You are on page 1of 6

Proceedings of the 29th Chinese Control Conference July 29-31, 2010, Beijing, China

Aircraft Category Based Genetic Algorithm for Aircraft Arrival Sequencing and Scheduling
MENG Xiangwei1, ZHANG Ping1, LI Chunjin2
1 School of Automation Science and Electrical Engineering, Beihang University, Beijing 100191, P.R.China E-mail: mxwbuaa@asee.buaa.edu.cn 2 Research Center of Air Traffic Control, Beihang University, Beijing 100191, P.R.China Abstract: Aircraft arrival sequencing and scheduling (ASS) at airports, which is an NP-hard problem, is a major issue in daily air traffic control (ATC) operations. Much effort has been made to tackle this problem using genetic algorithms (GAs) whose chromosomes are the permutation of the order of each aircraft in the arriving queues. This paper attempts to design an efficient GA whose chromosomes are constructed as the permutation of the categories of the arriving aircraft. The advantage of the resulting GA is the reduced encoding space so that the searching efficiency is promoted, which is very important for its real time application. The extensive comparative simulation study shows that this novel aircraft category based GA outperforms the aircraft order based GAs. Key Words:Air Traffic Control,Arrival Sequencing and Scheduling, Genetic Algorithms,Scheduling

INTRODUCTION
Tab. 1 Minimum LTI [3] Sij (seconds) Category of leading aircraft: i 1 2 3 4 Category of following aircraft: j 1 96 72 72 72 2 200 80 100 80 3 181 70 70 70 4 228 110 130 90

Of particular concern in air traffic control (ATC) are increased air traffic volume and aircraft delays. According to [1], world-wide air traffic is expected to grow to unexpected levels over the coming decades: revenue passenger miles worldwide of 1.7 trillion in 1996 are anticipated to reach 4.5 trillion by 2016. Existing demands on the air traffic system routinely exceed the capacity of airports leading to air-traffic-imposed ground and airborne delays of aircraft. In USA alone, this has been estimated to cost domestic airlines as much as $3.5 billion per year. In the increasingly competitive airline industry, with its market-driven pricing and very thin profit margins, such economic operating penalties are magnified [2]. Therefore, it becomes necessary to find a way to process aircraft faster without altering the safety of both the passengers and the airport staff. Increasing the number of runways and air traffic controllers is the most obvious solutions, but is not a realistic one in most cases because of cost and space limitations. Another solution is to improve the landing strategy in order to gain time without increasing the controller workload. Arrival sequencing and scheduling (ASS), which is a major issue in daily airport operations, is one of the standard problems in Air Traffic Control (ATC) Typically, after en-route flying an aircraft will enter a terminal area (TMA), which is around an airport and usually 30 to 50 nautical miles radius from the airport, and will be assigned a landing time and a runway on which to land, guided by the control tower. The landing time must be bounded from below by the Earliest Landing Time (ELT), which is the earliest time an aircraft could reach the runway if there were no traffic. The ELT is a function of parameters, such as the aircraft speed and its distance to the runway, and can be calculated when the aircraft enters the TMA. More details on the landing procedure in the TMA can be found in [3].

Landing time interval (LTI), which is a minimum permissible time interval between two successive landings at the same runway, must be observed in the ASS problem. LTI is used to prevent the trailing aircraft from losing aerodynamic stability because of the turbulence (wake vortices) generated by the leading aircraft and also prevent the runway from being occupied by two aircraft at the same time. Since turbulence generation, sensitivity and runway occupancy time strongly depend on the aircrafts weight and speed, it causes some asymmetry in the LTI. ATC regulations have classified the aircraft into several (3~4) main categories and fixed the LTI between them. An illustrative LTI is given in Tab. 1[3]. Generally, ATC controllers use a first-come-first-served (FCFS) policy based on the ELTs in handling ASS problem. Although the FCFS policy establishes a fair order in terms of ELT, it also ignores other useful information to efficiently make use of the capacity of the airport, reduce airborne delays, and/or improve the airport service to airlines and the passengers [4,[5].Apparently, shifting aircraft position can significantly reduce airborne delays during the arriving and landing phase because of the LTIs asymmetry. For example, a minimum LTI of 200s is required for an aircraft of category 2 to follow an aircraft of category 1, whereas a minimum LTI of only 72s needs to be satisfied for the same pair of aircraft in reverse order. However, like a double-edged sword, the LTIs asymmetry also makes the

5188

position shifting based ASS an NP hard problem [3], which has no known algorithm for finding a global optimal solution within a polynomial-bounded amount of time. As a large-scale parallel stochastic searching and optimizing algorithm, Genetic algorithms (GAs) are effective in solving NP-hard problems such as ASS problem [1], [6-10]. This paper aims to shed a little more light on how to design efficient genetic algorithms for the static version ASS problem for a single runway where the set of aircraft waiting to land is known. This contrasts with the dynamic case where decisions about each aircraft must be made as time passes, aircraft land and new aircraft appear. The static case is applicable when arrivals to an airport are batched over time, and the optimal sequence for each batch is computed separately. In most practical implementations, aircraft are scheduled in batches [11], and it is hence useful to be able to solve the static problem. Different with other GAs using the order or assigned landing time (ALT) of each aircraft in the arrival queue to construct the chromosomes in ASS problem, this paper uses the permutation of the categories of the aircraft to construct the chromosomes. The big advantage of the resulting GA is the reduced encoding space, so the searching efficiency is promoted, which is very important for its real-time application. Extensive literature review on ASS problem can be found in [12,13]. The rest of the paper is organized as follows. The ASS problem is formulated in the next section. The new GA based on aircraft categories is given in section 3. An extensive comparative simulation study is reported in section 4, which is followed by some conclusions in section 5.

emphasizes the level of service. In this case, the objective function is defined as J1 = Di
i =1 N

(3)

During congested periods, instead of considering the level of service, we emphasize more on the efficiency of using the runway capacity. The objective function is defined as the maximum sequence completion time J 2 = max( AQ ( N ) ) (4) With the above mathematical preparation, the ASS can be formulated as the following minimization problem:
Q (1), , Q ( N )

min

J1 J2

(5) (6)

or
Q (1), , Q ( N )

min

subject to Eqs. (1)~(4). With regard to the ASS problem, we have the following proposition: Proposition 1 The subsequence of the aircraft of same category in the optimized queues should follow the FCFS policy Proof: We prove this result by contradiction. Suppose the final optimized sequence Q1 isQ1(m-1)Q1 (m) Q1 (m+1) Q1 (n-1) Q1 (n) Q1 (n+1) in which CQ1 ( m ) = CQ1 ( n) = i ,and EQ1 ( m ) > EQ1 ( n ) so, according to Eq.(1),we can get , AQ1 ( m ) = max( EQ1 ( m ) , AQ1 ( m 1) + S CQ1 ( m 1) , i ,

ASS PROBLEM FORMULATION

AQ1 ( m +1) = max( EQ1 ( m +1) , AQ1 ( m ) + S i , CQ1 ( m +1) , , AQ1 ( n) = max( EQ1 ( n ) , AQ1 ( n 1) + S CQ1 ( n 1) , i ,

Simply speaking, ASS is the function of generating efficient landing sequences and landing times for arrivals at the airport so that the safety separation between arrival aircraft is guaranteed, the available capacity at the airport is efficiently used and airborne delays are significantly reduced. The ASS problem can be stated as follows: A set of N aircraft are planning to land on a single runway, and Ci , Ei , Ai denote the category, the ELT, and the ALT of the ith aircraft in the original predicted arrival sequence, respectively. Let Q(n) record the nth aircraft in the optimized arrival sequence, i.e., Q(n)=i means that ith aircraft in the original predicted arrival sequence turns out to be the nth aircraft in the optimized arrival sequence. With Q, A can be calculated as
AQ(n) n =1 EQ(n) , = max( EQ(n) , AQ(n1) + S (CQ(n1) , CQ(n) ), n > 1

. Now, consider another sequence Q2 which is same as Q1 except that Q1 (m) and Q1 (n) are swapped. i.e. Q2 (i ) = Q1 (i ), i = 1 N , i m, n , Q2 (m) = Q1 (n), Q2 (n) = Q1 (m) , It is obvious that AQ2 ( j ) = AQ1 ( j ) , j = 1 m 1 , and AQ2 ( m ) = max( EQ1 ( n ) , AQ1 ( m 1) + S CQ1 ( m 1) , i

AQ2 ( m +1) = max( EQ1 ( m +1) , AQ2 ( m ) + S i, CQ1 ( m +1) AQ2 ( n) , = max( EQ1 ( m ) , AQ2 ( n 1) + S CQ1 ( n 1) , i

( (

AQ1 ( m ) ,

AQ1 ( m +1) , AQ1 ( n ) ,

(1)

where S(i,j) is the LTI for an aircraft of category j to follow an aircraft of category of i. So, the airborne delay of the ith aircraft in the original predicated arrival sequence is Di = Ai Ei (2) Regarding the objective function, we present two different formulations to test our new GAs performance. In normal operating conditions (without congestion) we aim at minimizing the total delays suffered by aircraft which

When AQ1 ( m 1) + S CQ1 ( m 1) , i < EQ1 ( n ) , at least there is the

strict inequality AQ2 ( m ) < AQ1 ( m ) . Then, according to Eq. (2), Eq. (3) and (4), J1 (Q1 ) J1 (Q2 ), J 2 (Q1 ) J 2 (Q2 ) , and in some cases the inequalities can be strictly established. This contradiction implies that the aircraft of the same category should be sequenced into the queue according to ascending order of their ELTs. Besides, in real ATC operations, even

5189

though J i (Q1 ) = J i (Q2 ), i = 1, 2 , in view of the fairness, the aircraft of the same category should follow the FCFS policy. Q.E.D.

preserved in the two children, whereas the other genes are re-permutated in the children. An illustrative example is given in Fig. 2.

AIRCRAFT CATEGORY BASED GA

In this section, we follow the common practices of designing GAs to develop the new GA for the single runway static ASS problem. The chromosome structure, crossover operator, mutation operator, fitness function and algorithm parameters are described below. 3.1 Structure of Chromosomes The chromosome in the new GA is encoded as the permutation of the categories of the aircraft in the arriving queue. When decoded, the aircraft of same category are ordered according to their ELT, i.e. the aircraft of same category with earlier ELT will land before the aircraft with later ELT. By this way the landing sequence can be determined uniquely and its rationality is guaranteed by the Prop.1. For example, supposing that the aircraft are classified into 2 categories according to the ATC regulations, and a set of 5 aircraft {1, 2, 3, 4, and 5} is waiting to land in which C1=C4=1, E1<E4 C2=C3=C5=2, E2<E3<E5 A possible chromosome for this ASS is shown in Fig.1. Fig.2 crossover operator 3.3 Mutation Operator There are two types of mutation operator in this paper. The first mutation operator is the swap mutation which changes the positions of any two genes with different value. This mutation operator is different from that adopted in [6,9], and [15] which can swap any two different genes representing the order of the aircraft in the queue and sometimes make useless mutation by swap the orders of two aircraft of the same categories. The second mutation operator, inspired from the domain knowledge that successive landing of aircraft belonging to same category will help to reduce the delay, is the insert mutation which extracts a gene and inserts it before or after another gene with the same value. The two mutation operators are illustrated in Fig. 3 and Fig. 4 respectively.

Fig. 1 The chromosome

Regarding the encoding scheme, there is following proposition. Proposition 2 if there are N aircraft in arriving queue in which there are Ni aircraft of category i,i=1m. then the cardinality of the set of possible different chromosomes is N! M = . N1 ! N 2 ! N m ! Proof: it is easy derived from the theory of permutation and combination that , M = P( N , N1 ) P ( N N1 , N 2 ) = N! N1 ! N 2 ! N m ! P( N m , N m )

Fig. 3 swap mutation operator

Fig.4 insert mutation operator

where P(n,k) is the number of all combination of n items taken k at a time Q.E.D. If the chromosome is encoded as the permutation of the order of each aircraft in the arriving queue just like in [6,14,15], the M will equal N!. So, by the aircraft category based encoding scheme, the encoding space is significantly reduced. For example, if N=30, N1=10, N2=10, N3=5, N4=5, the order based encoding space will be 914457600 times as large as the category based encoding space. 3.2 Crossover Operator The crossover operator is adapted from the uniform crossover operator; the equal alleles of the two parents are

3.4 Algorithm Parameters and Flow With the minimization objective function, the normalized fitness of an individual chromosome s is simply defined for the two objective function respectively as J J1 ( s ) f1 ( s) = 1max (7) J1max J1min + or f 2 (s ) = J 2 max J 2 ( s ) J 2 max J 2 min (8)

Where Jmax, Jmin are the maximum and minimum J in the population of current generation respectively. The is a little number which is used to prevent the denominator from becoming zero. Just like [9], the population size of a generation, i.e. NP and the maximum number of generations in the evolutionary

5190

process, i.e. NG are adjusted according to the number of aircraft in the queue, i.e. N. thus, N P = 30 + 10(round (max(0, N 10) / 5)) (9) NG = 20 + 40(round (max(0, N 10) / 5)) (10) Because the LTI is sequence dependant, the crossover operator implemented in this paper seems to work as the mutation operator. So, crossover operations are only carried out for the top 20% optimal chromosomes so as to inherit and exploit the common optimal genes. The probability of the mutation is set to a relatively high value 0.6 because the experiments show that the evolution strength mainly comes from the mutation operations, which has been demonstrated in [9] in which there is only mutation operator. The selection operator is just implemented as the standard roulette wheel selection operator. To help the algorithm converge fast, not all of the chromosomes in the first generation is generated randomly, but some are generated by the FCFS procedure. The aircraft are also grouped into several subgroups according to the ELT, and some chromosomes are generated as the union of the permutations of the categories of the aircraft in the

subgroups. Besides, the elite preservation strategy is also used in the algorithm. The algorithm flow can be described using pseudo code as below: Procedure Input: GA parameters Output: best solution Begin t 1 //t:generation number; initialize P(t) by encoding routine;// P(t):population of chromosomes; fitness evel(P(t)) by decoding routine; while (not termination condition) do crossover P(t) to yield C(t); // C(t):offspring mutation P(t) to yield C(t); fitness eval(C(t)) by decoding routine; select P(t+1) from P(t) and C(t); t t+1; end output best solution; end

Initial traffic data No. Cat. ELT(s) 1 1 2 1 3 1 4 2 5 1 6 1 7 2 8 1 9 2 10 2 11 1 12 1 13 3 14 4 15 2 16 2 17 2 18 1 19 1 20 3 21 1 22 1 23 4 24 2 25 1 26 2 27 3 28 2 29 1 30 1 Total delay(s) 0 79 144 204 264 320 528 635 730 766 790 920 1046 1106 1136 1166 1233 1642 1715 1770 2074 2168 2259 2427 2481 2679 2883 2982 3046 3091

No. 1 2 3 5 6 7 4 9 10 11 12 8 17 16 15 14 13 20 18 19 21 22 23 24 25 26 27 28 29 30

Cat. 1 1 1 1 1 2 2 2 2 1 1 1 2 2 2 4 3 3 1 1 1 1 4 2 1 2 3 2 1 1

Tab.2 Result of Single Test CIH OGA ALT(s) Del(s) No. Cat. ALT(s) 0 96 192 288 384 584 664 744 824 896 992 1088 1288 1368 1448 1558 1628 1770 1842 1938 2074 2170 2398 2478 2550 2750 2883 2983 3055 3151 3800 0 17 48 24 64 380 136 14 58 261 202 168 55 202 312 452 582 0 200 223 0 2 139 51 69 71 0 1 9 60

Del(s) 0 17 48 24 64 380 136 14 58 261 202 168 152 202 215 452 582 58 81 207 0 2 139 51 69 71 0 1 9 60

No. 1 2 3 5 6 4 7 9 10 8 11 12 13 15 16 17 14 18 19 20 21 22 23 24 25 26 27 28 29 30

Cat. 1 1 1 1 1 2 2 2 2 1 1 1 3 2 2 2 4 1 1 3 1 1 4 2 1 2 3 2 1 1

CGA ALT(s) 0 96 192 288 384 584 664 744 824 896 992 1088 1269 1369 1449 1529 1639 1711 1807 1988 2074 2170 2398 2478 2550 2750 2883 2983 3055 3151 3721

Del(s) 0 17 48 24 64 380 136 14 58 261 202 168 223 233 283 296 533 69 92 218 0 2 139 51 69 71 0 1 9 60

1 2 3 5 6 4 7 9 10 8 11 12 15 16 17 14 13 18 19 20 21 22 23 24 25 26 27 28 29 30

0 1 1 1 1 2 2 2 2 1 1 1 2 2 2 4 3 1 1 3 1 1 4 2 1 2 3 2 1 1

0 96 192 288 384 584 664 744 824 896 992 1088 1288 1368 1448 1558 1628 1700 1796 1977 2074 2170 2398 2478 2550 2750 2883 2983 3055 3151
3723

5191

SIMULATION RESULTS

Tab. 4 congested cases with J2


case 1 2 3 4 CIH(s) 2958 2845 2814 3104 CGA Mean(s) 2950 2585.7 2767.2 3025.4 Std.(s) 5.5 34.9 17.61 34.38 OGA Mean(s) 2979 2692 2782 3046 Std (s) 10.1 40.2 30 45

To fully investigate the performance of the aircraft category base GA proposed in this paper, which hereafter is denoted as CGA, we compare it with the other two algorithms: a GA based on the permutation of the order of the arriving aircraft queue reported in [6], denoted as OGA below, and the cheapest insertion heuristic algorithm reported in [3], denoted as CIH. The OGA adopts the OCGT and OCGS crossover [16], swap mutation. Other critical parameters inOGA are the number of chromosomes in the population, the probabilities of crossover and mutation and the maximum number of iterations the values of which are 20, 0.8, 0.1 and 5000 respectively. The algorithms CGA, OGA and CIH were programmed in Matlab and run on a PC (with CPU of 2.6GHZ, 1.87GB main memory). To illustrate how these algorithms work, Tab. 2 gives the results of a realistic large-scale problem with 30aircraft. The data of the original arrival flow is borrowed directly from [3]. In Tab. 2, in the first three columns we list the aircraft serial number in the original sequence, categories and ELTs; in the next three groups of four columns, the actual sequences of landing aircraft, with their categories, ALTs and delays, issued by CIH, OGA and CGA respectively. It should be pointed out that the CPU time consumed by the OGA is 5s versus less than 1s by CGA. The time spent by CIH is very little because of its nature of deterministic algorithm. It should be also noted that in the sequence issued by CIH some aircraft of same category with earlier ELTs land after those with later ELTs, which violates the fairness discipline. However, because GAs are random searching algorithms, their performances can only be fully investigated by statistical analyses. Statistical analysis of aircraft arrivals at several major airports in the United States has shown that the distribution of times between estimated arrival times of successive aircraft (estimated when the aircraft are 100 miles from their final destinations) is nearly-exponential in character [17]. So, different from the assumption in [3] that the aircraft ELTs follow the uniform distribution, we generate eight algorithm cases in which the ELTs follow the negative exponential distribution. In each case, there are 30 aircraft which are classified into four categories: 1, 2, 3 and 4with the probabilities 0.3, 0.3, 0.3 and 0.1respectively. LTIs are listed in Tab.1. The mean inter-arrival time of the aircraft ELTs for the first four and second four cases, which correspond to the under congested periods and congested periods respectively, are 110s and 80s respectively. We conduct 20 simulations for each case with the two GAs and CIH. The mean value and standard deviation of J1 for the first four cases are listed below in Tab. 3. While those values of J2 for the second four cases are listed in Tab.4. Tab. 3 under congested cases with J1
case 1 2 3 4 CIH(s) 2862 2635 3861 1925 CGA Mean(s) 2402 2599 3777 1343 Std.(s) 0 12 32.6 0 OGA Mean(s) 2430 2764 3800 1967 Std (s) 12.2 26.2 40.5 17.95

It is evident from Tab.3 and Tab.4 that CGA statistically generate more efficient aircraft sequences than CIH. Its little standard deviations also mean that CGA is stable, which is also important for its real application. It is also evident that OGA in some cases generate less efficient aircraft sequences than CIH at the cost of spending more time than CGA and CIH. It should be clarified that although in Tab.3 the standard deviations of J1 for two cases equal zero, it doesnt contradict to the fact that GAs are random algorithms. It just implies that in each of the two cases all the best solutions evolve to a same value in 20 simulation runs. After all, the standard deviations of J1 for other two cases dont equal zero.

CONCLUSIONS

This paper aims to design an efficient GA based on aircraft categories to tackle the aircraft arrival sequencing and scheduling problem. Compared with the GAs the chromosomes of which are based on the order of each aircraft in the arriving queue, the new GA significantly reduces the searching space so as to promote the algorithms efficiency. Extensive simulation results show that the new GA outperforms not only in performances than CIH, a heuristic algorithm proposed in literature, and aircraft order based GAs, but also in stabilization than the order based GAs. It is also possible to exploit the peculiarity of the GAs to produce a nearly optimal set of solutions from which air traffic controllers can choose what they deem to be optimal. Further researches may include extending the new GA to dynamic version based on receding horizon control concept [9] or Constrained Position Shifting (CPS) concept [17], and extending it to include more than one runway.

REFERENCES
[1] J. V. Hansen. Genetic search methods in air traffic control. Computers & Operations Research, 2004, 31:445-459. [2] K. Heimerman. Air Traffic Control Modeling (www.mitre.or g/ support/paper/atcmodel). [3] L Bianco, P.DellOlmo, S.Giordani, Scheduling models and algorithms for TMA traffic management [M]//L.Bianco, P. DellOlmo A. R. Odoni, eds. Modeling and Simulation in Air Traffic Management. Berlin: Springer-Verlag, 1997:139-167. [4] G. C. Gregory, H. Erzberger, F. Neuman. Delay exchanges in arrival sequencing and scheduling. Journal of Aircraft, 1999, 36(5):785-791. [5] G. C. Gregory, H. Erzberger, F. Neuman. Fast-time study of aircraft-influenced arrival sequencing and scheduling. Journal of Guidance, Control and Dynamics, 2000, 23(3):526-531. [6] S. Capr,M. Ignaccolo. Genetic algorithms for solving the aircraft-sequencing problem: the introduction of departures into the dynamic model. Journal of Air Transport Management, 2004, 10(5):345-351.

5192

[7] V.H.L. Cheng,L.S.Crawford, P.K. Menon. Air traffic control using genetic search techniques[C]// IEEE International Conference on Control Application, Hawaii, USA, 1999.8. [8] J.E.Beasley, J.sonader,P.Havelock. Scheduling of aircraft landings at London Heathrow using a population heuristic. Journal of the Operational Research Society.2001,52(5):483-493. [9] X. B. Hu, E. D. Paolo. Binary-representation based genetic algorithm for aircraft arrival sequencing and scheduling. IEEE Trans. On Intelligent Transportation Systems, 2008, 9(2): 301-310. [10] X. B. Hu, E. D. Paolo. An efficient algorithm with uniform crossover for air traffic control. Computer & Operations Research, 2009,36(1):245-259. [11] F. Neuman, H. Erzberger. Analysis of delay reducing and fuel saving sequencing and spacing algorithms for arrival spacing. NASA Technical Memorandum 103880, 1991.10. [12] J. E. Beasley, M. Krishnamoorthy, Y.M. Sharaiha, D. Abramson. Displacement problem and dynamically scheduling aircraft

[13]

[14]

[15]

[16]

[17]

landings. Journal of the Operational Research Society, 2004, 34 (2):180-197. H. Pinol,J.E. Beasley. Scatter search and bionomic algorithms for the aircraft landing problem. European Journal of Operational Research, 2006, 171(2):439-462. X.B.Hu, W.H. Chen. Genetic algorithm based on receding horizon control for arrival sequencing and scheduling. Engineering Application of Artificial Intelligence, 2005, 18(5):633-642. X.B.Hu, W.H. Chen. Receding horizon control for aircraft arrival sequencing and scheduling. IEEE Transactions on Intelligent Transportation Systems, 2005, 6(2):189-197. D. Yip-Hoi, D. Dutta. A genetic algorithm application for sequencing operations in process planning for parallel machining.IIE transction, 1996, 28(1):55-68. H. Balakrishnan, B. Chandran. Scheduling aircraft landings under constrained position shifting.[C]//AIAA Guidance, Navigation, and Control Conference and Exhibit.Keystone, Colrado, 2006.8.

5193

You might also like