Flight quality evaluation method, device and storage medium integrating QAR and AI

Abstract

The present invention discloses a flight quality evaluation method, device, and storage medium that integrates QAR and AI. The method includes: obtaining historical QAR landing trajectory data and establishing a standard fitting curve; training an AI model using the historical QAR landing trajectory data, and having the AI ​​model learn and output prediction values ​​to construct a flight evaluation model; using the flight evaluation model to predict designated target QAR landing trajectory data to obtain a touchdown distance target prediction value; comparing the touchdown distance target prediction value with the touchdown distance target actual value to obtain a first comparison result; establishing a target trajectory curve based on the target QAR landing trajectory data; comparing the target trajectory curve with the standard fitting curve to obtain a second comparison result; and combining the first and second comparison results to generate a flight quality evaluation result for the target QAR landing trajectory data. The present invention can improve the effectiveness of flight quality evaluation and provide a reference for improving flight quality.

Description

Technical Field

[0001] The present invention relates to the field of computer software technology, and in particular to a flight quality evaluation method, device, and storage medium integrating QAR and AI. Background Art

[0002] During the approach and landing phases of a flight, pilots must precisely and synchronously control over ten parameters, including altitude, speed, and attitude, within just 2-3 minutes. This critical period, from 50 feet to 5 seconds after touchdown—a period of approximately 10 seconds—is crucial for flight safety. However, the current flight training system faces significant bottlenecks: First, due to the lack of standard reference points, flight instructors often rely on subjective experience to judge the correctness of trainees' maneuvers during training, lacking an objective basis for determining flight trajectories. Second, data application is fragmented. While QAR data records information such as altitude, speed, attitude, and pilot actions during landing, this data is fragmented across flights. The lack of an effective model to integrate this large amount of data and map it into a standard approach trajectory makes it difficult to use in guiding flight training. Therefore, improving flight quality evaluation and providing a reference for pilots' flight operations based on the evaluation results is a challenge facing those skilled in the art. Summary of the Invention

[0003] The embodiments of the present invention provide a flight quality evaluation method, device, and storage medium that integrate QAR and AI, aiming to improve the flight quality evaluation effect and provide a reference for improving flight quality.

[0004] In a first aspect, an embodiment of the present invention provides a flight quality evaluation method integrating QAR and AI, including:

[0005] Obtaining historical QAR landing trajectory data for different flights and establishing a standard fitting curve using the historical QAR landing trajectory data; wherein the historical QAR landing trajectory data includes touchdown distance at a preset position, flight level-off start altitude, throttle-down altitude, and vertical G-load at touchdown;

[0006] The AI ​​model is trained using the flight level-off start altitude, throttle-reduction altitude, and vertical overload at touchdown in the historical QAR landing trajectory data, and the AI ​​model learns and outputs corresponding touchdown distance prediction values ​​to construct a flight evaluation model;

[0007] predicting the designated target QAR landing trajectory data using the flight evaluation model to obtain a corresponding touchdown distance target prediction value, and comparing the touchdown distance target prediction value with a touchdown distance target true value in the target QAR landing trajectory data to obtain a first comparison result;

[0008] establishing a target trajectory curve based on the target QAR landing trajectory data, and comparing the target trajectory curve with the standard fitting curve to obtain a second comparison result;

[0009] The first comparison result and the second comparison result are combined to generate a flight quality evaluation result on the target QAR landing trajectory data.

[0010] In a second aspect, an embodiment of the present invention provides a flight quality evaluation device integrating QAR and AI, including:

[0011] a data acquisition unit, configured to acquire historical QAR landing trajectory data of different flights and establish a standard fitting curve using the historical QAR landing trajectory data; wherein the historical QAR landing trajectory data includes touchdown distance at a preset position, flight level-off start altitude, throttle-down altitude, and vertical G-load at touchdown;

[0012] a model building unit, configured to train an AI model using the flight level-off start altitude, throttle-reduction altitude, and vertical overload at touchdown in the historical QAR landing trajectory data, and to have the AI ​​model learn and output a corresponding touchdown distance prediction value to construct a flight evaluation model;

[0013] a prediction and comparison unit, configured to predict the designated target QAR landing trajectory data using the flight evaluation model to obtain a corresponding touchdown distance target prediction value, and compare the touchdown distance target prediction value with a touchdown distance target true value in the target QAR landing trajectory data to obtain a first comparison result;

[0014] a curve comparison unit, configured to establish a target trajectory curve based on the target QAR landing trajectory data, and compare the target trajectory curve with the standard fitting curve to obtain a second comparison result;

[0015] An evaluation generating unit is configured to generate a flight quality evaluation result regarding the target QAR landing trajectory data by combining the first comparison result and the second comparison result.

[0016] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the flight quality evaluation method integrating QAR and AI as described in the first aspect is implemented.

[0017] This embodiment of the present invention establishes a standard fitting curve by acquiring historical QAR landing trajectory data from different flights, including touchdown distance, flare start altitude, throttle reduction altitude, and vertical overload at touchdown. Furthermore, it uses a portion of this historical QAR landing trajectory data to train an AI model to construct a flight evaluation model. The flight evaluation model then predicts the touchdown distance for the target QAR landing trajectory data, which is then compared with the true value to obtain a first comparison result. Simultaneously, the target trajectory curve established from the target QAR landing trajectory data is compared with the standard fitting curve to obtain a second comparison result. These two comparison results are then combined to generate a flight quality evaluation result. This standard fitting curve provides a benchmark for flight quality evaluation. Furthermore, through the training and application of the AI ​​model, touchdown distance predictions are more accurate, improving the objectivity of the evaluation. Furthermore, the combination of the first and second comparison results enables a comprehensive evaluation of the flight's flight quality, providing a basis for flight operation improvements. Ultimately, by continuously optimizing flight operations, overall flight quality can be improved, ensuring flight safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 A flowchart of a flight quality evaluation method integrating QAR and AI provided by an embodiment of the present invention;

[0020] Figure 2 A schematic diagram of a standard fitting curve in a flight quality evaluation method integrating QAR and AI provided by an embodiment of the present invention;

[0021] Figure 3 A schematic block diagram of a flight quality evaluation device integrating QAR and AI provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0023] It will be understood that when used in this specification and the appended claims, the terms “comprises” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0024] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the present invention. As used in the specification and appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0025] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0026] See below Figure 1 An embodiment of the present invention provides a flight quality evaluation method integrating QAR and AI, specifically including: steps S101 to S105.

[0027] Step S101: Acquire historical QAR landing trajectory data of different flights, and establish a standard fitting curve using the historical QAR landing trajectory data; wherein the historical QAR landing trajectory data includes touchdown distance at a preset position, flight level-off start altitude, throttle-down altitude, and vertical G at touchdown;

[0028] Step S102: Using the flight level-off start altitude, throttle-off altitude, and vertical overload at touchdown in the historical QAR landing trajectory data to train an AI model, and having the AI ​​model learn and output a corresponding touchdown distance prediction value to construct a flight evaluation model;

[0029] Step S103: using the flight evaluation model to predict the designated target QAR landing trajectory data to obtain a corresponding touchdown distance target prediction value, and comparing the touchdown distance target prediction value with the touchdown distance target actual value in the target QAR landing trajectory data to obtain a first comparison result;

[0030] Step S104: establishing a target trajectory curve based on the target QAR landing trajectory data, and comparing the target trajectory curve with the standard fitting curve to obtain a second comparison result;

[0031] Step S105 : generating a flight quality evaluation result regarding the target QAR landing trajectory data by combining the first comparison result and the second comparison result.

[0032] In this embodiment, a standard fitting curve is established by acquiring historical QAR landing trajectory data from different flights, including touchdown distance, flare start altitude, throttle reduction altitude, and vertical G-force at touchdown. Furthermore, a portion of this historical QAR landing trajectory data is used to train an AI model to construct a flight evaluation model. This flight evaluation model is then used to predict the touchdown distance for the target QAR landing trajectory data, which is then compared with the true value to obtain a first comparison result. Simultaneously, a target trajectory curve established from the target QAR landing trajectory data is compared with the standard fitting curve to obtain a second comparison result. These two comparison results are then combined to generate a flight quality evaluation result.

[0033] This embodiment, on the one hand, utilizes the established standard fitting curve to provide a benchmark for flight quality evaluation. On the other hand, through the training and application of the AI ​​model, the touchdown distance prediction is made more accurate, thereby improving the objectivity of the evaluation. Furthermore, by combining the first and second comparison results, the flight quality of the flight can be comprehensively evaluated, providing a basis for improving flight operations. Ultimately, by continuously optimizing flight operations, the overall flight quality can be improved, ensuring flight safety.

[0034] It should be noted that the QAR landing trajectory data described in this embodiment specifically refers to the range from 100 feet radio altitude before touchdown to touchdown. The touchdown distance at a preset location refers to the distance from the preset radio altitude to touchdown, for example, 50 feet to touchdown. Here, the aircraft collects parameters such as altitude, speed, attitude, and time in seconds. For this purpose, the collected parameters are radio altitude parameters around 50 feet, and the ground speed integral distance during the second the landing gear changes from airborne to grounded is calculated. This provides the distance from 50 feet to touchdown. The touchdown distance must be within a reasonable range. Each airport has its own touchdown zone, and the touchdown distance cannot be too short or too long. Based on existing flight manuals and flight experience, a reasonable range needs to be set. Therefore, before training the model, this embodiment selected an integral distance range of 1000-2000 feet as the 50-foot touchdown distance. The flare start altitude refers to the time before landing, when the pilot adjusts the aircraft's attitude by adjusting the stick. Since the sensor data may fluctuate slightly due to aircraft motion even if the pilot takes no action, it's necessary to exclude these fluctuations and identify the significant stick changes that clearly indicate pilot control. This is then used to determine the corresponding position as the flare start location. The corresponding radio altitude is then obtained, which serves as the flare start altitude. The throttle-down altitude refers to the time when the pilot needs to reduce the throttle as they approach the ground. Failure to do so could cause the aircraft to plunge into the ground, resulting in a hard landing or damage. Throttle-down is a pilot action and is therefore also recorded by the data recorder. Therefore, a noticeable decrease in engine speed indicates the pilot's throttle reduction. The corresponding radio altitude is then obtained, representing the throttle-down altitude. The vertical overload at touchdown is the touchdown acceleration. When an aircraft descends from the air to the ground, the acceleration is necessarily greater than 1. However, because raw data collection is affected by sensor transmission, the moment of main wheel touchdown may not necessarily coincide with the moment of maximum vertical overload. Therefore, to align the data, this embodiment uses the period from 2 seconds before to 5 seconds after main wheel touchdown to identify the maximum vertical overload at touchdown, which is the acceleration at the moment of maximum ground impact. Based on these four data indicators, a pilot's flight performance can be evaluated. For example, in actual applications, based on existing flight manuals, the throttle should be reduced at around 30 feet. Therefore, the modeling of this embodiment can also determine that a flight in which the pilot reduces the throttle at 30 feet is a flight with relatively reasonable control.

[0035] In one embodiment, obtaining historical QAR landing trajectory data of different flights includes:

[0036] Based on the touchdown distance at the preset position, the flight level-off start altitude, the throttle reduction altitude, and the vertical overload at the touchdown, candidate QAR landing trajectory samples that meet the requirements are selected;

[0037] Performing missing value detection on the candidate QAR landing trajectory samples and filling in missing values ​​with the median when missing values ​​are detected;

[0038] And performing outlier detection on the candidate QAR landing trajectory samples and removing the detected outliers to obtain the historical QAR landing trajectory data.

[0039] After acquiring historical QAR landing trajectory data, we collected two types of flight sample data: samples with good flight quality and samples with flight quality that needs improvement. We ensured that each sample data contained the following key features: level-off start altitude, throttle-off altitude, vertical load at touchdown, and the target variable -50 feet to touchdown distance. Specifically, we performed a preliminary screening of the standard ranges for the four data indicators: touchdown distance at the preset position, level-off start altitude, throttle-off altitude, and vertical load at touchdown, according to the ranges recommended by the aircraft manual. For example:

[0040] (1) 50 feet to touchdown distance, select flights within the range of 1000 to 2000 feet;

[0041] (2) Level off the starting altitude, choose a flight altitude less than 40 feet and greater than 20 feet;

[0042] (3) Throttle down altitude: select a flight altitude less than 40 feet and greater than 20 feet;

[0043] (4) For vertical overload at the moment of touchdown, select an overload range less than 1.4g.

[0044] In actual application, a total of 22,000 flight samples were collected. Based on the aforementioned manual standards and experience, approximately 200 samples with good flight quality were selected, and approximately 200 samples with flight quality that could be improved were also selected. These approximately 400 samples were then trained using an AI model to produce an evaluation model, the flight evaluation model described above.

[0045] After data collection is complete, it is first preprocessed, including: ① Missing value processing: Check for missing values ​​in the data and fill them with the median. ② Outlier processing: Detect and process outliers using the statistical method Z-score to ensure data accuracy and model stability. ③ Standardization processing: Based on professional knowledge in the flight field, perform necessary transformations on the features, and here the features are standardized. Further, the preprocessed data can be divided into a training set and a test set, with the training set accounting for 80% and the test set accounting for 20%. Specifically, a random division method can be used to ensure the representativeness of the data and the generalization ability of the model.

[0046] In addition, the integrity of the flight can also be tested, that is, based on the full data, the landing data can be intercepted. Once the flight does not contain the landing data, it means that the flight data itself is missing. In this case, the flight itself is an unusable flight, and this part of the flight data is discarded.

[0047] In one embodiment, establishing a standard fitting curve using the historical QAR landing trajectory data includes:

[0048] Performing multiple spline interpolation and filter fitting processes on the historical QAR landing trajectory data to obtain smooth flight trajectory curves of different flights;

[0049] An average value is calculated for the flight trajectory smooth curve, and a result of the average value calculation is set as the standard fitting curve.

[0050] In establishing the standard fitting curve, this embodiment calculates a large number of flight data for the distance from 50 feet radio altitude to touchdown, the flight leveling start altitude, the throttle reduction altitude, and the vertical overload at the time of touchdown. Then, all flights whose four data indicators all meet the standard range are obtained. Then, the radio altitudes of all flights that meet the conditions (that is, the standard parameters describing the flight trajectory of the aircraft) are interpolated with cubic spline, and Savitzky-Golay filter fitting is used to obtain a smooth curve of the flight trajectory of all flights. Finally, the average of the smooth curves within the range of conditions is taken as the standard fitting curve. Figure 2 As shown, Figure 2 The flight curve from the time the flight's radio altitude is around 100 feet to the end of reverse thrust is shown, where the black line is the standard fitting curve finally found.

[0051] The flight quality of a flight can be determined by comparing the flight's radio altitude flight curve with the standard fitting curve. Specifically, the closer the flight's radio altitude flight curve matches the standard fitting curve, the higher the flight quality. Conversely, the further the flight's radio altitude flight curve deviates from the standard fitting curve, the lower the flight quality.

[0052] In one embodiment, the training of an AI model using the flight level-off start altitude, throttle-off altitude, and vertical overload at touchdown in the historical QAR landing trajectory data, and the learning and outputting of the corresponding touchdown distance prediction value by the AI ​​model to construct a flight evaluation model, includes:

[0053] Extracting characteristic values ​​corresponding to the flight's leveling start altitude, throttle reduction altitude, and vertical overload at touchdown;

[0054] Normalizing the eigenvalues ​​to obtain a characteristic matrix;

[0055] The feature matrix is ​​input into the XGBoost regression model for training, and the XGBoost regression model learns and outputs the corresponding ground distance prediction value.

[0056] This embodiment uses the XGBoost regression model to construct a flight evaluation model when training the AI ​​model using the historical QAR landing trajectory data. First, the XGBoost regression model's associated libraries are imported and initialized. Then, the eigenvalues ​​are filled with missing values, processed for outliers, and normalized. Here, the eigenvalues ​​are specifically the leveling start altitude, throttle-down altitude, and vertical load at touchdown. The touchdown distance serves as the target variable, i.e., the output value of the regression task. The aforementioned operations are not required; only the model-predicted values ​​need to be evaluated and predicted.

[0057] When the eigenvalue is normalized, it can be calculated according to the following formula:

[0058] ;

[0059] Among them, X 标准化 represents the characteristic matrix obtained after normalization, X represents the eigenvalues ​​(flare start altitude, throttle reduction altitude, vertical load at touchdown), μ is the mean of the eigenvalues, and σ is the standard deviation of the eigenvalues.

[0060] Then, the feature matrix is ​​input into the XGBoost regression model, and the XGBoost regression model predicts the output ground distance prediction value , for example, 50 feet to the ground distance value, where n is the number of samples.

[0061] This embodiment uses the leveling start altitude, the throttle reduction altitude, and the vertical load at the touchdown moment to interpret and predict the value of the target variable 50 feet to touchdown distance, so that the final output value of the XGBoost regression model is the predicted value of the flight's 50 feet to touchdown distance. In this way, a flight evaluation model capable of predicting the touchdown distance can be constructed. Subsequently, it is only necessary to input the flight data to be evaluated into the flight evaluation model, and the touchdown distance can be predicted by the flight evaluation model. The predicted touchdown distance can then be compared with the actual touchdown distance to determine whether the pilot of the corresponding flight has flight technical problems, thereby achieving flight quality evaluation. Furthermore, for pilots whose flight quality needs to be improved, detailed analysis can be conducted through manual detailed review and analysis of flight data, and targeted corrections can be made to improve subsequent flight quality.

[0062] In a specific embodiment, the method of training an AI model using the flight level-off start altitude, throttle-off altitude, and vertical overload at touchdown in the historical QAR landing trajectory data, and having the AI ​​model learn and output a corresponding touchdown distance prediction value to construct a flight evaluation model, further includes:

[0063] The model parameters were set using an open source directory tree generation tool, and the XGBoost regression model was cross-validated using Bayesian optimization to obtain the optimal parameter combination of the XGBoost regression model.

[0064] The XGBoost regression model is trained in combination with the optimal parameter combination.

[0065] In this example, parameters are set using GTree (an open-source directory tree generation tool). These parameters include the gamma splitting threshold, the maximum tree depth ( max_depth ), the lambda regularization parameter, the pre-training sample sampling ratio ( subsample ), the pre-training feature sampling ( colsample_bytree ), the eta learning rate, the seed random seed, and the number of threads for parallel computing ( nthread ). Bayesian optimization is also used to perform XGBoost CV cross-validation to find the optimal parameter combination in the parameter space. Model training is then performed based on this optimal parameter combination.

[0066] In practice, the current parameter combination is first obtained from the Bayesian optimization iteration. The full training set dtrain (containing features X and labels y) is then evenly divided into five subsets (nfold=5). In each iteration, four subsets are used as the training set and one subset is used as the validation set, and this cycle is repeated five times to form a 5-fold cross-validation. For each parameter combination, the model is trained on each of the five folds and the evaluation metric (RMSE) is calculated. Each training round is repeated for a maximum of 100 iterations (num_boost_round=100), and early stopping (early_stopping_rounds=20) is used to prevent overfitting: if the validation error does not decrease after 20 consecutive rounds, the model is terminated early. The results are then aggregated, and the evaluation results (RMSE) of the five folds are averaged to obtain the final performance metric for that parameter combination. The optimal number of iterations (i.e., the number of iterations that minimized the validation error) is returned.

[0067] In one embodiment, the method of training an AI model using the flight level-off start altitude, throttle-off altitude, and vertical overload at touchdown in the historical QAR landing trajectory data, and having the AI ​​model learn and output corresponding touchdown distance prediction values ​​to construct a flight evaluation model, further includes:

[0068] According to the following formula, the loss function is calculated using the mean square error, and the parameters of the flight evaluation model are updated according to the loss function:

[0069] ;

[0070] Among them, MSE represents mean square error, Indicates the true value of the ground distance, Indicates the predicted ground distance value.

[0071] In addition, the method of training an AI model using the flight level-off start altitude, throttle-off altitude, and vertical overload at touchdown in the historical QAR landing trajectory data, and having the AI ​​model learn and output a corresponding touchdown distance prediction value to construct a flight evaluation model, further includes:

[0072] According to the following formula, the flight evaluation model is evaluated using the R-squared coefficient of determination to obtain a model evaluation result, and the parameters of the flight evaluation model are updated according to the model evaluation result:

[0073] ;

[0074] where R 2 represents the R-squared coefficient of determination, Indicates the true value of the ground distance, represents the predicted value of ground distance, Represents the average of all true ground distance values.

[0075] This embodiment uses the mean squared error (MSE) to measure the difference between the predicted distance and the true distance, and the R-squared coefficient of determination (R-squared) to measure the proportion of variance explained by the model (the closer to 1, the better). In practical applications, the model can be evaluated using test data and the R-squared coefficient of determination and the mean squared error (MSE) can be calculated. The closer the R-squared coefficient of determination score is to 1 and the smaller the MSE, the better the model's predictive ability. Otherwise, the flight evaluation model needs to be adjusted and updated. For example, when R-squared is less than 0.8 and / or the MSE exceeds the service tolerance threshold, the flight evaluation model parameters are adjusted and updated.

[0076] Figure 3 A schematic block diagram of a flight quality evaluation device 300 integrating QAR and AI is provided in an embodiment of the present invention. The device 300 includes:

[0077] The data acquisition unit 301 is configured to acquire historical QAR landing trajectory data of different flights and establish a standard fitting curve using the historical QAR landing trajectory data; wherein the historical QAR landing trajectory data includes touchdown distance at a preset position, flight level-off start altitude, throttle reduction altitude, and vertical G at touchdown;

[0078] The model building unit 302 is configured to train an AI model using the flight level-off start altitude, throttle-off altitude, and vertical overload at touchdown in the historical QAR landing trajectory data, and to have the AI ​​model learn and output a corresponding touchdown distance prediction value to construct a flight evaluation model.

[0079] a prediction and comparison unit 303 configured to predict the designated target QAR landing trajectory data using the flight evaluation model to obtain a corresponding touchdown distance target prediction value, and compare the touchdown distance target prediction value with a touchdown distance target true value in the target QAR landing trajectory data to obtain a first comparison result;

[0080] a curve comparison unit 304, configured to establish a target trajectory curve based on the target QAR landing trajectory data, and compare the target trajectory curve with the standard fitting curve to obtain a second comparison result;

[0081] The evaluation generating unit 305 is configured to generate a flight quality evaluation result regarding the target QAR landing trajectory data by combining the first comparison result and the second comparison result.

[0082] In one embodiment, the data acquisition unit 301 includes:

[0083] The sample selection unit is used to select candidate QAR landing trajectory samples that meet the requirements based on four data indicators: touchdown distance at a preset position, flight level-off start altitude, throttle reduction altitude, and vertical overload at touchdown;

[0084] a missing value filling unit, configured to perform missing value detection on the candidate QAR landing trajectory sample and fill in the missing value using the median when a missing value is detected;

[0085] The anomaly elimination unit is used to perform outlier detection on the candidate QAR landing trajectory samples and eliminate the detected outliers to obtain the historical QAR landing trajectory data.

[0086] In one embodiment, the data acquisition unit 301 further includes:

[0087] an interpolation filtering unit, configured to perform multiple spline interpolation and filtering fitting processes on the historical QAR landing trajectory data to obtain smooth flight trajectory curves for different flights;

[0088] The curve averaging unit is used to calculate the average value of the flight trajectory smooth curve and set the result of the average value calculation as the standard fitting curve.

[0089] In one embodiment, the model building unit 302 includes:

[0090] a feature extraction unit for extracting feature values ​​corresponding to the leveling start altitude, the throttle reduction altitude, and the vertical overload at the touchdown moment of the flight;

[0091] A feature normalization unit, configured to perform normalization processing on the eigenvalues ​​to obtain a feature matrix;

[0092] The model training unit is used to input the feature matrix into the XGBoost regression model for training, and the XGBoost regression model learns and outputs the corresponding ground distance prediction value.

[0093] In one embodiment, the model building unit 302 includes:

[0094] A parameter setting unit is used to set model parameters using an open source directory tree generation tool, and to perform CV cross-validation on the XGBoost regression model using Bayesian optimization to obtain the optimal parameter combination of the XGBoost regression model;

[0095] A parameter training unit is used to train the XGBoost regression model in combination with the optimal parameter combination.

[0096] In one embodiment, the model building unit 302 further includes:

[0097] A parameter updating unit is configured to calculate a loss function using a mean square error according to the following formula, and to update parameters of the flight evaluation model according to the loss function:

[0098] ;

[0099] Among them, MSE represents mean square error, Indicates the true value of the ground distance, Indicates the predicted ground distance value.

[0100] In one embodiment, the model building unit 302 further includes:

[0101] The model evaluation unit is configured to evaluate the flight evaluation model using the R-squared coefficient of determination according to the following formula to obtain a model evaluation result, and update the parameters of the flight evaluation model according to the model evaluation result:

[0102] ;

[0103] where R 2 represents the R-squared coefficient of determination, Indicates the true value of the ground distance, represents the predicted value of ground distance, Represents the average of all true ground distance values.

[0104] Since the embodiments of the apparatus part correspond to the embodiments of the method part, please refer to the description of the embodiments of the method part for the embodiments of the apparatus part, and they will not be repeated here.

[0105] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When executed, the computer program can implement the steps provided in the above embodiment. The storage medium may include a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, an optical disk, or other medium capable of storing program code.

[0106] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the various embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of this application, several improvements and modifications can be made to this application, and these improvements and modifications also fall within the scope of protection of the claims of this application.

[0107] It should also be noted that, in this specification, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element.

Claims

1. A flight quality evaluation method integrating QAR and AI, characterized in that: include: Obtaining historical QAR landing trajectory data for different flights and establishing a standard fitting curve using the historical QAR landing trajectory data; wherein the historical QAR landing trajectory data includes touchdown distance at a preset position, flight level-off start altitude, throttle-down altitude, and vertical G-load at touchdown; The AI ​​model is trained using the flight level-off start altitude, throttle-reduction altitude, and vertical overload at touchdown in the historical QAR landing trajectory data, and the AI ​​model learns and outputs corresponding touchdown distance prediction values ​​to construct a flight evaluation model; predicting the designated target QAR landing trajectory data using the flight evaluation model to obtain a corresponding touchdown distance target prediction value, and comparing the touchdown distance target prediction value with a touchdown distance target true value in the target QAR landing trajectory data to obtain a first comparison result; establishing a target trajectory curve based on the target QAR landing trajectory data, and comparing the target trajectory curve with the standard fitting curve to obtain a second comparison result; generating a flight quality evaluation result regarding the target QAR landing trajectory data by combining the first comparison result and the second comparison result; The AI ​​model is trained using the flight level-off start altitude, throttle-down altitude, and vertical overload at touchdown in the historical QAR landing trajectory data, and the AI ​​model learns and outputs corresponding touchdown distance prediction values ​​to construct a flight evaluation model, including: Extracting characteristic values ​​corresponding to the flight's leveling start altitude, throttle reduction altitude, and vertical overload at touchdown; Normalizing the eigenvalues ​​to obtain a characteristic matrix; Inputting the feature matrix into an XGBoost regression model for training, and having the XGBoost regression model learn and output a corresponding ground distance prediction value; According to the following formula, the loss function is calculated using the mean square error, and the parameters of the flight evaluation model are updated according to the loss function: Among them, MSE represents the mean square error, y i Indicates the true value of the ground distance, y i represents the predicted value of ground distance; According to the following formula, the flight evaluation model is evaluated using the R-squared coefficient of determination to obtain a model evaluation result, and the parameters of the flight evaluation model are updated according to the model evaluation result: where R 2 represents the R-square determination coefficient, y i Indicates the true value of the ground distance, y i Denotes the predicted ground distance, y i Represents the average of all true ground distance values.

2. The flight quality evaluation method integrating QAR and AI according to claim 1, characterized in that: The acquisition of historical QAR landing trajectory data for different flights includes: Based on the touchdown distance at the preset position, the flight level-off start altitude, the throttle reduction altitude, and the vertical overload at the touchdown, candidate QAR landing trajectory samples that meet the requirements are selected; Performing missing value detection on the candidate QAR landing trajectory samples and filling in missing values ​​with the median when missing values ​​are detected; And performing outlier detection on the candidate QAR landing trajectory samples and removing the detected outliers to obtain the historical QAR landing trajectory data.

3. The flight quality evaluation method integrating QAR and AI according to claim 1, characterized in that: The method of establishing a standard fitting curve using the historical QAR landing trajectory data includes: Performing multiple spline interpolation and filter fitting processes on the historical QAR landing trajectory data to obtain smooth flight trajectory curves of different flights; An average value is calculated for the flight trajectory smooth curve, and a result of the average value calculation is set as the standard fitting curve.

4. The flight quality evaluation method integrating QAR and AI according to claim 1, characterized in that: The method further includes training an AI model using the flight level-off start altitude, throttle-off altitude, and vertical overload at touchdown in the historical QAR landing trajectory data, and having the AI ​​model learn and output a corresponding touchdown distance prediction value to construct a flight evaluation model. The model parameters were set using an open source directory tree generation tool, and the XGBoost regression model was cross-validated using Bayesian optimization to obtain the optimal parameter combination of the XGBoost regression model. The XGBoost regression model is trained in combination with the optimal parameter combination.

5. A flight quality evaluation device integrating QAR and AI, characterized in that: include: a data acquisition unit, configured to acquire historical QAR landing trajectory data of different flights and establish a standard fitting curve using the historical QAR landing trajectory data; wherein the historical QAR landing trajectory data includes touchdown distance at a preset position, flight level-off start altitude, throttle-down altitude, and vertical G-load at touchdown; a model building unit, configured to train an AI model using the flight level-off start altitude, throttle-reduction altitude, and vertical overload at touchdown in the historical QAR landing trajectory data, and to have the AI ​​model learn and output a corresponding touchdown distance prediction value to construct a flight evaluation model; a prediction and comparison unit, configured to predict the designated target QAR landing trajectory data using the flight evaluation model to obtain a corresponding touchdown distance target prediction value, and compare the touchdown distance target prediction value with a touchdown distance target true value in the target QAR landing trajectory data to obtain a first comparison result; a curve comparison unit, configured to establish a target trajectory curve based on the target QAR landing trajectory data, and compare the target trajectory curve with the standard fitting curve to obtain a second comparison result; an evaluation generating unit, configured to generate a flight quality evaluation result regarding the target QAR landing trajectory data in combination with the first comparison result and the second comparison result; The model building unit includes: a feature extraction unit for extracting feature values ​​corresponding to the leveling start altitude, the throttle reduction altitude, and the vertical overload at the touchdown moment of the flight; A feature normalization unit, configured to perform normalization processing on the eigenvalues ​​to obtain a feature matrix; A model training unit, configured to input the feature matrix into an XGBoost regression model for training, and have the XGBoost regression model learn and output a corresponding ground distance prediction value; A parameter updating unit is configured to calculate a loss function using a mean square error according to the following formula, and to update parameters of the flight evaluation model according to the loss function: Among them, MSE represents the mean square error, y i Indicates the true value of the ground distance, y i represents the predicted value of ground distance; The model evaluation unit is used to evaluate the flight evaluation model using the R-squared coefficient of determination according to the following formula to obtain a model evaluation result, and update the parameters of the flight evaluation model according to the model evaluation result: where R 2 represents the R-square determination coefficient, y i Indicates the true value of the ground distance, y i Denotes the predicted ground distance, y i Represents the average of all true ground distance values.

6. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the flight quality evaluation method integrating QAR and AI as described in any one of claims 1 to 4 is implemented.

Images