Title: Airspace sectorization
1Airspace sectorization
Philippe BAPTISTE, Vu DUONG Huy TRANDAC
2Overview
- Problem Definition
- Discretization
- Constraint Programming Formulation
- 2-step approach (local improvement)
- From discrete sectors to geometrical sectors
- Experimental results
- Future work
3Airspace
- Routes
- Waypoints
- Air traffic controllers
- monitor
- Conflict resolution
- coordination
4Current issues
- Overloaded system
- Unbalanced workload
- Dynamic load
5Whats a good sectorization ?
- Partition airspace in a set of sectors
- Workloads (conflict monitoring) are balanced
- Coordination is minimized
- Specific ATC constraints are met
6ATC constraints (1)
- Contrainte de temps de passage minimum
- Contrainte de distance minimum au sens des routes
7ATC constraints (2)
- Contrainte de connexité pour éviter la
fragmentation des secteurs.
- Contrainte de convexité au sens des routes un
avion passe une fois au maximum par secteur.
8Previous work
- Delahaye 1995, Delahaye et al. 1998, AENA
02, HADES - Voronoi diagrams
- Each sector corresponds to one  vertex and the
points in one region are closer to the
corresponding vertex than to any other one - Agregate diagrams
- GA, meta heuristics
- Do not take into account specific ATC constraints
- Unrealistic sector design
9Discretization of the Problem
- Airspace Valueted graph
- Vertices airports way points crosssing
points routes change - Edges planned routes that connect vertices
- Valuations
- Vertices monitoring conflict resolution
- Arête coordination
- Graph Partition (NP-Hard)
- Balanced sets of vertices
- Minimal total cut
- ATC constraints
- Then, compute geometrical bounds of sectors
USE CP
10Constraint programming
- State the decision problem in terms of
variables (domain of possible values) and of
constraints (either implicit or explicit). - Propagate constraints to reduce domains
- Build a search tree
Decision-making
new constraints (decision)
Partial solution
Problem Definition
Initial constraints
Constraint Propagation
new constraints (propagation)
11Constraint propagation A toy example
- X ? 1, , 10, Y ? 2, , 20, Z ? 3, , 10
- X Y 5, (a)
- Z ? X 6, (b)
- Z gt 10 X Y (c)
- Propagation
- (a) X ? 5 - max(Y) ? 5 - 20, X ? 5 - min(Y) ? 3
X ? 1, 2, 3 - (a) Y ? 5 - max(X) ? 5 - 3, Y ? 5 - min(X) ? 4
Y ? 2, 3, 4 - (c) Z gt 10 min(X) - max(Y) gt 10 - 4 gt Z ?
7 Z ? 7, 8, 9, 10 - (c) Y gt 10 min(X) - max(Z) gt 10 - 10 gt Y ? 1
- (c) X lt (max(Z) max(Y)) / 10 lt (10 4) / 10
X 1 Bound - (a) Y 5 - 1 4 Y 4 Bound
- (b) Z ? 7 gt Z ? 8 Z ? 8, 9, 10
12A dummy example
- X ? 0, 1, Y ? 0, 1, Z ? 0, 1,
- X ? Y
- Y ? Z
- Z ? X
- Propagation
- Consider each constraint one after another.
- No Propagation
- No contradiction is triggered...
13Local vs. Global Propagation
- Consider the all-different constraint.
- n vars x1, , xn must take pairwise distinct
values
Local propagation ?i, ?j ? i, xi ? xj If xi is
bound, remove the corresponding value from the
domain of xj.
Global propagation matching G (X, D) D
for Domains
Variables
Domains
14Constraint programming
- Framework to integrate
- OR
- AI
- Math Prog.
15A CP model for sectorization
- Variables
- For each vertex vi one variable xi
- domain D(xi)1,k
- xi j iff vi is in sector Vj.
- Constraints
- Either predefined arithmetic constraints
- Or global constraints specific to the problem
- Speed up search
- All ATC constraints can be formulated
16Search
- Heuristic based on estimated gain
- basée sur la notion de gain estimé g(xi,val) si
xi est instanciée à la valeur val (en terme de
minimisation de la somme des poids des routes
coupées) - à chaque noeud de l'arbre de recherche, la
variable xi ayant le gain estimé g(xi,val)
maximum est choisie et évidemment, elle est
instanciée à la valeur val.
g(x4,1) (?14 ?24)-(?45 ?46) g(x4,2)
?34-(?45 ?46) où ?ij poids de larête (xi,xj)
17What we can do with this formulation
- Many constraints, many variables
- Too many !!!
- No hope to solve the problem
- But CP works well to find for bi-partition (even
for large instances) and for small instances
(general problem) - So we use CP to compute a solution with recursive
bi-partition - Will be used latter as a starter for iterative
improvememnt
18Local Improvement (1)
- Restricted Kernighan/Lin Based Local Improvement
- Based on valid exchanges
-
19Local Improvement (2)
- Â Large Neighborhood Search (LNS)
- Randomly select a set of adjacent sectors
- Keep the solution outisde this set and reoptimize
locally inside the set (with the CP procedure) - Repeat while the solution is improved
Formulation de CP ? solution optimale
20Bounadries
- Start from the discrete sectors
- Naive idea
- Build Voronoï diagrams (or Delaunay
triangulation) and agregate them - May violate route convexity!
- Slightly better
- Use mid-points of edges
- Build small sectors and agregate
- Strange shape
21Software
- Build on top of Claire / Choco
- Less than 4000 lines
- 60 sectors, 6000 vertices, 1h
- Iterative imp. gain 40
22Contributions du travail
- Problématique
- Discrétisation de la sectorisation
- Formulation en programmation par contraintes pour
la sectorisation - Approche en deux temps
- Calcul denveloppes des secteurs
- Logiciel et résultats expérimentaux
- Contributions
- Perspectives
- Développement d'une formulation flexible de
programmation par contraintes pour le problème de
sectorisation une nouvelle contrainte peut y
être ajoutée, indépendante des autres contraintes
existantes. - Proposition d'algorithmes de propagation pour les
contraintes spécifiques la contrainte de temps
de passage minimum, la contrainte de convexité au
sens des routes et la contrainte de connexité du
secteur. - Développement d'une heuristique pour l'ordre de
la sélection des variables et des valeurs. - Proposition d'un schéma hybride de bisection
récursive pour trouver une bonne solution d'une
grande instance du problème.
23Contributions du travail (2)
- Problématique
- Discrétisation de la sectorisation
- Formulation en programmation par contraintes pour
la sectorisation - Approche en deux temps
- Calcul denveloppes des secteurs
- Logiciel et résultats expérimentaux
- Contributions
- Perspectives
- Adaptation de l'heuristique de Kernighan/Lin au
problème de partitionnement de graphes pour
s'appliquer au problème de sectorisation, en
respectant toutes les contraintes spécifiques. - Proposition d'une procédure d'optimisation locale
randomisée pour améliorer une solution. - Proposition d'une approche, basée sur la notion
de tessellation polygonale et de graphe de
relation de voisinage, pour calculer des
enveloppes des secteurs.
24Whats next
- Imrpoving search procedures
- Specific global constraints
- Parallelization of search
- Multi-level formulation (abstract, solve project)
- 3D
- Better shapes
- On site tests
25Some other optimization problem in ATM/ATC
- On the management of arrivals
- Pure scheduling problem
- Hybrid problems (trajectories sequencing)
- Missed time slots and delays Why ? Can we do
better ?
26On the management of arrivals Model 1
- When entering the TRACON area, aircraft either
land immediately or are put on stacks - Candidate landing times set of time windows
- Goal Schedule aircrafts so that there is enough
time between consecutive landings - Model proposed by Bayen Tomlin
- Contribution
- Establish complexity status of the problem
- Solve large instances of the problem by Branch
and Cut Techniques (LP)
27On the management of arrivals Model 2 (THALES
RT)
- Same situation Landing aircraft
- No more fixed routes Do what you want bu compute
safe trajectories and optimize the landing
sequence - Hybrid problem
- Combinatorial problem
- Diff equation
- We can only solve toy problems of a toy model
28Missed Time Slots
- Time slots are allocated to aircraft to avoid
congestion - Operations uncertainties
- gt missed time slots, network effect, more
congestion - Find regularities in past flight data and use
them to improve schedules