BOOSTING SIMULATED ANNEALING WITH FITNESS LANDSCAPE PARAMETERS FOR BETTER OPTIMALITY

Abstract

Multi Dimensional Knapsack problem is a widely studied NP hard problem requiring extensive processing to achieve optimality. Simulated Annealing (SA) unlike other is capable of providing fast solutions but at the cost of solution quality. This paper focuses on making SA robust in terms of solution quality while assuring faster convergence by incorporating effective fitness landscape parameters. For this it proposes to modify the ‘Acceptance Probability’ function of SA. The fitness landscape evaluation strategies are embedded to Acceptance Probability Function to identify the exploitation and exploration of the search space and analyze the behavior on the performance of SA. The basis of doing so is that SA in the process of reaching optimality ignores the association between the search space and fitness space and focuses only on the comparison of current solution with optimal solution on the basis of temperature settings at that point. The idea is implemented in two different ways i.e. by making use of Fitness Distance Correlation and Auto Correlation functions. The experiments are conducted to evaluate the resulting SA on the range of MKP instances available in the OR library.