KR102079040B1 - Device and method for aircraft landing sequence determination - Google Patents

Abstract

The aircraft landing order determination apparatus according to an embodiment of the present application, the reference arrival order based on a pair-based landing preference probability calculated in pairs with other aircraft for each of a plurality of aircraft having a different route approaching the target airport A reference arrival sequence generation unit to generate and a landing order generation unit determining a landing order of aircraft by applying a Constrained Position Shifting (CPS) technique to the reference arrival sequence, wherein the pair-based landing preference probability includes training including past track data Based on data, the model for each pair of aircraft to be compared is compared to a model trained by pairwise preference learning, which is performed by pairing two random aircrafts having different routes. It can be calculated by inputting the state information at the time point.

Description

DEVICE AND METHOD FOR AIRCRAFT LANDING SEQUENCE DETERMINATION

The present application relates to an aircraft landing order determination device and method.

The demand for air traffic has increased until recently. Air traffic flows are dependent on the capabilities of air traffic controllers. In particular, air traffic flows at major airports with frequent takeoffs and landings are saturated by many aircraft. This causes aircraft delays at airports and overworked tasks for aircraft pilots and air traffic controllers, resulting in a variety of problems.

One of the essential parts of air traffic management is the rapid handling of aircraft landing at the airport, minimizing the delay of other aircraft. Landing sequence management, the most important aspect of air traffic management, means maximizing the airport runway capacity by determining the landing sequence of two or more aircraft landing on the same runway, or the landing sequence at the confluence of different flight paths.

In this regard, conventionally, an arrival management system has been developed and used in the field to calculate the landing order based on the First Come First Served (FCFS) or optimization method and provide it to the air traffic controller. However, there is a problem in that the landing order calculated by the computer is different from the landing order that the air traffic controller thinks, and thus the utilization is inferior.

Background art of the present application is disclosed in Japanese Patent No. 3728263.

The present application is to solve the above-described problems of the prior art, by providing a landing sequence similar to the landing sequence of the aircraft predicted by the air traffic controller by learning the past track data to provide advice information for determining the final landing order of the air traffic controller An object of the present invention is to provide an aircraft landing order determination device that can be used.

An object of the present invention is to provide an aircraft landing order determination device capable of calculating a landing order having a high acceptance rate of an air traffic controller.

However, the technical problem to be achieved by the embodiments of the present application is not limited to the technical problems as described above, and other technical problems may exist.

As a technical means for achieving the above technical problem, the aircraft landing order determination apparatus according to an embodiment of the present application, is calculated in pairs with other aircraft for each of a plurality of aircraft approaching the target airport but having different routes And a landing order generation unit for generating a reference arrival order based on a pair-based landing preference probability and a landing order generation unit for determining an aircraft landing order by applying a Constrained Position Shifting (CPS) technique to the reference arrival order. Landing preference probability is a model trained by Pairwise Preference Learning that proceeds by pairing any two aircraft with different routes based on training data including historical track data. Enter the status information at each time point for each pair of aircraft to be compared. And it can be calculated.

According to one embodiment of the present application, the pair-based landing preference probability may be calculated by Equation 1.

According to an embodiment of the present application, Equation 1 may be calculated using Equation 2.

According to the exemplary embodiment of the present application, the reference arrival order generation unit may generate the reference arrival order based on a score obtained by accumulating the pair-based landing preference probability calculated in pairs with each other for each of the plurality of aircraft. Can be.

According to an embodiment of the present disclosure, a score obtained by accumulating the pair-based landing preference probability calculated in pairs with each of the other aircrafts except the one for one of the plurality of aircrafts may be calculated by Equation 3. have.

According to the exemplary embodiment of the present application, the reference arrival order generation unit may generate the reference arrival order through Equation 4.

According to the exemplary embodiment of the present application, the landing order generation unit changes the landing order in consideration of the maximum number of allowed position shifts (MPS) assigned to each of the plurality of aircraft based on the generated reference arrival order. Or by maintaining the aircraft landing order.

Aircraft landing order determination method according to an embodiment of the present application, (a) based on a pair-based landing preference probability calculated in pairs with other aircraft for each of a plurality of aircraft approaching the target airport but having different routes Generating an arrival sequence and (b) applying a Constrained Position Shifting (CPS) technique to the reference arrival sequence to determine the landing sequence of the aircraft, wherein the pair based landing preference probability includes historical track data. For a model trained by Pairwise Preference Learning, which is performed by pairing any two aircraft with different routes based on the training data, for each pair of aircraft to be compared It may be calculated by inputting state information at the reference time point.

According to an embodiment of the present disclosure, the pair-based landing preference probability may be calculated by Equation 1.

According to one embodiment of the present application, Equation 1 may be calculated using Equation 2.

According to an embodiment of the present disclosure, the step (a) may generate the reference arrival order based on a score obtained by accumulating a pair-based landing preference probability calculated in pairs with each other aircraft for each of the plurality of aircraft. Can be.

According to an embodiment of the present disclosure, a score obtained by accumulating the pair-based landing preference probability calculated in pairs with each of the other aircrafts except the one for one of the plurality of aircrafts may be calculated by Equation 3. have.

According to an embodiment of the present application, step (a) may generate the reference arrival sequence through Equation 4.

According to an embodiment of the present application, the step (b) is based on the generated reference arrival order, the landing order in consideration of the maximum number of allowed position shift (MPS) assigned for each of the plurality of aircraft By changing or maintaining, the aircraft landing order can be finally determined.

The above-mentioned means for solving the problems are merely exemplary, and should not be construed as limiting the present application. In addition to the above-described exemplary embodiments, additional embodiments may exist in the drawings and detailed description of the invention.

According to the above-described problem solving means of the present application, the final landing order of the air traffic controller is determined by providing a landing order similar to the landing order of the aircraft predicted by the air traffic controller through a model trained on pair-based preference learning of past track data. It is possible to provide an aircraft landing order determination device capable of providing advice information.

According to the above-described problem solving means of the present invention, by calculating the landing order of the air traffic controller high acceptance rate, the aircraft landing can maximize the aircraft capacity of the runway in the airport and reduce the delay of the aircraft An order determining device can be provided.

1 is a view showing the configuration of the aircraft landing order determination apparatus according to an embodiment of the present application.
2 is a diagram illustrating an example of an entry way point of an airport of the aircraft landing order determining apparatus according to an embodiment of the present application.
3 is a conceptual diagram schematically showing the flow of aircraft landing order determination of the aircraft landing order determining apparatus according to an embodiment of the present application.
4 is a diagram illustrating an example of an inter-aircraft score for generating a reference arrival order of the aircraft landing order determination device according to an embodiment of the present application.
5 is a diagram illustrating an example of a reference arrival order calculated according to the inter-aircraft score of the aircraft landing order determination device according to an embodiment of the present application.
6 is a diagram illustrating an example for explaining a method of extracting a pair of aircraft for learning the aircraft landing order determining apparatus according to an embodiment of the present application.
FIG. 7 is a diagram illustrating a coefficient for each variable according to a route in an experiment for evaluating the effectiveness of the aircraft landing order determination device according to an embodiment of the present application.
8 is a view comparing the landing order by the conventional method (conventional technique) and the landing order by the air traffic controller.
9 is a diagram illustrating the minimum separation criteria for each aircraft weight.
FIG. 10 is a diagram comparing a rank correlation coefficient by a hybrid sequencing model of an aircraft landing order colon apparatus and a pair-based landing preference model not applied to the CPS method, according to an exemplary embodiment.
11 is a view showing the flow of the aircraft landing order determination method according to an embodiment of the present application.

DETAILED DESCRIPTION Hereinafter, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings so that those skilled in the art may easily implement the present disclosure. As those skilled in the art would realize, the described embodiments may be modified in various different ways, all without departing from the spirit or scope of the present invention. In the drawings, parts irrelevant to the description are omitted for simplicity of explanation, and like reference numerals designate like parts throughout the specification.

Throughout this specification, when a part is "connected" to another part, this includes not only "directly connected" but also "electrically connected" with another element in between. do.

Throughout this specification, when a member is said to be located on another member "on", "upper", "top", "bottom", "bottom", "bottom", this means that any member This includes not only the contact but also the case where another member exists between the two members.

Throughout this specification, when a part is said to "include" a certain component, it means that it can further include other components, without excluding the other components unless otherwise stated.

1 is a view showing the configuration of the aircraft landing order determination apparatus according to an embodiment of the present application.

Arrival manager (AMAN), one of the tools to assist air traffic controllers in determining the landing sequence of aircraft arriving at the airport, predicts the expected arrival time of the aircraft to be landed based on the current status and flight plan of the aircraft first arriving. Based on the estimated arrival time, we calculate the efficient landing sequence to maximize the airport's aircraft throughput or minimize flight delays. In the past, optimization-based sequencing models have been proposed for the development of AMAN that maximize the objective function (eg, flight time or runway throughput) while maintaining minimum separation requirements between consecutive aircraft. In addition, mixed integer linear programming (MILP) has been proposed to determine the optimal arrival flight sequence and landing time for single and multiple runways. MILP has successfully considered various operational constraints, but the high computational cost has made it impossible to apply in real time despite reasonable air traffic.

In addition, as a method for optimizing the arrival sequence of the aircraft, a method for determining an optimal sequence through a heuristic approach has also been proposed. In the related art, a method of determining an efficient arrival sequence using a genetic algorithm and a method of determining an arrival sequence using an arrival scheduling algorithm based on dynamic programming have been proposed. However, despite this approach, there is still a problem that conventional algorithms are computationally efficient but cannot incorporate practical constraints such as time of arrival windows. In other words, even if a conventional algorithm provides a mathematically optimal or near optimal solution, it is not possible to take into account the actual problem affecting the air traffic controller's decision-making process or actual work, so it is actually used as a tool to assist the air traffic controller. There is a group.

Therefore, the present invention is to determine the landing order by applying the machine learning method to the past track data, the aircraft landing order determination device and method that can determine the landing order of the aircraft by applying the Constrained Position Shifting (CPS) technique to the reference arrival order To present.

Referring to FIG. 1, the aircraft landing order determining apparatus 100 may include a reference arrival order generation unit 110 and a landing order generation unit 120. The reference arrival order generation unit 110 may generate the reference arrival order based on a pair-based landing preference probability calculated in pairs with other aircraft for each of a plurality of aircrafts having different routes while approaching the target airport. Various routes exist for landing at one airport, and the reference arrival order generation unit 110 may determine landing priorities of a pair of planes landing at the airport on different routes, and landing priority of each pair of airplanes. The reference arrival order may be generated based on a ranking.

2 is a diagram illustrating an example of an entry way point of an airport of the aircraft landing order determining apparatus according to an embodiment of the present application.

For convenience of explanation, Incheon Airport will be described as an example. FIG. 2 shows Runway 15/16 of Incheon International Airport and Regional Navigation (RNAV) Standard STAR (Terminal Arrival Route). The Seoul TMA is bordered by other TMA and Regional Control Center (ACC) sectors. The Seoul TMA spans 60 nautical miles (NM) west, 70 nautical miles east, 25 nautical miles north, and 55 nautical miles south from Incheon International Airport (ICN). In addition, aircraft landing at Incheon Airport may enter the Seoul TMA through one of four entry waypoints (REBIT, OLMEN, GUKDO and KARBU). For example, the aircraft passing through the entry waypoint REBIT may be an aircraft departing from China and Europe, and the aircraft passing through the OLMEN may be an aircraft departing from Southeast Asia and Jeju. In addition, the aircraft passing through GUKDO may be from Japan and Oceania, and from KARBU may be from North America. Each aircraft passing the entry fix can access the Seoul TMA controller from the ACC sector controller. Thereafter, the aircraft may move to the ICN Control Tower after passing through the DANAN waypoint according to the designated STAR procedure.

3 is a conceptual diagram schematically showing the flow of aircraft landing order determination of the aircraft landing order determining apparatus according to an embodiment of the present application.

Referring to FIG. 3, some of the air trajectory data is used as training data for artificial intelligence learning. In addition, weather data may be considered in learning using training data. The reference arrival order generation unit 110 may predict the arrival order of the aircraft approaching the same destination (airport) in different directions based on the pair-based landing preference probability. The pair-based landing preference probability is a model trained by pairwise preference learning performed by pairing two random aircrafts having different routes based on training data including past track data. For, it can be calculated by inputting the state information at the reference time point for each of the pair of aircraft to be compared (S10). For example, the state information may include the longitude, latitude, altitude, and speed of the aircraft. In learning the aircraft landing sequence from historical track data, since the number of aircraft changes over time, it is difficult to simultaneously learn the aircraft landing sequence considering all possible arrival conditions. Therefore, the aircraft landing order is treated as an object prioritization method to determine the priority of an object pair through pair-based preferred learning, and further, to determine the priority of the entire object.

Flexible sequencing algorithms because the number of flights of an aircraft for a single airport is constantly changing, and the preference relationship between aircrafts, that is, landing priority, is determined by a combination of variables such as the direction of entry into the TMA and the number of aircraft. This requirement is important. Thus, a model based on pair-based preference learning (hereinafter referred to as a pair-based preference learning model) can be easily extended even in the presence of efficient computation and various variables. A landing sequence similar to the following) can be calculated.

The pair based landing preference probability may be calculated by Equation 1.

[Equation 1]

Here, AC _i , AC _j are a pair of aircraft to be compared, PREF ( AC _i , AC _j ) represents the probability that the aircraft AC _i will land before the aircraft AC _j . The closer the output of PREF ( AC _i , AC _j ) is to 1, the higher the probability that aircraft AC _i will land before aircraft AC _j . Among various binary classification algorithms, a logistic regression model can be used to estimate the binomial selection. The binary logistic regression model is a statistical method used to derive two results coded as '1' or '0'. The dependent variable of the logistic function has a pair-based landing preference probability AC _i > AC _j or AC _j > whether AC _i or not, and the independent variable represents status information about the aircraft. That is, the preference probability of a pair of aircraft may be calculated in consideration of the state information of the aircraft's longitude, latitude, altitude, and speed. In addition, AC _i A logistic function representing the probability that AC _j is true may be calculated through Equation 2 (S20).

[Equation 2]

는 상기 비교 대상이 되는 한 쌍의 항공기 각각에 대한 상태정보, 즉, k개의 독립 변수 벡터를 의미하고,

는 상기 쌍 기반 선호 학습에 의해 계산된 값, 즉, 학습 데이터로부터 추정된 계수 벡터를 의미한다. 쌍 기반 착륙 선호 확률은 항공기 각각이 이용하는 도착 경로(항로)의 조합이 고려될 필요가 있다. 따라서, 총 _m C ₂ +m 쌍의 모델이 구축될 필요가 있다.(여기서 m은 항로의 수를 나타낸다) 다시 말해, 모든 경로 조합에 대해 쌍 기반 착륙 선호 확률이 연산될 필요가 있다(S30). here,

Denotes state information of each pair of aircraft to be compared, that is, k independent variable vectors,

Denotes a value calculated by the pair-based preferred learning, that is, a coefficient vector estimated from the training data. The pair based landing preference probability needs to take into account the combination of arrival routes used by each aircraft. Therefore, a total model of _m C ₂ + m pairs needs to be constructed (where m represents the number of routes). In other words, pair based landing preference probabilities need to be calculated for all route combinations (S30). .

The reference arrival order generation unit 110 may generate the reference arrival order based on a score obtained by accumulating the pair-based landing preference probability calculated in pairs with each other for each of the plurality of aircraft. To generate a reference arrival sequence for the landing order across all n aircraft pairs, as described above, all aircraft pairs along the route { AC _i , AC _j | The calculation of pair based landing preference probabilities for 1 ≦ i ≦ j ≦ n} is done first. Then, the total landing order, that is, the base arrival order, that best matches the model trained by pair-based preference learning is determined. At this time, when the number of objects increases, the number of cases in the entire sequence increases exponentially, and the amount of calculation also increases. Accordingly, the reference arrival order generation unit 110 may assign a score to the pair-based landing preference probability of each aircraft pair, and determine the reference arrival order by combining them.

A score obtained by accumulating the pair-based landing preference probability calculated in pairs with each of the other aircrafts except the one aircraft with respect to one of the plurality of aircrafts may be calculated by Equation 3 (S40).

[Equation 3]

Here, Score ( i ) represents the score of one aircraft of the plurality of aircraft. The order of arrival may be determined in such a way that aircraft with higher scores come before aircraft with lower scores. The score function according to Equation 3 may provide an average for measuring and comparing preferences between the aircraft, but does not provide a probability of landing before other aircraft. Accordingly, the reference arrival order generation unit 110 may generate the reference arrival order by comparing the landing order between the aircrafts, which may be calculated by Equation 4 (S50).

[Equation 4]

Where Score ( i ) is the score of one aircraft AC _i of n aircraft, Score ( j ) is the score of another aircraft AC _j other than the one of n aircraft, AC _i > AC _j is relative This means that aircraft AC _i with a high score is given an earlier order in the reference arrival order than aircraft AC _j with a relatively low score.

4 is a diagram showing an example of an inter-aircraft score for generating a reference arrival order of the aircraft landing order determination device according to an embodiment of the present application, Figure 5 is a view of the aircraft landing order determination device according to an embodiment of the present application It is a figure which shows the example of the reference arrival order computed according to the score between aircrafts.

4 shows a pair-based landing preference probability calculated on the assumption that there are four aircraft, and FIG. 5 shows a reference arrival order determined according to scores of the four aircraft. The arrow in each aircraft pair { AC _i , AC _J } in FIG. 4 is directed to the aircraft of low pair based landing preference probability in the aircraft of high pair based landing preference probability. Referring to FIG. 5, a score of each aircraft may be calculated by summing pair-based landing preference probabilities of each of the four aircraft, and the reference arrival order may be determined according to the high scores.

The reference arrival order generation unit 110 may calculate the reference arrival order in consideration of Spearman's rank correlation between consecutive pairs of ranks, thereby minimizing the risk of inaccurate reference arrival order calculation. Can be. In exemplary embodiments, a similarity measurement function such as Kendall rank correlation may be used in addition to the similarity function (Spearman's rank correlation). In addition, the similarity function (Spearman's rank correlation) may be calculated by Equation 5.

[Equation 5]

는 실제 시퀀스(항공 관제사에 의해 결정된 착륙 순서) Seq와 추정 시퀀스(쌍 기반 선호 학습에 의한 기준도착순서) Seq^j 사이의 순위 거리이고, Seq(AC_i)는 항공기 AC_i의 실제 착륙 순서고, n은 항공기의 수를 나타낸다. 상기 수학식 5를 통해 연산되는 계수는 -1 와 +1 사이의 값을 가질 수 있다. 추정된 시퀀스가 실제 시퀀스의 반대일 경우에는 음의 계수가 산출되고, 추정된 시퀀스 및 실제 시퀀스가 동일할 경우에는 양의 계수가 산출될 수 있다.here,

Is the rank distance between the actual sequence (landing order determined by the air controller) Seq and the estimated sequence (baseline arrival order by pair-based preferred learning) Seq ^j , Seq (AC _i ) is the actual landing order of aircraft AC _i , n represents the number of aircraft. The coefficient calculated through Equation 5 may have a value between -1 and +1. A negative coefficient may be calculated when the estimated sequence is the opposite of the actual sequence, and a positive coefficient may be calculated when the estimated sequence and the actual sequence are the same.

Assuming that the pair-based landing preference probabilities calculated through Equations 1 and 2 provide an accurate estimate, the reference arrival order calculated through Equation 4 is D ( Seq ^j , Seq ) , which is an expected rank distance for the actual sequence. The prediction to minimize can be calculated. This may be represented by Equation 6.

[Equation 6]

Where A is a set of aircraft and S _n is a set of all permutational substitutions of A.

The landing order generator 120 may determine a landing order of the aircraft by applying a Constrained Position Shifting (CPS) technique to the reference arrival order. The CPS technique is a technique that assumes that each aircraft cannot be reordered more than the designated number by designating a specific number under the intuition that the reference arrival order will be close to the optimum. Specifically, the landing order generation unit 120 changes or maintains the landing order based on the generated reference arrival order in consideration of the maximum number of allowed position shifts (MPS) assigned to each of the plurality of aircraft. Thus, the aircraft landing order can be finally determined. The number of order variation allowances may be predetermined. For example, given constraints on landing, such as the status information of the aircraft and the allowable number of change in order, the landing order generator 120 may determine an optimal landing order that does not exceed the allowable number of change in order, and averages. The landing sequence can be determined in order to minimize delay, or to maximize runway throughput.

Hereinafter, an experiment for evaluating the performance of the aircraft landing order determination apparatus of the present application by applying actual aircraft trajectory data to the calculated reference arrival sequence will be described. First, we validate the model trained by pair-based preference learning using the training data set, and evaluate the performance of the pair-based preference learning model that yields the reference arrival order through static and dynamic analysis. For the experiment, 5700 track data that landed on 15L / R runway of Incheon Airport are used, 5200 of which are used as training data for constructing learning model and 500 are used as test data.

6 is a diagram illustrating an example for explaining a method of extracting a pair of aircraft for learning the aircraft landing order determining apparatus according to an embodiment of the present application.

Referring to Figure 6, for each aircraft pair all points on the trajectory are paired with points with the same time stamp on different orbits.

It is important to identify the appropriate set of variables in the pair-based preference learning model. Therefore, the variable set is set with reference to common opinions of professional air traffic controllers. According to the above opinion, the current position of the aircraft and the distance between the aircraft and the runway flying along the designated route were found to be the main factors affecting the landing sequence. However, each aircraft may deviate from a designated path within the TMA to maintain safe separation from other aircraft. In this case, the remaining flight distance of the aircraft along the trajectory pattern to the runway cannot be measured in real time. Thus, the distance between the current position of the aircraft and the runway is used as an independent variable in the pair based preference learning model. The main factors affecting the landing sequence are speed and altitude of the aircraft, and if the speed is high, the landing priority is likely to be high. In addition, low altitude aircraft have a higher landing priority than other aircraft.

When determining the landing order, the air traffic controller may determine the landing order by giving priority to the efficiency of the aircraft traffic flow in the airport. In addition, the designated or requested runway may be critical for determining the landing order of the air traffic controller. If the airport is busy, other aircraft may use the landing gate assigned to the landing aircraft. In this case, the air traffic controller may make changes that delay the order of some aircraft. However, the availability of the landing gate of Incheon Airport is not provided in real time, so the air traffic controller of Seoul TMA cannot take this into account. Incheon Airport, on the other hand, includes three parallel runways that are dependent on each other. Thus, the designation of runways does not affect the landing order because parallel runways that are dependent on each other are considered as a single runway with reduced separation requirements.

Another factor that affects the aircraft landing order is the aircraft type. The descent speed and altitude profile of an aircraft depends on aircraft performance and may vary by aircraft type. Thus, air traffic controllers may consider the type of aircraft when determining the order of landing of the aircraft. In addition, different segregation criteria apply depending on the type of preceding and following aircraft, and air traffic controllers may consider this to maximize runway capacity. Aircraft traffic conditions throughout the landing procedure may be another determinant of landing order. These independent variables can vary from airport to airport.

The binomial logistic regression model described above can be used to construct a pair-based preference learning model for each combination of airport traffic flows for aircraft landing. Since Incheon Airport has four arrival procedures as shown in FIG. 2, a pair-based preference learning model consisting of ten pairs can be constructed. In the model construction, the status information and actual landing sequence of two aircrafts AC-1 and AC2 flying Seoul TMA were used for learning. In models where other entry fixes were used (e.g. REBIT-OLMEN), AC1 represents the aircraft at the first entry point (REBIT) and AC2 represents the aircraft at the second entry point (OLMEN). In models where the entry fixes are all the same (e.g. REBIT-REBIT), AC1 and AC2 correspond to the preceding and next aircraft, respectively.

After standardizing the scale of all independent variables, Wald tests are performed on all models to examine the univariate associations between each variable and pair order. The Wald test is a hypothesis test based on the hypothesis that independent variables are not statistically significant. At the significance level α = 0.05, all features except one or two showed statistically similar results in most models. In addition, only statistically significant features were selected as independent variables for each paired model.

Compute variation influence factors (VIF) to further identify potential dependencies between variables. These values measure the degree of multicollinearity in the regression analysis, and independent variables that show the variance influencing factor can be considered to provide sufficient variability that is not generally explained. No significant multicollinearity was found in the model because the variance effect factor of the variables in all nodes was less than 5.

The probability that an aircraft using the KARBU route will have a higher priority than other aircraft is not necessarily low, even if the volume of traffic through the KARBU route is much less than that of other aircraft. For example, of the 75 pairs of aircraft for the OLMEN-KARBU route, the percentage of KARBU aircraft landing before OLMEN aircraft is 45%. Thus training data may not be considered imbalanced.

FIG. 7 is a diagram illustrating a coefficient for each variable according to a route in an experiment for evaluating the effectiveness of the aircraft landing order determination device according to an embodiment of the present application.

)를 도시하고, 통계적 유의성이 괄호안에 표시된다. 또한, 독립 변수로 선택되지 않은 계수는 음영 처리 된다. 쌍 기반 선호 학습 모델에서 양의 계수는 해당 독립변수가 종속변수(즉, 착륙 순서)와 양의 상관관계를 가짐을 의미하고, 음의 계수는 해당 독립변수가 종속변수에 음의 상관관계를 가짐을 의미한다. 또한 계수의 절대 값이 클수록 해당 변수와 종속변수 간의 상관 관계가 크다고 해석할 수 있다.Referring to FIG. 7, the estimated coefficient for each variable (

), And statistical significance is indicated in parentheses. Also, coefficients not selected as independent variables are shaded. In a pair-based preferred learning model, positive coefficients mean that the independent variable has a positive correlation with the dependent variable (ie, landing order), while negative coefficients have a negative correlation with the dependent variable. Means. It can also be interpreted that the greater the absolute value of the coefficient, the greater the correlation between the variable and the dependent variable.

Learning results show that aircraft close to the airport and those at lower altitudes have higher landing priorities. All coefficients for the DIST1 and ALT1 variables were negative (β ₁ , β ₂ <0) and all coefficients for DIST2 and ALT2 were positive (β ₄ , β ₅ > 0). It would be reasonable for an air traffic controller to first handle an aircraft approaching closer to the airport at a lower altitude. In the model for the same route (eg REBIT-REBIT), SP1 had a negative coefficient (β3 <0) and SP2 had a fixed coefficient (β6> 0). In general, aircraft closer to the airport at lower altitudes are more likely to fly at lower speeds, so they get higher landing priority. In models for other routes (eg OLMEN-GUID), all SPD variables have positive coefficients except one. All ALT coefficients showed greater absolute values than the corresponding SPD variables (| β2 |> | β3 |, | β5 |> | β6 |). This means that access to the aircraft inside the TMA will gradually lower the altitude from entry speed to landing, while the speed limit cannot be easily adjusted according to the speed limit. Thus, the altitude of the aircraft has a greater influence on the decision of the air traffic controller.

To verify the performance of the pair-based preference learning model, we performed 10 cross validations. The entire data set was randomly divided into 10 subsets of equal size. We then repeated the training and validation process ten times, using different subsets of data to test performance in each test. As a result, the accuracy of all trained models except KARBU-KARBU was over 88% and the average was 94.6% as shown in FIG.

To assess the accuracy of multiple aircraft, we generate a baseline arrival sequence for 140 sets of four consecutive arrivals based on a pair-based preference learning model. By way of example, for each set, six pairs of probability estimates are computed from the pair based preference learning model. The reference arrival order is generated based on Equation 3 and Equation 4 according to the score of each aircraft. The accuracy of the reference arrival order by the pair-based preference learning model is evaluated using the Spearman's rank correlation. For each aircraft landing, a correlation between the order according to the reference arrival order and the actual sequence (the landing order determined by the air controller) is calculated, and an average is calculated. The rank correlation coefficient with respect to the average may be calculated through Equation 7.

[Equation 7]

Here, k is 140, the number of sets, and ρs is a rank correlation coefficient by the similarity function of the s-th set defined in Equation (5). The rank correlation coefficient of the proposed pair-based preference learning model was calculated to be 0.8571.

We used the same algorithm to analyze the sensitivity of the results to feature selection, but built two reference models with different sets of variables. This model was applied to the same data set as the current model for evaluation. In addition, as a result of using only a variable for distance, a rank correlation coefficient according to the Spearman's rank correlation was calculated to be 0.7971. This means that the remaining distance is the most important factor in order of arrival, but other factors may also influence the decision of the air traffic controller.

In addition, tests were conducted using the latitude / longitude of the aircraft's current position as a variable instead of the distance to the runway. At this time, other variables were kept the same as the pair-based preferred learning model described above. The average accuracy of the modified modified pair-based preference learning model was 95.4%, which is almost the same as the above model considering distance variables. On the other hand, the accuracy for the entire sequence of multiple flights was slightly lower than the learning model with a rank correlation coefficient of 0.8086 according to Spearman's rank correlation, resulting in a 2.4% lower result on the scale of [-1, +1]. . That is, the pair-based preference learning model according to the experiment is similar to the landing order by the air traffic controller in terms of rank correlation coefficient according to Spearman's rank correlation, and in general traffic conditions, it is possible to obtain high accuracy results. However, since the actual air traffic controller can flexibly modify the landing order according to the airport's real-time traffic situation, the landing order prediction in a certain traffic situation is somewhat difficult. As shown in FIG. 2, the REBIT route has a shorter track length, which makes the arrival speed and descent rate larger than those of other routes at the same distance from Incheon International Airport. In addition, aircraft arriving from the REBIT route depart from the west and aircraft using other routes arrive east, which can cause difficulties in organizing traffic flows on the runway at REBIT. For these structural reasons, REBIT flight is relatively inflexible and air traffic controllers tend to give other priorities to aircraft using REBIT routes when performing traffic flow management.

8 is a view comparing the landing order by the conventional method (conventional technique) and the landing order by the air traffic controller, Figure 9 is a view showing the minimum separation criteria for each aircraft weight.

Referring to FIG. 8, the landing order calculated through optimization may be different from the landing order determined by the actual air traffic controller.

Because airspace near the airport has a complex traffic environment, it may be impractical to recommend an air traffic controller an optimal landing sequence that is different from the air traffic controller's decision. Conventionally, first-come-first-served FCFS and CPS techniques have been used to solve this problem. The landing sequence by the first-come-first-served technique is a method of landing an aircraft with a fast estimated time of arrival (ETA) first, similar to the landing sequence predicted by the air traffic controller, but not completely consistent. As such, the first-in-first-out service technique cannot fully reflect the intention of the air traffic controller, so there is still a difference in the landing order of the first-in-first-hand service technique and the air traffic controller. Accordingly, the aircraft landing order determining apparatus 100 may predict the landing order that is the same as the landing order predicted by the air traffic controller by using the pair-based preference learning model, and is more optimized by applying the aforementioned CPS technique. The landing order can be calculated. As described above, the landing order generation unit 120 changes the landing order in consideration of the maximum number of shifts allowed (Maximum Position Shift, MPS) assigned to each of the plurality of aircraft based on the generated reference arrival order. Or by maintaining the aircraft landing order. Hereinafter, the pair-based landing preference model to which the CPS technique is applied is referred to as a hybrid sequencing model. The number of order variation allowances may be predetermined. For example, given constraints on landing, such as the status information of the aircraft and the allowable number of change in order, the landing order generator 120 may determine an optimal landing order that does not exceed the allowable number of change in order, and averages. The landing sequence can be determined in order to minimize delay, or to maximize runway throughput.

Experiments were conducted to evaluate the accuracy and performance of the landing sequence calculated by applying the CPS technique. The data set used in the experiment used 5700 trajectories of data that landed at Incheon International Airport, and each test sample considered six aircraft arriving in succession to account for high traffic congestion. In this experiment, we evaluate by comparing the landing order by the pair-based landing preference model without the CPS technique and the landing order by the hybrid sequencing model. The landing order was determined in the direction of minimizing the average delay time, and the minimum separation criteria for each weight shown in FIG. 9 were applied. The aircraft's trajectory prediction is applied to the multiple linear regression model. In the regression model, altitude and speed are included as independent variables, and other variables are not considered.

FIG. 10 is a diagram comparing a rank correlation coefficient by a hybrid sequencing model of an aircraft landing order colon apparatus and a pair-based landing preference model not applied to the CPS method, according to an exemplary embodiment.

Referring to FIG. 10, the rank correlation coefficient may indicate the acceptability of the landing order generated by each model as an advice on determining the landing order of the air traffic controller. In other words, the higher the correlation coefficient, the more similar the landing order determined by the air traffic controller. Therefore, it may be interpreted that the air traffic controller is more likely to accept the landing order predicted by the model.

Referring to FIG. 10, the hybrid sequencing algorithm increases the rank correlation coefficient by 5.2% (0.6817 to 0.7174) at a cost that increases the average delay time by 5.6% compared to the optimal landing sequence by the conventional method. You can. On the other hand, the rank correlation coefficient of the reference sequence by proposed prediction model without applying the CPS technique is 0.8029, which is 17.8% higher than the rank correlation coefficient of the optimal landing order. The landing order of the aircraft is to be finally determined by the air traffic controller, but the landing order calculated by the aircraft landing order determining device 100 according to the embodiment of the present application described above is similar to the actual landing order. As a result, it can contribute to the decision of the air traffic controller. That is, a more optimized and accurate landing order of the aircraft can be derived through the interaction between the air traffic controller and the learning model.

11 is a view showing the flow of the aircraft landing order determination method according to an embodiment of the present application.

The aircraft landing order determination method illustrated in FIG. 11 is performed by the aircraft landing order determination apparatus 100 described with reference to FIGS. 1 to 10. Therefore, even if omitted below, the contents described with respect to the aircraft landing order determining apparatus 100 through FIGS. 1 to 10 may be equally applied to FIG. 11.

Referring to FIG. 11, in operation S1110, the reference arrival order generation unit 110 approaches a target airport based on a pair-based landing preference probability calculated in pairs with another aircraft for each of a plurality of aircrafts having different routes. Reference arrival order can be generated. The reference arrival order generation unit 110 may determine the landing priority of the pair of planes landing at the airport on different routes, and generate the reference arrival order based on the landing priority of each plane pair. .

The pair-based landing preference probability is a model trained by pairwise preference learning performed by pairing two random aircrafts having different routes based on training data including past track data. With respect to, can be calculated by inputting the state information at the reference time point for each of the pair of aircraft to be compared. For example, the state information may include the longitude, latitude, altitude, and speed of the aircraft. In learning the aircraft landing sequence from historical track data, since the number of aircraft changes over time, it is difficult to simultaneously learn the aircraft landing sequence considering all possible arrival conditions. Therefore, the aircraft landing order is treated as an object prioritization method to determine the priority of an object pair through pair-based preferred learning, and further, to determine the priority of the entire object.

The pair-based landing preference probability may be calculated by Equation 8.

[Equation 8]

Here, AC _i , AC _j are a pair of aircraft to be compared, PREF ( AC _i , AC _j ) represents the probability that the aircraft AC _i will land before the aircraft AC _j . The closer the output of PREF ( AC _i , AC _j ) is to 1, the higher the probability that aircraft AC _i will land before aircraft AC _j . Among various binary classification algorithms, a logistic regression model can be used to estimate the binomial selection. The binary logistic regression model is a statistical method used to derive two results coded as '1' or '0'. In addition, a logistic function representing the probability that AC _i > AC _j is true may be calculated through Equation 9.

[Equation 9]

는 상기 비교 대상이 되는 한 쌍의 항공기 각각에 대한 상태정보, 즉, k개의 독립 변수 벡터를 의미하고,

는 상기 쌍 기반 선호 학습에 의해 계산된 값, 즉, 학습 데이터로부터 추정된 계수 벡터를 의미한다.here,

Denotes state information of each pair of aircraft to be compared, that is, k independent variable vectors,

Denotes a value calculated by the pair-based preferred learning, that is, a coefficient vector estimated from the training data.

The reference arrival order generation unit 110 may generate the reference arrival order based on a score obtained by accumulating the pair-based landing preference probability calculated in pairs with each other for each of the plurality of aircraft. A score obtained by accumulating the pair-based landing preference probability calculated in pairs with each of the other aircrafts except the one aircraft for one of the plurality of aircrafts may be calculated by Equation 10:

[Equation 10]

Here, Score ( i ) represents the score of one aircraft of the plurality of aircraft. The order of arrival may be determined in such a way that aircraft with higher scores come before aircraft with lower scores. The score function according to Equation 10 may provide an average for measuring and comparing preferences between the aircraft, but does not provide a probability of landing before other aircraft. Accordingly, the reference arrival order generation unit 110 may generate the reference arrival order by comparing the landing order between the aircrafts, which may be calculated by Equation (11).

[Equation 11]

Where Score ( i ) is the score of one aircraft AC _i of n aircraft, Score ( j ) is the score of another aircraft AC _j other than the one of n aircraft, AC _i > AC _j is relative This means that aircraft AC _i with a high score is given an earlier order in the reference arrival order than aircraft AC _j with a relatively low score.

In operation S1120, the landing order generation unit 120 may determine an aircraft landing order by applying a Constrained Position Shifting (CPS) technique to the reference arrival order. The CPS technique is a technique that assumes that each aircraft cannot be changed more than the designated number by designating a specific number under the intuition that the reference arrival order will be close to the controller's decision. Specifically, the landing order generation unit 120 changes or maintains the landing order based on the generated reference arrival order in consideration of the maximum number of allowed position shifts (MPS) assigned to each of the plurality of aircraft. Thus, the aircraft landing order can be finally determined. The number of order variation allowances may be predetermined. For example, given constraints on landing, such as the status information of the aircraft and the allowable number of change in order, the landing order generator 120 may determine an optimal landing order that does not exceed the allowable number of change in order, and averages. The landing sequence can be determined in order to minimize delay, or to maximize runway throughput.

According to an embodiment of the present disclosure, the aircraft landing order determination method may be implemented in the form of program instructions that may be executed by various computer means and may be recorded in a computer readable medium. The computer readable medium may include program instructions, data files, data structures, and the like, alone or in combination. Program instructions recorded on the media may be those specially designed and constructed for the purposes of the present invention, or they may be of the kind well-known and available to those having skill in the computer software arts. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tape, optical media such as CD-ROMs, DVDs, and magnetic disks, such as floppy disks. Magneto-optical media, and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine code generated by a compiler, but also high-level language code that can be executed by a computer using an interpreter or the like. The hardware device described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

The above description of the present application is intended for illustration, and it will be understood by those skilled in the art that the present invention may be easily modified in other specific forms without changing the technical spirit or essential features of the present application. Therefore, it should be understood that the embodiments described above are exemplary in all respects and not restrictive. For example, each component described as a single type may be implemented in a distributed manner, and similarly, components described as distributed may be implemented in a combined form.

The scope of the present application is indicated by the following claims rather than the above description, and it should be construed that all changes or modifications derived from the meaning and scope of the claims and their equivalents are included in the scope of the present application.

100: aircraft landing order determination device
110: reference arrival order generation unit
120: landing sequence generator

Claims (15)

In the aircraft landing order determination device,
A reference arrival order generation unit generating a reference arrival order based on a pair-based landing preference probability calculated in pairs with other aircraft for each of a plurality of aircrafts having different routes, while approaching the target airport; And
It includes a landing order generation unit for determining the landing order of the aircraft by applying a Constrained Position Shifting (CPS) technique to the reference arrival sequence,
The pair-based landing preference probability is a model learned by pairwise preference learning, which is performed by pairing two random aircrafts having different routes based on training data including past track data. Is calculated by inputting state information at each reference point for each pair of aircraft to be compared,
The reference arrival order generation unit,
Generating the reference arrival order based on a score obtained by accumulating a pair-based landing preference probability calculated in pairs with each other aircraft for each of the plurality of aircraft,
The landing order generation unit finalizes the landing order of the aircraft by changing or maintaining the landing order in consideration of the maximum number of shift allowances (MPS) assigned to each of the plurality of aircraft based on the generated reference arrival order. Decide,
The reference arrival order generation unit
Equation 1 below to generate the reference arrival sequence,
[Equation 1]

Where Score ( i ) is the score of one aircraft AC _i of n aircraft, Score ( j ) is the score of another aircraft AC _j other than the one of n aircraft, AC _i > AC _j is relative The aircraft landing order determination device, which means that the aircraft AC _i with a high score is given an order prior to the reference arrival order than the aircraft AC _j with a relatively low score. The method of claim 1,
The pair based landing preference probability is calculated by Equation 2 below,
[Equation 2]

Here, AC _i , AC _j is a pair of aircraft to be compared, PREF ( AC _i , AC _j ) is the aircraft landing order determination device, the aircraft AC _i is the probability of landing before the aircraft AC _j . The method of claim 2,
Equation 2 is calculated using Equation 3 below,
[Equation 3]

here,

Means status information about each of the pair of aircraft to be compared,

Is a value calculated by the pair based preference learning. delete The method of claim 1,
A score obtained by accumulating the pair-based landing preference probability calculated in pairs with each of the other aircrafts except the one aircraft with respect to one of the plurality of aircrafts is calculated by Equation 4 below.
[Equation 4]

Wherein Score ( i ) is the score of one of the plurality of aircraft, aircraft landing order determination device. delete delete In the aircraft landing order determination method,
(a) generating a reference arrival order based on a pair-based landing preference probability calculated in pairs with other aircraft for each of a plurality of aircrafts having different routes, approaching the target airport; And
(b) determining a landing order of aircraft by applying a Constrained Position Shifting (CPS) technique to the reference arrival sequence,
The pair-based landing preference probability is a model learned by pairwise preference learning, which is performed by pairing any two aircraft having different routes based on training data including past track data. Is calculated by inputting state information at each reference point for each pair of aircraft to be compared,
In step (a),
The reference arrival order is generated based on a score obtained by accumulating pair-based landing preference probability calculated in pairs with each other aircraft for each of the plurality of aircraft,
Step (b) is,
Finally determining the landing order of the aircraft by changing or maintaining the landing order in consideration of the assigned maximum position shift (MPS) for each of the plurality of aircraft, based on the generated reference arrival sequence,
Step (a) is
Equation 5 below to generate the reference arrival sequence,
[Equation 5]

Where Score ( i ) is the score of one aircraft AC _i of n aircraft, Score ( j ) is the score of other aircraft AC _j other than the one of n aircraft, AC _i > AC _j is relative Means that the aircraft AC _i with a high score is given an order prior to the reference arrival order than the aircraft AC _j with a relatively low score. The method of claim 8,
The pair based landing preference probability is calculated by Equation 6 below,
[Equation 6]

Here, AC _i , AC _j is a pair of aircraft to be compared, PREF ( AC _i , AC _j ) is the aircraft AC _i is the probability of landing before the aircraft AC _j , aircraft landing order determination method. The method of claim 9,
Equation 6 is calculated using Equation 7 below,
[Equation 7]

here,

Means status information about each of the pair of aircraft to be compared,

Is a value calculated by the pair based preference learning. delete The method of claim 8,
A score obtained by accumulating the pair-based landing preference probability calculated in pairs with each of the other aircrafts except the one aircraft for one of the plurality of aircrafts is calculated by Equation 8 below.
[Equation 8]

Where Score ( i ) is the score of one of the plurality of aircraft, aircraft landing order determination method. delete delete A computer-readable recording medium having recorded thereon a program for executing the method of any one of claims 8 to 10 and 12 on a computer.

Images