Aircraft landing speed prediction method based on flight parameter data
By screening important flight parameters and using gradient lift regression models for real-time prediction, the problem of inaccurate landing speed prediction in the aircraft's approach phase is solved, the prediction accuracy and calculation efficiency are improved, and the safe and efficient operation management of the terminal area is supported.
Patent Information
- Application Number
- CN202411748482.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-02
- Publication Date
- 2025-03-25
AI Technical Summary
The prior art is difficult to accurately predict the landing speed of an aircraft during the approach phase, resulting in uncertain flight intervals, affecting landing safety and terminal area operation efficiency.
By collecting the aircraft's historical flight parameter data, using the recursive feature elimination method to filter out parameters that have a greater impact on landing speed prediction, construct an initial flight parameter set, and build a training data set based on real-time flight parameter data, and train a gradient-enhancing regression model for real-time prediction.
It improves the accuracy and calculation efficiency of aircraft landing speed prediction, and supports the safe and efficient operation of aircraft operation management in terminal areas.
Smart Images
Figure CN119691392A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a method for predicting an aircraft landing speed, in particular to a method for predicting an aircraft landing speed based on flight parameter data. Background Art
[0002] This section merely provides background information related to the present disclosure and is not necessarily prior art.
[0003] With the rapid growth of civil aviation transport volume, the congestion in terminal areas is becoming increasingly serious, and safety risks are also increasing. It is necessary to continuously increase the capacity of congested terminal areas to meet the growing air traffic demand. One of the prerequisites for the implementation of safe and efficient operation management of terminal areas is to ensure the safe interval of aviation operations, which relies on accurate prediction of the speed of aircraft during the approach and landing phase, thereby supporting the provision of precise intervals between consecutively arriving aircraft.
[0004] The current control procedures limit the flight speed of aircraft in most airspaces, while the landing speed of the aircraft during the approach phase is mainly determined by the pilot. The uncertainty of the landing speed will affect the size of the flight interval, which may affect the landing safety of the aircraft and the operating efficiency of the terminal area. Therefore, accurate prediction of the speed of the aircraft during the approach phase is of great significance to improving flight safety and efficiency. Currently, civil aircraft are generally equipped with Quick Access Recorders (QARs), which record the flight parameter data of various aircraft collected in real time and can be transmitted to the ground through ground-to-air data links and other methods. QAR data is easy to read and analyze, and can record and reflect the operating status of the aircraft in detail. Various flight parameters are also very helpful for predicting the landing speed of the aircraft.
[0005] Relevant research institutions have conducted a lot of research on aircraft speed prediction, among which landing speed prediction based on flight parameter data and using machine learning algorithms is one of the important research directions. However, there are many types of aircraft flight parameter data, and the actual operating environments of different aircraft are different. Simply selecting and using a few flight parameter data to carry out speed prediction is often difficult to achieve good actual prediction results, such as the paper "Aircraft Landing Speed Prediction Model Based on Neural Network". There are many types of aircraft flight parameters. How to identify and screen important flight parameters that have a greater impact on aircraft landing speed prediction, so as to construct an effective landing speed prediction data set, has a great impact on improving the prediction accuracy. In addition, different aircraft have different operating states and environments. If the historical flight parameter data of other aircraft is used to predict the landing speed of the current aircraft, it is inevitable that certain errors will occur. It is more effective to use the flight parameter data transmitted by the current aircraft in real time to make a real-time rolling prediction of its landing speed. Therefore, it is urgent to use a method that can effectively screen important flight parameter data and can perform dynamic, efficient and accurate predictions based on the real-time flight parameter data of the aircraft.
[0006] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention
[0007] Purpose of the invention: The technical problem to be solved by the present invention is to provide an aircraft landing speed prediction method based on flight parameter data in view of the shortcomings of the prior art.
[0008] In order to solve the above technical problems, the present invention discloses a method for predicting aircraft landing speed based on flight parameter data, comprising the following steps:
[0009] Step 1: collect historical flight parameter data of the aircraft, analyze and select parameters that affect the aircraft landing speed prediction, construct an initial flight parameter set, and obtain corresponding historical original data from the historical flight parameter data based on the initial flight parameter set;
[0010] Step 2: preprocess the historical raw data to obtain initial data;
[0011] Step 3, extracting flight parameter data of the landing phase from the initial data;
[0012] Step 4: Based on the flight parameter data of the landing phase, a recursive feature elimination method is used to screen and construct a parameter set affecting the aircraft landing speed prediction;
[0013] Step 5: According to the aircraft landing speed prediction influencing parameter set, a training data set is constructed based on the aircraft real-time flight parameter data, and the landing speed prediction model is trained and updated in real time to perform real-time prediction of the aircraft landing speed.
[0014] Furthermore, the parameters that affect the aircraft landing speed prediction described in step 1 include:
[0015] Radio altitude, pressure altitude, indicated airspeed, pitch angle, roll angle, heading angle, vertical acceleration, lateral acceleration, longitudinal acceleration, aileron left, aileron right, flaps left, flaps right, rudder and elevator.
[0016] Furthermore, the preprocessing of the historical raw data described in step 2 includes:
[0017] Step 2-1, check whether there are missing values in the flight parameter data in the historical original data. If there are missing values, use linear interpolation to fill in the missing flight parameter data;
[0018] Step 2-2, use the three sigma criterion to check the outliers in the historical raw data, discard the detected outliers, and use linear interpolation to fill in the discarded data.
[0019] Furthermore, in step 3, flight parameter data of the landing phase is extracted from the initial data, including:
[0020] Step 3-1, selecting a time interval during which the aircraft's radio altitude parameter continuously decreases from a preset value to 0 as a landing phase time interval;
[0021] Step 3-2, extracting the flight parameter data within the landing phase time interval from the initial data as the flight parameter data of the aircraft landing phase.
[0022] Furthermore, in step 4, a recursive feature elimination method is used to screen and construct a set of parameters affecting the aircraft landing speed prediction, including:
[0023] Step 4-1, construct a multiple linear regression model;
[0024] Step 4-2, using the flight parameter data of the landing phase extracted in step 3 to form a current flight parameter set;
[0025] Step 4-3, using the current flight parameter set to train the multiple linear regression model, and recording the regression coefficients corresponding to the flight parameters in the multiple linear regression model and the prediction accuracy of the multiple linear regression model during the training process;
[0026] Step 4-4, sort the importance of the flight parameters according to the absolute value of the regression coefficient. The larger the absolute value of the regression coefficient, the greater the importance of the corresponding flight parameter.
[0027] Step 4-5, delete the least important flight parameter in the current sorting, and use the remaining flight parameters to construct a new flight parameter set;
[0028] Step 4-6, determine whether there is only one flight parameter left in the new flight parameter set, if so, execute step 4-7, otherwise take the new flight parameter set as the current flight parameter set and return to execute step 4-3:
[0029] Step 4-7, according to the prediction accuracy of the multivariate linear regression model recorded in step 4-3, select the flight parameter set with the highest prediction accuracy and the least number of flight parameters to form the aircraft landing speed prediction influencing parameter set;
[0030] Furthermore, the multivariate linear regression model in step 4-1 is expressed as:
[0031]
[0032] in, is the predicted landing speed, x 1 , x 2 , ..., x m Represents flight parameter data, β 1 , β 2 , ..., β m is the regression coefficient corresponding to the flight parameters, β 0 is a constant compensation factor, m is the number of flight parameters;
[0033] β 0 and β 1 , β 2 , ..., β m The values of are obtained using the least squares estimation method based on flight parameter data.
[0034] Furthermore, the prediction accuracy of the multivariate linear regression model in step 4-1 is expressed by mean square error MSE, and the calculation method is:
[0035]
[0036] Among them, y i is the true value of the landing speed of the i-th flight parameter data sample, represents the predicted landing speed value of the i-th flight parameter data sample, and n is the number of flight parameter data samples used; the smaller the MSE, the higher the prediction accuracy of the multivariate linear regression model.
[0037] Furthermore, the real-time prediction of the aircraft landing speed described in step 5 includes:
[0038] Step 5-1, collect and record the actual flight parameter data and landing speed of the aircraft to be predicted in real time, and initialize the time window size to be W and the prediction time interval to be P;
[0039] Step 5-2, for the current time point t, according to the flight parameters included in the aircraft landing speed prediction influencing parameter set, extract a total of W groups of flight parameters corresponding to tWP seconds to tP-1 seconds as input data in the training data set, extract a total of W groups of landing speeds from tW seconds to t-1 seconds as output data in the training data set, and construct a training data set including W groups of training data;
[0040] Step 5-3, selecting a gradient boosting regression model as the landing speed prediction model;
[0041] Step 5-4, using the constructed W sets of training data to train the landing speed prediction model constructed in step 5-3;
[0042] Step 5-5, based on the trained landing speed prediction model, the flight parameters from t-P+1 to t seconds are used as inputs to predict the landing speed from t+1 to t+P;
[0043] Step 5-6, update time t=t+1, return and execute steps 5-2 to 5-6, and realize the real-time prediction of the landing speed of the aircraft.
[0044] Furthermore, the time window size W defined in step 5-1 is 15, and the prediction time interval P is 5.
[0045] Furthermore, the gradient boosting regression model described in step 5-3 is a standard gradient boosting regression model in the sklearn function library, and the number of weak learners n_estimators=100, the contribution rate of each weak learner learning_rate=0.1, and the random number generator seed random_state=42 are set.
[0046] Beneficial effects:
[0047] The present invention uses recursive feature elimination to screen the flight parameters in the landing phase, and then performs rolling prediction of the landing speed based on the real-time flight parameter data of the aircraft, thereby improving the calculation efficiency and prediction accuracy of the landing speed prediction, and providing support for the terminal area aircraft operation management, specifically including:
[0048] (1) Based on the flight parameter data of the historical landing phase, the recursive feature elimination algorithm is used to analyze and screen the flight parameters, and the key flight parameters that have a greater impact on the landing speed prediction are selected. This not only helps to reduce the calculation burden of the subsequent prediction model, but also helps to delete redundant and interfering flight parameter data and improve the prediction accuracy.
[0049] (2) During the landing speed prediction process, the real-time flight parameter data of the current aircraft is used to roll-train the prediction model, and the latest flight parameters are used to predict the landing speed of the aircraft. This can utilize the latest flight status information of the current aircraft, thereby improving the prediction accuracy of the landing speed. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments, and the above and / or other advantages of the present invention will become more clear.
[0051] Figure 1 It is a schematic diagram of the overall process of the present invention.
[0052] Figure 2 It is a schematic diagram of the change in model prediction accuracy during the flight parameter screening process.
[0053] Figure 3 This is a schematic diagram of the aircraft landing speed prediction test results. DETAILED DESCRIPTION
[0054] The present invention aims at the shortcomings of the existing aircraft landing speed prediction algorithm and provides an aircraft landing speed prediction method based on flight parameter data. The method can identify and select flight parameter data that has a greater impact on the landing speed, and dynamically roll the aircraft landing speed based on the real-time collected flight parameter data, thereby realizing a rapid and accurate prediction of the aircraft landing speed during the approach phase, and providing support for safe and efficient terminal area operation management. The technical solution of the present invention specifically includes the following steps:
[0055] Step 1: Collect historical flight parameter data from the aircraft's quick access recorder, sort out and analyze the parameters that affect the aircraft's landing speed prediction to construct an initial flight parameter set, and obtain the corresponding historical original data;
[0056] Step 2: Preprocess the historical raw data to obtain initial data;
[0057] Step 3: Extract the flight parameter data of the landing phase from the initial data;
[0058] Step 4: Based on the flight parameter data of the landing phase, a set of parameters influencing the aircraft landing speed prediction is constructed using recursive feature elimination screening;
[0059] Step 5: According to the aircraft landing speed prediction influencing parameter set, a training data set is constructed based on the aircraft real-time flight parameter data and a landing speed prediction model is trained to predict the aircraft landing speed.
[0060] Further, the parameters that affect the aircraft landing speed prediction described in step 1 are analyzed to construct the initial flight parameter set, which specifically includes: radio altitude, pressure altitude, indicated airspeed, pitch angle, roll angle, heading angle, vertical acceleration, lateral acceleration, longitudinal acceleration, aileron left, aileron right, flap left, flap right, rudder, elevator. These parameters are important operating status data during the operation of the aircraft.
[0061] Furthermore, the raw data described in step 2 is preprocessed, specifically, checking whether there are missing values in the flight parameter data in the raw data. If there are missing values, the missing flight parameter data is supplemented using linear interpolation. Then the three sigma criterion is used to check the outliers in the raw data, the detected outliers are discarded, and the discarded outlier parameter data is supplemented using linear interpolation. Linear interpolation and three sigma criteria are typical and commonly used data preprocessing methods in the field of machine learning. The calculation method is simple and the data processing effect is good.
[0062] Furthermore, in step 3, flight parameter data of the landing phase is extracted from the initial data. Specifically, the radio altitude parameter data of the aircraft is used as a reference, and the time interval in which the radio altitude continuously decreases from 1500 feet to 0 feet is selected as the landing phase time interval. Then, various flight parameter data within the landing phase time interval are extracted from the initial data, so as to obtain the flight data of the aircraft during the landing phase.
[0063] Radio altitude is one of the important reference data for dividing the aircraft operation phases. According to the general operation process of civil aircraft, the time interval when the radio altitude continuously decreases from 1500 feet to 0 feet is selected as the landing phase time interval. This division method is relatively simple to calculate and conforms to the actual operation process, and can quickly identify the landing phase of the aircraft.
[0064] Furthermore, in step 4, a recursive feature elimination algorithm is used to screen and construct a set of parameters that affect the prediction of aircraft landing speed. The recursive feature elimination algorithm selects the optimal feature subset by iteratively eliminating the features that contribute the least to the model performance and retraining the model to evaluate the importance of the remaining features. This method not only takes into account the relationship between features and reduces the risk of overfitting, but also improves the performance and generalization ability of the model. The recursive feature elimination algorithm does not require any assumptions about the distribution of features, so it is applicable to various types of data. It is a flexible and powerful feature selection algorithm that is widely used in the field of machine learning.
[0065] Furthermore, in step 4, a recursive feature elimination algorithm is used to screen and construct a set of parameters influencing the aircraft landing speed prediction. The specific calculation method is:
[0066] Step 4-1: Use the flight parameter data of the aircraft landing phase corresponding to the current flight parameter set to train a multiple linear regression model, and record the regression coefficients corresponding to each flight parameter in the current model and the prediction accuracy of the multiple linear regression model.
[0067] The function of the multiple linear regression model is expressed as:
[0068]
[0069] in, is the model's predicted landing speed, x 1 , x 2 , ..., x m Represents flight parameter data, β 1 , β 2 , ..., β m is the regression coefficient corresponding to the flight parameters, β 0 is a constant compensation factor, and m is the number of flight parameters. 0 and β 1 , β 2 , ..., β m The value of can be obtained using the least squares estimation method based on the flight parameter data.
[0070] The advantage of using multivariate linear regression as the basic model in the recursive feature elimination algorithm is that its model is simple and computationally efficient, easy to understand and implement, and the linear regression coefficient can directly indicate the importance of the feature, providing an intuitive basis for feature selection.
[0071] The mean square error (MSE) is used as a measure of the prediction accuracy of the multivariate linear regression model, and its calculation method is:
[0072]
[0073] Where y is the true value of landing speed, is the landing speed prediction value of the model, and n is the number of flight parameter data samples used. The smaller the MSE, the higher the prediction accuracy of the model.
[0074] The mean square error (MSE) is a commonly used indicator for evaluating the performance of regression models. Its advantage is that it can amplify the impact of errors in the form of squares, so that larger errors receive more attention during the model optimization process, thereby stimulating the model to be more sensitive to outliers. At the same time, the results of MSE are easy to interpret, the calculation is simple and direct, and it is easy to implement.
[0075] Step 4-2: Sort the importance of the flight parameters according to the absolute value of the regression coefficient corresponding to each recorded flight parameter. The larger the absolute value of the regression coefficient, the greater the importance of the corresponding flight parameter.
[0076] Step 4-3: Delete the flight parameter with the lowest current importance and construct a new flight parameter set using the remaining flight parameters.
[0077] Step 4-4: Determine whether there is only one flight parameter left in the current flight parameter set. If yes, proceed to step 4-5; otherwise, return to step 4-1.
[0078] Step 4-5: According to the recorded prediction accuracy of the multivariate linear regression model, select the flight parameter set with the highest prediction accuracy and the least number of features to construct the aircraft landing speed prediction influencing parameter set.
[0079] Furthermore, in step 5, a training data set is constructed based on the real-time flight parameter data of the aircraft and a landing speed prediction model is trained to predict the landing speed of the aircraft. This prediction method uses the real-time flight parameter data of the aircraft to be predicted for prediction. These flight parameter data can better reflect the current operating status of the aircraft, thereby improving the accuracy of the prediction.
[0080] Furthermore, in step 5, a training data set is constructed based on the real-time flight parameter data of the aircraft and a landing speed prediction model is trained to predict the landing speed of the aircraft. The specific prediction steps are as follows:
[0081] Step 5-1: Collect and record the actual flight parameter data and landing speed of the aircraft to be predicted every second, initialize the time window size W, and define the prediction time interval P.
[0082] The time window size W represents the use of W sets of flight parameter data in the window for model training. The prediction time interval P represents the use of the flight parameter data at the tth moment to predict the landing speed at the future (t+P)th second. Based on the air traffic management operation experience and considering the efficiency of real-time calculation, it is recommended to set it to W=15 and P=5.
[0083] Step 5-2: For the current time point t, according to the flight parameters included in the aircraft landing speed prediction influencing parameter set, a total of W groups of flight parameter data corresponding to (tWP) seconds to (tP-1) seconds are extracted as the input of the prediction model, and a total of W groups of actual landing speeds from (tW) seconds to (t-1) seconds are extracted as the output of the prediction model, thereby constructing W groups of training data.
[0084] Step 5-3: Use the constructed W set of training data to train the gradient boosting regression model as the aircraft landing speed prediction model.
[0085] The gradient boosting regression model gradually improves the prediction accuracy of the model by iteratively adding decision trees, each tree is fitted to the residual of the previous round. This method can not only handle complex nonlinear relationships, but also effectively control the risk of overfitting by adjusting parameters such as the learning rate and the number of trees. In addition, the gradient boosting model usually has good robustness and can resist the influence of outliers, which is very flexible and effective in practical applications.
[0086] Specifically, the standard gradient boosting regression model in the sklearn function library is used. The number of weak learners (decision trees) used in the model is n_estimators = 100, the contribution rate of each weak learner is learning_rate = 0.1, and the random number generator seed is random_state = 42.
[0087] Step 5-4: Based on the trained aircraft landing speed prediction model, the flight parameter data from (t-P+1) to t seconds are used as inputs of the prediction model to predict the landing speed from (t+1) to (t+P).
[0088] Step 5-5: Update time t=t+1, update and record the actual flight parameter data and landing speed of the aircraft, and return to step 5-2 to continue predicting the landing speed of the aircraft.
[0089] Embodiment 1:
[0090] The embodiments of the present invention will be described below in conjunction with the accompanying drawings.
[0091] The present invention discloses a method for predicting the landing speed of an aircraft based on flight parameter data. First, the historical flight parameter data in the aircraft quick access recorder is collected, the data samples are preprocessed, and the flight parameter data of the landing phase is extracted; then, a recursive feature elimination algorithm is used to calculate and filter the flight parameter data that has a greater impact on the aircraft landing speed, and a parameter set affecting the aircraft landing speed prediction is constructed; thus, a landing speed prediction model training data set is constructed based on the aircraft real-time flight parameter data, and a landing speed prediction model is obtained by training a gradient boosting regression model to dynamically predict the aircraft landing speed. The present invention uses a feature selection method to filter the flight parameter data that has a greater impact on the aircraft landing speed, and uses real-time flight parameter data to quickly and accurately predict the landing speed, which can effectively support the operation management of the aircraft landing phase in the terminal area.
[0092] The above technical solution specifically includes the following steps:
[0093] Step 1: Collect historical flight parameter data from the aircraft's quick access recorder, sort out and analyze the parameters that affect the aircraft's landing speed prediction to construct an initial flight parameter set, and obtain the corresponding historical original data;
[0094] The initial flight parameter set constructed by combing and analyzing the parameters that affect the aircraft landing speed prediction described in step 1 specifically includes: radio altitude, pressure altitude, indicated airspeed, pitch angle, roll angle, heading angle, vertical acceleration, lateral acceleration, longitudinal acceleration, aileron left, aileron right, flap left, flap right, rudder, and elevator.
[0095] Step 2: Preprocess the historical raw data to obtain initial data;
[0096] The raw data described in step 2 is preprocessed, specifically, checking whether there are missing values in the flight parameter data in the raw data. If there are missing values, the missing flight parameter data is supplemented using linear interpolation. Then, the three sigma criterion is used to check the outliers in the raw data, the detected outliers are discarded, and the discarded outlier parameter data is supplemented using linear interpolation.
[0097] Step 3: Extract the flight parameter data of the landing phase from the initial data;
[0098] In step 3, the flight parameter data of the landing phase is extracted from the initial data. Specifically, the radio altitude parameter data of the aircraft is used as a reference, and the time interval in which the radio altitude continuously decreases from 1500 feet to 0 feet is selected as the landing phase time interval. Then, various flight parameter data within the landing phase time interval are extracted from the initial data, so as to obtain the flight data of the aircraft during the landing phase.
[0099] Step 4: Based on the flight parameter data of the landing phase, a recursive feature elimination algorithm is used to screen and construct a parameter set affecting the aircraft landing speed prediction;
[0100] In step 4, a recursive feature elimination algorithm is used to screen and construct a set of parameters influencing the aircraft landing speed prediction. The specific calculation method is:
[0101] Step 4-1: Use the flight parameter data of the aircraft landing phase corresponding to the current flight parameter set to train a multiple linear regression model, and record the regression coefficients corresponding to each flight parameter in the current model and the prediction accuracy of the multiple linear regression model;
[0102] The multivariate linear regression model used in step 4-1 is expressed as:
[0103]
[0104] in, is the model's predicted landing speed, x 1 , x 2 , ..., x m Represents flight parameter data, β 1 , β2 , ..., β m is the regression coefficient corresponding to the flight parameters, β 0 is a constant compensation factor, and m is the number of flight parameters. 0 and β 1 , β 2 , ..., β m The value of can be obtained using the least squares estimation method based on the flight parameter data.
[0105] The prediction accuracy of the multivariate linear regression model described in step 4-1 uses the mean square error (MSE) as a measure of the model's prediction accuracy, and its calculation method is:
[0106]
[0107] Where y is the true value of landing speed, is the landing speed prediction value of the model, and n is the number of flight parameter data samples used. The smaller the MSE, the higher the prediction accuracy of the model.
[0108] Step 4-2: Sort the importance of the flight parameters according to the absolute value of the regression coefficient corresponding to each recorded flight parameter. The larger the absolute value of the regression coefficient, the greater the importance of the corresponding flight parameter;
[0109] Step 4-3: Delete the flight parameter with the lowest current importance, and use the remaining flight parameters to construct a new flight parameter set;
[0110] Step 4-4: Determine whether there is only one flight parameter left in the current flight parameter set. If yes, proceed to step 4-5; otherwise, return to step 4-1:
[0111] Step 4-5: According to the recorded prediction accuracy of the multivariate linear regression model, select the flight parameter set with the highest prediction accuracy and the least number of features to construct the aircraft landing speed prediction influencing parameter set;
[0112] Step 5: According to the aircraft landing speed prediction influencing parameter set, a training data set is constructed based on the aircraft real-time flight parameter data and a landing speed prediction model is trained to predict the aircraft landing speed.
[0113] In step 5, a training data set is constructed based on the real-time flight parameter data of the aircraft and a landing speed prediction model is trained to predict the landing speed of the aircraft. The specific prediction steps are as follows:
[0114] Step 5-1: Collect and record the actual flight parameter data and landing speed of the aircraft to be predicted every second, initialize the time window size W, and define the prediction time interval P;
[0115] The time window size W defined in step 5-1 represents the use of W sets of flight parameter data in the window for model training, and the prediction time interval P represents the use of the flight parameter data at the tth moment to predict the landing speed at the (t+P)th second in the future. Based on the experience of air traffic management operations and considering the efficiency of real-time calculation, it is recommended to set W=15 and P=5;
[0116] Step 5-2: For the current time point t, according to the flight parameters included in the aircraft landing speed prediction influencing parameter set, a total of W sets of flight parameter data corresponding to (tWP) seconds to (tP-1) seconds are extracted as the input of the prediction model, and a total of W sets of actual landing speeds from (tW) seconds to (t-1) seconds are extracted as the output of the prediction model, thereby constructing W sets of training data;
[0117] Step 5-3: Use the constructed W set of training data to train the gradient boosting regression model as the aircraft landing speed prediction model;
[0118] The gradient boosting regression model trained in step 5-3 specifically uses the standard gradient boosting regression model in the sklearn function library. The number of weak learners used in the model is n_estimators = 100, the contribution rate of each weak learner is learning_rate = 0.1, and the random number generator seed is random_state = 42.
[0119] Step 5-4: Based on the aircraft landing speed prediction model obtained through training, the flight parameter data from (t-P+1) to t seconds are used as inputs of the prediction model to predict the landing speed from (t+1) to (t+P);
[0120] Step 5-5: Update time t=t+1, update and record the actual flight parameter data and landing speed of the aircraft, and return to step 5-2 to continue predicting the landing speed of the aircraft;
[0121] Embodiment 2:
[0122] This embodiment provides a method for predicting aircraft landing speed based on flight parameter data.
[0123] The specific implementation process and steps are as follows: Figure 1 shown.
[0124] Step 1: Collect historical flight parameter data from the aircraft's quick access recorder, sort out and analyze the parameters that affect the aircraft's landing speed prediction to construct an initial flight parameter set, and obtain the corresponding historical original data;
[0125] The parameters that affect the aircraft landing speed prediction are sorted out and analyzed to construct the initial flight parameter set, including: radio altitude, pressure altitude, indicated airspeed, pitch angle, roll angle, heading angle, vertical acceleration, lateral acceleration, longitudinal acceleration, aileron left, aileron right, flap left, flap right, rudder, elevator. There are 15 flight parameters in total, that is, m=15;
[0126] Step 2: Preprocess the historical raw data to obtain initial data;
[0127] The raw data preprocessing specifically includes checking whether there are missing values in the flight parameter data in the raw data. If there are missing values, the missing flight parameter data is supplemented by linear interpolation. Then the three sigma criterion is used to check the outliers in the raw data, discard the detected outliers, and use linear interpolation to supplement the discarded outlier parameter data. After preprocessing the data recorded by the aircraft quick access recorder, 8050 raw flight parameter data are finally obtained.
[0128] Step 3: Extract the flight parameter data of the landing phase from the initial data;
[0129] Extract the flight parameter data of the landing phase. Specifically, take the radio altitude parameter data of the aircraft as a reference, select the time interval when the radio altitude continuously decreases from 1500 feet to 0 feet as the landing phase time interval, and then extract the flight parameter data within the landing phase time interval from the initial data to obtain the flight data of the aircraft landing phase. By processing the original flight parameter data, a total of 140 flight parameter data samples of the landing phase are extracted, that is, n = 140;
[0130] Examples of flight parameter data samples for the two landing phases are shown in Table 1:
[0131] Table 1 Examples of flight parameter data samples for two landing phases
[0132]
[0133] Step 4: Based on the flight parameter data of the landing phase, a recursive feature elimination algorithm is used to screen and construct a parameter set affecting the aircraft landing speed prediction;
[0134] The specific calculation method using the recursive feature elimination algorithm in step 4 is:
[0135] Step 4-1: Use the flight parameter data of the aircraft landing phase corresponding to the current flight parameter set to train a multiple linear regression model, and record the regression coefficients corresponding to each flight parameter in the current model and the prediction accuracy of the multiple linear regression model;
[0136] The multivariate linear regression model used in step 4-1 is expressed as:
[0137]
[0138] in, is the model's predicted landing speed, x 1 , x 2 , ..., x m Represents flight parameter data, β 1 , β 2 , ..., β m is the regression coefficient corresponding to the flight parameters, β 0 is a constant compensation factor, and m is the number of flight parameters. 0 and β 1 , β 2 , ..., β m The value of can be obtained using the least squares estimation method based on the flight parameter data.
[0139] The prediction accuracy of the multivariate linear regression model described in step 4-1 uses the mean square error (MSE) as a measure of the model's prediction accuracy, and its calculation method is:
[0140]
[0141] Where y is the true value of landing speed, is the model's predicted landing speed, and n is the number of flight parameters. The smaller the MSE, the higher the prediction accuracy of the model.
[0142] Step 4-2: Sort the importance of the flight parameters according to the absolute value of the regression coefficient corresponding to each recorded flight parameter. The larger the absolute value of the regression coefficient, the greater the importance of the corresponding flight parameter;
[0143] Step 4-3: Delete the flight parameter with the lowest current importance, and use the remaining flight parameters to construct a new flight parameter set;
[0144] Step 4-4: Determine whether there is only one flight parameter left in the current flight parameter set. If yes, proceed to step 4-5; otherwise, return to step 4-1:
[0145] Step 4-5: According to the recorded prediction accuracy of the multivariate linear regression model, select the flight parameter set with the highest prediction accuracy and the least number of features to construct the aircraft landing speed prediction influencing parameter set;
[0146] like Figure 2As shown in the figure, the prediction accuracy of the multivariate linear regression model trained when retaining different numbers of flight parameters in the process of screening flight parameters using the recursive feature elimination algorithm, the mean square error MSE is used as a measure of model accuracy, and the smaller the MSE, the higher the prediction accuracy of the model. As can be seen from the figure, in the process of flight parameter screening, as the number of retained flight parameters decreases, the MSE of the model generally shows an increasing trend. This shows that in the process of flight parameter screening, retaining too few flight parameters will lead to a significant decline in model performance, indicating that too few flight parameters cannot effectively predict the operating status of the aircraft. When the number of retained flight parameters is between 11 and 14, the MSE remains at the same level, indicating that the model can obtain comparable and better prediction capabilities. In order to maintain the high prediction ability of the model while minimizing the number of flight parameters used, 11 flight parameters are selected to construct the aircraft landing speed prediction influencing parameter set;
[0147] The selected flight parameters include: radio height, pitch angle, roll angle, heading angle, vertical acceleration, lateral acceleration, longitudinal acceleration, aileron left, aileron right, elevator, and rudder. The importance values of the selected flight parameters (i.e., the absolute value of the regression coefficient corresponding to each flight parameter in the multivariate linear regression model obtained by training) are: 0.4777, 0.2297, 0.1467, 0.0506, 0.0426, 0.0246, 0.0138, 0.0091, 0.0037, 0.0010, and 0.0003. It can be seen that flight parameters such as radio height, pitch angle, and roll angle have a greater influence on the prediction of landing speed.
[0148] Step 5: According to the aircraft landing speed prediction influencing parameter set, a training data set is constructed based on the aircraft real-time flight parameter data and a landing speed prediction model is trained to predict the aircraft landing speed.
[0149] In step 5, a training data set is constructed based on the real-time flight parameter data of the aircraft and a landing speed prediction model is trained to predict the landing speed of the aircraft. The specific prediction steps are as follows:
[0150] Step 5-1: Collect and record the actual flight parameter data and landing speed of the aircraft to be predicted every second, initialize the time window size W, and define the prediction time interval P;
[0151] The time window size W represents the use of W sets of flight parameter data in the window for model training, and the prediction time interval P represents the use of the flight parameter data at the tth moment to predict the landing speed at the (t+P)th second in the future. Based on the experience of air traffic management operations and considering the efficiency of real-time calculation, it is recommended to set W=15 and P=5;
[0152] Step 5-2: For the current time point t, according to the flight parameters included in the aircraft landing speed prediction influencing parameter set, a total of W sets of flight parameter data corresponding to (tWP) seconds to (tP-1) seconds are extracted as the input of the prediction model, and a total of W sets of real landing speeds from (tW) seconds to (t-1) seconds are extracted as the output of the prediction model, thereby constructing W sets of training data;
[0153] Step 5-3: Use the constructed W set of training data to train the gradient boosting regression model as the aircraft landing speed prediction model;
[0154] The gradient boosting regression model specifically uses the standard gradient boosting regression model in the sklearn function library (reference: Reference: Prettenhofer P, Louppe G. Gradient boosted regression trees in scikit-learn [C]. PyData, 2014.), the number of weak learners used in the model is n_estimators = 100, the contribution rate of each weak learner is learning_rate = 0.1, and the random number generator seed is random_state = 42.
[0155] Step 5-4: Based on the aircraft landing speed prediction model obtained through training, the flight parameter data from (t-P+1) to t seconds are used as inputs of the prediction model to predict the landing speed from (t+1) to (t+P);
[0156] Step 5-5: Update time t=t+1, update and record the actual flight parameter data and landing speed of the aircraft, and return to step 5-2 to continue predicting the landing speed of the aircraft;
[0157] like Figure 3 As shown, a schematic diagram comparing the actual landing speed using 10 flight parameter data samples and the predicted landing speed obtained by using the model prediction is shown. It can be seen from the figure that the landing speed of the aircraft decreases relatively smoothly during the landing process. The error between the predicted landing speed of the aircraft calculated by this method and the actual value is small, and basically maintains the same change trend. It has a good prediction accuracy and can effectively support auxiliary air traffic management personnel to manage aircraft in the terminal landing phase more reasonably and efficiently.
[0158] In a specific implementation, the present application provides a computer storage medium and a corresponding data processing unit, wherein the computer storage medium can store a computer program, and when the computer program is executed by the data processing unit, the invention content of the aircraft landing speed prediction method based on flight parameter data provided by the present invention and some or all steps in each embodiment can be executed. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.
[0159] Those skilled in the art can clearly understand that the technical solutions in the embodiments of the present invention can be implemented by means of computer programs and their corresponding general hardware platforms. Based on such an understanding, the technical solutions in the embodiments of the present invention can be essentially or partly contributed to the prior art in the form of computer programs, i.e., software products, which can be stored in a storage medium and include several instructions for enabling a device including a data processing unit (which can be a personal computer, a server, a single-chip microcomputer, an MCU or a network device, etc.) to execute the methods described in various embodiments of the present invention or certain parts of the embodiments.
[0160] The present invention provides a method and idea for predicting the landing speed of an aircraft based on flight parameter data. There are many methods and approaches to implement the technical solution. The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be considered as the protection scope of the present invention. All components not specified in this embodiment can be implemented by existing technologies.
Claims
1. A method for predicting aircraft landing speed based on flight parameter data, characterized in that: The following steps are involved: Step 1: collect historical flight parameter data of the aircraft, analyze and select parameters that affect the aircraft landing speed prediction, construct an initial flight parameter set, and obtain corresponding historical original data from the historical flight parameter data based on the initial flight parameter set; Step 2: preprocess the historical raw data to obtain initial data; Step 3, extracting flight parameter data of the landing phase from the initial data; Step 4: Based on the flight parameter data of the landing phase, a recursive feature elimination method is used to screen and construct a parameter set affecting the aircraft landing speed prediction; Step 5: According to the aircraft landing speed prediction influencing parameter set, a training data set is constructed based on the aircraft real-time flight parameter data, and the landing speed prediction model is trained and updated in real time to perform real-time prediction of the aircraft landing speed.
2. The method for predicting aircraft landing speed based on flight parameter data according to claim 1, characterized in that: The parameters that affect the aircraft landing speed prediction described in step 1 include: Radio altitude, pressure altitude, indicated airspeed, pitch angle, roll angle, heading angle, vertical acceleration, lateral acceleration, longitudinal acceleration, aileron left, aileron right, flaps left, flaps right, rudder and elevator.
3. The method for predicting aircraft landing speed based on flight parameter data according to claim 2, characterized in that: The preprocessing of historical raw data described in step 2 includes: Step 2-1, check whether there are missing values in the flight parameter data in the historical original data. If there are missing values, use linear interpolation to fill in the missing flight parameter data; Step 2-2, use the three sigma criterion to check the outliers in the historical raw data, discard the detected outliers, and use linear interpolation to fill in the discarded data.
4. The method for predicting aircraft landing speed based on flight parameter data according to claim 3, characterized in that: In step 3, the flight parameter data of the landing phase is extracted from the initial data, including: Step 3-1, selecting a time interval during which the aircraft's radio altitude parameter continuously decreases from a preset value to 0 as a landing phase time interval; Step 3-2, extracting the flight parameter data within the landing phase time interval from the initial data as the flight parameter data of the aircraft landing phase.
5. The method for predicting aircraft landing speed based on flight parameter data according to claim 4, characterized in that: In step 4, the recursive feature elimination method is used to screen and construct the aircraft landing speed prediction influencing parameter set, including: Step 4-1, construct a multiple linear regression model; Step 4-2, using the flight parameter data of the landing phase extracted in step 3 to form a current flight parameter set; Step 4-3, using the current flight parameter set to train the multiple linear regression model, and recording the regression coefficients corresponding to the flight parameters in the multiple linear regression model and the prediction accuracy of the multiple linear regression model during the training process; Step 4-4, sort the importance of the flight parameters according to the absolute value of the regression coefficient. The larger the absolute value of the regression coefficient, the greater the importance of the corresponding flight parameter. Step 4-5, delete the least important flight parameter in the current sorting, and use the remaining flight parameters to construct a new flight parameter set; Step 4-6, determine whether there is only one flight parameter left in the new flight parameter set, if so, execute step 4-7, otherwise take the new flight parameter set as the current flight parameter set and return to execute step 4-3: Step 4-7, according to the prediction accuracy of the multivariate linear regression model recorded in step 4-3, select the flight parameter set with the highest prediction accuracy and the least number of flight parameters to form the aircraft landing speed prediction influencing parameter set.
6. The method for predicting aircraft landing speed based on flight parameter data according to claim 5, characterized in that: The multivariate linear regression model in step 4-1 is expressed as: in, is the predicted landing speed, x1, x2, ..., x m Represents flight parameter data, β1, β2, ..., β m is the regression coefficient corresponding to the flight parameter, β0 is the constant compensation factor, and m is the number of flight parameters; β0 and β1, β2, ..., β m The values of are obtained using the least squares estimation method based on flight parameter data.
7. The method for predicting aircraft landing speed based on flight parameter data according to claim 6, characterized in that: The prediction accuracy of the multivariate linear regression model in step 4-1 is expressed by mean square error (MSE), and the calculation method is: Among them, y i is the true value of the landing speed of the i-th flight parameter data sample, represents the predicted landing speed value of the i-th flight parameter data sample, and n is the number of flight parameter data samples used; the smaller the MSE, the higher the prediction accuracy of the multivariate linear regression model.
8. The method for predicting aircraft landing speed based on flight parameter data according to claim 7, characterized in that: The real-time prediction of the aircraft landing speed described in step 5 includes: Step 5-1, collect and record the actual flight parameter data and landing speed of the aircraft to be predicted in real time, and initialize the time window size to be W and the prediction time interval to be P; Step 5-2, for the current time point t, according to the flight parameters included in the aircraft landing speed prediction influencing parameter set, extract a total of W groups of flight parameters corresponding to tWP seconds to tP-1 seconds as input data in the training data set, extract a total of W groups of landing speeds from tW seconds to t-1 seconds as output data in the training data set, and construct a training data set including W groups of training data; Step 5-3, selecting a gradient boosting regression model as the landing speed prediction model; Step 5-4, using the constructed W sets of training data to train the landing speed prediction model constructed in step 5-3; Step 5-5, based on the trained landing speed prediction model, the flight parameters from t-P+1 to t seconds are used as inputs to predict the landing speed from t+1 to t+P; Step 5-6, update time t=t+1, return and execute steps 5-2 to 5-6, and realize the real-time prediction of the landing speed of the aircraft.
9. The method for predicting aircraft landing speed based on flight parameter data according to claim 8, characterized in that: The time window size W defined in step 5-1 is 15, and the prediction time interval P is 5.
10. The method for predicting aircraft landing speed based on flight parameter data according to claim 8, characterized in that: The gradient boosting regression model described in step 5-3 is a standard gradient boosting regression model in the sklearn function library, and the number of weak learners n_estimators=100, the contribution rate of each weak learner learning_rate=0.1, and the random number generator seed random_state=42 are set.