to obtain meteorological variables, ground-based cloud macro-parameter variables and satellite cloud macro-parameter variables.

[0077] Since the ground-based and satellite cloud macro-parameter variables (including cloud amount, cloud base height, cloud shape and other factors) correspond to the measurement values ​​of multiple sensors or multiple channels of satellites, the information complexity is high, and the spatiotemporal matching of meteorological variables, ground-based cloud macro-parameter variables and satellite cloud macro-parameter variables is particularly important. Therefore, in this step, the correlation coefficient and information entropy are used to screen the data to select data with good spatiotemporal matching and high quality.

[0078] S105. Using meteorological variables, ground-based cloud macro-parameter variables and satellite cloud macro-parameter variables to train the particle cloud macro-parameter initial model, to obtain a trained particle cloud macro-parameter model.

[0079] Optionally, the particle cloud macro-parameter initial model includes: a particle cloud macro-parameter cloud height initial model and a particle cloud macro-parameter cloud amount initial model.

[0080] The initial model of particle cloud macro-parameter cloud height adopts UNet++ architecture; the initial model of particle cloud macro-parameter cloud amount adopts LGBM architecture, and the classifier of LGBM architecture is GBDT.

[0081] Optionally, S105 may specifically include:

[0082] Taking meteorological variables as input values ​​and cloud heights of ground-based cloud macro-parameter variables and satellite cloud macro-parameter variables as true values, the particle cloud macro-parameter cloud height initial model is trained, and the particle cloud macro-parameter cloud height initial model that meets the preset convergence conditions is used as the trained particle cloud macro-parameter cloud height model;

[0083] Taking meteorological variables as input values ​​and the cloud amounts of ground-based cloud macro-parameter variables and satellite cloud macro-parameter variables as true values, the particle cloud macro-parameter cloud amount initial model is trained, and the particle cloud macro-parameter cloud amount initial model that meets the preset convergence conditions is used as the trained particle cloud macro-parameter cloud amount model.

[0084] In this step, ground-based cloud macro-parameter variables and satellite cloud macro-parameter variables are used as true values ​​for model training, realizing the mutual complementation of ground observations and satellite observations.

[0085] The trained particle cloud macro-parameter cloud height model and the trained particle cloud macro-parameter cloud amount model are used together as the trained particle cloud macro-parameter model;

[0086] Among them, the initial model of particle cloud macroscopic parameter cloud height adopts Dice loss; the initial model of particle cloud macroscopic parameter cloud amount adopts L2 norm loss.

[0087] Optionally, the preset convergence condition includes: the number of iterations is greater than a preset iteration threshold, or the value of the loss function is continuously less than a preset loss threshold.

[0088] It should be noted that the initial model of the particle cloud macroscopic parameter cloud height uses the ResNeXt group convolution algorithm for encoding feature extraction, and the activation function uses Sigmoid.

[0089] S106, obtaining the meteorological data to be processed, and generating the particle cloud macro-parameter quantification results based on the meteorological data to be processed and the trained particle cloud macro-parameter model.

[0090] It should be noted that cloud particle environmental parameters can be predicted based on the quantified results of particle cloud macroscopic parameters. For example, new energy solar energy can be predicted or the flight trajectory of an aircraft can be planned through the quantified results of particle cloud macroscopic parameters.

[0091] The embodiment of the present invention provides a particle cloud macro-parameter quantification method based on machine learning, including: taking ground-based data, satellite data and meteorological data as benchmarks, training the cloud macro-parameter initial model at multiple scales, and avoiding the influence of abnormal data on the particle cloud macro-parameter initial model based on time-space matching and abnormal data elimination processing, thereby improving the comprehensiveness and accuracy of particle cloud macro-parameter quantification. Since the method of the present invention is a model processing based on overall level data, it avoids the prior art of calculating cloud amount based on a single relative variable, thereby improving the universality of the particle cloud macro-parameter quantification method and the accuracy of cloud particle environmental parameter prediction.

[0092] In order to verify the accuracy of the particle cloud macroscopic parameter quantification method based on machine learning provided by the embodiment of the present invention, the embodiment of the present invention also conducted a simulation experiment as follows:

[0093] The results of the trained particle cloud macro parameter model are tested using ARM observation data (meteorological verification data) and GOES geostationary satellite data (satellite verification data). In order to evaluate the accuracy of the output product of the trained particle cloud macro parameter model, the error and correlation between the data to be tested and the true value are calculated according to different output variables and different dimensions. The statistical indicators used in the embodiment of the present invention include: mean absolute error, root mean square error, correlation coefficient and accuracy.

[0094] For ARM ground-based data, data inversion and observation quality control analysis are performed according to the Quality Flag provided by different detection data sets, and only the highest quality products are selected to establish the particle cloud macro-parameter observation detection data set for the verification of the trained particle cloud macro-parameter model. Table 1 is the input data of the particle cloud macro-parameter initial model provided by an embodiment of the present invention. Table 2 is the verification data of the trained particle cloud macro-parameter model provided by an embodiment of the present invention.

[0095] Table 1 Input data of the initial model of particle cloud macroscopic parameters

[0096]

[0097] Table 2 Test data of the trained particle cloud macroscopic parameter model

[0098]

[0099] Figure 2 The following is a schematic diagram of cloud recognition time series results obtained based on the method of the present invention. Figure 2 It can be seen that the trained particle cloud macro-parameter model (also known as the optimization model) can make relatively accurate identification of clouds at different altitudes. The identification results of high, medium and low clouds are well consistent with the ARM observation data, and there is no obvious omission or misjudgment.

[0100] Figure 3 The schematic diagram of cloud height results obtained based on the method of the present invention is shown as an example. The cloud height includes the cloud top height and cloud bottom height. Figure 3 It can be seen that the cloud top and cloud bottom heights predicted by the optimization model are consistent with the trends of the observed true values. Except for the partial overestimation of the cloud bottom height near 10 km, there is no large deviation. The appearance of this deviation may be related to the observation method of ground-based observation equipment, that is, the appearance of thicker clouds will cause rapid attenuation of electromagnetic radiation, which will lead to inaccurate judgment of cloud tops. This error can be completely ignored in the field. Figure 4 The schematic diagram of cloud amount results obtained based on the method of the present invention is shown as an example. By comparison, it is found that the LGBM model (particle cloud macroscopic parameter cloud amount model) can predict the real cloud amount well, has a high consistency with ARM observations, and can better describe the changes of cloud amount with height and time, especially the cloud edge area with cloud amount less than 1, which reflects a more accurate quantitative effect than the diagnostic and statistical models.

[0101] The method provided in the embodiment of the present invention can be applied to an electronic device. Specifically, the electronic device can be: a desktop computer, a portable computer, an intelligent mobile terminal, a server, etc., which is not limited in the embodiment of the present invention.

[0102] Based on the same inventive concept, an embodiment of the present invention also provides a particle cloud macroscopic parameter quantification device based on machine learning. Figure 5 The present invention provides a schematic diagram of a particle cloud macro parameter quantification device based on machine learning. Figure 5 As shown, the particle cloud macro parameter quantification device based on machine learning includes: an acquisition unit 501, a matching unit 502, a data elimination unit 503, a variable screening unit 504, a training unit 505 and a forecasting unit 506;

[0103] The acquisition unit 501 is used to acquire ground-based data, satellite data and meteorological data; both the ground-based data and satellite data are particle cloud data.

[0104] The matching unit 502 is used to perform spatiotemporal matching processing on the ground-based data and the satellite data based on the meteorological data to obtain a matched ground-based data set and a matched satellite data set.

[0105] The data elimination unit 503 is used to eliminate abnormal data from the matched ground-based data set, the matched satellite data set and the meteorological data to obtain eliminated data; the eliminated data includes: the eliminated ground-based data set, the eliminated satellite data set and the eliminated meteorological data.

[0106] The variable screening unit 504 is used to screen the removed data to obtain meteorological variables, ground-based cloud macro-parameter variables and satellite cloud macro-parameter variables.

[0107] The training unit 505 is used to train the particle cloud macro-parameter initial model using meteorological variables, ground-based cloud macro-parameter variables and satellite cloud macro-parameter variables to obtain a trained particle cloud macro-parameter model.

[0108] The forecasting unit 506 is used to obtain the meteorological data to be processed, and generate the particle cloud macro-parameter quantification results based on the meteorological data to be processed and the trained particle cloud macro-parameter model.

[0109] Figure 6 A schematic diagram of the structure of a particle cloud macro parameter quantification device based on machine learning provided in an embodiment of the present invention includes: a processor 610, a storage medium 620 and a bus 630, wherein the storage medium 620 stores machine-readable instructions executable by the processor 610, and when the particle cloud macro parameter quantification device based on machine learning is running, the processor 610 communicates with the storage medium 620 through the bus 630, and the processor 610 executes the machine-readable instructions to execute the steps of the above method embodiment. The specific implementation method and technical effect are similar and will not be repeated here.

[0110] The storage medium may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk storage. Optionally, the storage medium may also be at least one storage device located away from the aforementioned processor.

[0111] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0112] It should be noted that the terms "first", "second", etc. are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention.

[0113] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.

[0114] Although the present invention is described herein in conjunction with various embodiments, in the process of implementing the claimed invention, those skilled in the art can understand and implement other changes of the above disclosed embodiments by viewing the drawings and the disclosed content. In the description of the present invention, the term "comprising" does not exclude other components or steps, "one" or "an" does not exclude multiple situations, and the meaning of "multiple" is two or more, unless otherwise clearly and specifically limited. In addition, certain measures are recorded in different embodiments, but this does not mean that these measures cannot be combined to produce good results.

[0115] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the scope of protection of the present invention.

Claims

1. A particle cloud macroscopic parameter quantification method based on machine learning, characterized in that: include: Access to ground-based data, satellite data, and meteorological data; The ground-based data and the satellite data are both particle cloud data; Performing spatiotemporal matching processing on the ground-based data and the satellite data based on the meteorological data to obtain a matched ground-based data set and a matched satellite data set; Performing abnormal data elimination processing on the matched ground-based data set, the matched satellite data set and the meteorological data to obtain eliminated data; The removed data include: removed ground-based data sets, removed satellite data sets and removed meteorological data; Performing variable screening on the eliminated data to obtain meteorological variables, ground-based cloud macro-parameter variables, and satellite cloud macro-parameter variables; The meteorological variables, the ground-based cloud macro-parameter variables and the satellite cloud macro-parameter variables are used to train the particle cloud macro-parameter initial model to obtain a trained particle cloud macro-parameter model; Acquire the meteorological data to be processed, and generate the particle cloud macro-parameter quantification result based on the meteorological data to be processed and the trained particle cloud macro-parameter model.

2. The particle cloud macroscopic parameter quantification method based on machine learning according to claim 1 is characterized in that: The performing spatiotemporal matching processing on the ground-based data and the satellite data based on the meteorological data to obtain a matched ground-based data set and a matched satellite data set includes: Performing spatiotemporal matching processing on the meteorological data and the ground-based data according to the principle of temporal proximity and spatial proximity to obtain a matching ground-based data set; The meteorological data and the satellite data are subjected to time-space matching processing according to the time proximity principle and the space proximity principle to obtain a matching satellite data set.

3. The particle cloud macroscopic parameter quantification method based on machine learning according to claim 1 is characterized in that: The performing abnormal data elimination processing on the matched ground-based data set, the matched satellite data set and the meteorological data to obtain the eliminated data includes: The matched ground-based data set, the matched satellite data set and the meteorological data are subjected to abnormal data elimination processing by using a preset elimination threshold or an isolation forest algorithm to obtain the eliminated ground-based data set, the eliminated satellite data set and the eliminated meteorological data.

4. The particle cloud macroscopic parameter quantification method based on machine learning according to claim 1, characterized in that: The matched ground-based data set and the matched satellite data set both include: cloud amount, cloud base height and cloud top height; The meteorological data include: temperature, vertical speed, specific humidity, relative humidity, liquid water content, rain water content, snow water content, ice water content, longitude U wind component, latitude V wind component, month, hour, dew point temperature difference and altitude.

5. The particle cloud macroscopic parameter quantification method based on machine learning according to claim 1, characterized in that: The step of performing variable screening on the eliminated data to obtain meteorological variables, ground-based cloud macro-parameter variables, and satellite cloud macro-parameter variables includes: Calculate the correlation coefficient and information entropy between the eliminated meteorological data and the eliminated ground-based data set, and obtain the ground-based correlation coefficient and ground-based information entropy accordingly; Calculate the correlation coefficient and information entropy between the eliminated meteorological data and the eliminated satellite data set, and obtain the satellite correlation coefficient and satellite information entropy accordingly; The eliminated data are subjected to variable screening through the ground-based correlation coefficient, the ground-based information entropy, the satellite correlation coefficient and the satellite information entropy to obtain the meteorological variables, the ground-based cloud macro-parameter variables and the satellite cloud macro-parameter variables.

6. The particle cloud macroscopic parameter quantification method based on machine learning according to claim 5, characterized in that: The step of performing variable screening on the eliminated data by using the ground-based correlation coefficient, the ground-based information entropy, the satellite correlation coefficient, and the satellite information entropy to obtain the meteorological variables, the ground-based cloud macro-parameter variables, and the satellite cloud macro-parameter variables includes: The ground-based information entropy and the satellite information entropy are sorted by size to obtain an information entropy sorting result; The ground-based correlation coefficient and the satellite correlation coefficient are greater than 0.3, and the eliminated data in the top 75% of the information entropy sorting results are screened out to obtain the meteorological variables, the ground-based cloud macro-parameter variables, and the satellite cloud macro-parameter variables.

7. The particle cloud macroscopic parameter quantification method based on machine learning according to claim 1, characterized in that: The particle cloud macro-parameter initial model includes: the particle cloud macro-parameter cloud height initial model and the particle cloud macro-parameter cloud amount initial model; The particle cloud macro-parameter cloud height initial model adopts the UNet++ architecture; the particle cloud macro-parameter cloud amount initial model adopts the LGBM architecture, and the classifier of the LGBM architecture is GBDT; The method of training the particle cloud macro parameter initial model by using the meteorological variables, the ground-based cloud macro parameter variables and the satellite cloud macro parameter variables to obtain a trained particle cloud macro parameter model comprises: Taking the meteorological variables as input values ​​and the cloud heights of the ground-based cloud macro-parameter variables and the satellite cloud macro-parameter variables as true values, the particle cloud macro-parameter cloud height initial model is trained, and the particle cloud macro-parameter cloud height initial model that meets the preset convergence conditions is used as the trained particle cloud macro-parameter cloud height model; Taking the meteorological variables as input values ​​and the cloud amounts of the ground-based cloud macro-parameter variables and the satellite cloud macro-parameter variables as true values, the particle cloud macro-parameter cloud amount initial model is trained, and the particle cloud macro-parameter cloud amount initial model that meets the preset convergence conditions is used as the trained particle cloud macro-parameter cloud amount model; The trained particle cloud macro-parameter cloud height model and the trained particle cloud macro-parameter cloud amount model are used together as the trained particle cloud macro-parameter model; Among them, the initial model of the cloud height of the particle cloud macroscopic parameter adopts Dice loss; the initial model of the cloud amount of the particle cloud macroscopic parameter adopts L2 norm loss.

8. The particle cloud macroscopic parameter quantification method based on machine learning according to claim 7 is characterized in that: The preset convergence condition includes: the number of iterations is greater than a preset iteration threshold, or the value of the loss function is continuously less than a preset loss threshold.

9. A particle cloud macroscopic parameter quantification device based on machine learning, characterized in that: The particle cloud macro parameter quantification device based on machine learning includes: an acquisition unit, a matching unit, a data elimination unit, a variable screening unit, a training unit and a forecasting unit; The acquisition unit is used to: acquire ground-based data, satellite data and meteorological data; the ground-based data and the satellite data are both particle cloud data; The matching unit is used to: perform spatiotemporal matching processing on the ground-based data and the satellite data based on the meteorological data to obtain a matched ground-based data set and a matched satellite data set; The data elimination unit is used to eliminate abnormal data from the matched ground-based data set, the matched satellite data set and the meteorological data to obtain eliminated data; the eliminated data includes: the eliminated ground-based data set, the eliminated satellite data set and the eliminated meteorological data; The variable screening unit is used to: perform variable screening on the eliminated data to obtain meteorological variables, ground-based cloud macro-parameter variables and satellite cloud macro-parameter variables; The training unit is used to: train the particle cloud macro parameter initial model using the meteorological variables, the ground-based cloud macro parameter variables and the satellite cloud macro parameter variables to obtain a trained particle cloud macro parameter model; The forecasting unit is used to obtain the meteorological data to be processed, and generate the particle cloud macro-parameter quantification result based on the meteorological data to be processed and the trained particle cloud macro-parameter model.

10. A particle cloud macroscopic parameter quantification device based on machine learning, characterized in that: include: A processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor. When the device for quantifying particle cloud macro parameters based on machine learning is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the method for quantifying particle cloud macro parameters based on machine learning as described in any one of claims 1 to 8.

Images