See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/267453361

The Algorithm of the Time-Dependent Shortest Path Problem with Time

Windows

Article  in  Applied Mathematics · October 2014

DOI: 10.4236/am.2014.517264

CITATIONS READS
18 515

1 author:

Nasser El-Sherbeny
Al-Azhar University
26 PUBLICATIONS   300 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Optimization Network View project

Multi-objective VRPTW View project

All content following this page was uploaded by Nasser El-Sherbeny on 03 January 2019.

The user has requested enhancement of the downloaded file.

El-Sherbeny, Scientific Research and Impact, 2(1): 1-8, Febuary 2013
Available online at http://www.scienceparkjournals.org/SRI
(ISSN 2315-5396) © 2013 Science Park Journals

Full Length Research Paper

A new algorithm of A* algorithm for the time-dependent

shortest path problem with time windows

Nasser A. El-Sherbeny
Department of Mathematics, Faculty of Sciences, Al-Azhar University, (11884)-Nasr City, Cairo, Egypt
Department of Mathematics, Faculty of Applied Medical Sciences, Taif University, Turaba, KSA

Accepted 15 January 2013

Abstract: This paper present a new algorithm for the time-dependent shortest path problem with time windows
without a waiting time. Given G = (V , E ), where V is a set of nodes, E is a set of edges be directed graph, a
nonnegative transit-time function c e (t ) for each edge e = (v, w) ∈ E , where t ∈T is the time to leave v ,
av ≤ t ≤ bv , [ av , bv ] is a time window for each node v ∈ V . A source node s , a destination node d and a departure
time t o , the time-dependent shortest path problem time windows asks to find an s, t − path that leaves s at time t o
and minimizes the arrival time at d . This formulation generalizes the classical shortest path problem in which ce are
constants. In this problem, the generalized Dijkstra algorithm was suggested by (Dreyfus, 1969), there was no
significant advancement despite of many studies. This paper presents a new generalized algorithm and, as an
application, also gives a generalization of the ALT algorithm for the classical problem due to (Goldberg and Harrelson,
2005).

Keywords: shortest path, time-dependent shortest path, ALT algorithm, time windows.
_________________________________________
Correspondence Email: [email protected]
1. Introduction

The shortest path problem on graphs is a problem with many real-life applications like route planning in an internet, car
navigation system, traffic simulation or logistic optimization. The shortest path problem is a classical combinatorial
optimization problem. It has countless applications and so far numerous algorithm have been proposed (see Ahuja et
al., 1993) including the well-known Dijkstra’s algorithm. Recently, partly because new improvement becomes fairly
difficult, researchers began to study variants of this problem, which include the time-dependent and the time windows
generalization.
Given a directed graph G = (V , E ), where V is a set of nodes, E is a set of edges, a nonnegative transit-time
function c e (t ) for each edge e = (v, w) ∈ E , where t ∈T is the time to leave
v , av ≤ t ≤ bv ,[av , bv ] is a time window
for each node v ∈ V , a source node s ∈ V with time window [ as , bs ], a destination node d ∈ V with time window
[ad , bd ], and departure time to , the time-dependent shortest path problem with time windows and without a waiting
time, asks to find an s , t − path that leaves s at time t o and minimizes the arrival time at d which satisfies the time
windows condition (see Figure 1 for an illustration) (see El-Sherbeny 2001; El-Sherbeny and Tuyttens 2001; Tuyttene et
al., 2004). Notice undirected graphs can be treated by replacing each undirected edge with two reverse directed edges.
Without loss of generality, we suppose d is reachable from s . For simplicity, we suppose the domain of definition for all
ce (t ) is ℜ + , but our algorithms work for the discrete version too. We also assume the time complexity to calculate a
c e (t ) is bounded by some constant α . This formulation generalizes the classical shortest path problem with time
windows, constant c e (t ) and t o . It can further handle time-variable edge costs, thus has more application than the
classical one, which is also referred to as the static problem in contrast.

ce (t )

time windows: [ as , bs ] s d time windows: [ad , bd ]

departure time: to

Figure 1: the time-dependent shortest path problem with time windows. The difference from the static problem is that
the edge length is generalized from a constant to a time-variable function, hence a departure time to , a s ≤ to ≤ bs at
the source s with time window [ as , bs ] is also needed as an input.
In (Cook and Halsey, 1966), considered it and gave a dynamic programming algorithm which is not polynomial-time at
all. (Dreyfus, 1969), then suggested a polynomial-time straightforward generalization of the Dijkstra algorithm (see
Section 2). However, he did not notice that it works correctly only for instances satisfying the FIFO (First-In First-Out)
property, i.e., for any edge e = (v, w) ∈ E and t1 ≤ t 2 , it holds that t1 + ce (t1 ) ≤ t2 + ce (t2 ), in other words, the arrival-
time function t + ce (t ) is non-decreasing. With this property, we can ensure that there is no cycle of negative transit-
time, hence a simple optimal solution exists. This was pointed out and discussed later (see Halpern 1977; Kaufman and
Smith 1993; Orda and Rom 1990).
On the other hand, the general problem without the FIFO constraint is NP-hard if waiting at nodes is not allowed (see
Sherali et al., 1998). In (Orda and Rom, 1990) showed that, however, if waiting at nodes is allowed, which is natural
in transportation systems, then any instance can be converted to an equivalent instance that satisfies the FIFO property,
hence no waiting is needed, and that can be done in polynomial time (if c e (t ) can be calculated in polynomial time).
Thus in the following, we will only consider instances that satisfy the FIFO property.
Even with the FIFO constraint, unlike the static case, studies are not rich. Dreyfus’s proposal of the generalized Dijkstra
algorithm, despite of many studies (see Dean 1999; Ding et al. 2008; Halpern 1977; Kanoulas et al. 2006; Kaufman and
Smith 1993; Orda and Rom 1990), there was no significant advancement in solving the problem more efficiently.
*
In this paper, we give a new algorithm that generalizes the A algorithm (see Hart et al. 1986) for the static problem.
Unlike the generalized Dijkstra algorithm, in Section 2. We note that Kanoulas et al., (2006) considered, for piecewise
linear functions, a slightly-generalized A* algorithm which we will discuss later in Section 2 also. In section 3, we give an
application instance of our algorithm, we will give a generalization of the ALT algorithm (see Goldberg and Harralson,
2005) that is based on the static A* algorithm and is faster than the Dijkstra algorithm using preprocessing. Thus we
have found the first algorithm for the time-dependent problem that speeds up the calculation using preprocessing, which
is observed to be several time faster than the generalized Dijkstra algorithm. Finally, the conclusion is given in Section
4.

2. A new algorithm of A* algorithm for the time-dependent shortest path problem with time windows

We start from the classical and well-known Dijkstra algorithm. Suppose ce (t ) = ce is a constant for each edge
e = (v, w) ∈ E , where t ∈T is the time to leave v , av ≤ t ≤ bv , [av , bv ] is a time window for each node v ∈ V , if
t o = 0 (the value of t o is not important in this static case), the Dijkstra algorithm tries to find a shortest s, d -path in
greedy manner. Let p(v) denote the precedent node of a node v where v is satisfies the time window condition of the
shortest s, v -path found so far. The Dijkstra algorithm maintains for each node v a status (v) ∈ {“unlabeled”, “labeled”,
“finished”} and a distance label g (v) . At the beginning, g (s) is set to 0 and all status (v) are initialized to “unlabeled”
except that s is “labeled”. Then it repeatedly find a “labeled” node v with the smallest g (v) (such v is called the
active node) until v = d ; then it tries to relax all non-“finished” neighbours w of v , i.e., if status (w) = “unlabeled” then
set it to “labeled” and let g ( w) = g (v) + c( v ,w ) , p( w) = v; otherwise status (w) = “labeled”, with time window [ aw , bw ],
then let g ( w) = g (v) + c( v ,w ) , p( w) = v if g ( w) f g (v) + c(v ,w ) ; after all these have done, set status (v) to “finished”
and continue. See Table 1 for the pseudo-code.
Table 1: Pseudo-code of the Dijkstra algorithm for the static shortest path problem with time windows:
______________________________________________________________________________________
1. status (s) := “labeled”, g ( s) := 0, status (v) := “unlabeled” for all v ≠ s
2. Let v be a “labeled” node with satisfies the time window condition and the smallest g (v) (the active node). IF
v = d GOTO 11
3. FOR all edges (v, w) ∈ E DO
4. IF status (w) = “unlabeled” THEN
5. status (w) := “labeled” with satisfies the time window condition, then g ( w) := g (v) + c( v, w) , p ( w) := v
6. ELSE IF status (w) = “labeled” AND g ( w) f g ( v ) + c( v , w ) THEN

7. g ( w) := g (v) + c( v, w) , p ( w) := v
8. END IF
9. DONE
10. status (v) := “finished”. GOTO 2
11. OUTPUT g (d ) and the s, d − path found (i.e. the reverse of d , p(d ), p( p (d )),..., s).
_____________________________________________________________________________________
A new algorithm of the A* algorithm given in (Table 2) follows the same fashion except that it employs an estimator
h(v) for all v with the time windows [av , bv ] and chooses the active node by the smallest g (v) + h(v). Notice that
how to determine h(v) is not part of algorithm. It must be obtained by some other method, and the choice of h
*
determines the correctness and the efficiency of the new algorithm of the A algorithm (a good lower-bound on the v, d -
distance is preferred). Clearly the Dijkstra algorithm is a special case with h = 0 .
Table 2: Pseudo-code of the new algorithm of the A* algorithm for the static problem:
We note that the Dijkstra algorithm is a special case of h = 0. For general h, however, the correctness is not
guaranteed.
________________________________________________________________________________________
Table 1
2. Let v be a “labeled” with time window [av , bv ] and the smallest g (v) + h(v). IF v = d GOTO 11
Table 1
________________________________________________________________________________________
Now we are ready to describe our generalized A* algorithm. It generalizes h(v) by time dependent version h(v, t ) ,
where av ≤ t ≤ bv is the time at node v . Thus in table 3, we use h(v, g (v)) to replace h(v) . Notice the rule for
choosing the active nod (Line 2) has been changed in addition.
Table 3: Pseudo-code of a new algorithm of A* algorithm for the time-dependent shortest path problem time windows.
_______________________________________________________________________________________
1. status (s) := “labeled”, g ( s) := t0 , status (v) := “unlabeled” for all v ≠ s
2. Let v be a “labeled” node with av ≤ t ≤ bv , t is the time at node v and the smallest
g (v) + h(v, g (v)). In the case that there are multiple candidates, choose one with the smallest g (v).
IF v = d GOTO 11
3. FOR all edges (v, w) ∈ E DO
4. IF status (w) is “unlabeled” THEN
5. status (w) := “labeled”, g ( w) := g (v) + c( v, w) ( g (v)), p( w) := v
6. ELSE IF status (w) is “labeled” AND g ( w) f g ( v) + c( v, w) ( g (v )) THEN
7. g ( w) := g ( v) + c(v ,w ) ( g ( v)) , p( w) := v
8. END IF
9. DONE
10. status (v) := “finished”. GOTO 2
11. OUTPUT g (d ) and the s, d -path found (i.e. the reverse of d , p (d ), p ( p(d )),..., s ).
______________________________________________________________________________________
In general, the above algorithm may fail to find an optimal solution. The next theorem gives two sufficient (but
reasonable) conditions for the correctness.
Definition 2.1 Given a directed graph G = (V , E ), a nonnegative transit-time function ce (t ) of each edge
e = (v, w) ∈ E , where av ≤ t ≤ bv , is the time to leave v , then for all edges e = (v, w) and
t , h(v, t ) ≤ ce (t ) + h( w, t + ce (t )) is called a triangle condition.
Theorem 2.1 Given a directed graph G = (V , E ), a nonnegative transit-time function ce (t ) of each edge
e = (v, w) ∈ E , where av ≤ t ≤ bv , is the time to leave v , a source node s ∈V , a destination node d ∈ V and a
departure time t 0 at the source s of the time-dependent shortest path problem time windows such that satisfies the
FIFO properly and d is reachable from s , the new generalized A* algorithm in Table 3 finds an optimal solution if h
satisfies the three conditions:
• For all vertex v ∈ V and t1 ≤ t2 , t1 + h(v, t1 ) ≤ t2 + h(v, t 2 ) is FIFO condition.
• For all edges e = (v, w) ∈ E and t , h(v, t ) ≤ ce (t ) + h( w, t + ce (t )) is a triangle condition.
• For all vertex v ∈ V and t1 ≤ t2 , [ av , bv ] ≤ [av , bv ] is a time windows condition.
t1 t1 t2 t2

We note that the triangle condition (Figure 2) is a natural generalization from the classical A* algorithm whereas the
FIFO condition is only available in the time-dependent and time window case. The generalized Dijkstra algorithm is
nothing but the simplest case with h = 0, and the generalization of (Kanoulas et al., 2006), on the other hand, simply
uses a constant function h(v, t ) = h(v), with av ≤ t ≤ bv , t ∈ T , thus it also a simple special-case of our algorithm.

time windows: [aw , bw ] w h( w, t + ce (t ))

ce (t ) time windows: [ad , bd ]

time windows: [av , bv ] v h(v, t )

time: t

Figure 2: The triangle condition with time windows for function h. Roughly speaking, it says the supposed transit-time
h(v, t ) from v to d is no more than ce (t ) + h( w, t + ce (t ), i.e. the supposed transit-time of the v, d − path
v → w → d . Notice that, h( w, t + ce (t )) is the supposed transit-time from w to d by leaving w at time t + ce (t ) with
time windows [aw , bw ], aw ≤ t ≤ bw .
It is easy to see the next lemma by Triangle condition using the induction on k .
Lemma 2.1 Let p = v1 , v2 ,..., vk be a path and t , av1 ≤ t ≤ bv1 be a departure time at v1 with time windows [ av1 , bv1 ].
i −1
Define σ1 = 0 and σ i = ∑ j =1 c( v ,v ) (t + σ j )
j j −1
be the transit-time from v1 to vi , i = 2,...k. Then it holds that
h(v1 , t ) ≤ σ k + h(vk , t + σ k ) , triangle condition is the case of k = 2 .
Proof for the theorem 1. We show by induction that, every active node v must get the optimal distance label, i.e., the
earliest arrival time at v for leaving s at time t 0 . Obviously this well prove the theorem (notice we have supposed d is
reachable from s ).
Let v be an active node satisfies the time window condition and consider when v is active. If v = s, we are done.
Otherwise, let p be a simple optimal s, v -path (it exists!) and w be the first node on p such that status (w) ≠
“finished”. Clearly w must exist and w ≠ s (it can be v ) see Figure 3.

s u w v
s ,u : finished nodes w : the first non-finished node v : the active node
Figure 3: Illustration for the proof of Theorem 1 where an optimal s, v -path with time windows is being considered.
g (w) = g * ( w) because w was
Let g* denote the optimal distance (i.e. the earliest arrival time). It is obvious that
relaxed when the precedent node u of w was active and at that time g (u ) = g * (u ) by the induction hypothesis). Let
σ = g * (v) − g * ( w) be the shortest transit-time from w to v at departure time g * ( w) (notice σ ≥ 0 ). By applying
Lemma 2.1 to the w, v -path with time windows on p with t = g * ( w), we have
h( w, g * ( w)) ≤ σ + h(v, g * ( w) + σ ) = g * (v) − g * ( w) + h( g * (v)). (2.1)
That is equivalent to
g * ( w) + h( w, g * ( w)) ≤ g * (v) + h(w, g * (v)). (2.2)
Then, since v is the active node (thus has the smallest g (v) + h(v, g (v)), we have
g (v) + h(v, g (v)) ≤ g (w) + h(w, g ( w)) = g * (w) + h(w, g * (w)) ≤ g * (v) + h(w, g * (v)). (2.3)
On the other hand, by the FIFO condition and g * (v) ≤ g (v) (the optimality of g*), we have
g * (v) + h(v, g * (v)) ≤ g (v) + h(v, g (v)) (2.4)
Therefore we get the next fact by combining (2.3) and (2.4).
g (v) + h(v, g (v)) ≤ g ( w) + h( w, g ( w)) = g * ( w) + h( w, g * ( w)) ≤ g (v) + h(v, g (v)) . (2.5)
This means the equalities hold, hence g (v) + h(v, g (v)) = g (w) + h( w, g (w)). Then by our choice of the active node,
g (v) ≤ g ( w) must hold. Thus g (v) ≤ g * ( w) ≤ g * (v), hence g (v) = g * (v).
• We remark that analogously to the static version, an h with h(d , t ) = 0 implies h(v, t ) where v satisfies the
time windows condition is a lower bound on the shortest transit-time from v to d with departure time g (v) (by
Lemma 1). Moreover, it is not difficult to show that with an h satisfying h(d , t ) ≡ 0 and h ≥ 0, the search
*
space (the set of active nodes) of the generalized A algorithm is no longer than that of the generalized Dijkstra
algorithm. Using this observation, we will give an algorithm in the next section that is practically faster than the
generalized Dijkstra algorithm.

3. Application instance

The time complexity of the generalized Dijkstra algorithm is O( n log+ mα ) by using a Fibonacci heap (we note it was
O(( m + n logn)α ) in (Ding et al., 2008), where m, n , α are the number of edges, the number of nodes, and the time
complexity to calculate ce (t ) , respectively. While we cannot improve this theoretical bound, let us give a practically
faster algorithm that is based on our A* algorithm and generalizes the static landmark-based ALT algorithm (Goldberg
and Harrelsin, 2005).
The ALT algorithm is such as an algorithm that is supposed to answer (unknown) shortest-path queries for a known
graph. This means we can preprocess the graph beforehand and use it to answer a query faster than a normal
calculation by the Dijkstra algorithm. Of course there is a trivial method of saving solutions for all possible queries and
answers a query in O(1) time, but the n2 order for the static case is big (if not impossible) for large graphs, usually a
road network is spares (i.e., m ≤ kn for some small k ) and has several millions of nodes. So researchers are seeking
efficient algorithm that uses O(n) storage, (see Wagnar and Willhalm, 2007) for a review. While this is an extremely
hot topic for the static problem these several years, for the time-dependent case as far as we know, there was no
proposal before our work.
Now let us describe the detail of our generalized ALT algorithm. Let τ * (v, w, t ) denote the shortest transit-time from a
node v to anther node w with satisfies a time windows condition and a departure time t , av ≤ t ≤ bv , hence we want
to find an s, d -path of transit-time τ * (s, d , t0 ). Suppose we have a node z with time windows [a z , bz ] and values
τ * ( z , v, t ) for all nodes v and all t ( z is called a landmark). Also suppose we can calculate a tˆ (if exists) that
tˆ = max{t ′ : t ′ + τ * ( z, v, t ′) ≤ t}. (3.1)
In other words, tˆ is the latest departure time in order to get v before t (from z ). Define h by:
hz (v, t ) = max{τ * ( z , d , tˆ) − τ * ( z , v, tˆ),0} if tˆ exists, 0 otherwise (i.e., tˆ does not exist). (3.2)
It is clear that hz (d , t ) = 0 and hz ≥ 0. Actually this definition is a generalization from the static case, i.e., hz is an
estimation a lower bound on the v, d transit-time, which is no shorter than the right side of (3.2) (by the triangle
inequality due to the optimality of τ * ). Moreover, we can show that hz satisfies the FIFO condition, the Triangle
condition and the Time windows condition at the same time, too. The proof is not trivial nor difficult, but due to the page
limit, we omit it in this paper. We not it is important to choose tˆ to be the maximum.
We still have to show how to calculate tˆ , which usually is difficult if there is no explicit expression for τ * ( z, v, t ).
Moreover, in general it is difficult to hold all values of τ * ( z, v, t ). Fortunately, however, we can show that sampling of
time works, i.e., we can calculate and hold values τ * ( z , v, ti ) only for some t1 p t2 p ... p tk and define tˆ, if it exists,
by
tˆ = max{ti : ti + τ * ( z , v, ti ) ≤ t} (3.3)
Again, we can show the function hz defined by (3) with the above tˆ satisfies the FIFO condition, the Triangle condition,
= 0, hz ≥ 0. Moreover, we can employ more than one landmarks to get a
the Time windows condition and hz (d , t )
better estimation (notice the maximum of all hz s works). Applying this generalized ALT algorithm to a number of US
road networks (obtained from the web site of the 9th DIMACS implementation challenge
http://www.dis.uniromal.it/~challenge9/, where 320,000 ≤ n ≤ 1,210,000 and m ≤ 3n) with periodic piecewise-linear
transit-time functions with 9 samples a day.
The comparison of the search space between the generalized Dijkstra algorithm and our ALT algorithm for an instance
with 321, 270 nodes and 800, 172 edges. The number of landmarks is 16 and the number of time samplings is 2. The
search space of ALT algorithm is 1/18.2 smaller and the running time is 7.4 times faster.

4. Conclusion
In this paper, we have given a generalized framework of A* algorithm for the time-dependent shortest path problem with
time windows. By constructing some appropriate estimator h, it is possible to get an algorithm that is faster than a
normal generalized Dijkstra algorithm. As an example, we have generalized the landmark–based ALT algorithm, which
we believe is the first algorithm that uses preprocessing to speed up the calculation of time dependent shortest paths
time windows. Our experimental result shows it is several times faster than a normal generalized Dijekstra algorithm,
which requires no preprocessing for large road networks.

Acknowledgment
The author would like to thank the anonymous referees for their comments which improved the original manuscript.

REFERENCES

1. Ahuja R., Magnanti T., Orlin J. (1993). Network flows: theory, algorithms, and applications, Prentice-Hall.
2. Cook K., Halsey E. (1966). The shortest route through a network with time-dependent Intermodal transit, J. Math.
Anal. Appl., 14: 493-498.
3. Denan B. (1999). Continuous-time dynamic shortest path algorithms, Master’s thesis, MIT.
 4. Dreyfus S. (1969). An appraisal of some shortest-path algorithms, Operations Research, 17(3): 395-412.
5. Ding B., Xu J., Qin L. (2008). Finding time-dependent shortest paths over large graphs, Proc. EDBT 08, ACM Intl.
Conf. Proc. 261: 205-216.
6. El-Sherbeny, N. (2001). Resolution of a vehicle routing problem with multiobjective simulated annealing method.
Ph.D. Dissertation of Faculty of Science, Mons University, Mons, Belgique.
7. El-Sherbeny N. and Tuyttens D. (2001). Optimization multicriteria of routing problem, Troisieme Journee de
Travail sur la Programming Mathematique Multi-Objective, Faculte Polytechnique de Mons, Mons, Belgique.
8. Goldberg A., Harrelson C. (2005). Computing the shortest path: A* search meets graph theory, Proc. SODA, 156-
165.
9. Halpern H. (1977). Shortest route with time dependent length of edges and limited delay possibilities in nodes,
Operations Research, 21: 117-124.
10. Hart P., Nilsson N., RAPHAEL B. (1968). A formal basis for the heuristic determination of minimum cost paths,
IEEE Transactions Systems Science and Cybernetics, 4(2):100-107.
11. Kanoulse E., Du Y., Xia T., ZHANG D. (2006). Finding fastest paths on a road network with speed patterns, Proc.
ICDE ’06: 10-19.
12. KAUFMAN D., Smith R. (1993). Fastest paths in time-dependent networks for intelligent vehicle-highway
systems application, J. Intelligent Transportation Systems, 1 (1): 1-11.
13. Orda A., R. Rom R. (1990). Shortest-path and minimum-delay algorithms in networks with time-dependent edge-
length, J. ACM , 37 (3): 607-625.
14. Sherali H., Ozbay K., S. Subramanian S. (1998). The time-dependent shortest pair of disjoint paths problem:
complexity, models, and algorithms, Networks, 31(4): 259-272.
15. Tuyttens D., Teghem J., El-Sherbeny N. (2004). A particular multiobjective vehicle routing problem solved by
simulated annealing, Lecture Notes in Economics and Mathematical Systems, Springer-Verlag, Berlin, Germany,
535: 133-152,
16. Wagner D., Willhalm T. (2007). Speed-up techniques for shortest-path computations, Proc. STACS LNCS,
4393: 23-36.
Cite this article as:
El-Sherbeny (2013). A new algorithm of A* algorithm for the
.
time-dependent shortest path problem with time windows Sci.
Res. Impact. 2(1): 1-8.

Submit your manuscript at:

http://www.scienceparkjournals.org/SRI/submit

View publication stats