Summary: | 碩士 === 國立交通大學 === 土木工程研究所 === 83 === Threshold accepting(TA) is a new general purpose algorithm
presented by Dueck and Scheuer in 1990 for the solution of
combinatorial optimization problems. This new method is even
simpler structures than the well-known simulated annealing (SA)
approach. While traditional exchange heuristic methods that
were limit to local search, TA method can avoid caving into
local optimum. We have studied the threshold sequence of TA on
TSP, and we have generalized TA to VRP. While the earlier Dueck
and Scheuer study reported only one test problem result of
Grotschel''s 442-cities TSP , this research attempts to apply TA
more comprehensively to a set of TSP and VRP test problems. TSP
has 13 test problems with 42 to 442 nodes selected from TSPLIB.
VRP has 11 test problems with 51 to 200 nodes selected from
literature. The average deviation from the best available
solutions of 13 TSP test problem is only 1.80%; the average CPU
time is 9.842 seconds. For VRP, the average deviation from the
best available solutions of the 11 test problems is 2.796%, and
the average CPU time is 115.17 seconds. Most of all, amount the
11 VRP test problems, this research using TA has updated the
best available solutions of 4 test problems.
|