A Recourse Approach for the Capacitated Vehicle Routing Problem with Evidential Demands

Size: px

Start display at page:

Download "A Recourse Approach for the Capacitated Vehicle Routing Problem with Evidential Demands"

Georgia Edwards
5 years ago
Views:

Daniel Porumbel 2, David Mercier 1 and Éric Lefèvre1 1

Conservatoire National des Arts et Métiers, EA 4629 CEDRIC,

1 A Recourse Approach for the Capacitated Vehicle Routing Problem with Evidential Demands Nathalie Helal 1, Frédéric Pichon 1, Daniel Porumbel 2, David Mercier 1 and Éric Lefèvre1 1 Université d Artois, LGI2A, F Béthune, France 2 Conservatoire National des Arts et Métiers, EA 4629 CEDRIC, Paris, France Helal et al. A Recourse Approach for the CVRP with Evidential Demands 1 / 24

2 Background - CVRP / CVRPSD The Capacitated Vehicle Routing Problem (CVRP) Finding the least cost routes to serve customers deterministic demands while respecting problem constraints, in particular vehicles capacity constraints. The CVRP with Stochastic Demands (CVRPSD) Customers have stochastic demands. Helal et al. A Recourse Approach for the CVRP with Evidential Demands 2 / 24

3 The CVRPSD may be addressed by two main approaches: Chance Constrained Programming (CCP). Stochastic Programming with Recourse (SPR). Alternative uncertainty framework In a previous work [1], the CVRP with Evidential Demands () modelled by a belief function based extension of CCP. In this paper, we model the by a belief function based extension of the SPR and solve it using a metaheuristic algorithm. The first papers that handle discrete NP-hard problem involving uncertainty represented by belief functions. Helal et al. A Recourse Approach for the CVRP with Evidential Demands 3 / 24

4 The CVRPSD may be addressed by two main approaches: Chance Constrained Programming (CCP). Stochastic Programming with Recourse (SPR). Alternative uncertainty framework In a previous work [1], the CVRP with Evidential Demands () modelled by a belief function based extension of CCP. In this paper, we model the by a belief function based extension of the SPR and solve it using a metaheuristic algorithm. The first papers that handle discrete NP-hard problem involving uncertainty represented by belief functions. Helal et al. A Recourse Approach for the CVRP with Evidential Demands 3 / 24

5 Outline 1 CVRP with Stochastic Demands The CVRP The CVRPSD modelled by SPR 2 CVRP with Evidential Demands Belief function theory Formalisation Uncertainty on recourses (failure situations) Interval Demands Experiments 3 Conclusions & perspectives Helal et al. A Recourse Approach for the CVRP with Evidential Demands 4 / 24

6 CVRPSD 1 CVRP with Stochastic Demands The CVRP The CVRPSD modelled by SPR 2 CVRP with Evidential Demands Belief function theory Formalisation Uncertainty on recourses (failure situations) Interval Demands Experiments 3 Conclusions & perspectives Helal et al. A Recourse Approach for the CVRP with Evidential Demands 5 / 24

7 The CVRP CVRPSD The CVRP Given: n = number of customers including the depot, m = number of vehicles, Q = vehicle capacity, d i = (known) demand of client i, c i,j = cost { of travelling from client } i to client j, w i,j,k = 1 if k travels from i to j, 0 otherwise., R k = the route associated to vehicle k. Objective function: min m C(R k ), k=1 where: C(R k ) = n n c i,j w i,j,k, the travel cost of route R k. i=0 j=0 Helal et al. A Recourse Approach for the CVRP with Evidential Demands 6 / 24

8 CVRPSD The CVRPSD modelled by SPR The CVRPSD d i represents the stochastic demand of i (cannot exceed Q). Need to verify the capacity constraints of the CVRPSD for all realizations of d i unrealistic. A SPR approach for the CVRPSD Clients demands are collected until remaining vehicle capacity is not sufficient to pick up entire customer demand failure. If failure recourse (a return trip to the depot). Failure can happen at multiple customers except the first one. Helal et al. A Recourse Approach for the CVRP with Evidential Demands 7 / 24

9 CVRPSD The CVRPSD modelled by SPR The CVRPSD d i represents the stochastic demand of i (cannot exceed Q). Need to verify the capacity constraints of the CVRPSD for all realizations of d i unrealistic. A SPR approach for the CVRPSD Clients demands are collected until remaining vehicle capacity is not sufficient to pick up entire customer demand failure. If failure recourse (a return trip to the depot). Failure can happen at multiple customers except the first one. Helal et al. A Recourse Approach for the CVRP with Evidential Demands 7 / 24

10 CVRPSD The CVRPSD modelled by SPR The CVRPSD d i represents the stochastic demand of i (cannot exceed Q). Need to verify the capacity constraints of the CVRPSD for all realizations of d i unrealistic. A SPR approach for the CVRPSD Clients demands are collected until remaining vehicle capacity is not sufficient to pick up entire customer demand failure. If failure recourse (a return trip to the depot). Failure can happen at multiple customers except the first one. Helal et al. A Recourse Approach for the CVRP with Evidential Demands 7 / 24

11 CVRPSD The CVRPSD modelled by SPR The CVRPSD modelled by SPR The objective function becomes: m min Ce(R k ), k=1 where Ce(R k ) is the expected cost of R k defined by with Ce(R k ) = C(R k ) + Cp(R k ), C(R k ) the travel cost on R k when no recourse action is performed; Cp(R k ) the expected penalty cost on R k induced by failures. Helal et al. A Recourse Approach for the CVRP with Evidential Demands 8 / 24

12 Belief function theory 1 CVRP with Stochastic Demands The CVRP The CVRPSD modelled by SPR 2 CVRP with Evidential Demands Belief function theory Formalisation Uncertainty on recourses (failure situations) Interval Demands Experiments 3 Conclusions & perspectives Helal et al. A Recourse Approach for the CVRP with Evidential Demands 9 / 24

13 Belief function theory Needed concepts A variable x taking values in a finite domain X. A MF m X : 2 X [0, 1] s.t. m X (A) = 1. A X A variable whose true value is known in the form of a MF is called an evidential variable. Given a MF m X and a function h: X R +, then the upper expected value of h relative to m X is : E (h, m X ) = A X m X (A) max x A h(x). Helal et al. A Recourse Approach for the CVRP with Evidential Demands 10 / 24

14 1 CVRP with Stochastic Demands The CVRP The CVRPSD modelled by SPR 2 CVRP with Evidential Demands Belief function theory Formalisation Uncertainty on recourses (failure situations) Interval Demands Experiments 3 Conclusions & perspectives Helal et al. A Recourse Approach for the CVRP with Evidential Demands 11 / 24

15 The d i represents the evidential demand of i (cannot exceed Q). The recourse approach: failure situations Suppose a route R having N customers. r i = { } 1 if failure occurs at the i-th client on R and r1 = 0. 0 otherwise Possible failure situations on R represented by vectors (r 2, r 3,..., r N ) Ω s.t. Ω = {0, 1} N 1. Cost and uncertainty of each failure situation ω Ω Cost of each ω Ω determined by g : Ω R +. The penalty cost upon failure on i is 2c 0,i g(ω) = N r i 2c 0,i. A MF m Ω representing uncertainty on failure situations on R. i=2 Helal et al. A Recourse Approach for the CVRP with Evidential Demands 12 / 24

16 The d i represents the evidential demand of i (cannot exceed Q). The recourse approach: failure situations Suppose a route R having N customers. r i = { } 1 if failure occurs at the i-th client on R and r1 = 0. 0 otherwise Possible failure situations on R represented by vectors (r 2, r 3,..., r N ) Ω s.t. Ω = {0, 1} N 1. Cost and uncertainty of each failure situation ω Ω Cost of each ω Ω determined by g : Ω R +. The penalty cost upon failure on i is 2c 0,i g(ω) = N r i 2c 0,i. A MF m Ω representing uncertainty on failure situations on R. i=2 Helal et al. A Recourse Approach for the CVRP with Evidential Demands 12 / 24

17 The d i represents the evidential demand of i (cannot exceed Q). The recourse approach: failure situations Suppose a route R having N customers. r i = { } 1 if failure occurs at the i-th client on R and r1 = 0. 0 otherwise Possible failure situations on R represented by vectors (r 2, r 3,..., r N ) Ω s.t. Ω = {0, 1} N 1. Cost and uncertainty of each failure situation ω Ω Cost of each ω Ω determined by g : Ω R +. The penalty cost upon failure on i is 2c 0,i g(ω) = N r i 2c 0,i. A MF m Ω representing uncertainty on failure situations on R. i=2 Helal et al. A Recourse Approach for the CVRP with Evidential Demands 12 / 24

18 The d i represents the evidential demand of i (cannot exceed Q). The recourse approach: failure situations Suppose a route R having N customers. r i = { } 1 if failure occurs at the i-th client on R and r1 = 0. 0 otherwise Possible failure situations on R represented by vectors (r 2, r 3,..., r N ) Ω s.t. Ω = {0, 1} N 1. Cost and uncertainty of each failure situation ω Ω Cost of each ω Ω determined by g : Ω R +. The penalty cost upon failure on i is 2c 0,i g(ω) = N r i 2c 0,i. A MF m Ω representing uncertainty on failure situations on R. i=2 Helal et al. A Recourse Approach for the CVRP with Evidential Demands 12 / 24

19 A pessimistic attitude: penalty cost and upper expected cost The upper expected penalty cost of R is Cp (R) = E (g, m Ω ). The Objective of the : min m Ce (R k), with C e (R k) = C(R k ) + C p (R k). k=1 Similarities with robust optimisation. Bayesian evidential demands CVRPSD via SPR. Helal et al. A Recourse Approach for the CVRP with Evidential Demands 13 / 24

20 A pessimistic attitude: penalty cost and upper expected cost The upper expected penalty cost of R is Cp (R) = E (g, m Ω ). The Objective of the : min m Ce (R k), with C e (R k) = C(R k ) + C p (R k). k=1 Similarities with robust optimisation. Bayesian evidential demands CVRPSD via SPR. Helal et al. A Recourse Approach for the CVRP with Evidential Demands 13 / 24

21 1 CVRP with Stochastic Demands The CVRP The CVRPSD modelled by SPR 2 CVRP with Evidential Demands Belief function theory Formalisation Uncertainty on recourses (failure situations) Interval Demands Experiments 3 Conclusions & perspectives Helal et al. A Recourse Approach for the CVRP with Evidential Demands 14 / 24

22 Example: no uncertainty on clients demands A route R with N = 3 clients; θ 1 = 3 and m 1 (θ 1 ) = 1; θ 2 = 3 and m 2 (θ 2 ) = 1; θ 3 = 5 and m 3 (θ 3 ) = 1; Capacity limit Q = 5; q i, i = 1,..., N, the vehicle load after serving i-th client: r 1 = 0 and q 1 = θ 1 = 3. r 2 = 1 since q 1 +θ 2 > Q, and q 2 = q 1 +θ 2 Q = 1. r 3 = 1 since q 2 +θ 3 > Q, and q 3 = q 2 +θ 3 Q = 1. f (θ 1, θ 2, θ 3 ) = ω and ω (r 2 = 1, r 3 = 1) Helal et al. A Recourse Approach for the CVRP with Evidential Demands 15 / 24

23 Example: no uncertainty on clients demands A route R with N = 3 clients; θ 1 = 3 and m 1 (θ 1 ) = 1; θ 2 = 3 and m 2 (θ 2 ) = 1; θ 3 = 5 and m 3 (θ 3 ) = 1; Capacity limit Q = 5; q i, i = 1,..., N, the vehicle load after serving i-th client: r 1 = 0 and q 1 = θ 1 = 3. r 2 = 1 since q 1 +θ 2 > Q, and q 2 = q 1 +θ 2 Q = 1. r 3 = 1 since q 2 +θ 3 > Q, and q 3 = q 2 +θ 3 Q = 1. f (θ 1, θ 2, θ 3 ) = ω and ω (r 2 = 1, r 3 = 1) Helal et al. A Recourse Approach for the CVRP with Evidential Demands 15 / 24

24 Example: no uncertainty on clients demands A route R with N = 3 clients; θ 1 = 3 and m 1 (θ 1 ) = 1; θ 2 = 3 and m 2 (θ 2 ) = 1; θ 3 = 5 and m 3 (θ 3 ) = 1; Capacity limit Q = 5; q i, i = 1,..., N, the vehicle load after serving i-th client: r 1 = 0 and q 1 = θ 1 = 3. r 2 = 1 since q 1 +θ 2 > Q, and q 2 = q 1 +θ 2 Q = 1. r 3 = 1 since q 2 +θ 3 > Q, and q 3 = q 2 +θ 3 Q = 1. f (θ 1, θ 2, θ 3 ) = ω and ω (r 2 = 1, r 3 = 1) Helal et al. A Recourse Approach for the CVRP with Evidential Demands 15 / 24

25 Example: no uncertainty on clients demands A route R with N = 3 clients; θ 1 = 3 and m 1 (θ 1 ) = 1; θ 2 = 3 and m 2 (θ 2 ) = 1; θ 3 = 5 and m 3 (θ 3 ) = 1; Capacity limit Q = 5; q i, i = 1,..., N, the vehicle load after serving i-th client: r 1 = 0 and q 1 = θ 1 = 3. r 2 = 1 since q 1 +θ 2 > Q, and q 2 = q 1 +θ 2 Q = 1. r 3 = 1 since q 2 +θ 3 > Q, and q 3 = q 2 +θ 3 Q = 1. f (θ 1, θ 2, θ 3 ) = ω and ω (r 2 = 1, r 3 = 1) Helal et al. A Recourse Approach for the CVRP with Evidential Demands 15 / 24

26 Precise clients demands f (θ 1,..., θ N ) = (r 2, r 3,..., r N ). Imprecise clients demands MF mi Θ, i = 1,..., N, on R, s.t mi Θ (A i ) = 1, A i Θ. Then failure situation on R belongs to B Ω B = f (A 1,..., A N ) = f (θ 1,..., θ N ). (θ 1,...,θ N ) A 1 A N m Θ i, i = 1,..., N have arbitrary number of focal sets The joint probability that θ i A i Θ, i = 1,..., N is N mi Θ (A i ) m Ω (B) = i=1 f (A 1,...,A N )=B i=1 N mi Θ (A i ). Computing m Ω : evaluating f (A 1,..., A N ) for all combinations of focal sets of mi Θ worst-case complexity O(Q N ) (intractable). Helal et al. A Recourse Approach for the CVRP with Evidential Demands 16 / 24

27 Precise clients demands f (θ 1,..., θ N ) = (r 2, r 3,..., r N ). Imprecise clients demands MF mi Θ, i = 1,..., N, on R, s.t mi Θ (A i ) = 1, A i Θ. Then failure situation on R belongs to B Ω B = f (A 1,..., A N ) = f (θ 1,..., θ N ). (θ 1,...,θ N ) A 1 A N m Θ i, i = 1,..., N have arbitrary number of focal sets The joint probability that θ i A i Θ, i = 1,..., N is N mi Θ (A i ) m Ω (B) = i=1 f (A 1,...,A N )=B i=1 N mi Θ (A i ). Computing m Ω : evaluating f (A 1,..., A N ) for all combinations of focal sets of mi Θ worst-case complexity O(Q N ) (intractable). Helal et al. A Recourse Approach for the CVRP with Evidential Demands 16 / 24

28 Precise clients demands f (θ 1,..., θ N ) = (r 2, r 3,..., r N ). Imprecise clients demands MF mi Θ, i = 1,..., N, on R, s.t mi Θ (A i ) = 1, A i Θ. Then failure situation on R belongs to B Ω B = f (A 1,..., A N ) = f (θ 1,..., θ N ). (θ 1,...,θ N ) A 1 A N m Θ i, i = 1,..., N have arbitrary number of focal sets The joint probability that θ i A i Θ, i = 1,..., N is N mi Θ (A i ) m Ω (B) = i=1 f (A 1,...,A N )=B i=1 N mi Θ (A i ). Computing m Ω : evaluating f (A 1,..., A N ) for all combinations of focal sets of mi Θ worst-case complexity O(Q N ) (intractable). Helal et al. A Recourse Approach for the CVRP with Evidential Demands 16 / 24

29 Precise clients demands f (θ 1,..., θ N ) = (r 2, r 3,..., r N ). Imprecise clients demands MF mi Θ, i = 1,..., N, on R, s.t mi Θ (A i ) = 1, A i Θ. Then failure situation on R belongs to B Ω B = f (A 1,..., A N ) = f (θ 1,..., θ N ). (θ 1,...,θ N ) A 1 A N m Θ i, i = 1,..., N have arbitrary number of focal sets The joint probability that θ i A i Θ, i = 1,..., N is N mi Θ (A i ) m Ω (B) = i=1 f (A 1,...,A N )=B i=1 N mi Θ (A i ). Computing m Ω : evaluating f (A 1,..., A N ) for all combinations of focal sets of mi Θ worst-case complexity O(Q N ) (intractable). Helal et al. A Recourse Approach for the CVRP with Evidential Demands 16 / 24

30 1 CVRP with Stochastic Demands The CVRP The CVRPSD modelled by SPR 2 CVRP with Evidential Demands Belief function theory Formalisation Uncertainty on recourses (failure situations) Interval Demands Experiments 3 Conclusions & perspectives Helal et al. A Recourse Approach for the CVRP with Evidential Demands 17 / 24

31 Binary recourse tree example: route R with N = 3 clients θ 1 4; 8 and m 1 ( 4; 8 ) = 1; θ 2 5; 7 and m 2 ( 5; 7 ) = 1; θ 3 7; 9 and m 3 ( 7; 9 ) = 1; Capacity limit Q = 10; q i, i = 1,..., N, the vehicle load after visiting i-th client: ( 4; 8, 0) 1 st level ( 9; 10, 0) ( 6; 9, 1) ( 1; 5, 1) ( 8; 10, 0) ( 1; 4, 1) 2 nd level 3 rd level Helal et al. A Recourse Approach for the CVRP with Evidential Demands 18 / 24

32 Binary recourse tree example: route R with N = 3 clients θ 1 4; 8 and m 1 ( 4; 8 ) = 1; θ 2 5; 7 and m 2 ( 5; 7 ) = 1; θ 3 7; 9 and m 3 ( 7; 9 ) = 1; Capacity limit Q = 10; q i, i = 1,..., N, the vehicle load after visiting i-th client: 4; 8 no failure and q 1 4; 8. Helal et al. A Recourse Approach for the CVRP with Evidential Demands 18 / 24

33 Binary recourse tree example: route R with N = 3 clients θ 1 4; 8 and m 1 ( 4; 8 ) = 1; θ 2 5; 7 and m 2 ( 5; 7 ) = 1; θ 3 7; 9 and m 3 ( 7; 9 ) = 1; Capacity limit Q = 10; q i, i = 1,..., N, the vehicle load after visiting i-th client: 4; 8 no failure and q 1 4; 8. ( 4; 8, 0) 1 st level Helal et al. A Recourse Approach for the CVRP with Evidential Demands 18 / 24

34 Binary recourse tree example: route R with N = 3 clients θ 1 4; 8 and m 1 ( 4; 8 ) = 1; θ 2 5; 7 and m 2 ( 5; 7 ) = 1; θ 3 7; 9 and m 3 ( 7; 9 ) = 1; Capacity limit Q = 10; q i, i = 1,..., N, the vehicle load after visiting i-th client: { } no failure q2 9;10 4; 8 + 5; 7 = 9; 15. a failure q ;15 10 ( 4; 8, 0) 1 st level Helal et al. A Recourse Approach for the CVRP with Evidential Demands 18 / 24

35 Binary recourse tree example: route R with N = 3 clients θ 1 4; 8 and m 1 ( 4; 8 ) = 1; θ 2 5; 7 and m 2 ( 5; 7 ) = 1; θ 3 7; 9 and m 3 ( 7; 9 ) = 1; Capacity limit Q = 10; q i, i = 1,..., N, the vehicle load after visiting i-th client: { } no failure q2 9;10 4; 8 + 5; 7 = 9; 15. a failure q ;15 10 ( 4; 8, 0) ( 9; 10, 0) ( 1; 5, 1) 1 st level 2 nd level Helal et al. A Recourse Approach for the CVRP with Evidential Demands 18 / 24

36 Binary recourse tree example: route R with N = 3 clients θ 1 4; 8 and m 1 ( 4; 8 ) = 1; θ 2 5; 7 and m 2 ( 5; 7 ) = 1; θ 3 7; 9 and m 3 ( 7; 9 ) = 1; Capacity limit Q = 10; q i, i = 1,..., N, the vehicle load after visiting i-th client: If 9; ; 9 = 16; 19 a failure, q ; 19 10, ( 4; 8, 0) ( 9; 10, 0) ( 1; 5, 1) 1 st level 2 nd level Helal et al. A Recourse Approach for the CVRP with Evidential Demands 18 / 24

37 Binary recourse tree example: route R with N = 3 clients θ 1 4; 8 and m 1 ( 4; 8 ) = 1; θ 2 5; 7 and m 2 ( 5; 7 ) = 1; θ 3 7; 9 and m 3 ( 7; 9 ) = 1; Capacity limit Q = 10; q i, i = 1,..., N, the vehicle load after visiting i-th client: If 9; ; 9 = 16; 19 a failure, q ; 19 10, ( 4; 8, 0) 1 st level ( 9; 10, 0) ( 6; 9, 1) ( 1; 5, 1) 2 nd level 3 rd level Helal et al. A Recourse Approach for the CVRP with Evidential Demands 18 / 24

38 Binary recourse tree example: route R with N = 3 clients θ 1 4; 8 and m 1 ( 4; 8 ) = 1; θ 2 5; 7 and m 2 ( 5; 7 ) = 1; θ 3 7; 9 and m 3 ( 7; 9 ) = 1; Capacity limit Q = 10; q i, i = 1,..., N, the vehicle load after visiting i-th client: { } no failure q3 8;10 else 1; 5 + 7; 9 = 8; 14. a failure q ;14 10 ( 4; 8, 0) 1 st level ( 9; 10, 0) ( 6; 9, 1) ( 1; 5, 1) 2 nd level 3 rd level Helal et al. A Recourse Approach for the CVRP with Evidential Demands 18 / 24

39 Binary recourse tree example: route R with N = 3 clients θ 1 4; 8 and m 1 ( 4; 8 ) = 1; θ 2 5; 7 and m 2 ( 5; 7 ) = 1; θ 3 7; 9 and m 3 ( 7; 9 ) = 1; Capacity limit Q = 10; q i, i = 1,..., N, the vehicle load after visiting i-th client: { } no failure q3 8;10 else 1; 5 + 7; 9 = 8; 14. a failure q ;14 10 ( 4; 8, 0) 1 st level ( 9; 10, 0) ( 6; 9, 1) ( 1; 5, 1) ( 8; 10, 0) ( 1; 4, 1) 2 nd level 3 rd level Helal et al. A Recourse Approach for the CVRP with Evidential Demands 18 / 24

40 Binary recourse tree example: route R with N = 3 clients θ 1 4; 8 and m 1 ( 4; 8 ) = 1; θ 2 5; 7 and m 2 ( 5; 7 ) = 1; θ 3 7; 9 and m 3 ( 7; 9 ) = 1; Capacity limit Q = 10; q i, i = 1,..., N, the vehicle load after visiting i-th client: Worst case complexity is O(2 N 1 ). ( 4; 8, 0) 1 st level ( 9; 10, 0) ( 6; 9, 1) ( 1; 5, 1) ( 8; 10, 0) ( 1; 4, 1) 2 nd level 3 rd level Helal et al. A Recourse Approach for the CVRP with Evidential Demands 18 / 24

41 Experiments 1 CVRP with Stochastic Demands The CVRP The CVRPSD modelled by SPR 2 CVRP with Evidential Demands Belief function theory Formalisation Uncertainty on recourses (failure situations) Interval Demands Experiments 3 Conclusions & perspectives Helal et al. A Recourse Approach for the CVRP with Evidential Demands 19 / 24

42 Experiments Metaheuristic Simulated annealing to solve the via recourse. Benchmarks Transformed each deterministic demand d det in CVRP data sets, into an evidential demand with associated MF m Θ ({d det }) = α, m Θ ( d det γ d det ; d det + γ d det ) = 1 α with α (0, 1) and γ [0, 1]. Proposition The optimal solution upper expected cost is non decreasing in γ a lower bound on the optimal solution upper expected cost γ = 0. Helal et al. A Recourse Approach for the CVRP with Evidential Demands 20 / 24

43 Experiments Metaheuristic Simulated annealing to solve the via recourse. Benchmarks Transformed each deterministic demand d det in CVRP data sets, into an evidential demand with associated MF m Θ ({d det }) = α, m Θ ( d det γ d det ; d det + γ d det ) = 1 α with α (0, 1) and γ [0, 1]. Proposition The optimal solution upper expected cost is non decreasing in γ a lower bound on the optimal solution upper expected cost γ = 0. Helal et al. A Recourse Approach for the CVRP with Evidential Demands 20 / 24

44 Experiments Table: Simulated annealing results when α = 0.8 and γ = 0.1 Best Penalty Avg Stand. Avg Best cost Instance cost cost cost dev. runtime γ = 0 A-n32-k5 843, % 874,18 9, s. 839,18 A-n33-k5 705, % 724,11 8, s. 697,12 A-n33-k6 773, % 793,07 10, s. 758,36 A-n34-k5 820, % 837,04 9, s. 812,16 A-n36-k5 884, % 914,85 13, s. 869,10 A-n37-k5 722,57 0% 753,51 12, s. 720,85 A-n37-k6 1044, % 1071,27 12, s. 995,07 A-n38-k5 781, % 816,67 18, s. 748,64 A-n39-k5 890, % 935, s. 885,04 A-n39-k6 896, % 916, s. 884,09 A-n44-k6 1051, % 1104,58 24, s. 1019,07 A-n45-k6 1091, % 1129,21 18, s. 1006,90 A-n45-k7 1296, % 1348,57 23, s. 1246,14 A-n46-k7 1060, % 1087, s. 1045,93 A-n48-k7 1241, % 1274,24 20, s. 1227,79 Helal et al. A Recourse Approach for the CVRP with Evidential Demands 21 / 24

45 Conclusions & perspectives 1 CVRP with Stochastic Demands The CVRP The CVRPSD modelled by SPR 2 CVRP with Evidential Demands Belief function theory Formalisation Uncertainty on recourses (failure situations) Interval Demands Experiments 3 Conclusions & perspectives Helal et al. A Recourse Approach for the CVRP with Evidential Demands 22 / 24

46 Conclusions & perspectives Conclusions The modelled by an evidential extension of the SPR. A technique making computations tractable in realistic cases. Experiments using a simulated annealing algorithm. Perspectives More recourse policies. Improving the solving algorithm. References N. Helal, F. Pichon, D. Porumbel, D. Mercier, and E. Lefèvre. The capacitated vehicle routing problem with evidential demands: a belief-constrained programming approach. In Belief Functions: Theory and Applications, volume 9861 of LNCS, pages Springer, Helal et al. A Recourse Approach for the CVRP with Evidential Demands 23 / 24

47 Conclusions & perspectives Conclusions The modelled by an evidential extension of the SPR. A technique making computations tractable in realistic cases. Experiments using a simulated annealing algorithm. Perspectives More recourse policies. Improving the solving algorithm. References N. Helal, F. Pichon, D. Porumbel, D. Mercier, and E. Lefèvre. The capacitated vehicle routing problem with evidential demands: a belief-constrained programming approach. In Belief Functions: Theory and Applications, volume 9861 of LNCS, pages Springer, Helal et al. A Recourse Approach for the CVRP with Evidential Demands 23 / 24

48 Conclusions & perspectives Conclusions The modelled by an evidential extension of the SPR. A technique making computations tractable in realistic cases. Experiments using a simulated annealing algorithm. Perspectives More recourse policies. Improving the solving algorithm. References N. Helal, F. Pichon, D. Porumbel, D. Mercier, and E. Lefèvre. The capacitated vehicle routing problem with evidential demands: a belief-constrained programming approach. In Belief Functions: Theory and Applications, volume 9861 of LNCS, pages Springer, Helal et al. A Recourse Approach for the CVRP with Evidential Demands 23 / 24

49 Conclusions & perspectives Thank you for your attention. Helal et al. A Recourse Approach for the CVRP with Evidential Demands 24 / 24

The Capacitated Vehicle Routing Problem with Evidential Demands

The Capacitated Vehicle Routing Problem with Evidential Demands Nathalie Helal a,, Frédéric Pichon a, Daniel Porumbel b, David Mercier a, Éric Lefèvre a a Univ. Artois, EA 3926, Laboratoire de Génie Informatique