A method for early warning of aircraft icing

Abstract

This invention discloses an aircraft icing early warning method, relating to the field of icing early warning technology. The method acquires cloud image data and cloud product file data for a specified aircraft. Based on the cloud product data, it extracts the cloud field operation status feature set of the specified aircraft and analyzes the cloud liquid water path value. Based on a pre-trained cloud image error mapping model and combined with the cloud image data of the specified aircraft, it analyzes the cloud shape temperature correction factor of the specified aircraft and, combined with the cloud field operation status feature set, analyzes the optimized cloud top temperature value of the specified aircraft. Based on the cloud liquid water path value and the optimized cloud top temperature value, it analyzes the icing index of the specified aircraft. This invention performs icing early warning processing on the specified aircraft based on the icing index, thereby making the final generated icing index result more environmentally adaptable and achieving dynamic correction and high-precision characterization of icing risk, thus improving the accuracy of icing early warning.

Description

Technical Field

[0001] This invention relates to the field of icing warning technology, specifically to a method for warning of aircraft icing. Background Technology

[0002] When an aircraft passes through tropospheric clouds or humid weather environments such as cumulus and stratocumulus, the outside air often contains a large number of supercooled water droplets. When these water droplets come into contact with the aircraft, they freeze rapidly and form ice layers on the surfaces of the wings, air intakes, or measuring devices, resulting in reduced lift, increased drag, and instrument errors. In severe cases, it can even cause safety accidents such as engine stall or engine failure.

[0003] In recent years, with the development of meteorological satellite multispectral remote sensing technology, the microphysical properties of clouds can be obtained through inversion. However, existing icing identification methods based on satellite data have failed to fully consider the radiation deviation between cloud top brightness temperature and actual cloud temperature under different cloud types, and also lack a mechanism for quantitatively correcting satellite inversion errors, resulting in underestimation of temperature, which makes icing warnings prone to misjudgment.

[0004] Existing technology, such as the aviation hazardous weather diagnosis and analysis system disclosed in patent application CN114927009A, includes an aircraft icing diagnosis module and a cloud analysis module. The aircraft icing diagnosis module is used to obtain the aircraft's icing status based on an icing algorithm. The icing algorithm includes at least one of the following: IC icing algorithm, RAP icing algorithm, and RAOB icing algorithm. The cloud analysis module is used to determine the cloud conditions of the radiosonde based on the ambient temperature and dew point temperature of the radiosonde. This invention provides an aviation hazardous weather diagnosis and analysis system that, through icing and cloud analysis algorithms, comprehensively judges possible icing and cloud conditions during flight to provide early warnings of major factors contributing to hazardous aviation flight, thereby improving the early warning level for aviation flights.

[0005] Based on the above findings, the limitations of existing technologies include at least the following problems: Existing technologies do not fully utilize high spatiotemporal resolution cloud product file data, lack comprehensive analysis of parameters such as cloud top temperature, cloud optical thickness, and effective particle radius, making it difficult to accurately characterize the thermal state and icing environment of the cloud field. Furthermore, they are difficult to adaptively correct for different cloud types, resulting in large deviations in cloud temperature estimation and insufficient accuracy in icing index calculation. In addition, existing technologies lack multi-source joint analysis integrating cloud type temperature correction systems, cloud product file data, and flight parameters, making it difficult to achieve dynamic response analysis of cloud liquid water paths, cloud top temperature, and flight altitude differences, thus leading to insufficient accuracy in icing warnings. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides an aircraft icing early warning method, which solves the problem of insufficient icing early warning accuracy caused by the lack of multi-source fusion in existing technologies.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an aircraft icing early warning method, comprising the following steps: acquiring cloud image data and cloud product file data of a designated aircraft; extracting the cloud field operation status feature set of the designated aircraft based on the cloud product file data of the designated aircraft, and analyzing the cloud liquid water path value of the designated aircraft; analyzing the cloud shape temperature correction factor of the designated aircraft based on a pre-trained cloud image error mapping model and combined with the cloud image data of the designated aircraft, and analyzing the optimized cloud top temperature value of the designated aircraft based on the cloud field operation status feature set; analyzing the icing index of the designated aircraft based on the cloud liquid water path value and the optimized cloud top temperature value; and performing icing early warning processing on the designated aircraft based on the icing index.

[0008] Further, the specific steps for extracting the cloud field operation status feature set of the designated aircraft are as follows: parse and process the cloud product file data of the designated aircraft; obtain the latitude and longitude of the designated aircraft and match it with the parsed cloud product file data of the designated aircraft to obtain the grid data of the designated aircraft; based on the grid data of the designated aircraft, output the cloud field operation status feature set of the designated aircraft, including cloud top temperature value, effective particle radius value, cloud optical thickness value, and cloud top height value.

[0009] Furthermore, the specific steps for analyzing the cloud liquid water path value of the set aircraft are as follows: read the effective particle radius value and optical thickness value of the set aircraft; obtain the water density value stored in the pre-established database, and combine it with the effective particle radius value and optical thickness value of the set aircraft to analyze the cloud liquid water path value of the set aircraft.

[0010] Furthermore, the cloud image data specifically includes the albedo value and channel brightness temperature set of each pixel in the cloud image, as well as the corresponding two-dimensional coordinates. The cloud image error mapping model consists of an input layer, a cloud image deconstruction layer, and a mapping output layer.

[0011] Further, the specific steps for analyzing the cloud shape temperature correction factor of the set aircraft are as follows: input the cloud image data of the set aircraft into the pre-trained cloud image error mapping model, analyze the temperature correction mapping feature set of the set aircraft, including spectral difference temperature mapping value, cloud undulation temperature mapping value, and absorption-scattering co-shift temperature mapping value; based on the temperature correction mapping feature set of the set aircraft, analyze the cloud shape temperature correction value of the set aircraft.

[0012] Furthermore, the specific steps for analyzing the temperature correction mapping feature set of the target aircraft are as follows: In the input layer of the cloud map error mapping model, the cloud map data of the target aircraft is received and preprocessed; in the cloud map deconstruction layer of the cloud map error mapping model, the preprocessed cloud map data of the target aircraft is subjected to feature extraction processing to obtain the cloud map radiation feature vector of the target aircraft; in the mapping output layer of the cloud map error mapping model, the temperature correction mapping feature set of the target aircraft is output based on the cloud map radiation feature vector of the target aircraft.

[0013] Furthermore, the specific steps for analyzing the optimized cloud top temperature value of the set aircraft are as follows: read the cloud top temperature value, cloud top altitude value, and cloud shape temperature correction value of the set aircraft; obtain the flight altitude value of the set aircraft; and combine the cloud top temperature value, cloud top altitude value, and cloud shape temperature correction value of the set aircraft for comprehensive processing to obtain the optimized cloud top temperature value of the set aircraft.

[0014] Furthermore, the specific formula for calculating the optimal cloud top temperature value for the aircraft is as follows: ;in, To set the optimal cloud top temperature value for the aircraft, To set the cloud top temperature value for the aircraft, To set the cloud top altitude value for the aircraft, To set the aircraft's flight altitude, The temperature decay adjustment coefficient is stored in the database. To set the cloud cover temperature correction value for the aircraft.

[0015] Furthermore, the specific formula for calculating the icing index of a given aircraft is as follows: ;in, To set the aircraft icing index, To set the cloud liquid water path value for the aircraft, This refers to the ice accumulation amplification factor stored in the database. To set the optimal cloud top temperature value for the aircraft, The temperature response normalization coefficient is stored in the database.

[0016] Furthermore, the specific steps for conducting icing warning processing for a designated aircraft based on the icing index are as follows: compare the icing index of the designated aircraft with the preset warning interval set; and send the corresponding warning level to the designated aircraft based on the comparison processing results.

[0017] The present invention has the following beneficial effects:

[0018] (1) The aircraft icing warning method constructs an icing index analysis system based on multi-source information by integrating cloud map data and cloud product file data. It not only utilizes the cloud field operation status feature set obtained from the cloud product file, but also uses a pre-trained cloud map error mapping model to adaptively output cloud type temperature correction factors for different cloud types, thereby obtaining optimized cloud top temperature values. On this basis, it integrates the cloud liquid water path values ​​generated by the cloud field operation status feature set to generate an icing index, thereby effectively solving the cloud temperature estimation deviation caused by a single data source and realizing dynamic correction and high-precision quantification of icing risk, thus significantly improving the accuracy of icing warning.

[0019] (2) The aircraft icing early warning method uses cloud liquid water path generated by cloud field operation status characteristics and optimized cloud top temperature for joint analysis, thereby avoiding the neglect of cloud field complexity and thermal unevenness, making the final generated icing index result more environmentally adaptable. By parsing and processing cloud product file data, the cloud field operation status feature set is automatically extracted and cloud liquid water path value is calculated. On this basis, based on the optimized cloud top temperature value, the coupling effect of liquid water content and temperature conditions on icing formation is comprehensively analyzed to obtain the icing index, thereby realizing dynamic assessment of icing risk and effectively improving the safety of aircraft flight.

[0020] (3) The aircraft icing warning method realizes an intelligent icing warning feedback mechanism by automatically comparing the icing index with the preset warning range. After analyzing the set icing index of the aircraft, it can realize quantitative and graded warning of icing risk according to the corresponding warning range based on the value. When the icing index is in different ranges, it automatically outputs the corresponding warning level information. In turn, it can respond to changes in the cloud field during flight, and thus maintain high sensitivity and high reliability under complex weather conditions, and significantly improve the safety assurance level of aviation operations.

[0021] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0022] Figure 1 This is a flowchart of an aircraft icing early warning method according to the present invention.

[0023] Figure 2 This is a flowchart illustrating the specific steps involved in analyzing and setting the cloud-type temperature correction factor for an aircraft in an aircraft icing early warning method according to the present invention.

[0024] Figure 3 This is a flowchart illustrating the analysis and setting of the temperature correction mapping feature set of an aircraft in an aircraft icing early warning method according to the present invention. Detailed Implementation

[0025] Please see Figure 1 , an embodiment of the present invention provides a technical solution: an aircraft icing warning method, including the following steps: obtaining cloud map data and cloud product file data of a set aircraft, extracting a set of cloud field operation state characteristics of the set aircraft based on the cloud product file data of the set aircraft, and analyzing the cloud liquid water path value of the set aircraft; based on a pre-trained cloud map error mapping model, and combining the cloud map data of the set aircraft, analyzing the cloud type temperature correction factor of the set aircraft, and combining the set of cloud field operation state characteristics, analyzing the optimized cloud top temperature value of the set aircraft; based on the cloud liquid water path value and the optimized cloud top temperature value of the set aircraft, analyzing the icing index of the set aircraft; and performing icing warning processing on the set aircraft based on the icing index.

[0026] The specific formula for calculating the icing index of the set aircraft is as follows: ; where is the icing index of the set aircraft, is the cloud liquid water path value of the set aircraft, is the icing amplification factor stored in the database (when the cloud liquid water path value is low, a higher value is used to improve sensitivity, and when the cloud liquid water path value is high, a lower value is used to maintain stability), is the optimized cloud top temperature value of the set aircraft, is the temperature response normalization coefficient stored in the database (and in this embodiment, it can take the value of ).

[0027] It should be noted that the icing amplification factor stored in the database is obtained as follows: obtaining the cloud liquid water path values of several historical times, and extracting its 25th percentile (Q25) and 75th percentile (Q75), and using these as interval boundaries, and dividing them into a low liquid water interval (cloud liquid water path value < Q25), a medium liquid water interval (Q25 ≤ cloud liquid water path value < Q75), and a high liquid water interval (cloud liquid water path value ≥ Q75), where Q25 and Q75 respectively correspond to the 25% and 75% distribution positions of the cloud liquid water path value, to reflect the physical stratification characteristics of different cloud water contents. If the current cloud liquid water path value is within the low liquid water interval, the icing amplification factor can take 6.2849. If the current cloud liquid water path value is within the medium liquid water interval, the icing amplification factor can take 4.1899. If the current cloud liquid water path value is within the high liquid water interval, the icing amplification factor can take 3.3519.

[0028] The temperature response normalization coefficient stored in the database The steps are as follows: Obtain sample datasets of several historical icing environments. The sample datasets include multidimensional meteorological feature data and physical observation data related to the calculation of the icing index. The sample data are stored in time series form and are divided into training set and validation set in an 8:2 ratio using a time sequence partitioning strategy. The training set is used for model training and parameter search, and the validation set is used for model performance evaluation.

[0029] During the training phase, the temperature response normalization coefficient was adjusted. Initial values ​​are set, and automatic optimization search is performed within the parameter space of [-80, -5]. An objective function is constructed based on a Bayesian optimization framework, using the icing index calculation error as the optimization constraint. A sliding time window cross-validation mechanism is introduced, dividing the later time series data into multiple continuous time windows (e.g., 5). Each window serves as a validation sample, and the remaining windows serve as training samples. After each round of training, based on the validation set prediction results and the corresponding measured label values, the following training metrics are extracted: critical success index (representing the degree of matching between the model's predicted icing events and the measured icing events), false negative rate (the proportion of real icing samples missed by the model), false alarm rate (the probability that non-icing samples are misclassified as icing), and overall accuracy (evaluating the overall prediction consistency). Through iterative training and validation window by window, the accuracy of each window is calculated. The corresponding overall performance score;

[0030] The overall performance score is obtained through a weighted calculation of multiple indicators, and its evaluation function is as follows: ;in, The critical success index, with a weight of 40%, reflects the core accuracy of ice accumulation prediction. The false negative rate, weighted at 30%, emphasizes the ability to identify real ice accumulation; The false alarm rate is set at a weight of 20% to control false alarms in scenarios without ice buildup. The overall accuracy is weighted at 10% and measures the consistency of overall predictions.

[0031] The generated data in each iteration during the Bayesian optimization process The values ​​are used for performance evaluation, and the highest score is given. The corresponding value was determined as the final temperature response normalization coefficient, and the target value obtained after optimization was... = .

[0032] The specific steps for icing warning processing of a designated aircraft based on the icing index are as follows: The icing index of the designated aircraft is compared with a preset set of warning intervals. The set of warning intervals includes: First warning interval: when the icing index < 0, it is determined to be a no-icing risk area, corresponding to clear skies or thin low-cloud areas; Second warning interval: when 20 ≤ icing index < 50, it is determined to be a mild icing risk area, corresponding to areas with low cloud liquid water content and local supercooled water droplets; Third warning interval: when 50 ≤ icing index < 80, it is determined to be a moderate icing risk area, corresponding to areas with thicker clouds, moderate liquid water paths, and lower temperatures; Fourth warning interval: when the icing index ≥ 80, it is determined to be a strong icing risk area, corresponding to areas with high cloud liquid water content, extremely low temperatures, and active ice particles.

[0033] Based on the comparison and processing results, corresponding warning levels are sent to the designated aircraft. Specifically: if the icing index of the designated aircraft is within the first warning range, normal monitoring is maintained and no alarm is issued; if the icing index of the designated aircraft is within the second warning range, an icing attention signal is generated to remind the pilot to pay attention to cloud changes and continuously monitor the icing index; if the icing index of the designated aircraft is within the third warning range, an icing warning signal is generated, and a visual warning prompt is sent to the flight control terminal, suggesting that the pilot take altitude fine-tuning or heading avoidance measures; if the icing index of the designated aircraft is within the fourth warning range, a severe icing warning signal is generated, triggering a mandatory alarm mechanism and sending it to the ground control center through the flight communication link, while recording the warning time, location coordinates, and relevant parameters for subsequent analysis and emergency response.

[0034] Specifically, the cloud product file data is the raw field data read from the L2-level cloud product file obtained from the Himawari-8 / 9 satellite. This file is stored in NetCDF format and contains multiple cloud microphysical parameter fields. Each file uses a latitude and longitude grid as its basic unit, with each grid accompanied by corresponding latitude and longitude coordinates and multiple cloud microphysical parameters. (It should be noted that this L2 cloud product file is automatically generated by the meteorological satellite ground inversion system based on L1-level satellite observation data. This system directly obtains the inversion result by calling the meteorological data interface or satellite data sharing service, without needing to perform radiometric inversion calculations within this design.) The specific steps for extracting the cloud field operational status feature set of the designated aircraft are as follows:

[0035] The cloud product file data of the aircraft is parsed and processed. That is, the structure of the file is identified and the fields are decoded through the data parsing interface. The (raw binary) field data in the cloud product file is parsed into a processable numerical matrix form. Preprocessing operations such as outlier removal, missing measurement point interpolation and data format unification are performed. After parsing, a standardized dataset containing cloud microphysical fields in multiple latitude and longitude grids is obtained.

[0036] The latitude and longitude of the designated aircraft are obtained (which can be obtained through airborne GNSS or BDS) and matched with the cloud product file data of the designated aircraft after parsing and processing to obtain the grid data of the designated aircraft. That is, the latitude and longitude of the designated aircraft are compared with the latitude and longitude coordinate information in each latitude and longitude grid to obtain the latitude and longitude grid corresponding to the designated aircraft, and the cloud microphysical fields, such as cloud top temperature field, effective particle radius field, cloud optical thickness field, cloud top height field, etc., are read as grid data.

[0037] Based on the grid data of the set aircraft, the cloud field operation status feature set of the set aircraft is output, including cloud top temperature value (the temperature value at the upper boundary of the cloud layer), effective particle radius value (the equivalent average radius of droplets or ice crystals in the cloud), cloud optical thickness value (the overall extinction ability of the cloud layer to solar radiation, reflecting the intensity of the cloud layer's blocking of incident radiation; the thicker the cloud layer and the higher the particle concentration, the larger the optical thickness value; the thinner the cloud layer and the sparser the particles, the smaller the optical thickness value), and cloud top height value (the geometric height of the upper boundary of the cloud layer from the ground). In other words, the cloud microphysical fields in the grid data, such as cloud top temperature field, effective particle radius field, cloud optical thickness field, cloud top height field, etc., are output with their corresponding values.

[0038] The specific steps for analyzing the cloud liquid water path value of the set aircraft are as follows: Read the effective particle radius and optical thickness values ​​from the cloud field operation state feature set of the set aircraft; obtain the water density values ​​stored in the pre-established database; and combine these with the effective particle radius and optical thickness values ​​of the set aircraft to analyze the cloud liquid water path value of the set aircraft (which is the total mass of liquid water per unit area when vertically passing through the entire cloud layer). The specific formula is as follows: ;in, To set the cloud liquid water path value for the aircraft, To pre-establish the water density values ​​stored in the database, To set the effective particle radius value for the aircraft, To set the optical thickness value for the aircraft.

[0039] It should be explained that the calculation of cloud liquid water path value is based on the theory of radiative transfer and optical thickness of cloud bodies. In the cloud microphysics model, the radius distribution function and volume distribution of cloud droplets determine the relationship between optical thickness and liquid water content. When the effective radius of cloud droplets is approximately constant in the vertical direction, a quantitative mapping relationship between the optical thickness of the cloud body, the average radius of droplets, and the density of liquid water can be established through integration simplification, thus obtaining this formula. It reflects the coupling law between the total mass of liquid water in the cloud layer and the cloud droplet scale and optical extinction capability. Moreover, the effective particle radius value and optical thickness value are obtained by parsing the L2 level cloud product file of Himawari-8 / 9 satellite. The parameters in this file have been inverted based on multi-channel radiation observation through the meteorological satellite ground inversion system, so there is no need to repeat the inversion calculation. The cloud liquid water path value of the set aircraft is calculated only based on this physical formula.

[0040] In this implementation plan, this step enables structured analysis and quantitative utilization of satellite-inverted cloud product file data, significantly improving the availability and accuracy of cloud field microphysical parameters. By directly calling the L2-level cloud product file generated by the meteorological satellite ground inversion system, complex radiometric inversion calculations are avoided locally, ensuring data reliability and reducing computational burden. By extracting and setting grid data corresponding to the aircraft position through latitude and longitude matching, the precise correspondence between cloud field spatial distribution and flight position is achieved, thereby reflecting the cloud field thermal state and structural characteristics of the area where the aircraft is located in real time. Secondly, by using the effective particle radius value and optical thickness value to calculate the cloud liquid water path, a dynamic quantitative expression from cloud microscopic particle characteristics to macroscopic water content is achieved, thereby reflecting the cloud water vapor content and potential icing conditions, and thus improving the accuracy of overall icing identification and risk assessment.

[0041] Specifically, such as Figure 2-3 As shown, the cloud image data (the cloud image is a local cloud image area with a radius of up to 10km centered on the latitude and longitude of the aircraft) specifically consists of the albedo value and channel brightness temperature set (including water vapor channel brightness temperature value, infrared window brightness temperature value, and shortwave infrared brightness temperature) of each pixel in the cloud image, as well as the corresponding two-dimensional coordinates. The cloud image error mapping model consists of an input layer, a cloud image deconstruction layer, and a mapping output layer.

[0042] The specific steps for analyzing the cloud shape temperature correction factor of the target aircraft are as follows: Input the cloud image data of the target aircraft into the pre-trained cloud image error mapping model, analyze the temperature correction mapping feature set of the target aircraft, including spectral difference temperature mapping value, cloud undulation temperature mapping value, and absorption-scattering co-shift temperature mapping value; Based on the temperature correction mapping feature set of the target aircraft, analyze the cloud shape temperature correction value of the target aircraft, specifically by weighting the spectral difference temperature mapping value, cloud undulation temperature mapping value, and absorption-scattering co-shift temperature mapping value of the target aircraft to obtain the cloud shape temperature correction value of the target aircraft.

[0043] It should be noted that, in this weighted processing, the weighting coefficients corresponding to the spectral difference temperature mapping value, cloud undulation temperature mapping value, and absorption-scattering co-shift temperature mapping value can be obtained through the following steps: obtain historical spectral difference temperature mapping values, historical cloud undulation temperature mapping values, and historical absorption-scattering co-shift temperature mapping values ​​(and it should be noted that all the above values ​​are absolute values), and extract the mean values ​​of the historical spectral difference temperature mapping, historical cloud undulation temperature mapping, and historical absorption-scattering co-shift temperature mapping respectively, and sum them to obtain a corrected sum value. Ratio the mean values ​​of the historical spectral difference temperature mapping, historical cloud undulation temperature mapping, and historical absorption-scattering co-shift temperature mapping with the corrected sum value, and use the corresponding results as the corresponding weighting coefficients.

[0044] The specific steps for analyzing the temperature correction mapping feature set of the set aircraft are as follows: In the input layer of the cloud map error mapping model, the cloud map data of the set aircraft is received and preprocessed. Specifically, the albedo value and the brightness temperature value of each channel are normalized to the same numerical range, such as between 0 and 1, so as to facilitate the uniformity of the feature scale of the model input. For the missing pixels in the cloud map caused by occlusion, noise or data loss, compensation is made by neighborhood interpolation or spatial smoothing methods to maintain the continuity and integrity of the data. After that, 3×3 neighborhood mean convolution can be used to perform spatial filtering on the pixels in the local area to reduce random noise and enhance the spatial consistency of cloud brightness temperature and albedo distribution.

[0045] In the cloud image deconstruction layer of the cloud image error mapping model, feature extraction processing is performed on the preprocessed cloud image data of the target aircraft to obtain the cloud image radiation feature vector of the target aircraft. Specifically, this involves reading the brightness temperature value of the water vapor channel, the brightness temperature value of the infrared window area, and the brightness temperature value of the short-wave infrared for each pixel in the cloud image, extracting the brightness temperature difference between the water vapor channel and the infrared window area, and the brightness temperature difference between the short-wave infrared and the infrared window area for each pixel, and performing ratio processing to obtain the brightness temperature difference ratio for each pixel. The two-dimensional coordinates of each pixel are then read, averaged, and the center coordinates of the center pixel are obtained. Based on the Euclidean distance method, the center pixel and the... The Euclidean distance value of each pixel is calculated and represented by reciprocal normalization, such as 1 / (1+Euclidean distance value). The sum of these values ​​is then obtained to obtain the Euclidean distance sum. The Euclidean distance values ​​of the center pixel and each pixel are then compared with the Euclidean distance sum to obtain the weight ratio of each pixel. This weight ratio is then multiplied by the brightness temperature difference ratio of the corresponding pixel to extract the spectral difference ratio feature. This feature is used to characterize the thickness distribution characteristics of clouds over a given aircraft in different spectral channels. A higher value indicates thicker clouds, higher water vapor content, and significant radiation attenuation, corresponding to a lower cloud top brightness temperature.

[0046] The infrared window brightness temperature value of each pixel is read, and for each pixel, the difference in infrared window brightness temperature between it and its neighboring pixels in the horizontal and vertical directions is calculated. The squares of the difference in infrared window brightness temperature in the horizontal and vertical directions are summed and squared to obtain the brightness temperature gradient value of each pixel. The weight ratio value of each pixel is read and multiplied with the brightness temperature gradient value of the corresponding pixel to extract the cloud undulation energy feature, which is used to characterize the spatial undulation degree of the cloud over the set aircraft. When it is high, it indicates that the cloud top height changes drastically, the cloud structure is complex, and the uneven distribution of infrared radiation leads to a low cloud top brightness temperature.

[0047] The albedo value and infrared window brightness temperature value of each pixel are read, and the mean albedo, standard deviation of albedo, mean infrared window brightness temperature, and standard deviation of infrared window brightness temperature are extracted respectively. The correlation coefficient method is used, that is, based on the albedo value and infrared window brightness temperature value of each pixel, as well as the mean albedo and the mean infrared window brightness temperature, their covariance is calculated, and the ratio is processed with the product of the standard deviation of albedo and the standard deviation of infrared window brightness temperature to extract the absorption and scattering co-shift feature. This feature is used to correct the absorption and scattering co-shift performance of clouds over the aircraft during the phase transition process. When the value is small, it indicates that the visible light albedo is high and the infrared brightness temperature is low, that there is ice-water mixing or phase inhomogeneity in the cloud, the infrared radiation absorption is weakened, and the cloud top temperature value is lower. The spectral difference ratio feature, cloud undulation energy feature, and absorption and scattering co-shift feature are concatenated into a cloud image radiation feature vector.

[0048] In the mapping output layer of the cloud image error mapping model, based on the cloud image radiation feature vector of the set aircraft, the temperature correction mapping feature set of the set aircraft is output. Specifically, the feature-temperature correction correspondence established during the training phase is called, and the current values ​​of the spectral difference ratio feature, cloud undulation energy feature, and absorption-scattering co-shift feature are mapped to specific spectral difference temperature mapping values, cloud undulation temperature mapping values, and absorption-scattering co-shift temperature mapping values. When the spectral difference temperature mapping value, cloud undulation temperature mapping value, and absorption-scattering co-shift temperature mapping value represent the specific values ​​of the spectral difference ratio feature, cloud undulation energy feature, and absorption-scattering co-shift feature, respectively, the corresponding feature is the specific correction value for the current cloud top temperature value.

[0049] The pre-training steps of the cloud map error mapping model are as follows:

[0050] The labeled dataset consists of multi-channel cloud image product files from several historical Himawari-8 / 9 satellites and historical meteorological radiosonde observation temperature data from the same time and region. Each sample was labeled by meteorological observation experts based on the deviation between the cloud top brightness temperature retrieved from historical satellites and the measured cloud top temperature. Each sample group includes the visible light albedo value, water vapor channel brightness temperature value, shortwave infrared brightness temperature value, infrared window brightness temperature value, and the corresponding real cloud top temperature label value for each pixel in the set area.

[0051] The labeled dataset is preprocessed, including standardizing the range of brightness temperature and albedo values ​​for each channel, compensating for missing measurement points, and removing outliers to maintain data continuity and spatial consistency. The preprocessed historical dataset is divided into training, validation, and test sets according to the proportions, and the samples are ensured to cover different cloud types, climate zones, and altitudes to enhance the model's generalization ability.

[0052] During the training phase, the spectral difference ratio, cloud undulation energy, and absorption-scattering co-shift features of each sample are extracted and sequentially concatenated into a cloud image comprehensive radiation feature vector. At the same time, the deviation between the satellite-inverted cloud top brightness temperature and the measured cloud top temperature in the sample is calculated and used as the training label input model to supervise the training of the cloud image error mapping model.

[0053] The model employs a regression learning mechanism based on historical data, with the goal of minimizing the mean square error (MSE) between the predicted temperature correction value and the historical measured temperature error. The mapping function parameters are iteratively updated through the backpropagation algorithm, and convergence control is achieved using an adaptive optimization algorithm (such as the Adam optimizer). The model performance is evaluated and hyperparameters are adjusted through a validation set to ensure that the model can accurately learn the relationship between different cloud map features and temperature deviations.

[0054] After training, statistical analysis is performed on the feature vectors of all historical samples and their corresponding correction values ​​to form a historical correspondence table between feature value intervals and temperature correction values. That is, a historical mapping relationship is established between spectral difference ratio features, cloud undulation energy features, absorption and scattering co-shift features and cloud top temperature correction values, and this relationship is solidified into the model parameter file.

[0055] Finally, the trained cloud map error mapping model is saved, and its internal historical correspondence table will be called during the runtime phase for fast temperature correction calculation of the real-time input cloud map feature vector, thereby realizing adaptive correction of cloud top temperature based on historical experience data.

[0056] In this implementation scheme, a pre-trained cloud image error mapping model is introduced to perform deep analysis of cloud image data, thereby achieving intelligent correction of cloud top temperature deviation and significantly improving the accuracy and adaptability of cloud field temperature. Secondly, using the pre-trained cloud image error mapping model, parameters such as albedo, water vapor channel brightness temperature, shortwave infrared brightness temperature, and infrared window brightness temperature are jointly deconstructed to extract spectral difference ratio features, cloud undulation energy features, and absorption-scattering co-shift features. The model achieves quantitative correction of cloud top temperature through the feature-temperature correction correspondence established during the training phase. This model can adaptively output correction values ​​based on different cloud types and cloud thicknesses. Finally, by introducing historical data for supervised learning, the model learns the brightness temperature deviation patterns under multiple cloud types, enabling it to quickly and accurately correct real-time cloud images during operation. This achieves automated mapping from features to temperature correction values, which not only improves the physical authenticity and regional adaptability of cloud top temperature inversion but also significantly enhances the prediction accuracy of icing.

[0057] Specifically, the steps for analyzing the optimal cloud top temperature value of the set aircraft are as follows: read the cloud top temperature value, cloud top altitude value, and cloud shape temperature correction value of the set aircraft; obtain the flight altitude value of the set aircraft (which can be obtained through a barometric altitude sensor, i.e., the barometric altitude sensor detects the external static pressure of the aircraft in real time, and converts the relationship between the atmospheric static pressure value and the standard atmospheric pressure altitude using the barometric altitude equation to obtain the flight altitude value); and combine the cloud top temperature value, cloud top altitude value, and cloud shape temperature correction value of the set aircraft for comprehensive processing to obtain the optimal cloud top temperature value of the set aircraft.

[0058] The specific formula for calculating the optimal cloud top temperature value for the aircraft is as follows: ;in, To set the optimal cloud top temperature value for the aircraft, To set the cloud top temperature value for the aircraft, To set the cloud top altitude value for the aircraft, To set the aircraft's flight altitude, The temperature decay adjustment coefficient is stored in the database (in this implementation example, the value can be 0.65). To set the cloud cover temperature correction value for the aircraft.

[0059] It needs to be explained that the temperature decay adjustment coefficient stored in the database The acquisition steps are as follows: Obtain historical cloud top altitude values, historical flight altitude values, and corresponding measured temperature values ​​for several historical periods. Extract the measured temperature difference (i.e., the difference between the measured temperature values ​​corresponding to the historical cloud top altitude value and the historical flight altitude value) and altitude difference (i.e., the absolute value of the difference between the historical cloud top altitude value and the historical flight altitude value) for each historical period, and perform ratio processing to obtain the temperature decay value for each historical period. Then, perform moving average processing and use the result as the temperature decay adjustment coefficient. .

[0060] In this implementation scheme, based on the cloud top temperature, the aircraft's flight altitude and cloud shape temperature correction values ​​are combined to achieve dynamic optimization and correction of the cloud top temperature, thereby significantly improving the accuracy of cloud field thermal parameters. A barometric altimeter is used to acquire the flight altitude in real time, and the precise spatial location is obtained through the pressure-altitude conversion relationship. Difference analysis is performed between the altitude and cloud top temperature to establish a thermal coupling relationship between altitude and temperature differences. Secondly, statistical analysis is conducted on multiple historical sets of cloud top altitude, flight altitude, and measured temperature data to calculate a temperature decay adjustment coefficient, reflecting the changing characteristics of the temperature gradient between different altitude layers. This allows the temperature correction to be dynamically adjusted according to the actual vertical atmospheric structure. Finally, this step achieves adaptive correction of the cloud top temperature, avoiding the deviation problem of the fixed temperature decay rate model under different climatic conditions. Therefore, the optimized cloud top temperature value can more realistically reflect the thermal state of the aircraft's environment, thus significantly enhancing the regional adaptability of icing identification.

[0061] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0062] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for early warning of aircraft icing, characterized in that, Includes the following steps: Obtain cloud image data and cloud product file data for the designated aircraft. The cloud product file data is the raw field data read from the L2 level cloud product file obtained from Himawari-8 / 9 satellites. Based on the cloud product file data of the designated aircraft, the cloud field operation status feature set of the designated aircraft is extracted, and the cloud liquid water path value of the designated aircraft is analyzed. Based on the pre-trained cloud map error mapping model, which consists of an input layer, a cloud map deconstruction layer, and a mapping output layer, and combined with the cloud map data of the set aircraft, the cloud shape temperature correction factor of the set aircraft is analyzed, and combined with the cloud field operation state feature set, the optimized cloud top temperature value of the set aircraft is analyzed. Based on the set cloud liquid water path value and optimized cloud top temperature value, the icing index of the set aircraft is analyzed. Icing warnings are issued for designated aircraft based on the icing index.

2. The aircraft icing early warning method according to claim 1, characterized in that, The specific steps for extracting the cloud field operation status feature set of a given aircraft are as follows: The cloud product file data for the aircraft is parsed and processed. Obtain the latitude and longitude of the designated aircraft and match it with the parsed cloud product file data of the designated aircraft to obtain the grid data of the designated aircraft. Based on the grid data of the specified aircraft, the cloud field operation status feature set of the specified aircraft is output, including cloud top temperature value, effective particle radius value, cloud optical thickness value, and cloud top height value.

3. The aircraft icing early warning method according to claim 2, characterized in that, The specific steps for analyzing and setting the cloud liquid water path values ​​for an aircraft are as follows: Read the effective particle radius and optical thickness values ​​of the set aircraft; Obtain the water density value stored in the pre-established database, and combine it with the effective particle radius value and optical thickness value of the set aircraft to analyze the cloud liquid water path value of the set aircraft.

4. The aircraft icing early warning method according to claim 1, characterized in that, The cloud image data specifically includes the albedo value and channel brightness temperature set of each pixel in the cloud image, as well as the corresponding two-dimensional coordinates.

5. The aircraft icing early warning method according to claim 4, characterized in that, The specific steps for analyzing and setting the cloud type temperature correction factor for an aircraft are as follows: The cloud image data of the set aircraft is input into the pre-trained cloud image error mapping model, and the temperature correction mapping feature set of the set aircraft is analyzed, including spectral difference temperature mapping value, cloud undulation temperature mapping value, and absorption and scattering co-shift temperature mapping value. Based on the temperature correction mapping feature set of the set aircraft, the cloud temperature correction value of the set aircraft is analyzed.

6. The aircraft icing early warning method according to claim 5, characterized in that, The specific steps for analyzing and setting the temperature correction mapping feature set of the aircraft are as follows: In the input layer of the cloud image error mapping model, cloud image data of the specified aircraft is received and preprocessed. In the cloud map deconstruction layer of the cloud map error mapping model, feature extraction processing is performed on the preprocessed cloud map data of the specified aircraft to obtain the cloud map radiation feature vector of the specified aircraft. In the mapping output layer of the cloud map error mapping model, the temperature correction mapping feature set of the set aircraft is output based on the cloud map radiation feature vector of the set aircraft.

7. The aircraft icing early warning method according to claim 5, characterized in that, The specific steps for analyzing and setting the optimal cloud top temperature value for the aircraft are as follows: Read the cloud top temperature value, cloud top altitude value, and cloud shape temperature correction value of the set aircraft; The system obtains the flight altitude of the set aircraft and combines it with the cloud top temperature value, cloud top altitude value, and cloud shape temperature correction value to obtain the optimized cloud top temperature value of the set aircraft.

8. The aircraft icing early warning method according to claim 7, characterized in that, The specific formula for calculating the optimal cloud top temperature value for the aircraft is as follows: ； in, To set the optimal cloud top temperature value for the aircraft, To set the cloud top temperature value for the aircraft, To set the cloud top altitude value for the aircraft, To set the aircraft's flight altitude, The temperature decay adjustment coefficient is stored in the database. To set the cloud cover temperature correction value for the aircraft.

9. The aircraft icing early warning method according to claim 1, characterized in that, The specific formula for calculating the icing index of a given aircraft is as follows: ； in, To set the aircraft icing index, To set the cloud liquid water path value for the aircraft, This refers to the ice accumulation amplification factor stored in the database. To set the optimal cloud top temperature value for the aircraft, The temperature response normalization coefficient is stored in the database.

10. The aircraft icing early warning method according to claim 1, characterized in that, The specific steps for issuing icing warnings for designated aircraft based on the icing index are as follows: The icing index of the aircraft is set and compared with the preset warning interval set. Based on the comparison and processing results, the corresponding early warning level is sent to the designated aircraft.

Images