Weight Annealing Heuristics for Solving the Two-Dimensional Bin Packing Problem

Transcription

1 Weight Annealing Heuristics for Solving the Two-Dimensional Bin Packing Problem Kok-Hua Loh, Nanyang Technological University Bruce Golden, University of Maryland Edward Wasil, American University th ICS Conference Charleston, Jan 009

2 Outline of Presentation Introduction Concept of Weight Annealing One-Dimensional Bin Packing Problems Classic Bin Packing Two-Dimensional Bin Packing Problems Computational Results Conclusions

3 Weight Annealing Concept Assigning different weights to different parts of a combinatorial problem to guide computational effort to poorly solved regions Ninio and Schneider (005) Allowing both uphill and downhill moves to escape from a poor local optimum Tracking changes in objective function value and how well every region is being solved Applied to the Traveling Salesman Problem (Ninio and Schneider 005) Weight annealing led to mostly better results than simulated annealing

4 Solving the TSP with Weight Annealing () Local Optimum (visit sequence: ) 3

5 One-Dimensional Bin Packing Problem A classic combinatorial optimization problem Pack a set of N = {,,, n} items, each with size t i, i=,,, n, into identical bins, each with capacity C Minimize the number of bins without violating the capacity constraints Large literature on solving this NP-hard problem 4

6 Outline of Weight Annealing Algorithm Construct an initial solution using first-fit decreasing Compute and assign weights to items to distort sizes according to the packing solutions of individual bins Perform local search by swapping items between all pairs of bins Carry out re-weighting based on the result of the previous optimization run Reduce weight distortion according to a cooling schedule 5

7 Neighborhood Search for Bin Packing Problem From a current solution, obtain the next solution by swapping items between bins with the following objective function (suggested by Fleszar and Hindi 00) p = i= Maximize f ( l ) i l i q i j= bin i load p = number of bins q i t j = t j = number of items in bin = size of item j i f = (5 + 3) + = 68 f = + + = new (5 3 ) 00 6

8 Neighborhood Search for Bin Packing Problem Swap (,) Uphill move Bin α Bin β Bin α (f = 6) ( f new = 64) Bin β Swap (,) Uphill move Bin α Bin β Bin α (f = 6) ( f new = 64) Bin β 7

9 Weight Annealing for Bin Packing Problem Weight of bin i ω i = + K r i residual capacity C li ri = C C = capacity l = load of bin i i An item in a not-so-well-packed bin, with large r i, will have its size distorted by a large amount No size distortions for items in fully packed bins K controls the size distortion, given a fixed r i 8

10 Weight Annealing for Bin Packing Problem Weight annealing allows downhill moves in a maximization problem Example C = 00, K= 0.5, 00 li ωi = ω =.05 ω =.087 transformation 9.5 (90) (80) 0.50 (00) 9.44 (85) Bin Bin Transformed space f = Original space f = 6335 Transformed space f new = 703. Original space f new = 635 Transformed space - uphill move Original space - downhill move 9

11 Weight Annealing for Bin Packing Problem Step 0. Step. Step. Initialization Parameters are K (scaling parameter), nloop, T (temperature), and Tred. Inputs are number of items (n), item sizes (t j ), bin capacity (C), and lower bound (LB). Construct an initial solution using first-fit decreasing procedure. Compute residual capacity r i. Improve the current solution. For i = to nloop Compute the weights w it = ( + Kr i ) T. Do for all pairs of bins Perform Swap (,0){ Evaluate feasibility and f(,0). If f(,0) 0 Move the item. Exit Swap(,0) and, Exit i loop if LB is reached. Exit Swap (,0) if no feasible move with f(,0) 0 is found. } Step 3. Perform Swap (,) Perform Swap (,) Perform Swap (,) T := T Tred End of i loop Output the results. Outputs are the number of bins used, the final distribution of items, and r i. 0

12 Two-Dimensional Bin Packing Problems Problem statement Allocate, without overlapping, n rectangular items to identical rectangular bins Pack items such that the edges of bins and items are parallel to each other Minimize the total number of rectangular bins (NP-hard) Classifications Guillotine cutting BP O G BP R G Free cutting BP O F BP R F Fixed Orientation (O), Guillotine Cutting (G) Allowable 90 Rotation (R), Guillotine Cutting (G) Fixed Orientation (O), Free Cutting (F) Allowable 90 Rotation (R), Free Cutting (F)

13 Two-Dimensional Bin Packing Problems (BP O G) Hybrid first-fit algorithm Phase One (one-dimensional horizontal level packing) Arrange the items in the order of non-increasing height Pack the items from left to right into levels; each level i with a maximum width W Pack an item (left justified) on the first level that can accommodate it; initiate a new level if no level can accommodate it Phase Two (one-dimensional vertical bin packing) Arrange the levels in the order of non-increasing height h i ; this is the height of the first item on the left Solve one-dimensional bin packing problems, each item i with size h i and bin size H

14 Two-Dimensional Bin Packing Problems (BP O G) Example: Arrange items in the order of non-increasing height 3

15 Two-Dimensional Bin Packing Problems (BP O G) An example of hybrid first-fit Phase - One Dimensional Horizontal Level Packing h h h 5 8 h 3 w 4

16 Two-Dimensional Bin Packing Problems (BP O G) Hybrid first-fit algorithm Phase One (one-dimensional horizontal level packing) Arrange the items in the order of non-increasing height Pack the items from left to right into levels; each level i with the same width W Pack an item (left justified) on the first level that can accommodate it; initiate a new level if no level can accommodate it Phase Two (one-dimensional vertical bin packing) Solve one-dimensional bin packing problems, each item i with size h i and bin size H 5

17 Two-Dimensional Bin Packing Problems (BP O G) An example of hybrid first-fit Phase - One Dimensional Horizontal Level Packing Phase - One Dimensional Vertical Bin Packing h h h h 5 8 h 5 8 Bin h 3 h H w h 3 w Bin 6

18 Two-Dimensional Bin Packing Problems (BP O G) Weakness of hybrid first-fit 7

19 Weight Annealing Algorithm(BP O G) Phase One (one-dimensional horizontal level packing) Construct an initial solution Arrange the items in the order of non-increasing height Swap items between levels to minimize the number of levels Objective function p p Maximize f = ( bi ) - ( Whi - Ai ) i= i= b w i ij = m i j= w ij = width of item j in level i mi = number of items in level i p = number of levels A = sum of item areas in level i i h i = height of level i W = bin width 8

20 Weight Annealing Algorithm(BP O G) p Example Maximize f = ( bi ) i= Level Level b = (5 + 3) = 64 b = =4 f = b + b = 68 (two levels) Level f = ( ) = 00 (one level) 9

21 Weight Annealing Algorithm(BP O G) p p Example = i Maximize f ( b ) - ( Wh - A ) i= i= i i 0

22 Weight Annealing Algorithm(BP O G) Phase Two (one-dimensional vertical bin packing) Construct initial solution with first-fit decreasing using level height h i as item sizes and bin height H Swap levels between bins to minimize the number of bins Objective function Maximize f = ( d ) q i= i d i = m i j= h ij h = height of level j in bin i ij mi = number of levels in bin i q = number of bins

23 Weight Annealing Algorithm(BP O G) Phase Three q Filling unused space in each level Maximize f = ( Ai ) i=

24 Weight Annealing Algorithm(BP O G) Phase Three q Filling unused space at the top of each bin Maximize f = ( Ai ) i= 3

25 Weight Annealing Algorithm(BP O G) Weight assignments Phase One W bi ωi = + Kri ri = W Phase Two H di ωi = + Kri ri = H Phase Three HW Ai ωi = + Kri ri = HW 4

26 Solution Procedures Tabu search by Lodi, Martello and Toth (999) for BP O G, BP R G, BP O F, BP R F Guided local search by Faroe, Pisinger and Zachariasen (003) for BP O F Exact algorithm by Martello and Vigo (998) for BP O F Constructive algorithm (HBP) by Boschetti and Mingozzi (003) for BP O F Set covering heuristics by Monaci and Toth (006) for BP O F 500 test instances in all for BP O F 5

27 Computational Results for BP O F Procedures Number of bins Running Time (sec) Tabu Search Guided Local Search Exact Algorithm Constructive Algorithm (HBP) Set Covering Heuristics Weight Annealing Best Lower Bound (LB): 773 bins Weight Annealing is.% above LB Timed out on several problems 6

28 Conclusions The application of weight annealing to bin packing and knapsack problems is new Weight annealing algorithms produce high-quality solutions Weight annealing algorithms are fast and competitive Easy to understand Simple to code Small number of parameters 7