Optimization method and device of cloud prediction model

Abstract

The optimization method and device of the cloud prediction model provided by the invention comprises the following steps: realizing accurate identification of the cloud and obtaining the liquid water content of the same height layer by constructing a cloud liquid water content prediction model and a cloud boundary prediction model, and obtaining the liquid water content of the same height layer according to the cloud bottom height and the cloud top height of the same height layer and the cloud liquid water content of the same height layer. And carrying out inversion to obtain a liquid water path, and if the liquid water path does not meet a preset condition, respectively optimizing the cloud liquid water content prediction model and the cloud boundary prediction model so as to enable the liquid water path to meet the preset condition. And based on the liquid water path, verifying the accuracy of the cloud liquid water content obtained by the cloud liquid water content prediction model and the cloud recognition accuracy of the cloud boundary prediction model, and optimizing the cloud liquid water content prediction model and the cloud boundary prediction model. The accuracy of obtaining the water content of the cloud liquid by the cloud liquid water content prediction model and the accuracy of identifying the cloud by the cloud boundary prediction model are improved.

Description

Technical Field

[0001] The present application relates to the technical field of prediction models, and in particular to a method and device for optimizing a cloud prediction model. Background Art

[0002] So far, clouds are still one of the main uncertainties in atmospheric models. They play an important role in the propagation attenuation of microwave signals in remote sensing research. Even thin clouds with less liquid water content can have a large impact on radiation transmission. The important role of clouds in the earth's energy balance comes mostly from the liquid water content in them, and their feedback has a very important impact on global climate change. Therefore, it is extremely important to accurately obtain the physical quantity of liquid water in clouds.

[0003] At present, real observation data such as sounding are used to train neural networks to obtain prediction models. Unlike physical parameters such as temperature, humidity, and water vapor, liquid water content cannot be directly read from sounding data. It is generally estimated by cloud models. Therefore, the accuracy of cloud models is crucial to the inversion results. In addition, cloud models will have a certain impact on the inversion accuracy of liquid water paths, water vapor, and humidity profiles. Microwave radiometers have been deployed in large numbers in China. Improving their liquid water content detection accuracy can make up for the high cost of in-situ aircraft measurements and the lack of data acquisition. It is of great significance to the establishment and correction of three-dimensional cloud fields. Therefore, it is necessary to establish a more accurate cloud model to improve the inversion accuracy of radiometers.

[0004] The cloud model includes two parts: one is to identify clouds based on certain conditions, that is, cloud boundary identification, and the other is to calculate the cloud liquid water content on the vertical profile of the atmosphere. Since the appearance of clouds is particularly sensitive to temperature and humidity conditions, the cloud models in the existing technology all use relative humidity thresholds (or combined with temperature) to identify clouds, but this requires that the sounding measurement of temperature and humidity has a high degree of accuracy. In fact, different regions use different models of sounding instruments, and even different sounding instruments have been used in the same region in history. This makes the relative humidity recorded by the sounding different in different regions and at different times. These differences will inevitably affect cloud identification. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiments of the present application provide a method and device for optimizing a cloud prediction model, which solves the problem that the cloud model in the prior art cannot accurately identify clouds and calculate the accurate cloud liquid water content.

[0006] According to one aspect of the present application, a method for optimizing a cloud prediction model is provided, comprising:

[0007] Acquire training samples; wherein the training samples include a training data set, the training data set includes a plurality of historical observation data, each of the historical observation data includes a training cloud liquid water content and its corresponding relative humidity, temperature, dew point, first cloud thickness and second cloud thickness, the first cloud thickness is the height difference between the cloud top height and the cloud bottom height, the second cloud thickness is the height difference between the aircraft height and the cloud bottom height;

[0008] Based on the BP neural network algorithm, the plurality of training cloud liquid water contents and their corresponding relative humidity, temperature, dew point, first cloud thickness and second cloud thickness are trained to construct a cloud liquid water content prediction model;

[0009] Constructing a cloud boundary prediction model; wherein the cloud boundary prediction model includes cloud probability values ​​of multiple altitude layers and their corresponding relative humidity, temperature, altitude, air pressure and altitude air pressure;

[0010] Inputting the relative humidity, temperature, altitude, air pressure and altitude air pressure of the same altitude layer into the cloud boundary prediction model to determine the cloud base height and cloud top height when there are clouds in the same altitude layer;

[0011] Inputting the relative humidity, temperature, dew point, first cloud thickness and second cloud thickness of the same altitude layer into the cloud liquid water content prediction model to obtain the cloud liquid water content of the same altitude layer;

[0012] Inverting the liquid water path according to the cloud base height and cloud top height of the same altitude layer and the cloud liquid water content of the same altitude layer;

[0013] If the liquid water path does not meet the preset conditions, the cloud liquid water content prediction model and the cloud boundary prediction model are optimized respectively to make the liquid water path meet the preset conditions.

[0014] In one embodiment, after the plurality of training cloud liquid water contents and their corresponding relative humidity, temperature, dew point, first cloud thickness, and second cloud thickness are trained based on the BP neural network algorithm to construct a cloud liquid water content prediction model, the cloud prediction model optimization method further includes:

[0015] Acquire a verification data set in the training sample; wherein the verification data set includes verification humidity, verification temperature, verification dew point, first verification cloud thickness, second verification cloud thickness and real cloud liquid water content;

[0016] Inputting the verification humidity, verification temperature, verification dew point, first verification cloud thickness and second verification cloud thickness in the verification data set into the cloud liquid water content prediction model to obtain the verification cloud liquid water content of the verification data set;

[0017] Determining, according to the verification cloud liquid water content of the verification data set and the actual cloud liquid water content, a degree of deviation between the verification cloud liquid water content of the verification data set and the actual cloud liquid water content;

[0018] Based on the degree of deviation, the cloud liquid water content prediction model is optimized.

[0019] In one embodiment, determining the degree of deviation between the verified cloud liquid water content of the verification data set and the actual cloud liquid water content according to the verified cloud liquid water content of the verification data set comprises:

[0020] Calculating a root mean square error based on the verification cloud liquid water content of the verification data set and the true cloud liquid water content;

[0021] Wherein, optimizing the cloud liquid water content prediction model based on the degree of deviation includes:

[0022] If the mean root mean error is greater than a first preset error threshold, the cloud liquid water content prediction model is adjusted so that the mean root mean error between the verified cloud liquid water content predicted by the cloud liquid water content prediction model and the actual cloud liquid water content is less than the first preset error threshold.

[0023] In one embodiment, determining the degree of deviation between the verified cloud liquid water content of the verification data set and the actual cloud liquid water content according to the verified cloud liquid water content of the verification data set comprises:

[0024] Calculating a mean absolute error based on the verification cloud liquid water content of the verification data set and the true cloud liquid water content;

[0025] Wherein, optimizing the cloud liquid water content prediction model based on the degree of deviation includes:

[0026] If the mean absolute error is less than a second preset error threshold, the cloud liquid water content prediction model is adjusted so that the mean absolute error between the verified cloud liquid water content predicted by the cloud liquid water content prediction model and the actual cloud liquid water content is less than the second preset error threshold.

[0027] In one embodiment, constructing a cloud boundary prediction model includes:

[0028] Obtain cloud boundary training samples; wherein the cloud boundary training samples include a cloud boundary training set, the cloud boundary training set includes a cloud training data set and a cloud-free training data set, the cloud training data set includes cloud probability values ​​of multiple training altitude layers and their corresponding relative humidity, temperature, altitude, air pressure and ground pressure, and the cloud-free training data set includes cloud-free probability values ​​of the multiple training altitude layers and their corresponding relative humidity, temperature, altitude, air pressure and ground pressure;

[0029] Based on the MLP multi-layer perceptron, the probability values ​​of the presence of clouds and the corresponding relative humidity, temperature, altitude, air pressure and ground pressure of the multiple training altitude layers and the probability values ​​of the absence of clouds and the corresponding relative humidity, temperature, altitude, air pressure and ground pressure of the multiple training altitude layers are trained to build a cloud boundary prediction model.

[0030] In one embodiment, if the liquid water path does not meet a preset condition, respectively optimizing the cloud liquid water content prediction model and the cloud boundary prediction model includes:

[0031] Based on the microwave radiation transmission equation, the brightness temperature values ​​of the four channels are calculated according to the observation data detected by the airborne microwave radiometer at the same altitude layer;

[0032] Based on the BP neural network, the inverted liquid water path is calculated according to the brightness temperature values ​​of the four channels, the lowest aircraft flight altitude at the same altitude layer, the temperature corresponding to the lowest aircraft flight altitude, and the liquid water content at the same altitude layer;

[0033] According to the inverted liquid water path, constructing a preset condition;

[0034] If the liquid water path does not meet the preset condition, the cloud liquid water content prediction model and the cloud boundary prediction model are optimized respectively.

[0035] In one embodiment, constructing a preset condition according to the inverted liquid water path includes:

[0036] Obtaining a real liquid water path at the same altitude layer;

[0037] calculating a first deviation between the inverted liquid water path and the true liquid water path;

[0038] Wherein, if the liquid water path does not meet the preset conditions, respectively optimizing the cloud liquid water content prediction model and the cloud boundary prediction model includes:

[0039] calculating a second deviation between the liquid water path and the actual liquid water path;

[0040] If the second deviation is greater than the first deviation, the cloud liquid water content prediction model and the cloud boundary prediction model are optimized respectively.

[0041] In one embodiment, constructing a preset condition according to the inverted liquid water path includes:

[0042] Calculate the reference liquid water path at the same altitude layer based on the vertical cumulative liquid water formula;

[0043] calculating a third deviation between the inverted liquid water path and the reference liquid water path;

[0044] Wherein, if the liquid water path does not meet the preset conditions, respectively optimizing the cloud liquid water content prediction model and the cloud boundary prediction model includes:

[0045] calculating a fourth deviation between the liquid water path and the reference liquid water path;

[0046] If the fourth deviation is greater than the third deviation, the cloud liquid water content prediction model and the cloud boundary prediction model are optimized respectively.

[0047] In one embodiment, after the cloud probability values ​​of the plurality of test altitude layers and their corresponding relative humidity, temperature, altitude, air pressure and ground pressure and the cloud-free probability values ​​of the plurality of test altitude layers and their corresponding relative humidity, temperature, altitude, air pressure and ground pressure are trained based on the MLP multilayer perceptron to construct a cloud boundary prediction model, the cloud prediction model optimization method further includes:

[0048] Obtain a cloud boundary test set from the cloud boundary training samples; wherein the cloud boundary test set includes a cloud test data set, and the cloud test data set includes relative humidity, temperature, altitude, air pressure and ground air pressure of multiple test altitude layers;

[0049] Inputting the relative humidity, temperature, altitude, air pressure and ground pressure corresponding to the plurality of test altitude layers into the cloud boundary prediction model to obtain a test probability value of cloudiness at each test altitude layer;

[0050] Selecting a cloud-presence test probability value greater than a preset probability value as a target test probability value; obtaining a target cloud base height and a target cloud top height corresponding to the target test probability value;

[0051] Calculating the cloud thickness corresponding to the target test probability value according to the target cloud base height and the target cloud top height;

[0052] Acquire ground-based cloud radar data corresponding to the target test probability value at the same time;

[0053] Read the cloud base height and cloud top height of the ground-based cloud radar data;

[0054] Calculating the cloud thickness corresponding to the ground-based cloud radar data according to the cloud base height and cloud top height of the ground-based cloud radar data;

[0055] If the deviation between the cloud thickness corresponding to the target test probability value and the cloud thickness corresponding to the ground-based cloud radar data is greater than a preset deviation threshold, the cloud boundary prediction model is optimized.

[0056] According to another aspect of the present application, there is provided a cloud prediction model optimization device, comprising:

[0057] An acquisition module is used to acquire training samples; wherein the training samples include a training data set, and the training data set includes a plurality of historical observation data, each of which includes a training cloud liquid water content and its corresponding relative humidity, temperature, dew point, first cloud thickness, and second cloud thickness, wherein the first cloud thickness is the height difference between the cloud top height and the cloud bottom height, and the second cloud thickness is the height difference between the aircraft height and the cloud bottom height;

[0058] A training module, for training the plurality of training cloud liquid water contents and their corresponding relative humidity, temperature, dew point, first cloud thickness and second cloud thickness based on a BP neural network algorithm, so as to construct a cloud liquid water content prediction model;

[0059] A construction module is used to construct a cloud boundary prediction model; wherein the cloud boundary prediction model includes cloud probability values ​​of multiple altitude layers and their corresponding relative humidity, temperature, altitude, air pressure and altitude air pressure;

[0060] A first input module is used to input the relative humidity, temperature, altitude, air pressure and altitude air pressure of the same altitude layer into the cloud boundary prediction model to determine the cloud base height and cloud top height when there are clouds in the same altitude layer;

[0061] A second input module is used to input the relative humidity, temperature, dew point, first cloud thickness and second cloud thickness of the same altitude layer into the cloud liquid water content prediction model to obtain the cloud liquid water content of the same altitude layer;

[0062] An inversion module, used for inverting a liquid water path according to the cloud base height and cloud top height of the same altitude layer and the cloud liquid water content of the same altitude layer;

[0063] The optimization module is used for respectively optimizing the cloud liquid water content prediction model and the cloud boundary prediction model if the liquid water path does not meet the preset conditions, so that the liquid water path meets the preset conditions.

[0064] The optimization method and device of the cloud prediction model provided by the present application include: obtaining training samples, wherein the training samples include a training data set, the training data set includes a plurality of historical observation data, each of the historical observation data includes a training cloud liquid water content and its corresponding relative humidity, temperature, dew point, first cloud thickness and second cloud thickness, the first cloud thickness is the height difference between the cloud top height and the cloud bottom height, and the second cloud thickness is the height difference between the aircraft height and the cloud bottom height, based on the BP neural network algorithm, the plurality of training cloud liquid water contents and their corresponding relative humidity, temperature, dew point, first cloud thickness and second cloud thickness are trained to construct a cloud liquid water content prediction model, and a cloud boundary prediction model is constructed, wherein the cloud boundary prediction model includes multiple The cloud probability value of each altitude layer and its corresponding relative humidity, temperature, altitude, air pressure and altitude pressure are input into the cloud boundary prediction model to determine the cloud base height and cloud top height when there are clouds at the same altitude layer. The relative humidity, temperature, dew point, first cloud thickness and second cloud thickness of the same altitude layer are input into the cloud liquid water content prediction model to obtain the cloud liquid water content of the same altitude layer. According to the cloud base height and cloud top height of the same altitude layer and the cloud liquid water content of the same altitude layer, the liquid water path is inverted. If the liquid water path does not meet the preset conditions, the cloud liquid water content prediction model and the cloud boundary prediction model are optimized respectively to make the liquid water path meet the preset conditions. By constructing a cloud liquid water content prediction model and a cloud boundary prediction model, clouds can be identified and the liquid water content of different altitude layers can be obtained. Then, the liquid water path is inverted through the cloud liquid water content prediction model and the cloud boundary prediction model. Based on the liquid water path, the accuracy of the cloud liquid water content prediction model and the cloud boundary prediction model in identifying clouds and obtaining the liquid water content of different altitude layers is verified. The cloud liquid water content prediction model and the cloud boundary prediction model are optimized to improve the accuracy of the cloud liquid water content prediction model in obtaining the cloud liquid water content and the accuracy of the cloud boundary prediction model in identifying clouds. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0066] Figure 1 It is a flowchart of a cloud prediction model optimization method provided by an exemplary embodiment of the present application.

[0067] Figure 2It is a flowchart of a method for constructing a cloud boundary prediction model provided by an exemplary embodiment of the present application.

[0068] Figure 3 It is a structural diagram of an optimization device for a cloud prediction model provided by an exemplary embodiment of the present application.

[0069] Figure 4 It is a structural diagram of an optimization device for a cloud prediction model provided by another exemplary embodiment of the present application.

[0070] Figure 5 It is a structural diagram of an electronic device provided by an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0071] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0072] Figure 1 FIG. 1 is a flow chart of a method for optimizing a cloud prediction model provided by an exemplary embodiment of the present application. Figure 1 As shown, the optimization method of the cloud prediction model includes:

[0073] Step 110: Obtain training samples, wherein the training samples include a training data set, the training data set includes a plurality of historical observation data, each of the historical observation data includes a training cloud liquid water content and its corresponding relative humidity, temperature, dew point, first cloud thickness and second cloud thickness, the first cloud thickness is the height difference between the cloud top height and the cloud bottom height, and the second cloud thickness is the height difference between the aircraft height and the cloud bottom height.

[0074] In an embodiment of the present application, in order to improve the applicability of the cloud liquid water content prediction model, it is necessary to construct a training sample to train the cloud liquid water content prediction model. The training sample includes a training data set and a validation data set. The training data set includes multiple historical observation data. The method for obtaining the historical observation data includes: obtaining historical observation data of Tianjin rain-making aircraft King Air B-10KL, Beijing rain-making aircraft, and Shanxi rain-making aircraft, screening historical observation cases with clouds but no precipitation from the historical observation data, reading the cloud liquid water content (LWC) and the corresponding relative humidity, temperature, dew point, cloud thickness, and distance to cloud base height from each observation case, thereby constructing a training sample based on the cloud liquid water content (LWC), and the corresponding relative humidity, temperature, dew point, cloud thickness, and distance to cloud base height.

[0075] Among them, the CDP cloud particle probe is used to measure cloud particles with a scale range of 2-50 microns to calculate the liquid water content. The cloud particles can be quality controlled by two-dimensional particle images, including: observing the morphological characteristics of cloud particles on two-dimensional images, such as size, shape, edge clarity, etc., to preliminarily judge their quality. Particles with abnormal shapes or blurs may need to be excluded or marked. The density distribution of cloud particles in the image is analyzed to check whether there are abnormal conditions, such as too dense or too sparse areas, and whether there are obvious missing or overlapping phenomena. When CDP is missing, a hot wire water content meter (Hotwire-LWC) can be used to measure the water vapor content in the atmosphere as supplementary data. The cloud top height and cloud bottom height data are read by KPR, and the difference between the cloud top height and the cloud bottom height is calculated to obtain the first cloud thickness. The aircraft altitude data is read by AIMMS-20, and the cloud bottom height data is read by KPR. The difference between the aircraft altitude and the cloud bottom altitude is calculated to obtain the second cloud thickness.

[0076] Step 120: Based on the BP neural network algorithm, multiple training cloud liquid water contents and their corresponding relative humidity, temperature, dew point, first cloud thickness and second cloud thickness are trained to construct a cloud liquid water content prediction model.

[0077] In an embodiment of the present application, based on the BP neural network algorithm, and through relative humidity, temperature, dew point, first cloud thickness, and second cloud thickness, the cloud liquid water content value can be accurately predicted. This prediction belongs to regression prediction. Specifically, the training sample set data is preprocessed to check the distribution of the data set and whether there is data bias. It can be checked by drawing histograms and scatter plots of the characteristic parameter values. If the data is unevenly distributed, the data is normalized. The data set is randomly divided into a training set and a validation data set, and the ratio is set to 7:3. Determine the network structure, such as the number of input layer nodes is set according to the number of features, the hidden layer has one or more nodes, and the number of nodes needs to be adjusted to determine the parameter, and 1 node in the output layer corresponds to the cloud liquid water content; select the activation function, loss function, and optimization algorithm; set hyperparameters, such as learning rate, number of iterations, etc., which require multiple rounds of experiments to find the optimal value. After each round of iteration, calculate the change curve of the loss function on the training set to observe whether the loss decreases smoothly; evaluate the accuracy on the validation set every few rounds to find the direction of parameter adjustment; appropriately reduce the learning rate to avoid overfitting; increase the number of hidden layer nodes to increase the complexity of the model; try different values ​​for the more influential hyperparameters multiple times to find the optimal value.

[0078] Step 130: Construct a cloud boundary prediction model, wherein the cloud boundary prediction model includes cloud probability values ​​of multiple altitude layers and their corresponding relative humidity, temperature, altitude, air pressure and altitude air pressure.

[0079] Step 140: Inputting the relative humidity, temperature, altitude, air pressure and altitude air pressure of the same altitude layer into the cloud boundary prediction model to determine the cloud base height and cloud top height when there are clouds at the same altitude layer.

[0080] Step 150: inputting the relative humidity, temperature, dew point, first cloud thickness and second cloud thickness of the same altitude layer into a cloud liquid water content prediction model to obtain the cloud liquid water content of the same altitude layer.

[0081] Step 160: Invert and obtain the liquid water path according to the cloud base height and cloud top height of the same altitude layer and the cloud liquid water content of the same altitude layer.

[0082] Step 170: If the liquid water path does not meet the preset conditions, the cloud liquid water content prediction model and the cloud boundary prediction model are optimized respectively to make the liquid water path meet the preset conditions.

[0083] The optimization method and device of the cloud prediction model provided by the present application include: obtaining training samples, wherein the training samples include a training data set, the training data set includes a plurality of historical observation data, each of the historical observation data includes a training cloud liquid water content and its corresponding relative humidity, temperature, dew point, first cloud thickness and second cloud thickness, the first cloud thickness is the height difference between the cloud top height and the cloud bottom height, and the second cloud thickness is the height difference between the aircraft height and the cloud bottom height, based on the BP neural network algorithm, the plurality of training cloud liquid water contents and their corresponding relative humidity, temperature, dew point, first cloud thickness and second cloud thickness are trained to construct a cloud liquid water content prediction model, and a cloud boundary prediction model is constructed, wherein the cloud boundary prediction model includes multiple The cloud probability value of each altitude layer and its corresponding relative humidity, temperature, altitude, air pressure and altitude pressure are input into the cloud boundary prediction model to determine the cloud base height and cloud top height when there are clouds at the same altitude layer. The relative humidity, temperature, dew point, first cloud thickness and second cloud thickness of the same altitude layer are input into the cloud liquid water content prediction model to obtain the cloud liquid water content of the same altitude layer. According to the cloud base height and cloud top height of the same altitude layer and the cloud liquid water content of the same altitude layer, the liquid water path is inverted. If the liquid water path does not meet the preset conditions, the cloud liquid water content prediction model and the cloud boundary prediction model are optimized respectively to make the liquid water path meet the preset conditions. By constructing a cloud liquid water content prediction model and a cloud boundary prediction model, clouds can be identified and the liquid water content of different altitude layers can be obtained. Then, the liquid water path is inverted through the cloud liquid water content prediction model and the cloud boundary prediction model. Based on the liquid water path, the accuracy of the cloud liquid water content prediction model and the cloud boundary prediction model in identifying clouds and obtaining the liquid water content of different altitude layers is verified. The cloud liquid water content prediction model and the cloud boundary prediction model are optimized to improve the accuracy of the cloud liquid water content prediction model in obtaining the cloud liquid water content and the accuracy of the cloud boundary prediction model in identifying clouds.

[0084] In one embodiment, after step 170, the optimization method of the cloud prediction model can be specifically implemented as follows: obtaining a verification data set in the training sample; wherein the verification data set includes a verification humidity, a verification temperature, a verification dew point, a first verification cloud thickness, a second verification cloud thickness and a real cloud liquid water content; inputting the verification humidity, the verification temperature, the verification dew point, the first verification cloud thickness and the second verification cloud thickness in the verification data set into the cloud liquid water content prediction model to obtain the verification cloud liquid water content of the verification data; determining the degree of deviation between the verification cloud liquid water content of the verification data set and the real cloud liquid water content according to the verification cloud liquid water content of the verification data set and the real cloud liquid water content; and optimizing the cloud liquid water content prediction model based on the degree of deviation.

[0085] In the embodiment of the present application, the optimal model learned is evaluated on the validation data set, and the accuracy and confusion matrix results are observed. If the performance is not ideal, consider increasing the number of hidden layers to expand the model capabilities, adding Dropout to prevent overfitting, transforming some key features, and using ensemble learning methods such as Dropout+BN to improve performance.

[0086] Specifically, the validation data set can be the historical observation data of the King Air B-10KL, Beijing rainmaking aircraft, and Shanxi rainmaking aircraft. The historical observation cases with clouds but no precipitation are selected from the historical observation data. The cloud liquid water content (LWC) and the corresponding relative humidity, temperature, dew point, cloud thickness, and distance to the cloud base height are read from each observation case. Thus, according to the cloud liquid water content (LWC), and the corresponding relative humidity, temperature, dew point, cloud thickness, and distance to the cloud base height, a training sample is constructed, and a validation data set is selected from the training sample. The validation humidity, validation temperature, validation dew point, first validation cloud thickness, and second validation cloud thickness in the validation data set are input into the cloud liquid water content prediction model to obtain the validation cloud liquid water content of the validation data set. Then, the validation cloud liquid water content is compared with the actual cloud liquid water content in the validation data set, and the deviation between the two is determined. If the deviation is large, it means that the cloud liquid water content prediction model predicts the cloud liquid water content inaccurately, and the predicted cloud liquid water content needs to be optimized.

[0087] In one embodiment, the optimization method of the cloud prediction model can be specifically implemented as follows: calculating the root mean square error based on the verified cloud liquid water content and the actual cloud liquid water content of the verification data set; if the root mean square error is greater than a first preset error threshold, adjusting the cloud liquid water content prediction model so that the root mean square error between the verified cloud liquid water content predicted by the cloud liquid water content prediction model and the actual cloud liquid water content is less than the first preset error threshold.

[0088] In the embodiment of the present application, the root mean square error can be calculated based on the verification cloud liquid water content of the verification data set and the actual cloud liquid water content to verify the accuracy of the cloud liquid water content predicted by the cloud liquid water content prediction model. The calculation formula of the root mean square error is:

[0089] x(i) represents the verified cloud liquid water content input to the cloud liquid water content prediction model, y(i) represents the true cloud liquid water content, m represents the number of samples, and i represents the i-th sample. The root mean square error refers to the square root of the ratio of the sum of the squares of the deviations between the calculated value and the true value to the number of samples.

[0090] In one embodiment, the optimization method of the cloud prediction model can be specifically implemented as follows: calculating the mean absolute error based on the verified cloud liquid water content and the actual cloud liquid water content of the verification data set; if the mean absolute error is less than a second preset error threshold, adjusting the cloud liquid water content prediction model so that the mean absolute error between the verified cloud liquid water content predicted by the cloud liquid water content prediction model and the actual cloud liquid water content is less than the second preset error threshold.

[0091] In the embodiment of the present application, the average absolute error can be calculated by comparing the verification cloud liquid water content of the verification data set with the actual cloud liquid water content to verify the accuracy of the cloud liquid water content prediction model. The calculation formula of the average absolute error is:

[0092] Where x(i) represents the verified cloud liquid water content input to the cloud liquid water content prediction model, y(i) represents the true cloud liquid water content, m represents the number of samples, and i represents the i-th sample. The mean absolute error refers to the average of the absolute values ​​of the errors between the calculated value and the true value.

[0093] Figure 2 FIG. 1 is a flow chart of a method for constructing a cloud boundary prediction model provided by an exemplary embodiment of the present application. Figure 2 As shown, step 130 may include:

[0094] Step 131: Obtain cloud boundary training samples, wherein the cloud boundary training samples include a cloud boundary training set, the cloud boundary training set includes a cloud training data set and a cloud-free training data set, the cloud training data set includes cloud probability values ​​of multiple training altitude layers and their corresponding relative humidity, temperature, altitude, air pressure and ground pressure, and the cloud-free training data set includes cloud-free probability values ​​of multiple training altitude layers and their corresponding relative humidity, temperature, altitude, air pressure and ground pressure.

[0095] In an embodiment of the present application, continuous observation data of the Ka-band millimeter wave (35GHz) cloud radar of Beijing Nanjiao Observatory (BNWO, 39.81°N, 116.47°E, 32m above sea level) from 2022 to 2024 and sounding data of the Beijing Observatory at two times a day are collected as cloud boundary training samples for the cloud boundary prediction model. Since the continuous release time of the sounding is 90 minutes, the cloud radar observes the vertical structure of the cloud at a certain moment. There are temporal and spatial differences between the two. In order to reduce the impact of the difference, first, the time for the sounding to reach a specific height is estimated. The cloud radar at the corresponding height uses the detection data at the corresponding time, supplemented by satellite observation data, to remove data that does not belong to the same cloud system. The sounding balloons are launched at 07:15 and 19:15 (Beijing time) every day. In order to facilitate the processing of temporal and spatial differences, the observation data of the cloud radar in the periods of 07:00-09:00 and 19:00-21:00 are selected for analysis. During these two periods, the sounding data of relative humidity, temperature, altitude, air pressure and ground pressure of each altitude layer were obtained based on the cloud radar data with clouds but no precipitation. At the same time, the information of the cloud radar data at the corresponding time of each altitude layer was extracted to record whether there were clouds at that altitude at that time, so as to construct two sets of training sample data sets, one with clouds and the other without clouds. Such a data processing process helps to ensure the accuracy and consistency of the data and establish reliable training samples for subsequent analysis.

[0096] Step 132: Based on the MLP multi-layer perceptron, the probability values ​​of clouds in multiple training altitude layers and their corresponding relative humidity, temperature, altitude, air pressure and ground pressure, as well as the probability values ​​of cloudlessness in multiple training altitude layers and their corresponding relative humidity, temperature, altitude, air pressure and ground pressure are trained to build a cloud boundary prediction model.

[0097] In an embodiment of the present application, the probability values ​​of clouds at multiple training altitude layers and their corresponding relative humidity, temperature, altitude, air pressure and ground pressure, as well as the probability values ​​of cloudlessness at multiple training altitude layers and their corresponding relative humidity, temperature, altitude, air pressure and ground pressure are trained by an MLP multilayer perceptron to construct a cloud boundary prediction model. The input of the MLP multilayer perceptron is defined as X={x1, x2, x3, x4, x5}, where xi represents the five features of humidity, temperature, altitude, air pressure and ground pressure, respectively, and the output category of the MLP multilayer perceptron is 2, i.e., the probability of clouds and clouds.

[0098] Specifically, the distribution of cloud boundary training samples is analyzed and data is standardized. The cloud boundary training samples are randomly divided into cloud boundary training set and cloud boundary test set, with a ratio of 7:3. The input layer of the MLP multi-layer perceptron is set to 5 nodes, namely humidity, temperature, altitude, air pressure, and ground air pressure, and the output layer is set to 2 nodes, corresponding to cloud and no cloud, respectively. There are one or more hidden layers, and the number of nodes is determined according to the parameter adjustment. Select the activation function, loss function, and optimization algorithm, and set the hyperparameters to find the best value. After each round of iteration, calculate the change curve of the loss function on the training set to observe whether the loss decrease trend is smooth; evaluate the accuracy on the validation set every few rounds to find the parameter adjustment direction; appropriately reduce the learning rate to avoid overfitting; increase the hidden layer nodes to increase the complexity of the model; try different values ​​for the hyperparameters with greater influence multiple times to find the best. Evaluate the optimal model learned on the cloud edge test set, and observe the accuracy and confusion matrix results; use the following indicators to evaluate the model: Accuracy: the ratio of the number of correctly predicted samples in the test set to the total number of samples, which directly reflects the classification ability of the model; Confusion matrix: observe the prediction results of various types of samples to determine whether there is classification bias; ROC curve and AUC value: evaluate the discrimination ability of the model under different thresholds, and the AUC value reflects the overall classification quality.

[0099] In one embodiment, after step 132, the optimization method of the cloud prediction model can be specifically implemented as follows: obtaining a cloud boundary test set in the cloud boundary training sample; wherein the cloud boundary test set includes a cloud test data set, and the cloud test data set includes relative humidity, temperature, altitude, air pressure and ground air pressure of multiple test altitude layers; inputting the corresponding relative humidity, temperature, altitude, air pressure and ground air pressure of the multiple test altitude layers into the cloud boundary prediction model to obtain a cloud test probability value for each test altitude layer; selecting a cloud test probability value greater than a preset probability value as a target test probability value; obtaining a target cloud base height and a target cloud top height corresponding to the target test probability value; calculating a cloud thickness corresponding to the target test probability value according to the target cloud base height and the target cloud top height; obtaining ground-based cloud radar data corresponding to the target test probability value at the same time; reading the cloud base height and the cloud top height of the ground-based cloud radar data; calculating the cloud thickness corresponding to the ground-based cloud radar data according to the cloud base height and the cloud top height of the ground-based cloud radar data; if the deviation between the cloud thickness corresponding to the target test probability value and the cloud thickness corresponding to the ground-based cloud radar data is greater than a preset deviation threshold, the cloud boundary prediction model is optimized.

[0100] In an embodiment of the present application, the cloud boundary test set can be the 3-month sounding data of Beijing South Suburb Observatory (54511), and the 3-month sounding data is a cloud test data set. The corresponding relative humidity, temperature, altitude, air pressure and ground air pressure of each test altitude layer are input into the cloud boundary prediction model to obtain the test probability value of clouds at each test altitude layer, and the test probability value of clouds is selected as the target test probability value, that is, the data with clouds is selected. The ground-based cloud radar data of the same station of Beijing South Suburb Observatory at the same time of the target test probability value, the cloud base height and cloud top height of the ground-based cloud radar data are read and the cloud thickness corresponding to the ground-based cloud radar data is calculated, and the cloud thickness corresponding to the target test probability value is calculated according to the target cloud base height and the target cloud top height. If the deviation between the cloud thickness corresponding to the target test probability value and the corresponding cloud thickness of the ground-based cloud radar data is greater than the preset deviation threshold, the cloud boundary prediction model is optimized, that is, the cloud boundary prediction model predicts that the probability value of clouds is low, and the cloud boundary prediction model needs to be optimized. The preset deviation threshold is 200 meters. Since the target test probability value can be multiple, the probability value of the cloud boundary model correctly judging the cloud bottom height and cloud top height can be calculated at the end. If the probability value is low, continue to optimize the cloud boundary prediction model.

[0101] In one embodiment, step 170 can be specifically implemented as follows: based on the microwave radiation transmission equation, the brightness temperature values ​​of the four channels are calculated according to the observation data detected by the airborne microwave radiometer at the same altitude layer; based on the BP neural network, the inverted liquid water path is calculated according to the brightness temperature values ​​of the four channels, the lowest cloud base height value at the same altitude layer, the temperature corresponding to the lowest cloud base height value, and the liquid water content at the same altitude layer; based on the inverted liquid water path, the preset conditions are constructed; if the liquid water path does not meet the preset conditions, the cloud liquid water content prediction model and the cloud boundary prediction model are optimized respectively.

[0102] In an embodiment of the present application, a cloud-sky sample data set is obtained; wherein the cloud-sky sample data set includes sounding data at each time, the sounding data includes the minimum aircraft flight altitude and its corresponding cloud top altitude value and cloud bottom altitude value, based on the Decker model, according to the minimum cloud base altitude value at each time and its corresponding cloud top altitude value, the cloud thickness at each time is determined and based on the cloud thickness, the density of cloud water is calculated; wherein cloud water represents cloud solid water and cloud liquid water, according to the functional relationship between the total amount of cloud water and cloud liquid water and cloud solid water, respectively, the density of supercooled water in the cloud is calculated, according to the density of supercooled water in the cloud, a vertical distribution profile of liquid water is established, based on the height integral of the liquid water vertical distribution profile, the vertical accumulated liquid water is calculated to obtain the liquid water content at different altitudes.

[0103] The airborne microwave radiometer conducts observations at the aircraft's flight altitude and receives microwave radiation from above it. The observation method is the sweep frequency method, and the airborne microwave radiation transmission equation can be used:

[0104]

[0105] Among them, T0 (unit: K) represents the cosmic background brightness temperature, which is generally taken as a constant of 2.7K, z0 (unit: m) represents the flight altitude of the aircraft, f (unit: Ghz) represents the operating frequency of the radiometer, and α (unit: K·m-1) represents the sum of the absorption coefficients of water vapor, oxygen and cloud liquid water corresponding to 183GHz. The above microwave radiation transmission equation can be used to calculate the sky brightness temperature values ​​of the four channels received by the microwave radiometer at the aircraft flight altitude. The brightness temperatures of the four channels, the lowest altitude of the cloud sky samples (simulating the flight altitude of the aircraft) and the temperature corresponding to the lowest altitude, and the liquid water content at the same altitude layer are used as input variables, and the liquid water path is used as the output variable. The BP neural network algorithm is used to train all cloud sky samples to establish the liquid water path inversion equation. The above-mentioned same altitude layer includes the maximum altitude of the aircraft flight and the minimum altitude of the aircraft flight.

[0106] In one embodiment, step 170 can be specifically implemented as follows: obtaining a real liquid water path at the same altitude layer; calculating a first deviation between the inverted liquid water path and the real liquid water path; calculating a second deviation between the liquid water path and the real liquid water path; if the second deviation is greater than the first deviation, respectively optimizing the cloud liquid water content prediction model and the cloud boundary prediction model.

[0107] In an embodiment of the present application, generally speaking, the maximum altitude of an aircraft flying in the same altitude layer is the altitude of the aircraft at the cloud top, and the actual liquid water path obtained when the aircraft flies at the cloud top should be 0. Then, a first deviation between the inverted liquid water path and the actual liquid water path is calculated, and a second deviation between the liquid water path and the actual liquid water path is calculated. If the second deviation is greater than the first deviation, the cloud liquid water content prediction model and the cloud boundary prediction model are optimized respectively.

[0108] In one embodiment, step 170 can be specifically implemented as follows: based on the vertical cumulative liquid water formula, calculate the reference liquid water path of the same altitude layer; calculate the third deviation between the inverted liquid water path and the reference liquid water path; calculate the fourth deviation between the liquid water path and the reference liquid water path; if the fourth deviation is greater than the third deviation, optimize the cloud liquid water content prediction model and the cloud boundary prediction model respectively.

[0109] In the embodiment of the present application, a cloud band with no precipitation is selected as the observation area through observation data such as satellites and weather radars. After the aircraft enters the area and reaches the cloud base height, it first performs a level flight at the cloud base, and selects a liquid water area for the next step of detection based on the detection results. Drawing on the vertical cumulative liquid water comparison observation scheme of Pazmany (2007) (comparison of detection results of hot-wire water content meter and airborne microwave radiometer), a zigzag up and down movement is made in the cloud area from the cloud base to the cloud top. Perform another level flight to the cloud top height. During the flight, the detection data of GVR and hot-wire water content meter LWC-100 are collected simultaneously.

[0110] Assuming that the liquid water content detected by the hot wire water content meter LWC-100 at a certain height h is g / cm3, the vertical accumulated liquid water above this height can be calculated using the following formula: Among them, h1 refers to the lowest altitude of the aircraft, and h2 refers to the maximum altitude of the aircraft. The vertical accumulated liquid water above this altitude is the true same altitude layer of the same altitude layer.

[0111] The third deviation between the inverted liquid water path and the reference liquid water path is calculated. The fourth deviation between the liquid water path and the reference liquid water path is calculated. If the fourth deviation is greater than the third deviation, the cloud liquid water content prediction model and the cloud boundary prediction model are optimized respectively. This indicates that the prediction results of the cloud liquid water content prediction model and the cloud boundary prediction model are low, so they need to be optimized.

[0112] In addition, the verification method of the cloud liquid water content prediction model and the cloud boundary prediction model may also include: using the Beijing Nanjiao Observatory (54511) sounding and ground-based microwave radiometer MP3000A observation data for comparative verification. Specifically, the sounding data under the condition of cloud without precipitation at the 54511 site is selected according to the ground-based cloud radar observation data, and the ground-based microwave radiometer observation data of the same period is collected. Since the sounding release time is 90 minutes, for the convenience of comparison, the average brightness temperature observation data of the 90 minutes starting from the sounding release time of the ground-based microwave radiometer is taken as a comparison. That is, the verification data set includes the cloud without precipitation data observed by the ground-based cloud radar and the ground-based microwave radiometer sounding data. The sounding data is the average brightness temperature observation data during the sounding period of the ground-based microwave radiometer, and the sounding period is within 90 minutes when the ground-based microwave radiometer starts sounding. The average value of brightness temperature observation data is the brightness temperature value of water vapor and liquid water channels detected by the ground-based microwave radiometer MP3000A, and the average value of brightness temperature observation data is obtained by averaging the brightness temperature values ​​of water vapor and liquid water channels. The ground-based microwave radiometer MP3000A has 8 channels in the water vapor and liquid water frequency band, and the corresponding frequencies are 22.23GHz, 22.50GHz, 23.03GHz, 23.83GHz, 25.00GHz, 26.23GHz, 28.00GHz, and 30.00GHz. The brightness temperature observation data of these 8 channels within 90 minutes are read respectively, so the average value of brightness temperature observation data includes the average value of each channel.

[0113] The MonoRTM model and sounding data are used to calculate the brightness temperature. The specific calculation method is as follows: the sounding data is processed to obtain the temperature and humidity profile data, the sounding data is input into the cloud boundary prediction model and the cloud liquid water content prediction model to calculate the liquid water profile, the liquid water profile information is added to each height layer of the temperature and humidity profile, and the MonoRTM model is used to simulate and calculate the brightness temperature values ​​of the 8 channels corresponding to the above MP3000A. According to the above method, the brightness temperature simulation calculation values ​​of the 8 channels under multiple cloudy and non-precipitation weather conditions are calculated, and compared with the MP3000A brightness temperature observation values ​​for verification. The evaluation indicators also use the root mean square error RMSE and the mean absolute error MAE.

[0114] Figure 3 Schematic diagram of the structure of an optimization device for a cloud prediction model provided by an exemplary embodiment of the present application. Figure 3As shown, the optimization device of the cloud prediction model includes: an acquisition module, which is used to acquire training samples; wherein the training samples include a training data set, and the training data set includes multiple historical observation data, each of which includes a training cloud liquid water content and its corresponding relative humidity, temperature, dew point, first cloud thickness and second cloud thickness, the first cloud thickness is the height difference between the cloud top height and the cloud bottom height, and the second cloud thickness is the height difference between the aircraft height and the cloud bottom height; a training module, which is used to train multiple training cloud liquid water contents and their corresponding relative humidity, temperature, dew point, first cloud thickness and second cloud thickness based on a BP neural network algorithm to construct a cloud liquid water content prediction model; a construction module, which is used to construct a cloud boundary prediction model; wherein the cloud boundary prediction model includes cloud probability values ​​of multiple altitude layers The corresponding relative humidity, temperature, altitude, air pressure and altitude air pressure; a first input module, used to input the relative humidity, temperature, altitude, air pressure and altitude air pressure of the same altitude layer into the cloud boundary prediction model to determine the cloud base height and cloud top height when there are clouds at the same altitude layer; a second input module, used to input the relative humidity, temperature, dew point, first cloud thickness and second cloud thickness of the same altitude layer into the cloud liquid water content prediction model to obtain the cloud liquid water content of the same altitude layer; an inversion module, used to invert the liquid water path according to the cloud base height and cloud top height of the same altitude layer and the cloud liquid water content of the same altitude layer; an optimization module, used to optimize the cloud liquid water content prediction model and the cloud boundary prediction model respectively if the liquid water path does not meet the preset conditions, so that the liquid water path meets the preset conditions.

[0115] Figure 4 FIG. 1 is a schematic diagram of a structure of an optimization device for a cloud prediction model provided by another exemplary embodiment of the present application. Figure 4 As shown, after the training module, the optimization device of the cloud prediction model can be specifically configured as follows: obtaining a verification data set in the training samples; wherein the verification data set includes verification humidity, verification temperature, verification dew point, first verification cloud thickness, second verification cloud thickness and real cloud liquid water content; inputting the verification humidity, verification temperature, verification dew point, first verification cloud thickness and second verification cloud thickness in the verification data set into the cloud liquid water content prediction model to obtain the verification cloud liquid water content of the verification data set; determining the degree of deviation between the verification cloud liquid water content of the verification data set and the real cloud liquid water content according to the verification cloud liquid water content of the verification data set and the real cloud liquid water content; and optimizing the cloud liquid water content prediction model based on the degree of deviation.

[0116] In one embodiment, the optimization device of the cloud prediction model can be specifically configured as follows: calculating the root mean square error based on the verified cloud liquid water content and the actual cloud liquid water content of the verification data set; wherein, based on the degree of deviation, optimizing the cloud liquid water content prediction model includes: if the mean root mean square error is greater than a first preset error threshold, adjusting the cloud liquid water content prediction model so that the mean root mean square error between the verified cloud liquid water content predicted by the cloud liquid water content prediction model and the actual cloud liquid water content is less than the first preset error threshold.

[0117] In one embodiment, the optimization device of the cloud prediction model can be specifically configured as follows: calculating the mean absolute error based on the verified cloud liquid water content and the actual cloud liquid water content of the verification data set; if the mean absolute error is less than a second preset error threshold, adjusting the cloud liquid water content prediction model so that the mean absolute error between the verified cloud liquid water content predicted by the cloud liquid water content prediction model and the actual cloud liquid water content is less than the second preset error threshold.

[0118] In one embodiment, the construction module can be specifically configured as follows: obtaining cloud boundary training samples; wherein the cloud boundary training samples include a cloud boundary training set, the cloud boundary training set includes a cloud training data set and a cloud-free training data set, the cloud training data set includes the probability values ​​of cloud in multiple training altitude layers and their corresponding relative humidity, temperature, altitude, air pressure and ground pressure, and the cloud-free training data set includes the probability values ​​of cloud-free in multiple training altitude layers and their corresponding relative humidity, temperature, altitude, air pressure and ground pressure; based on the MLP multi-layer perceptron, the probability values ​​of cloud in multiple training altitude layers and their corresponding relative humidity, temperature, altitude, air pressure and ground pressure, as well as the probability values ​​of cloud-free in multiple training altitude layers and their corresponding relative humidity, temperature, altitude, air pressure and ground pressure are trained to construct a cloud boundary prediction model.

[0119] In one embodiment, the optimization module may include: a calculation unit, which is used to calculate the brightness temperature values ​​of four channels based on the microwave radiation transmission equation and the observation data detected by the airborne microwave radiometer at the same altitude layer; an inversion unit, which is used to calculate the inverted liquid water path based on the BP neural network according to the brightness temperature values ​​of the four channels, the minimum aircraft flight altitude at the same altitude layer, the temperature corresponding to the minimum aircraft flight altitude and the liquid water content at the same altitude layer; a construction unit, which is used to construct preset conditions based on the inverted liquid water path; and an optimization unit, which is used to optimize the cloud liquid water content prediction model and the cloud boundary prediction model respectively if the liquid water path does not meet the preset conditions.

[0120] In one embodiment, the construction unit can be specifically configured to: obtain the actual liquid water path of the same altitude layer; calculate the first deviation between the inverted liquid water path and the actual liquid water path; the optimization unit can be specifically configured to: calculate the second deviation between the liquid water path and the actual liquid water path; if the second deviation is greater than the first deviation, then optimize the cloud liquid water content prediction model and the cloud boundary prediction model respectively.

[0121] In one embodiment, the construction unit can be specifically configured as follows: based on the vertical cumulative liquid water formula, calculate the reference liquid water path of the same altitude layer; calculate the third deviation between the inverted liquid water path and the reference liquid water path; the optimization unit can be specifically configured as follows: calculate the fourth deviation between the liquid water path and the reference liquid water path; if the fourth deviation is greater than the third deviation, optimize the cloud liquid water content prediction model and the cloud boundary prediction model respectively.

[0122] In one embodiment, the optimization device of the cloud prediction model can be specifically configured as follows: obtaining a cloud boundary test set in a cloud boundary training sample; wherein the cloud boundary test set includes a cloud test data set, and the cloud test data set includes relative humidity, temperature, altitude, air pressure and ground air pressure of multiple test altitude layers; inputting the corresponding relative humidity, temperature, altitude, air pressure and ground air pressure of the multiple test altitude layers into the cloud boundary prediction model to obtain a test probability value of cloud at each test altitude layer; selecting a test probability value of cloud greater than a preset probability value as a target test probability value; obtaining a target cloud base height and a target cloud top height corresponding to the target test probability value; calculating a cloud thickness corresponding to the target test probability value based on the target cloud base height and the target cloud top height; obtaining ground-based cloud radar data corresponding to the target test probability value at the same time; reading the cloud base height and cloud top height of the ground-based cloud radar data; calculating the cloud thickness corresponding to the ground-based cloud radar data based on the cloud base height and cloud top height of the ground-based cloud radar data; if the deviation between the cloud thickness corresponding to the target test probability value and the cloud thickness corresponding to the ground-based cloud radar data is greater than a preset deviation threshold, the cloud boundary prediction model is optimized.

[0123] Figure 5 A block diagram of an electronic device according to an embodiment of the present application is illustrated.

[0124] like Figure 5 As shown, the electronic device 10 includes one or more processors 11 and a memory 12 .

[0125] The processor 11 may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 10 to perform desired functions.

[0126] The memory 12 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 11 may run the program instructions to implement the optimization method of the cloud prediction model of each embodiment of the present application described above and / or other desired functions. Various contents such as input signals, signal components, noise components, etc. may also be stored in the computer-readable storage medium.

[0127] In one example, the electronic device 10 may further include: an input device 13 and an output device 14, and these components are interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0128] When the electronic device is a stand-alone device, the input device 13 may be a communication network connector, which is used to receive the collected input signals from the first device and the second device.

[0129] In addition, the input device 13 may also include, for example, a keyboard, a mouse, and the like.

[0130] The output device 14 can output various information to the outside, including the determined distance information, direction information, etc. The output device 14 can include, for example, a display, a speaker, a printer, a communication network and a remote output device connected thereto, and the like.

[0131] Of course, to simplify, Figure 5 Only some of the components related to the present application in the electronic device 10 are shown, and components such as a bus, an input / output interface, etc. are omitted. In addition, according to specific application situations, the electronic device 10 may also include any other appropriate components.

[0132] The computer program product may be written in any combination of one or more programming languages ​​to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages, such as Java, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0133] The computer readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can include, for example, but is not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0134] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although multiple example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

1. A method for optimizing a cloud prediction model, characterized in that: include: Acquire training samples; wherein the training samples include a training data set, the training data set includes a plurality of historical observation data, each of the historical observation data includes a training cloud liquid water content and its corresponding relative humidity, temperature, dew point, first cloud thickness and second cloud thickness, the first cloud thickness is the height difference between the cloud top height and the cloud bottom height, the second cloud thickness is the height difference between the aircraft height and the cloud bottom height; Based on the BP neural network algorithm, the plurality of training cloud liquid water contents and their corresponding relative humidity, temperature, dew point, first cloud thickness and second cloud thickness are trained to construct a cloud liquid water content prediction model; Constructing a cloud boundary prediction model; wherein the cloud boundary prediction model includes cloud probability values ​​of multiple altitude layers and their corresponding relative humidity, temperature, altitude, air pressure and altitude air pressure; Inputting the relative humidity, temperature, altitude, air pressure and altitude air pressure of the same altitude layer into the cloud boundary prediction model to determine the cloud base height and cloud top height when there are clouds in the same altitude layer; Inputting the relative humidity, temperature, dew point, first cloud thickness and second cloud thickness of the same altitude layer into the cloud liquid water content prediction model to obtain the cloud liquid water content of the same altitude layer; Inverting the liquid water path according to the cloud base height and cloud top height of the same altitude layer and the cloud liquid water content of the same altitude layer; If the liquid water path does not meet the preset conditions, the cloud liquid water content prediction model and the cloud boundary prediction model are optimized respectively to make the liquid water path meet the preset conditions.

2. The cloud prediction model optimization method according to claim 1, characterized in that: After training the plurality of training cloud liquid water contents and their corresponding relative humidity, temperature, dew point, first cloud thickness and second cloud thickness based on the BP neural network algorithm to construct a cloud liquid water content prediction model, the method further includes: Acquire a verification data set in the training sample; wherein the verification data set includes verification humidity, verification temperature, verification dew point, first verification cloud thickness, second verification cloud thickness and real cloud liquid water content; Inputting the verification humidity, verification temperature, verification dew point, first verification cloud thickness and second verification cloud thickness in the verification data set into the cloud liquid water content prediction model to obtain the verification cloud liquid water content of the verification data set; Determining, according to the verification cloud liquid water content of the verification data set and the actual cloud liquid water content, a degree of deviation between the verification cloud liquid water content of the verification data set and the actual cloud liquid water content; Based on the degree of deviation, the cloud liquid water content prediction model is optimized.

3. The cloud prediction model optimization method according to claim 2, characterized in that: Determining the degree of deviation between the verified cloud liquid water content of the verification data set and the actual cloud liquid water content according to the verified cloud liquid water content of the verification data set comprises: Calculating a root mean square error based on the verification cloud liquid water content of the verification data set and the true cloud liquid water content; Wherein, optimizing the cloud liquid water content prediction model based on the degree of deviation includes: If the mean root mean error is greater than a first preset error threshold, the cloud liquid water content prediction model is adjusted so that the mean root mean error between the verified cloud liquid water content predicted by the cloud liquid water content prediction model and the actual cloud liquid water content is less than the first preset error threshold.

4. The method for optimizing the cloud prediction model according to claim 2, characterized in that: Determining the degree of deviation between the verified cloud liquid water content of the verification data set and the actual cloud liquid water content according to the verified cloud liquid water content of the verification data set comprises: Calculating a mean absolute error based on the verification cloud liquid water content of the verification data set and the true cloud liquid water content; Wherein, optimizing the cloud liquid water content prediction model based on the degree of deviation includes: If the mean absolute error is less than a second preset error threshold, the cloud liquid water content prediction model is adjusted so that the mean absolute error between the verified cloud liquid water content predicted by the cloud liquid water content prediction model and the actual cloud liquid water content is less than the second preset error threshold.

5. The model verification method based on the inversion algorithm according to claim 1, characterized in that: The cloud boundary prediction model is constructed as follows: Obtain cloud boundary training samples; wherein the cloud boundary training samples include a cloud boundary training set, the cloud boundary training set includes a cloud training data set and a cloud-free training data set, the cloud training data set includes cloud probability values ​​of multiple training altitude layers and their corresponding relative humidity, temperature, altitude, air pressure and ground pressure, and the cloud-free training data set includes cloud-free probability values ​​of the multiple training altitude layers and their corresponding relative humidity, temperature, altitude, air pressure and ground pressure; Based on the MLP multi-layer perceptron, the probability values ​​of the presence of clouds and the corresponding relative humidity, temperature, altitude, air pressure and ground pressure of the multiple training altitude layers and the probability values ​​of the absence of clouds and the corresponding relative humidity, temperature, altitude, air pressure and ground pressure of the multiple training altitude layers are trained to build a cloud boundary prediction model.

6. The model verification method based on the inversion algorithm according to claim 1, characterized in that: If the liquid water path does not meet the preset condition, respectively optimizing the cloud liquid water content prediction model and the cloud boundary prediction model includes: Based on the microwave radiation transmission equation, the brightness temperature values ​​of the four channels are calculated according to the observation data detected by the airborne microwave radiometer at the same altitude layer; Based on the BP neural network, the inverted liquid water path is calculated according to the brightness temperature values ​​of the four channels, the lowest aircraft flight altitude at the same altitude layer, the temperature corresponding to the lowest aircraft flight altitude, and the liquid water content at the same altitude layer; According to the inverted liquid water path, constructing a preset condition; If the liquid water path does not meet the preset condition, the cloud liquid water content prediction model and the cloud boundary prediction model are optimized respectively.

7. The model verification method based on the inversion algorithm according to claim 6, characterized in that: The constructing of preset conditions according to the inverted liquid water path includes: Obtaining a real liquid water path at the same altitude layer; calculating a first deviation between the inverted liquid water path and the true liquid water path; Wherein, if the liquid water path does not meet the preset conditions, respectively optimizing the cloud liquid water content prediction model and the cloud boundary prediction model includes: calculating a second deviation between the liquid water path and the actual liquid water path; If the second deviation is greater than the first deviation, the cloud liquid water content prediction model and the cloud boundary prediction model are optimized respectively.

8. The model verification method based on the inversion algorithm according to claim 6, characterized in that: The constructing of preset conditions according to the inverted liquid water path includes: Calculate the reference liquid water path at the same altitude layer based on the vertical cumulative liquid water formula; calculating a third deviation between the inverted liquid water path and the reference liquid water path; Wherein, if the liquid water path does not meet the preset conditions, respectively optimizing the cloud liquid water content prediction model and the cloud boundary prediction model includes: calculating a fourth deviation between the liquid water path and the reference liquid water path; If the fourth deviation is greater than the third deviation, the cloud liquid water content prediction model and the cloud boundary prediction model are optimized respectively.

9. The method for optimizing a cloud prediction model according to claim 5, characterized in that: After training the probability values ​​of the cloudiness of the multiple test altitude layers and the corresponding relative humidity, temperature, altitude, air pressure and ground pressure, and the probability values ​​of the cloudlessness of the multiple test altitude layers and the corresponding relative humidity, temperature, altitude, air pressure and ground pressure based on the MLP multilayer perceptron to construct a cloud boundary prediction model, the method further includes: Obtain a cloud boundary test set from the cloud boundary training samples; wherein the cloud boundary test set includes a cloud test data set, and the cloud test data set includes relative humidity, temperature, altitude, air pressure and ground air pressure of multiple test altitude layers; Inputting the relative humidity, temperature, altitude, air pressure and ground pressure corresponding to the plurality of test altitude layers into the cloud boundary prediction model to obtain a test probability value of cloudiness at each test altitude layer; Selecting a cloud-presence test probability value greater than a preset probability value as a target test probability value; obtaining a target cloud base height and a target cloud top height corresponding to the target test probability value; Calculating the cloud thickness corresponding to the target test probability value according to the target cloud base height and the target cloud top height; Acquire ground-based cloud radar data corresponding to the target test probability value at the same time; Read the cloud base height and cloud top height of the ground-based cloud radar data; Calculating the cloud thickness corresponding to the ground-based cloud radar data according to the cloud base height and cloud top height of the ground-based cloud radar data; If the deviation between the cloud thickness corresponding to the target test probability value and the cloud thickness corresponding to the ground-based cloud radar data is greater than a preset deviation threshold, the cloud boundary prediction model is optimized.

10. An optimization device for a cloud prediction model, characterized in that: include: An acquisition module is used to acquire training samples; wherein the training samples include a training data set, and the training data set includes a plurality of historical observation data, each of which includes a training cloud liquid water content and its corresponding relative humidity, temperature, dew point, first cloud thickness, and second cloud thickness, wherein the first cloud thickness is the height difference between the cloud top height and the cloud bottom height, and the second cloud thickness is the height difference between the aircraft height and the cloud bottom height; A training module, for training the plurality of training cloud liquid water contents and their corresponding relative humidity, temperature, dew point, first cloud thickness and second cloud thickness based on a BP neural network algorithm, so as to construct a cloud liquid water content prediction model; A construction module is used to construct a cloud boundary prediction model; wherein the cloud boundary prediction model includes cloud probability values ​​of multiple altitude layers and their corresponding relative humidity, temperature, altitude, air pressure and altitude air pressure; A first input module is used to input the relative humidity, temperature, altitude, air pressure and altitude air pressure of the same altitude layer into the cloud boundary prediction model to determine the cloud base height and cloud top height when there are clouds in the same altitude layer; A second input module is used to input the relative humidity, temperature, dew point, first cloud thickness and second cloud thickness of the same altitude layer into the cloud liquid water content prediction model to obtain the cloud liquid water content of the same altitude layer; An inversion module, used for inverting a liquid water path according to the cloud base height and cloud top height of the same altitude layer and the cloud liquid water content of the same altitude layer; The optimization module is used for respectively optimizing the cloud liquid water content prediction model and the cloud boundary prediction model if the liquid water path does not meet the preset conditions, so that the liquid water path meets the preset conditions.

Images