Feature selection method for predicting intention of aircraft in climbing stage

Abstract

The invention provides a feature selection method for predicting the intention of an aircraft in a climbing stage, and the method comprises the steps: analyzing radar data, a plan message and an aircraft performance model, and obtaining a flight path, plan and aircraft performance model data set; performing multi-source data matching on the data set to obtain comprehensive track data and extract a feature sequence; setting a speed threshold value, and obtaining a constant-speed section set; setting a speed constraint and a height constraint, updating a constant-speed section set, calculating a constant-speed value, and obtaining a climbing section aircraft intention as a predicted target value; discretizing an indicated airspeed sequence and a vacuum speed sequence in the characteristic sequence; calculating correlation coefficients of various features and the intention of the aircraft, and making different feature combination strategies; and establishing a neural network prediction model, and calculating errors of prediction models corresponding to different feature combination strategies to obtain an optimal feature combination strategy. The method has a wide application prospect in the field of flight four-dimensional trajectory prediction.

Description

Technical Field

[0001] The present invention relates to a feature selection method, in particular to a feature selection method for predicting the intention of an aircraft in the climb phase. Background Art

[0002] The information provided in this section is only background information related to the present disclosure, and it does not necessarily represent prior art.

[0003] With the continuous growth of air traffic flow, problems such as flight delays, capacity imbalances, and airspace congestion occur frequently. To ensure the flight safety of aircraft and improve the operational efficiency of air traffic control, various air traffic control automation and intelligent decision-making support tools have emerged. Aircraft trajectory prediction is the basis and guarantee of such decision-making support tools. Accurate trajectory prediction can greatly reduce the uncertainty of future aircraft flights and improve the predictability of air traffic.

[0004] The trajectory prediction model based on dynamics is one of the most widely used trajectory prediction algorithms at present, mainly studying the relationship between the forces acting on the aircraft and the aircraft's motion. The aircraft performance used in the dynamics model is provided by the Base of Aircraft Data (BADA), including a set of differential equations and a parameter data set of a series of aircraft type coefficients. When determining the current state of the aircraft (such as mass, drag, thrust, true airspeed, etc.) and the aircraft intention (such as target speed), the aircraft trajectory is predicted by performing integral operations on time slices. The aircraft performance model parameters in different flight phases are also different. Therefore, when performing trajectory prediction based on the dynamics model, it is necessary to distinguish the aircraft intention in each flight phase. In the climb phase, when selecting speed as the constraint condition, the flight phase of the aircraft can be refined into the initial acceleration climb phase, initial constant-speed climb, acceleration climb phase, constant-speed climb phase, and constant-Mach climb phase. Correspondingly, the aircraft intention in the climb phase includes the initial constant-speed value, the second constant-speed value, and the constant-Mach value.

[0005] Since the dynamics model rarely considers the influence of actual control constraints and pilot behavior in order to simplify the model, it is impossible to guarantee the accuracy of trajectory prediction, which reduces the generality of the model. To enhance the robustness of the dynamics model, it is necessary to construct a speed constraint prediction model for aircraft. Machine learning methods can mine the speed constraints of the aircraft in the climb phase from historical trajectory data. However, due to the extremely complex trajectory data, it is necessary to compare and analyze the prediction accuracy of different feature combinations, eliminate the interference to the prediction results, and extract important features.

[0006] It should be noted that the information disclosed in the above background art section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute prior art known to those of ordinary skill in the art. Summary of the Invention

[0007] Purpose of the invention: The technical problem to be solved by the present invention is to provide a feature selection method for predicting the intention of an aircraft in the climbing phase in view of the deficiencies of the prior art.

[0008] To solve the above technical problem, the present invention discloses a feature selection method for predicting the intention of an aircraft in the climbing phase, including the following steps:

[0009] Step 1, parse the radar data to obtain a track data set; parse the planned message to obtain a planned data set; parse the aircraft performance model file to obtain an aircraft performance model data set;

[0010] Step 2, perform multi-source data matching on the track data set, the planned data set and the aircraft performance model data set, obtain comprehensive track data and extract a feature sequence;

[0011] Step 3, set a speed threshold to obtain an isovelocity segment set;

[0012] Step 4, set speed constraints and altitude constraints, update the isovelocity segment set according to the speed constraints, calculate the isovelocity value according to the altitude constraints, and obtain the intention of the aircraft in the climbing phase as the target value of the prediction;

[0013] Step 5, perform discretization processing on the indicated airspeed IAS sequence and the true airspeed TAS sequence in the feature sequence in Step 2, including numerical normalization, time discretization and numerical discretization;

[0014] Step 6, calculate the correlation coefficient between various features and the intention of the aircraft, and make different feature combination strategies;

[0015] Step 7, build a neural network prediction model, calculate the errors of the prediction models corresponding to different feature combination strategies, and obtain the optimal feature combination strategy.

[0016] Further, the track data set described in Step 1 is expressed as follows:

[0017] TP = {tp x , x = 1, 2, …, N TP}

[0018] where N TP is the total number of tracks, and the x-th track is tp x , which is expressed as follows:

[0019] tp x = {pt i , i = 1, 2,..., N PT}

[0020] where N PTis the total number of track points, and the i-th track point is pt i , and the vector form of the track point is expressed as follows:

[0021] pt i = [date i , acid i , t i , hp i , ias i , tas i , mach i , θ i , lon i , lat i

[0022] Among them, date i , acid i , t i , hp i , ias i , tas i , mach i , θ i , lon i and lat i respectively represent the date, flight number, timestamp, altitude, indicated airspeed, true airspeed, Mach number, heading angle, longitude and latitude of the i-th track point;

[0023] The planned data set mentioned above is expressed as follows:

[0024] PL = {pl y , y = 1, 2,..., N PL}

[0025] Among them, N PL is the total number of plans, and the y-th plan is pl y , and its vector form is expressed as follows:

[0026] pl y = [date y , acid y , plane y , speed y , fl y , sid y , air y , dep y , arr y

[0027] Among them, date y , acid y , plane y , speed y , fl​​y , sid y , air y , dep y and arr y respectively represent the date, flight number, aircraft type, cruising speed, cruising altitude, departure procedure, airline, departure airport and arrival airport of the plan pl y ;

[0028] The aircraft performance model dataset mentioned above is represented as follows:

[0029] AP = {ap z , z = 1, 2, …, N AP}

[0030] where N AP is the total number of aircraft types, and the performance model of the z-th aircraft is ap z , and its vector form is represented as follows:

[0031] ap z = [plane z , mass z

[0032] where plane z and mass z respectively represent the aircraft type and the aircraft mass.

[0033] Furthermore, the obtaining of the comprehensive track data and the extraction of the feature sequence in step 2 include:

[0034] Step 2-1, perform data matching between the track dataset and the plan dataset according to the date and flight number;

[0035] Step 2-2, perform data matching between the plan dataset and the aircraft performance model dataset according to the aircraft type;

[0036] Step 2-3, after cleaning the data with failed matching, obtain the comprehensive track dataset TP′, which is represented as follows:

[0037] TP′ = {tp x ′, x = 1, 2, …, M TP}

[0038] where M TP is the total number of comprehensive tracks, and the x-th comprehensive track is tp x ′, which is represented as follows:

[0039] tp x ’ = {pt j ’, j = 1, 2, …, M PT} ​

[0040] Among them, the j-th integrated track point is tp j ′, M PT is the total number of integrated track points. The total number of track points of the same flight is equal to the total number of integrated track points. The vector form of the integrated track point pt j is as follows:

[0041] pt j = [t j , hp j , ias j , tas j , mach j , θ j , lon j , lat j , plane j , sid j , air j , dep j , arr j , mass j , dis j

[0042] Among them, t j , hp j , ias j , tas j , mach j , θ j , lon j , lat j , plane j , sid j , air j , dep j , arr j , mass j and dis j respectively represent the timestamp, altitude, indicated airspeed, true airspeed, Mach number, course angle, longitude, latitude, aircraft type, departure procedure, airline, departure airport, arrival airport, aircraft reference mass and flight range of the j-th integrated track point. The flight range dis j is calculated through the longitude and latitude of the track point, that is, lon j and lat j ;

[0043] Step 2-4, extract the feature sequence from the integrated track, including:

[0044] Construct the indicated airspeed feature sequence IAS, which is expressed as follows:

[0045] IAS = {ias j , j = 1, 2, …, M PT ​}

[0046] Construct the true airspeed feature sequence TAS, which is expressed as follows:

[0047] TAS = {tas j , j = 1, 2, …, M PT}

[0048] Construct the Mach number feature sequence MACH, which is expressed as follows:

[0049] MACH = {ias j , j = 1, 2,..., M PT}

[0050] Construct the altitude feature sequence H p , which is expressed as follows:

[0051] H p = [h j , j = 1, 2,..., M PT}.

[0052] Furthermore, the obtaining of the constant-speed segment set described in step 3 includes:

[0053] Step 3-1, set the indicated airspeed threshold η v ;

[0054] Step 3-2, calculate the derivative set IAS' of all elements in the indicated airspeed feature sequence IAS;

[0055] Step 3-3, calculate the difference between the elements in the derivative set IAS' and the indicated airspeed threshold η v The elements with the absolute value of the difference less than the indicated airspeed threshold η v constitute the constant-speed segment set SEG V , which is expressed as follows:

[0056] SEG V = {seg k , k = 1, 2, …, K}

[0057] where the k-th constant-speed segment is seg k , and K is the number of constant-speed segments.

[0058] Furthermore, the k-th constant-speed segment seg k described in step 3-3 is expressed as follows:

[0059] seg k = [t beg , t end

[0060] where t beg and t end ​They are the start time and end time of the constant-speed segment seg k .

[0061] Furthermore, obtaining the intention of the aircraft in the climb segment described in step 4 includes:

[0062] Step 4-1, set speed constraints, specifically including:

[0063] Set the threshold η of the mean square error of the indicated airspeed mse , the time interval η t and the starting altitude threshold η h ;

[0064] Step 4-2, calculate the mean square error seg k of the constant-speed segment seg mse , the duration seg t and the starting altitude seg h ;

[0065] Step 4-3, calculate the differences between the mean square error seg k of the constant-speed segment seg mse , the duration seg t , the starting altitude seg h and the threshold η of the mean square error of the speed, the time interval η mse and the starting altitude threshold η t . Retain the constant-speed segments with the absolute value of the difference less than the corresponding threshold, and remove other constant-speed segments to obtain a new set of constant-speed segments SEG h ′, which is expressed as: V SEG

[0066] ′={seg V ′, k = 1, 2,..., K′} k

[0067] Step 4-4, set the altitude threshold η H , calculate the derivative of all elements in the altitude feature sequence H p to obtain a derivative set, compare the elements in the derivative set with the altitude threshold, and retain the elements with the comparison result less than the threshold to obtain the level flight segment set SEG H , which is expressed as follows:

[0068] SEG H ={seg q , q = 1, 2,…, Q}

[0069] where seg q represents the qth level flight segment, and Q represents the number of level flight segments;

[0070] Calculate the level flight segment seg q ​Average height The average height The longest level flight segment is denoted as seg fL , which is expressed as follows:

[0071]

[0072] Among them, represents the start time of the level flight segment, represents the end time of the level flight segment. Traverse the set of constant-speed segments, and use the weighted average of the indicated airspeed of the last constant-speed segment before the start time of the level flight segment as the aircraft's intended CAS2 value.

[0073] Furthermore, in step 5, the indicated airspeed IAS sequence and the true airspeed TAS sequence are discretized, specifically including:

[0074] Step 5-1, normalization processing, which is specifically as follows:

[0075]

[0076] Among them, x j , and σ are respectively the actual value, the feature average value, and the feature variance of the i-th feature in the indicated airspeed feature sequence IAS or the true airspeed feature sequence TAS, and x j ′ is the result obtained from the normalization processing;

[0077] Step 5-2, set the time compression ratio α, then the length of the new feature sequence is:

[0078]

[0079] Perform time dimension reduction on the new indicated airspeed feature sequence IAS or the true airspeed feature sequence TAS, specifically as follows:

[0080]

[0081] Among them, y iv represents the result after time dimension reduction;

[0082] Step 5-3, set the dimension M BP after numerical dimension reduction, and determine the breakpoint sequence BP according to the Gaussian distribution breakpoint list, which is expressed as follows:

[0083] BP = {bp ib , ib = 1, 2,..., M BP}

[0084] Among them, bp i represents the ib-th breakpoint;

[0085] Mapping is performed according to the word list, which is expressed as follows:

[0086] c ia = alpha(y iv ), bp ib-1 ≤ y iv < bp ib

[0087] Among them, c ia represents the ia-th word in the word list;

[0088] The feature sequence is mapped into an alphabet sequence. After double dimensionality reduction, the indicated airspeed discretization sequence IAS_SAX and the true airspeed discretization sequence TAS_SAX are obtained.

[0089] Furthermore, the specific methods for making different feature combination strategies described in step 6 include:

[0090] Step 6-1, according to the comprehensive track data set TP′, create a feature data set FE = {fe j , j = 1, 2,..., M TP}, construct the target vector sequence CAS2 = {cas2 j , j = 1, 2,..., M TP}, and the feature vector cas2 j is the one-dimensional vector of the aircraft intention CAS2; among them, the feature fe j in the feature data set FE includes: aircraft type, departure airport, arrival airport, IAS discretization sequence, TAS discretization sequence, distance, cruise speed, and cruise altitude;

[0091] Step 6-2, select each feature in the comprehensive track feature data set FE, and calculate its feature similarity R with each aircraft intention;

[0092] Step 6-3, sort according to the feature similarity R, and select the preset number of features ranked in the front for combination to obtain the feature combination strategy.

[0093] Furthermore, the feature similarity R described in step 6-2 is specifically as follows:

[0094]

[0095] Among them, fe j represents the feature value of the j-th track, represents the average value of all track features, cas2 j represents the aircraft intention value of the j-th track, represents the average value of all aircraft intentions.

[0096] Further, calculating the errors of different feature combination strategies corresponding to the prediction model in step 7 includes:

[0097] Step 7-1: Build a neural network prediction model for predicting the intention of the aircraft. Use the feature combination strategy obtained in step 6-3 as the input and the intention of the aircraft as the output for training.

[0098] Step 7-2: Calculate the mean square error mse of different feature combination strategies corresponding to the neural network prediction model x , and select the combination with the smallest error as the optimal feature combination strategy.

[0099] Beneficial effects:

[0100] The present invention fully considers the influence of the feature combination strategy on the intention of the aircraft. By performing a double dimensionality reduction operation on the speed feature, compressing the effective information of the data, and transforming the time series data into string information, it is thus incorporated into the aircraft intention prediction model. Through feature correlation analysis and calculation, a reliable feature combination strategy is made. By comparing the errors of the aircraft intention prediction model, a more perfect model feature combination is obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0101] The following further describes the present invention in detail with reference to the drawings and specific embodiments, and the above and / or other advantages of the present invention will become clearer.

[0102] Figure 1 is the overall process schematic diagram of the present invention.

[0103] Figure 2 is the scatter diagram of the correlation between aircraft features and aircraft intentions in the embodiment of the present invention.

[0104] Figure 3 is the curve diagram of the prediction error of aircraft intentions corresponding to different feature combination strategies in the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0105] The present invention proposes a feature selection method for aircraft intention prediction in the climb phase, which fully considers the influence of different feature combination strategies on the aircraft intention prediction result, so as to obtain a better-performing aircraft feature combination.

[0106] The specific technical solution is as Figure 1 shown, and includes the following steps:

[0107] Step 1: Parse the radar data to obtain the track data set TP = {tp x , x = 1, 2,..., N TP}, where N TP is the total number of tracks, and the x-th track is denoted as tpx , the flight track tp x = {pt i , i = 1, 2,..., N pT}, N pT is the total number of flight track points. Denote the i-th flight track point as pt i , the flight track point vector pt i = [date i , acid i , t i , hp i , ias i , tas i , mach i , θ i , lon i , lat i , where date i , acid i , t i , hp i , ias i , tas i , mach i , θ i , lon i and lat i respectively represent the date, flight number, timestamp, altitude, indicated airspeed, true airspeed, Mach number, course angle, longitude and latitude of the current flight track point. Analyze the planned message to obtain the planned data set PL = {pl y , y = 1, 2,..., N pL}, N pL is the total number of plans. Denote the y-th plan as pl y , the plan vector pl y = [date y , acid y , plane y , speed y , fl y , sid y , air y , dep y , arr y , where date y , acid y , plane y , speed y , fl y , sid y , air y , dep y , arr y respectively represent the plan pl yThe date, flight number, aircraft type, departure procedure, airline, departure airport and arrival airport are parsed from the aircraft performance model file to obtain the aircraft performance model dataset AP = {ap z , z = 1, 2,..., N AP}, where N AP is the total number of aircraft types. Denote the z tracks as ap z . The aircraft performance vector ap z = [plane z , mass z , where plane k and mass k represent the aircraft type and the aircraft mass respectively;

[0108] Step 2: Perform multi-source data matching to obtain comprehensive track data, and extract feature sequences from the comprehensive tracks;

[0109] Step 3: Set a speed threshold, compare the speed feature derivative with the speed threshold to obtain the constant-speed segment set;

[0110] Step 4: Set speed constraints, update the constant-speed segment set according to the speed constraints, calculate the constant-speed value according to the altitude constraints, and calculate the intention of the aircraft in the climb segment;

[0111] Step 5: Discretize the indicated airspeed IAS sequence and the true airspeed TAS sequence, including numerical normalization, time discretization, and numerical discretization;

[0112] Step 6: Calculate the correlation coefficients between various features and the constant-speed value, and formulate different feature combination strategies;

[0113] Step 7: Build a neural network prediction model, calculate the errors of the prediction models corresponding to different feature combination strategies, and obtain the optimal feature combination strategy;.

[0114] In one implementation, in Step 2, the specific methods for data pairing and feature extraction are as follows:

[0115] Step 2.1, complete the pairing of radar data and planned data according to the date and flight number, complete the pairing of planned data and aircraft performance files according to the aircraft type, clean the data with failed pairing, and obtain the comprehensive track dataset TP′ = {tp x ′, x = 1, 2,..., M TP}, where M TP is the total number of comprehensive tracks. Denote the x-th comprehensive track as tp x ′;

[0116] Step 2.2, the comprehensive track tp z ′ = {pt j ′, j = 1, 2,..., M PT}, denote the j-th integrated track point as pt j ′, M PT is the total number of integrated track points. The total number of track points of the same flight is equal to the total number of integrated track points. The integrated track point vector pt j = [t j , hp j , ias j , tas j , mach j , θ j , lon j , lat j , plane j , sid j , air j , dep j , arr j , mass j , dis j , where t j , hp j , ias j , tas j , mach j , θ j , lon j , lat j , plane j , sid j , air j , dep j , arr j , mass j and dis j respectively represent the timestamp, altitude, indicated airspeed, true airspeed, Mach number, heading angle, aircraft type, departure procedure, airline, departure airport and arrival airport, aircraft reference mass and flight range of the current track point. The flight range dis j is calculated from the longitude and latitude of the track point;

[0117] Step 2.2, extract the feature sequence from the integrated track, construct the IAS feature sequence IAS = {ias j , j = 1, 2,..., M PT}, construct the TAS feature sequence: TAS = {tas j , j = 1, 2,..., M PT}, construct the MACH feature sequence: MACH = {ias j , j = 1, 2,..., M PT}, construct the H p feature sequence: H p = {h j , j = 1, 2,..., MPT}.

[0118] In one implementation, in step 3, the method for detecting the constant-speed section is as follows:

[0119] Step 3.1, taking the indicated airspeed feature as an example, set the indicated airspeed threshold η v , and calculate the derivative IAS′ of the indicated airspeed IAS;

[0120] Step 3.2, compare the size of the indicated airspeed derivative IAS′ and the indicated airspeed threshold η v to obtain the constant-speed section set SEG V ={seg k , k = 1, 2,..., K}, and denote the k-th constant-speed section as seg k , and there is seg k =[t beg , t end , where t beg and t end are the start and end times respectively.

[0121] In one implementation, in step 4, the specific process of screening the constant-speed section set according to the speed constraint and height constraint and calculating the aircraft intention is as follows:

[0122] Step 4.1, taking the indicated airspeed feature as an example, set the speed constraint, including the mean square error threshold η mse of the indicated airspeed, the time interval η t and the starting height threshold η h , and calculate the mean square error seg k , the duration seg mse and the starting height seg t h h of the constant-speed section seg

[0123] Step 4.2, compare the mean square error seg k , the duration seg mse , the starting height seg t h h of the constant-speed section seg mse with the mean square error threshold η t of the speed, the time interval η h and the starting height threshold η V ′={seg k ′, k = 1, 2,..., K′};

[0124] Step 4.3, set the height threshold η H , and repeat the operations of steps 3.1 and 3.2 for the height feature Hp to obtain the level flight section set SEGH = {seg q , q = 1, 2, ..., Q}, calculate the average height of seg q Record the level flight segment to which the one with the largest value belongs as seg FL = [seg FL_beq , seg FL_end , traverse the constant speed segment set, and use the average value of the last constant speed segment before seg FL_beq as the predicted target CAS2 value;

[0125] Step 4.4, update the constraint conditions and thresholds according to the flight phase, repeat Steps 2 and 3, and calculate the intentions of all aircraft in the climb segment, including the CAS1 value and the MACH value.

[0126] In one implementation, in Step 5, the process of discretizing the speed characteristics is as follows:

[0127] Step 5.1, according to the formula i = 1, 2, ..., M PT perform numerical normalization on the indicated airspeed IAS sequence and the true airspeed TAS sequence, where x, σ are the actual value of the feature, the average value of the feature, and the variance of the feature, respectively;

[0128] Step 5.2, set the time compression ratio α, then the length of the new feature sequence is According to the formula i = 1, 2, ..., M pt ' perform time dimension reduction on the indicated airspeed IAS sequence and the true airspeed TAS sequence;

[0129] Step 5.3, set the dimension M after numerical dimension reduction BP , determine the breakpoint sequence BP = {bp i , i = 1, 2, ..., M BP} according to the Gaussian distribution breakpoint list, and perform mapping c i = alpha(y i ), bp j-1 ≤ z i < bp j , map the feature sequence to an alphabet sequence, and after double dimension reduction, obtain the indicated airspeed discretized sequence IAS_SAX and the true airspeed discretized sequence TAS_SAX.

[0130] In one implementation, in Step 6, make different feature combination strategies according to the correlation coefficient, and the specific method is as follows:

[0131] ​​Step 6.1, create a feature dataset FE = {fe i , i = 1, 2,..., M TP}, construct a target vector sequence CAS2 = {cas2 i , i = 1, 2,..., M TP}, and the feature vector cas2 i is a one-dimensional vector of the aircraft intention;

[0132] Step 6.2, select a feature from the comprehensive track feature dataset FE, such as aircraft type, departure airport, arrival airport, IAS discretization sequence, TAS discretization sequence, distance, cruise speed, and cruise altitude, etc., and calculate the feature similarity between each feature vector and the target aircraft intention according to the similarity calculation formula ;

[0133] Step 6.3, according to the feature similarity, select several features with high similarity for combination and matching to make a feature combination strategy.

[0134] In one implementation, in step 7, it is necessary to calculate the errors of the prediction models corresponding to different feature combination strategies to obtain the optimal feature combination strategy. The specific process is as follows:

[0135] Step 7.1, build an aircraft intention prediction model, use the feature combination strategy made in step 6.3 as the input and the aircraft intention as the output to train the model. The number of feature types included in the combination strategy should be selected appropriately to avoid overfitting of the prediction model or too large errors;

[0136] Step 7.2, calculate the mean square error mse x of the prediction models corresponding to different feature combination strategies, and select the combination with the smallest error as the optimal feature combination strategy.

[0137] According to the results of step 7 of the present invention, the accuracy of the prediction model under different feature combination strategies can be verified to obtain a more optimal feature combination strategy.

[0138] Example 1:

[0139] As Figure 1 shown, the present invention includes the following steps:

[0140] Step 1: Parse the radar data to obtain a track dataset TP = {tp x , x = 1, 2,..., N TP}, where N TP is the total number of tracks, and the x-th track is denoted as tp x , and the track tp x = {pt i , i = 1, 2,..., NPT}, N PT is the total number of waypoints. Denote the i-th waypoint as pt i , and the waypoint vector pt i = [date i , acid i , t i , hp i , ias i , tas i , mach i , θ i , lon i , lat i . Among them, date i , acid i , t i , hp i , ias i , tas i , mach i , θ i , lon i and lat i respectively represent the date, flight number, timestamp, altitude, indicated airspeed, true airspeed, Mach number, heading angle, longitude, and latitude of the current waypoint. Analyze the planned message to obtain the planned dataset PL = {pl y , y = 1, 2,..., N PL} where N PL is the total number of plans. Denote the y-th plan as pl y , and the plan vector pl y = [date y , acid y , plane y , speed y , fl y , sid y , air y , dep y , arr y . Among them, date y , acid y , plane y , speed y , fl y , sid y , air y , dep y , arr y respectively represent the date, flight number, aircraft type, departure procedure, airline, departure airport, and arrival airport of plan pl y . Analyze the aircraft performance model file to obtain the aircraft performance model dataset AP = {ap z, z = 1, 2, ..., N AP}, N AP is the total number of aircraft models. Denote the z flight tracks as ap z , and the aircraft performance vector ap z = [plane z , mass z , where plane k and mass k represent the aircraft model and the aircraft mass respectively;

[0141] Step 2: Perform multi-source data matching to obtain comprehensive flight track data. The specific methods for extracting feature sequence data pairing and feature extraction from the comprehensive flight track are as follows:

[0142] Step 2.1, Complete the pairing of radar data and planned data according to the date and flight number, complete the pairing of planned data and aircraft performance files according to the aircraft model, and clean the data with failed pairing to obtain the comprehensive flight track data set TP′ = {tp x ′, x = 1, 2, ..., M TP}, where M TP is the total number of comprehensive flight tracks. Denote the x-th comprehensive flight track as tp x ′;

[0143] Step 2.2, The comprehensive flight track tp x ′ = {pt j ′, j = 1, 2, ..., M PT}, Denote the j-th comprehensive flight track point as pt j ′, where M PT is the total number of comprehensive flight track points. The total number of flight track points of the same flight is equal to the total number of comprehensive flight track points. The comprehensive flight track point vector pt j = [t j , hp j , ias j , tas j , mach j , θ j , lon j , lat j , plane j , sid j , air j , dep j , arr j , mass j , dis j , where t j , hp j , ias j , tas j , mach j , θ j , lonj , lat j , plane j , sid j , air j , dep j , arr j , mass j and dis j respectively represent the timestamp, altitude, indicated airspeed, true airspeed, Mach number, course angle, aircraft type, departure procedure, airline, departure airport and arrival airport, aircraft reference mass and flight range of the current track point. The flight range dis j is calculated through the longitude and latitude of the track point;

[0144] Step 2.2, extract the feature sequence from the comprehensive track. Taking the indicated airspeed feature as an example, construct the IAS feature sequence IAS = {ias j , j = 1, 2,..., M PT}, and so on, create feature sequences such as TAS, MACH, H p .

[0145] Step 3: Set the speed threshold, compare the derivative of the speed feature with the speed threshold, and obtain the constant speed segment set. The method for constant speed segment detection is:

[0146] Step 3.1, taking the indicated airspeed feature as an example, set the indicated airspeed threshold η v , and calculate the derivative IAS' of the indicated airspeed IAS;

[0147] Step 3.2, compare the derivative IAS' of the indicated airspeed and the indicated airspeed threshold η v to obtain the constant speed segment set SEG V = {seg k , k = 1, 2,..., K}, and denote the kth constant speed segment as seg k , and there is seg k = [t beg , t end , where t beg and t end are the start and end times respectively.

[0148] Step 4: Set the speed constraint, update the constant speed segment set according to the speed constraint, calculate the constant speed value according to the altitude constraint, and calculate the intention of the aircraft in the climb segment. The specific process is as follows:

[0149] Step 4.1, taking the indicated airspeed feature as an example, set the speed constraint, including the mean square error threshold η mse of the indicated airspeed, the time interval η t and the starting altitude threshold η h, calculate the mean square error of the constant speed segment seg k , the duration of seg mse , and the starting height of seg t respectively; h ;

[0150] Step 4.2, compare the mean square error of the constant speed segment seg k , the duration of seg mse , the starting height of seg t with the mean square error threshold η of speed h , the time interval η mse , and the starting height threshold η t , and update the set SEG of constant speed segments that meet the conditions h ′ = {seg V ′, k = 1, 2,..., K′}; k ;

[0151] Step 4.3, set the height threshold η H , and repeat the operations of Step 3.1 and Step 3.2 for the height feature Hp to obtain the set SEG of level flight segments H = {seg q , q = 1, 2,..., Q}, calculate the average height of seg q Record the level flight segment to which the one with the largest value belongs as seg FL = [seg FL_beg , seg FL_end , traverse the set of constant speed segments, and use the average value of the last constant speed segment before seg FL_beg as the predicted target CAS2 value;

[0152] Step 4.4, update the constraint conditions and thresholds according to the flight phase, and repeat Step 2 and Step 3 to calculate the intentions of all aircraft in the climb segment, including the CAS1 value and the MACH value.

[0153] Step 5: Discretize the indicated airspeed IAS sequence and the true airspeed TAS sequence, including numerical normalization, time discretization, and numerical discretization. The process of discretization is as follows:

[0154] Step 5.1, according to the formula i = 1, 2,..., M PT perform numerical normalization on the indicated airspeed IAS sequence and the true airspeed TAS sequence, where x, σ are the actual value of the feature, the average value of the feature, and the variance of the feature respectively;

[0155] Step 5.2, set the time compression ratio α, then the length of the new feature sequence is​​ According to the formula i = 1, 2, ..., M pt 'Perform temporal dimensionality reduction on the indicated airspeed IAS sequence and the true airspeed TAS sequence;

[0156] Step 5.3, set the dimension M after numerical dimensionality reduction BP , determine the breakpoint sequence BP = {bp i , i = 1, 2, ..., M BP} according to the Gaussian distribution breakpoint list, and perform mapping c i = aplha(y i ), bp j-1 ≤ z i < bp j , map the feature sequence to an alphabet sequence. After double dimensionality reduction, obtain the discretized indicated airspeed sequence IAS_SAX and the discretized true airspeed sequence TAS_SAX.

[0157] Step 6: Calculate the correlation coefficients between various features and the constant airspeed value, and formulate different feature combination strategies. The specific method is as follows:

[0158] Step 6.1, create a feature dataset FE = {fe i , i = 1, 2, ..., M TP} according to the comprehensive track set, and construct the target vector sequence CAS2 = {cas2 i , i = 1, 2, ..., M TP}, where the feature vector cas2 i is a one-dimensional vector of the constant airspeed value;

[0159] Step 6.2, select a feature from the comprehensive track feature dataset FE, such as aircraft type, departure airport, arrival airport, IAS discretized sequence, TAS discretized sequence, distance, cruise speed, and cruise altitude, etc., and calculate the feature similarity between each feature vector and the target constant airspeed value vector according to the similarity calculation formula ;

[0160] Step 6.3, according to the feature similarity, select several features with high similarity for combination and formulate a feature combination strategy.

[0161] Step 7: Build a neural network prediction model, calculate the errors of the prediction models corresponding to different feature combination strategies, and obtain the optimal feature combination strategy. The specific process is as follows:

[0162] Step 7.1: Build an aircraft intention prediction model. Use the feature combination strategy created in Step 6.3 as the input and the aircraft intention as the output to train the model. The number of feature types included in the combination strategy should be appropriately selected to avoid overfitting or excessive error in the prediction model.

[0163] Step 7.2: Calculate the mean squared error (mse) of the prediction model corresponding to different feature combination strategies. x Select the combination with the minimum error as the optimal feature combination strategy.

[0164] Example 2:

[0165] The present invention will be further illustrated by the examples of simulation experiments and their effect evaluations.

[0166] In this example, as Figure 2 and Figure 3 The curves in show the scatter plot of the correlation between aircraft features and aircraft intentions and the aircraft intention prediction error curves corresponding to different feature combination strategies respectively. The experimental objective is to create the optimal feature combination strategy through the feature selection method for aircraft intention prediction during the climb phase. Figure 2 In the scatter plot of the corresponding mass, the horizontal axis represents the cruise speed feature of the aircraft, and the vertical axis represents the Mach speed intention of the aircraft. It can be observed from the calculation of the correlation coefficient and the scatter plot that the cruise speed feature and the aircraft intention show a positive correlation. Figure 3 Describes the aircraft intention prediction error curves corresponding to different feature combination strategies. The vertical axis represents different feature combination strategies, including single feature, double feature combination, triple feature combination, quadruple feature combination, and quintuple feature combination. The horizontal axis represents the absolute mean error of the aircraft intention prediction model corresponding to the feature combination. Observing Figure 3 It can be seen that the contribution degrees of different features to the aircraft intention prediction model are different, and compared with single features, the multi-feature combination strategy has a positive gain for the aircraft intention prediction model. This result shows that the method of the present invention can select beneficial features to participate in aircraft intention prediction through the calculation results of correlation, and obtain the optimal feature combination strategy by making different feature combination strategies and analyzing and comparing the errors of the aircraft intention prediction model.

[0167] In specific implementation, the present application provides a computer storage medium and a corresponding data processing unit. Among them, the computer storage medium can store a computer program, and when the computer program is executed by the data processing unit, it can run the inventive content of a feature selection method for predicting the intention of an aircraft in the climb phase and some or all of the steps in each embodiment. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), or the like.

[0168] 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 a computer program and its corresponding general hardware platform. Based on such an understanding, the essence of the technical solutions in the embodiments of the present invention, or the part that contributes to the prior art, can be embodied in the form of a computer program, that is, a software product. The computer program software product can be stored in a storage medium and includes several instructions for causing 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 each embodiment or some parts of the embodiments of the present invention.

[0169] The present invention provides an idea and method for a feature selection method for predicting the intention of an aircraft in the climb phase. There are many methods and ways to specifically implement this technical solution. The above is only the preferred implementation manner of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention. Each component not clearly defined in this embodiment can be implemented by the prior art.

Claims

1. A feature selection method for predicting aircraft intention during climb phase, characterized in that: The following steps are involved: Step 1: parse radar data to obtain a track data set; parse the plan message to obtain a plan data set; parse the aircraft performance model file to obtain an aircraft performance model data set; Step 2, perform multi-source data matching on the track data set, the plan data set and the aircraft performance model data set to obtain the comprehensive track data and extract the feature sequence; Step 3, set the speed threshold and obtain a set of constant speed segments; Step 4, setting speed constraints and altitude constraints, updating the constant speed segment set according to the speed constraints, calculating the constant speed value according to the altitude constraints, and obtaining the aircraft intention in the climb phase as the predicted target value; Step 5, discretizing the indicated airspeed IAS sequence and the true airspeed TAS sequence in the characteristic sequence in step 2, including numerical normalization, time discretization and numerical discretization; Step 6: Calculate the correlation coefficients between various features and aircraft intentions, and make different feature combination strategies; Step 7: Build a neural network prediction model, calculate the error of the prediction model corresponding to different feature combination strategies, and obtain the optimal feature combination strategy.

2. A feature selection method for predicting aircraft intention in the climb phase according to claim 1, characterized in that: The track dataset described in step 1 is represented as follows: TP={tp x ,x=1,2,...,N TP } Among them, N TP is the total number of tracks, the xth track is tp x , which is expressed as follows: type x ={pt i ,i=1,2,...,N PT } Among them, N PT is the total number of track points, the i-th track point is pt i , the vector form of the track point is as follows: pt i =[date i ,acid i ,t i ,hp i ,clean i ,the i ,mach i ,θ i ,lon i ,lat i ] Among them, date i 、acid i ,t i , hp i ,ias i ,tas i 、mach i ,θ i ,lon i and lat i Represent the date, flight number, timestamp, altitude, indicated airspeed, true airspeed, Mach number, heading angle, longitude and latitude of the ith track point respectively; The planned data set is represented as follows: PL={pl y ,y=1,2,...,N PL } Among them, N PL is the total number of plans, the yth plan is pl y , its vector form is as follows: pl y =[date y ,acid y ,plane y ,speed y ,fl y ,sid y ,air y ,dep y ,arr y ] Among them, date y 、acid y 、plane y 、speed y 、fl y ,sid y 、air y 、dep y and arr y Respectively represent the plan pl y Date, flight number, aircraft type, cruising speed, cruising altitude, departure procedure, airline, departure airport and landing airport; The aircraft performance model data set is represented as follows: AP={ap z ,z=1,2,...,N AP , Among them, N AP is the total number of aircraft models, and the performance model of the zth aircraft is ap z , its vector form is as follows: ap z =[plane z ,mass z ] Among them, plane z and mass z Represent aircraft model and aircraft mass respectively.

3. A feature selection method for predicting aircraft intention in the climb phase according to claim 2, characterized in that: The step 2 of obtaining comprehensive track data and extracting feature sequences includes: Step 2-1, matching the data in the track data set with the data in the plan data set according to the date and flight number; Step 2-2, matching the data in the planned data set and the aircraft performance model data set according to the aircraft type; Step 2-3, after cleaning the data that failed to match, the comprehensive track data set TP′ is obtained, which is expressed as follows: TP′={tp x ′,x=1,2,...,M TP } Among them, M TP is the total number of integrated tracks, and the xth integrated track is tp x ′, which is expressed as follows: tp x ’={pt j ’,j=1,2,...,M PT } Among them, the jth integrated track point is pt j ′,M PT is the total number of integrated track points. The total number of track points of the same flight is equal to the total number of integrated track points. The vector form of the integrated track points is pt j as follows: pt j =[t j ,hp j ,ias j ,tas j ,mach j ,θ j ,lon j ,lat j ,plane j ,sid j ,air j ,dep j ,arr j ,mass j ,dis j ] Among them, t j , hp j ,ias j ,tas j 、mach j ,θ j ,lon j ,lat j 、plane j ,sid j 、air j 、dep j 、arr j 、mass j and dis j Respectively represent the timestamp, altitude, indicated airspeed, true airspeed, Mach number, heading angle, longitude, latitude, aircraft type, departure procedure, airline, departure airport, landing airport, aircraft reference mass and flight range of the jth integrated track point. The flight range dis j The longitude and latitude of the track point, i.e. lon j and lat j Calculated; Step 2-4, extracting feature sequences from the integrated track, includes: Construct the indicated airspeed characteristic sequence IAS, which is expressed as follows: IAS={ias j ,j=1,2,...,M PT } Construct the true airspeed characteristic sequence TAS, which is expressed as follows: TAS={tas j ,j=1,2,...,M PT } Construct the Mach number characteristic sequence MACH, which is expressed as follows: MACH={ias j ,j=1,2,...,M PT } Construct a highly characteristic sequence H p , which is expressed as follows: H p ={h j ,j=1,2,...,M PT }。 4. A feature selection method for predicting aircraft intention in the climb phase according to claim 3, characterized in that: The step 3 of obtaining a set of constant velocity segments includes: Step 3-1, setting the indicated airspeed threshold ηv; Step 3-2, calculating the derivative set IAS′ of all elements in the indicated airspeed characteristic sequence IAS; Step 3-3, calculate the elements in the derivative set IAS′ and the indicated airspeed threshold η v The absolute value of the difference is less than the indicated airspeed threshold η v The element composition isospeed segment set SEG V , which is expressed as follows: SEG V ={self k ,k=1,2,...,K} Among them, the kth constant speed segment is seg k , K is the number of constant velocity segments.

5. A feature selection method for predicting aircraft intention in the climb phase according to claim 4, characterized in that: The kth constant velocity segment seg described in step 3-3 k , which is expressed as follows: hello k =[t beg ,t end ] Among them, t beg and t end The constant speed segments are seg k The start time and end time of the 6. A feature selection method for predicting aircraft intention in the climb phase according to claim 5, characterized in that: Obtaining aircraft intent for the climb phase as described in Step 4 includes: Step 4-1, set speed constraints, including: Set the indicated airspeed mean square error threshold η mse , time interval η t and the starting height threshold η h ; Step 4-2, calculate the constant speed segment seg respectively k The speed mean square error seg mse , duration seg t and starting height segh; Step 4-3, calculate the constant speed segment seg k The speed mean square error seg mse , duration seg t , starting height seg h and the speed mean square error threshold η mse , time interval η t and the starting height threshold η h The constant speed segments whose absolute value of the difference is less than the corresponding threshold are retained, and the other constant speed segments are removed to obtain a new constant speed segment set SEG V ′, expressed as: SEG V ′={self k ′,k=1,2,...,K′} Step 4-4, set the height threshold η H , for the highly characteristic sequence H p Calculate the derivatives of all elements to obtain a derivative set, compare the elements in the derivative set with the height threshold, retain the elements whose comparison results are less than the threshold, and obtain the level flight segment set SEG H , which is expressed as follows: SEG H ={self q ,q=1,2,...,Q} Among them, seg q represents the qth level flight segment, and Q represents the number of level flight segments; Calculate level flight segment seg q Average height The average height The largest level flight segment is denoted as seg FL , which is expressed as follows: in, Indicates the start time of the level flight segment. Indicates the end time of the level flight segment, traverses the set of constant speed segments, and sets the start time of the level flight segment The weighted average of the indicated airspeed during the last constant speed segment before the aircraft is taken as the aircraft's intended CAS2 value.

7. A feature selection method for predicting aircraft intention in the climb phase according to claim 6, characterized in that: In step 5, the indicated airspeed IAS sequence and the true airspeed TAS sequence are discretized, specifically including: Step 5-1, normalization processing, is as follows: Among them, x j , and σ are the actual value, feature mean value and feature variance of the i-th feature in the indicated airspeed feature sequence IAS or true airspeed feature sequence TAS, respectively. j ' is the result obtained by the normalization process; Step 5-2, set the time compression ratio α, then the new feature sequence length is: The time dimension reduction is performed on the new indicated airspeed characteristic sequence IAS or true airspeed characteristic sequence TAS as follows: Among them, y iv Represents the result after time dimension reduction; Step 5-3, set the dimension M after numerical dimension reduction BP , the breakpoint sequence BP is determined according to the Gaussian distribution breakpoint list, which is expressed as follows: BP={bp ib ,ib=1,2,...,M BP } Among them, bp i Indicates the breakpoint; Mapping is performed according to the word list, as follows: c ia =alpha(y iv ),bp ib-1 ≤y iv <bp ib Among them, c ia Represents the iath word in the vocabulary; The characteristic sequence is mapped into a letter sequence, and after double dimensionality reduction, the indicated airspeed discretization sequence IAS_SAX and the true airspeed discretization sequence TAS_SAX are obtained.

8. A feature selection method for predicting aircraft intention in the climb phase according to claim 7, characterized in that: The specific methods for making different feature combination strategies described in step 6 include: Step 6-1: Create a feature dataset FE = {fe j , j = 1, 2, ..., M TP }, construct the target vector sequence CAS2 = {cas2 j , j = 1, 2, ..., M TP }, feature vector cas2 j is a one-dimensional vector of aircraft intention CAS2; wherein the feature fe in the feature data set FE j , including: aircraft type, take-off airport, landing airport, IAS discretization sequence, TAS discretization sequence, distance, cruising speed and cruising altitude; Step 6-2, select each feature in the comprehensive track feature dataset FE and calculate its feature similarity R with each aircraft intention; Step 6-3, sorting according to feature similarity R, selecting a preset number of features that are ranked first for combination, and obtaining a feature combination strategy.

9. A feature selection method for predicting aircraft intention in the climb phase according to claim 8, characterized in that: The feature similarity R described in step 6-2 is as follows: Among them, fe j represents the characteristic value of the j-th track, Represents the average value of all track characteristics, cas2 j represents the aircraft intention value of the jth track, Represents the average of all aircraft intentions.

10. A feature selection method for predicting aircraft intention in the climb phase according to claim 9, characterized in that: The calculation of the prediction model errors corresponding to different feature combination strategies described in step 7 includes: Step 7-1, build a neural network prediction model to predict aircraft intentions, using the feature combination strategy obtained in step 6-3 as input and aircraft intentions as output for training; Step 7-2: Calculate the mean square error (mse) of the neural network prediction model corresponding to different feature combination strategies x , select the combination with the smallest error as the optimal feature combination strategy.

Images