Improved simulated annealing algorithm solving for 0/1 knapsack problem

The 0/1 knapsack problem belongs to an NP-complete problem. Usually its search space may include substantially large infeasible regions. From theory, unlike other SA-based (SA:Simulated Annealing) algorithms that get the approximate results from all search space, the improved SA algorithm proposed in this paper can obtain the better approximate results only searching from little feasible space, additionally, this algorithm also can improve approximate solution by adding several new variables that record the local maximum, by which the loss of better solution can be avoided caused by Metropolis rule during annealing process. We apply our SA algorithm and the other two algorithms, general SA and greedy-based SA algorithm, to 0/1 knapsack problem test datum generated randomly at the same seed and report our experimental results. Comparing the experiment results of the three algorithms we can conclude that our SA algorithm not only can obtain better approximate results but also can largely decrease the computation time overhead for knapsack problem, especially for relatively large knapsack problem.