An information entropy-based evolutionary computation for multi-factorial optimization

Recently, a new category of problems known as multi-factorial optimization (MFO) is gaining momentum in the field of evolutionary computation (EC). This paper aims at improving the aggregate performance of an underlying EC model in a multi-tasking environment by implementing simple strategies with minimal parameter tuning effort. Firstly, an enhanced Simulated Binary Crossover (SBX)-based unary variation operator to improve EC performance is devised. Secondly, to overcome the challenge in parameter tuning and operators selection, an adaptive control strategy underpinned by information entropy is proposed. We study the measure of entropy to quantify the uncertainty of evolutionary search and use the information to adapt the algorithmic parameters. Thirdly, the MFO problems are solved using the proposed methods. Three experiments are carried out to attest the methodology efficacy. The first experiment benchmarks the proposed method against twelve state-of-the-art single objective optimization algorithms in the CEC2014 competition. The second experiment compares the performance of the proposed method using a recent benchmark MFO problem. We further extend the investigation into the analysis of parameter sensitivity and solicit insights pertaining to the algorithm characteristics. Overall, the empirical results on various benchmark problems are promising. The third experiment offers a solution to a multi-mode resource-constrained project scheduling problem in the real-world construction industry. The results indicate improvements of 2.29% in project quality and 8.23% in project duration subject to an increase of 4.78% in project cost, which are well within the acceptance limits of the decision makers.