Simulated annealing, weighted simulated annealing and genetic algorithms at work (Q1389252)

The main aim of this article is the selection of the best algorithm for finding an optimal test sequence in quality control. This task leads to a combinatorial optimization problem with an objective function consisting of the expectation of a random variable. It is necessary to find \(\min_{\gamma\in \Phi}H(\gamma)=\min_{\gamma\in \Phi}E[f(x,\gamma)]\), where \(x\) is a random variable with an unknown distribution and \(\Phi\) is a finite set. Sample average is used to estimate \(H(\gamma)\). The authors compare two well known stochastic optimization algorithms (simulated annealing and genetic algorithm) and use the design of the experiment for adjusting the parameters of these algorithms. It turns out that simulated annealing is more efficient for test optimization problems. The randomly weighted version of simulated annealing is proposed to reduce over-fit bias. The convergence of a weighted version is proved and numerical experiments show that this version really reduces the bias.