Abstract
Most heuristic methods for VRP and its variants are based on the partial exploration of large neighborhoods, typically by means of single, simple moves applied to the current solution. In this paper we define an extended concept of independent moves and show how even a very standard heuristic method can significantly improve when considering the simultaneous application of carefully chosen sets of moves. We show in particular that, when choosing a set such that the total cost variation is equal to the sum of the variations induced by each single move, the quality of solutions obtained is in general very high. When compared with numerical results obtained by some of the best available heuristics on challenging, large scale, problems, our simple algorithm equipped with the application of optimally chosen independent moves displayed very good quality.
This is a preview of subscription content, log in to check access.
References
Boschetti M, Maniezzo V (2015) A set covering based matheuristic for a realworld city logistics problem. Int Trans Oper Res 22:169–196
Bosco A, Laganà D, Musmanno R, Vocaturo F (2014) A matheuristic algorithm for the mixed capacitated general routing problem. Networks 64(4):262–281
Bräysy O, Gendreau M (2005) Vehicle routing problem with time windows, part i: route construction and local search algorithms. Trans Sci 39(1):104–118
Congram RK, Potts CN, van de Velde SL (2002) An iterated dynasearch algorithm for the singlemachine total weighted tardiness scheduling problem. INFORMS J Comput 14(1):52–67
Corman F, Voß S, Negenborn RR (eds) (2015) An ant colonybased matheuristic approach for solving a class of vehicle routing problems. Springer International Publishing, Cham
Dayarian I, Crainic TG, Gendreau M, Rei W (2016) An adaptive large neighborhood search heuristic for a multiperiod vehicle routing problem. Transp Res Part E: Logist Transp Rev 95:95–123
De Franceschi R, Fischetti M, Toth P (2006) A new ILPbased refinement heuristic for vehicle routing problems. Math Program 105(2–3):471–499
Ergun Ö, Orlin JB, SteeleFeldman A (2006) Creating very large scale neighborhoods out of smaller ones by compounding moves. J Heuristics 12(1):115–140
Foster BA, Ryan DM (1976) An integer programming approach to the vehicle scheduling problem. J Oper Res Soc 27(2):367–384
Gurobi Optimization Inc (2016) Gurobi optimizer reference manual. http://www.gurobi.com. Accessed 24 Mar 2017
Kelly JP, Xu J (1999) A setpartitioningbased heuristic for the vehicle routing problem. INFORMS J Comput 11(2):161–172
Koç Ç, Bektaş T, Jabali O, Laporte G (2015) A hybrid evolutionary algorithm for heterogeneous fleet vehicle routing problems with time windows. Comput Oper Res 64:11–27
Mancini S (2016) A reallife multi depot multi period vehicle routing problem with a heterogeneous fleet: Formulation and adaptive large neighborhood search based matheuristic. Transportation Research Part C: Emerging Technologies, pp. 100–112
Nemhauser GL, Wolsey LA (1988) Integer programming and combinatorial optimization. Wiley, New York
Pillac V, Guéret C, Medaglia AL (2013) A parallel matheuristic for the technician routing and scheduling problem. Optim Lett 7(7):1525–1535
Potts CN, van de Velde SL (1995) DynasearchIterative local improvement by dynamic programming. Part I. The traveling salesman problem. Tech. rep., University of Twente
Riise A, Burke EK (2014) On parallel local search for permutations. J Oper Res Soc 66(5):822–831
Rochat Y, Taillard ÉD (1995) Probabilistic diversification and intensification in local search for vehicle routing. J Heuristics 1(1):147–167
Rousseau LM, Gendreau M, Pesant G (2002) Using constraintbased operators to solve the vehicle routing problem with time windows. J Heuristics 8(1):43–58
Schmid V, Doerner KF, Hartl RF, Savelsbergh MW, Stoecher W (2009) A hybrid solution approach for readymixed concrete delivery. Transp Sci 43(1):70–85
Subramanian A, Uchoa E, Ochi LS (2013) A hybrid algorithm for a class of vehicle routing problems. Comput Oper Res 40(10):2519–2531
Toth P, Vigo D (2014) Vehicle routing: problems, methods, and applications, second, edition edn. SIAM/MOS, Philadelphia
Uchoa E, Pecin D, Pessoa A, Poggi M, Vidal T, Subramanian A (2017) New benchmark instances for the capacitated vehicle routing problem. Eur J Oper Res 257(3):845–858
Vidal T, Crainic TG, Gendreau M, Prins C (2014) A unified solution framework for multiattribute vehicle routing problems. Eur J Oper Res 234(3):658–673
Acknowledgements
We are grateful to both reviewers and the associate editor for their stimulating comments on the first version of this paper: answering those comments helped us to significantly improve the quality of this paper.
Author information
Affiliations
Corresponding author
Additional information
Disclaimer
Some of the authors are affiliated with Fleetmatics, a Verizon company, which is a leading global provider of mobile workforce solutions for servicebased businesses. The views set forth in this article do not necessarily represent the views of Fleetmatics, Verizon or any of their respective affiliates.
Appendix
Appendix
In this section we will provide the conditions required in order to safely combine sets of legal moves. First we show that, under suitable conditions, each of the moves considered in TSIS is legal.
Proposition 2
Given a wellformed solution s, a relocate(v,{y,z}) move m that moves the order v into the edge \(\{y,z\}\in E_s\), with \(v\ne y\), \(v \ne z\), is legal.
Proof
By definition of the relocate operator, \(R_m = \{ \{u,v\}, \{v,w\}, \{y,z\}\}\) where u and w are the two nodes adjacent to v, and \(I_m = \{\{y,v\}, \{v,z\}, \{u,w\}\}\). It is easy to verify that \(R_m \subseteq E_s\), and \(I_m \cap (E_s {\setminus } R_m) = \emptyset\), even in the case where \(u = y\) and/or \(w=z\), otherwise s could not be a wellformed solution. It is also trivial to verify that the move preserves the degree of the involved nodes.
Finally, suppose by contradiction that the move creates subtours when applied on a wellformed solution s. Let T be the set edges comprising such subtour. Note that, if \(\{y,v\} \in T\), then also \(\{v,z\} \in T\), otherwise the degree of v would not be preserved. Then three cases can happen

\(T \cap I_m = \emptyset\). Then also \(T \subseteq E_s\), so s is not a wellformed solution.

\(\{u, w\} \in T\). Then consider the set \(T' = (T {\setminus } \{u, w\}) \cup \{\{u,v\} \cup \{v,w\}\). Since this operation replaces the arc \(\{u, w\}\) with the pair \(\{u,v\}\) and \(\{v,w\}\), also \(T'\) contains a tour. But we can see that by construction \(T' \subseteq E_s\), so s is not a wellformed solution.

\(\{y,v\} \in T\) and \(\{v,z\} \in T\). Then consider \(T' = (T {\setminus } \{ y, v\} {\setminus } \{v,z\}) \cup \{y,z\}\). Following the same reasoning as the previous case, we can see that also \(T'\) contains a tour, and that \(T' \subseteq E_s\), so again s is not a wellformed solution.
Proposition 3
Given a wellformed solution s, an exchange(v,y) move that swaps two orders v and y is legal.
Proof
By definition of the exchange operator, \(R_m = \{\{u,v\}, \{v,w\}, \{x,y\}, \{y,z\}\}\) where u, w and x, z are the nodes adjacent to v and y, respectively, and \(I_m = \{\{u,y\}, \{y,w\}, \{x,v\}, \{v,z\}\}\). It is easy to verify that \(R_m \subseteq E_s\), and \(I_m \cap (E_s {\setminus } R_m) = \emptyset\), even in the case where \(w = x\) and/or \(u=z\), otherwise s could not be a wellformed solution.
The move preserves the degree of the involved nodes, and does not create subtours. The proof is trivial and follows the exact same structure as the one for the relocate operator. \(\square\)
Proposition 4
Given a wellformed solution s, a relocatepair({v,w},{y,z}) move m that relocates the edge \(\{v,w\}\in E_s\) into the edge \(\{y,z\}\in E_s\) is legal.
Proof
Let us denote by \(u \ne w\) the other node adjacent to v in \(E_s\) (i.e., \(\{u,v\} \in E_s\)) and analogously \(x \ne u\) is adjacent to w. Then this move is defined through
and the proof proceeds similarly to the previous one. \(\square\)
Notice that in the definition of this move we chose to insert edge \(\{v,w\}\) both in \(R_m\) and in \(I_m\). This will prove useful in order to to ensure that the edge \(\{v,w\}\) remains in \(E_s\) when combining this move with other ones, as we will see later.
Proposition 5
Given a wellformed solution s, an exchangepairs({v,w},{y,z}) move that swaps two edges \(\{v,w\}\in E_s\) and \(\{y,z\}\in E_s\) is legal.
The proof is omitted, as trivial and similar to the previous ones. As before, we assume that both edges \(\{v,w\}\) and \(\{y,z\}\) appear both in \(R_m\) and in \(I_m\) in order to ensure that they are not removed by other moves, when we will combine them.
Theorem 1
A set M of legal moves of the type relocate, exchange, relocatepair, exchangepairs with no edge overlap over a wellformed solution s is independent in the sense of Definition 3.
Proof
We need to show that the solution obtained after applying all the moves in M does not contain subtours. To do so, consider any sequence of moves in M, in any order. The first move \(m_1\) can be applied to s, and \(m_1(s)\) is still wellformed, by hypothesis and hence does not contain subtours.
Consider the nth move in the sequence and consider the \(n1\) moves applied before. Let \(s_{n1}\) denote the solution obtained after the application of these moves, and assume that it does not contain subtours. The nth move can be applied since the edges affected by \(m_n\) are, by hypothesis, nonoverlapping with all the other ones, so the edges in \(R_{m_n}\) belong to \(s_{n1}\) and \(I_{m_n}\) do not. Then by Propositions 2–5 move \(m_n\) is legal, \(s_n\) is wellformed, so it does not contain subtours. The claim follows by induction. \(\square\)
Extension to 2opt moves
In the previous subsection we have proved that several classes of moves can be safely combined. More specifically, it is always safe to combine nonoverlapping relocate, exchange, relocatepair, and exchangepairs moves, due to the fact that the only requirements for them to be legal is that the edges in \(R_m\) are in the solution they are applied to (and those in \(I_m\) are not). More complex moves require some additional preconditions in order to avoid subtours. For 2opt moves, these conditions depend on the relative order of the nodes in the tour.
Let us define a path p as an ordered sequence of nodes which are pairwise adjacent in the solution s. A path is simple if no node appears more than once in it. For any two nodes u, v in the same route, the path p induces a partial ordering \(\prec _p\) such that \(u\prec _p v\) if u precedes v in the path p. Observe that u and v are not required to be adjacent.
Proposition 6
Given a wellformed solution s, a 2opt({u,v},{y,z}) move m over s, defined as \(R_m = \{\{u,v\},\{y,z\}\} \subset E_s\) and \(I_m = \{\{u,y\},\{v,z\}\} \not \subset E_s\), with u, v, y, z belonging to the same tour (route), is legal if and only if there is a simple path p where \(u \prec _p v \prec _p y \prec _p z\).
Proof
If a simple path p with \(u \prec _p v \prec _p y \prec _p z\) exists, then cutting the edges \(R_m=\{\{u,v\},\{y,z\}\}\) in the tour creates two disconnected components: a path \(P_1\) that contains u and z at the two ends, and a path \(P_2\) with v and y at the two ends. The nodes u, v, y, z have degree 1 after the cut. Inserting the edges \(I_m = \{\{u,y\},\{v,z\}\}\) fulfills the degree condition, and it reconnects the two components, creating a single tour. Thus the move is legal.
Let us now assume that there is no simple path p over the considered route with \(u \prec _p v \prec _p y \prec _p z\). Since u must be adjacent to v, and y to z, it is easy to verify that there must be a path q such that \(v \prec _q u \prec _q y \prec _q z\). If such path exists, removing the edges \(R_m=\{\{u,v\},\{y,z\}\}\) creates two disconnected components: a path \(p_1\) with endpoints v and z, and a path \(p_2\) with endpoints u and y. Inserting the edges \(I_m = \{\{u,y\},\{v,z\}\}\) fulfills the degree condition, but it does not reconnect the components, creating two disconnected subtours. Then the precedence must hold, and the claim follows. \(\square\)
When combining 2opt moves, then, we must pay particular care to maintain the necessary and sufficient condition in Proposition 6. To show how this can be achieved, let us start with the definition of nested 2opt moves (see Fig. 3).
Definition 4
Given two 2opt moves \(m_i = \texttt {2opt}(\{u_i,v_i\},\{y_i,z_i\})\) and \(m_j = \texttt {2opt}(\{u_j,v_j\},\{y_j,z_j\})\) over the same route of a solution s, let \(\bar{p}_i\) be the set of nodes in the simple path that connects \(v_i\) to \(y_i\) not passing through the depot, and let \(\bar{p}_j\) be the set of nodes in the path that connects \(v_j\) to \(y_i\) not passing through the depot. Three cases can occur:

if \(\bar{p}_i \subset \bar{p}_j\), we say that \(m_i\) is nested into \(m_j\);

if \(\bar{p}_i \cap \bar{p}_j = \emptyset\), we say that \(m_i\) and \(m_j\) are disjoint;

if \(\bar{p}_i \cap \bar{p}_j \ne \emptyset\), but neither is nested in the other, we say that \(m_i\) and \(m_j\) are intertwined.
Two 2opt moves are intertwined if neither is nested into the other. Applying simultaneously two intertwined 2opt moves does not guarantee the absence of subtours, as shown in Fig. 4.
Definition 5
Let \(\bar{E} \subseteq E_s\) be a set of edges belonging to the same route. Consider any partial ordering on the nodes of the edges in \(\bar{E}\) and a simple directed path following that order that connects all the nodes in the route.
We say here that a move preserves the relative order of \(\bar{E}\) if, after its application to s, the edges in \(\bar{E}\) still belong to the same tour, and it is still possible to find a path over such tour so that the chosen partial ordering of the nodes of the edges in \(\bar{E}\) is preserved.
Note that no partial ordering can be defined on edges that belong to different tours.
Two preliminary results are needed before we can finally prove how 2opt moves can be safely combined with the others.
Lemma 1
Given a wellformed solution s, a move m of the type relocate, exchange, relocatepair, exchangepairs, preserves the relative order of any pair of edges \(\{u,v\}\), \(\{x,y\}\) not affected by the move.
Proof
Consider a legal relocate move m that moves the order b with adjacent nodes a, c into the edge \(\{f,g\}\). Assume that a simple path p induces the ordering \(u \prec _p v \prec _p x \prec _p y\). We can distinguish two cases:

p does not include the edges in \(R_m\). Then the same path also exists in m(s), so the order is trivially preserved.

p includes \(\{a,b,c\}\). Assume that p is defined as \(p = (u, v, \dots , a, b, c, \dots , x, y)\). Then consider in m(s) the path \(q = (u, v, \dots , a, c, \dots , x, y)\), where the edge \(\{a,c\}\) has been inserted by applying m. Over q, \(u \prec _q v \prec _q x \prec _q y\).

p includes \(\{f,g\}\). Assume that p is defined as \(p = (u, v, \dots , f, g, \dots , x, y)\). Then consider in m(s) the path \(q = (u, v, \dots , f, b, g, \dots , x, y)\), where the edges \(\{f,b\}\) and \(\{b,g\}\) have been inserted by applying m. Over q, \(u \prec _q v \prec _q x \prec _q y\).
Analogous arguments can be applied to show the claim holds also for the other types of moves. \(\square\)
Lemma 2
Given a wellformed solution s, a legal 2opt move m preserves the relative order of any two edges \(\{u,v\}\), \(\{x,y\}\)
if they belong to the same of the two connected components in \(E_s{\setminus } R_m\).
Proof
Assume that a simple path p induces the ordering \(u \prec _p v \prec _p x \prec _p y\). Taking into account the move m, we can distinguish two cases:

p does not include the edges in \(R_m\). Then the same path also exists in m(s), so the order is trivially preserved.

p include both edges in \(R_m = \{\{a,b\},\{c,d\}\}\). Assume w.l.o.g. that p is defined as \(p = (u, v, \dots , a, b, \dots , c,d, \dots , x, y)\). Then consider in m(s) the path \(q = (u, v, \dots , a, c, \dots , b, d, \dots , x, y)\), where the nodes between a and d are reversed due to the application of the move. Over q, \(u \prec _q v \prec _q x \prec _q y\).
The case where p includes only one of the edges in \(R_m\) is excluded by the hypothesis that \(\{u,v\}\), \(\{x,y\}\) lie in the same connected component of \(E_s{\setminus } R_m\). Then the claim follows. \(\square\)
We can now show that, if we restrict ourselves to nested moves, as in Fig. 3, the following result holds:
Theorem 2
Given a set M of legal moves of the type relocate, exchange, relocatepair, exchangepairs, 2opt with no edge overlap over a wellformed solution s, if all pairs \(m_i,m_j\) of 2opt moves are nonintertwined, then the set is independent.
Proof
We need to show that the solution obtained after applying all the moves in M does not contain subtours. To do so, consider any sequence of moves in M, in any order. The first move of the sequence \(m_1\) can be applied to s, and \(m_1(s)\) is still wellformed, by hypothesis, thus no subtour exists. Moreover, by Lemmas 1–2 the relative order of any two edges affected by any other 2opt move \(m_k\) with \(k > 1\) is preserved (since all 2opt are pairwise nested in s), so all pairwise nonintertwined 2opt moves are still so.
Let \(m_n\) be the nth move in the sequence, and
let \(s_{n1}\) denote the solution obtained after the application of the first \(n1\) moves. By inductive assumption, \(s_{n1}\) does not contain subtours, the relative order of any two edges affected by any 2opt move \(m_k\) with \(k > n1\) is preserved, and all pairs of 2opt moves that were nonintertwined on s, are still so in \(s_{n1}\). Then:

If the nth move is not a 2opt, it can be legally applied since the edges affected by \(m_n\) are, by hypothesis, nonoverlapping with all the other moves in M, so the edges in \(R_{m_n}\) still belong to \(s_{n1}\), and those in \(I_{m_n}\) do not. The solution \(s_n\) does not contain subtours. By applying Lemma 1, the order of any two edges not affected by \(m_n\) is preserved and all pairwise nonintertwined 2opt moves are still so.

If on the contrary the nth move is a 2opt, it can be applied without introducing subtours, since the relative order of the edges in \(R_{m_n}\) is preserved in \(s_{n1}\) by the inductive assumption. We must now show that the order is preserved also in \(s_n\). Consider any 2opt move \(m_k\) with \(k > n+1\). By the inductive assumption, \(m_n\) and \(m_k\) are still nonintertwined in \(s_{n1}\) so the move \(m_n\) preserves the order of the two edges in \(R_{m_k}\) by Lemma 2. By applying the same lemma, all nonintertwined moves are still so in \(s_n\).
The claim follows by induction. \(\square\)
Extension to 2opt \(^*\) moves
The classes of moves that can be combined can be further extended to include the 2opt* operator.
Proposition 7
Given a wellformed solution s, a 2opt* move m over s, defined as \(R_m = \{\{u,v\},\{y,z\}\} \subset E_s\) and \(I_m = \{\{u,y\},\{v,z\}\} \not \subset E_s\), with \(\{u,v\}\) and \(\{y,z\}\) belonging to two different tours (routes), is legal.
Proof
Let us consider only the subset of \(E_s\) that contains the two tours with the nodes u, v and y, z, respectively. Cutting the edges \(R_m=\{\{u,v\},\{y,z\}\}\) creates a tree with the depot as the root node, and four linear branches with the nodes u, v, y, z as leaves. Inserting back in the solution the edges \(I_m = \{\{u,y\},\{v,z\}\}\) reestablishes the degree condition and reconnects the dangling branches, thus obtaining two tours, connected by the depot, which contain the nodes u, y and v, z, respectively. The solution is wellformed. \(\square\)
The additional assumption here is that the edges in \(R_m\) must be in two different routes. To safely combine them, then, it is sufficient to ensure that no route is affected by more than one 2opt*. Concerning the interaction with 2opt moves, similar steps to what is shown in the previous subsection can be followed.
Definition 6
Given a 2opt move \(m_i = \texttt {2opt}(\{u_i,v_i\},\{y_i,z_i\})\) and a 2opt \(^*\) move \(m_j = \texttt {2opt}^*(\{u_j,v_j\},\{y_j,z_j\})\) with \(\{u_i,v_i\},\{y_i,z_i\},\{u_j,v_j\}\) belonging to the same route of a solution s, and \(\{y_j,z_j\})\) belonging to a different one, let \(\bar{p}_i\) be the set of nodes in the simple path that connects \(v_i\) to \(y_i\) not passing through the depot. Two cases can occur:

if \(\{u_j,v_j\} \subset \bar{p}_i\), we say that \(m_j\) is nested into \(m_i\);

if \(\{u_j,v_j\} \not \subset \bar{p}_i\), we say that \(m_i\) and \(m_j\) are disjoint.
Lemma 3
Given a wellformed solution s, a legal 2opt* move m preserves the relative order of any pair of edges \(\{u,v\}\), \(\{x,y\}\) in the same tour in s, if there is a tour in m(s) that still contains them both.
Proof
Consider the path \(p = (u, v, \dots , x, y)\) that induces the ordering \(u \prec _p v \prec _p x \prec _p y\). Let m be a legal 2opt* move. The set \(R_m\) consists of two edges, belonging to different routes. If \(\{u,v\}\), \(\{x,y\}\) still belong to the same tour in m(s), then p does not include the edge in \(R_m\) – otherwise they would be in two different routes. This means that p also exists in m(s), and the order is preserved. \(\square\)
Theorem 3
Given a set M of nonoverlapping legal moves over a wellformed solution s, if:

all pairs \(m_i, m_j \in M\) of 2opt moves are nonintertwined (either one is nested in the other, or they are disjoint).

for each route in s, there can be at most one 2opt* affecting any edge of that route, and it must be disjoint from all the 2opt moves over that route.
then the set is independent.
Proof
We need to show that the solution obtained after applying all the moves in M does not contain subtours.
An induction proof that follows the same idea used in Theorem 2 can be used. For sake of readability, we will omit the details. To prove the thesis, it is sufficient to guarantee, applying Lemmas 1, 2 and 3, that at each step of the induction:

the relative order of any two edges affected by any 2opt is preserved

all pairwise nonintertwined 2opt moves are still so

for any 2opt* move \(m\in M\), the edges \(R_m\) are still in different routes

all pairs of 2opt* and 2opt moves are still disjoint.\(\square\)
Rights and permissions
About this article
Cite this article
Bianconcini, T., Di Lorenzo, D., Lori, A. et al. Exploiting sets of independent moves in VRP. EURO J Transp Logist 7, 93–120 (2018). https://doi.org/10.1007/s136760170110y
Received:
Accepted:
Published:
Issue Date:
Keywords
 VRP
 Tabu search
 Matheuristic
 Independent moves