Research And Application Of Problem-oriented Multi-fidelity KH Algorithm

In the real world,there are many complex problems with high computational complexity and long computational times.For such problems,the computational cost of algorithmic optimization processes is an area of great interest.It is a common method to build agent fidelity models with different degrees of accuracy for the problem.Multi-fidelity technique is an effective strategy to control the fidelity.However,in many studies of algorithms,there is a lack of effective ways to combine them.The krill herd algorithm(KH)is a high-performance algorithm that has been developed in recent years,but it also suffers from the disadvantage of long computational time in solving such problems.In this paper,three MFKH algorithms for multi-fidelity problem are proposed based on the krill herd algorithm by combining Multi-fidelity technique.They are MFKH-I,MFKH-C and MFKH-D,respectively.MFKH-I is a MFKH algorithm that incorporates an iterative-based multi-fidelity adjustment strategy that allows the fidelity level of the problem to be adjusted directly during the iterative optimization process through four fidelity control functions.The MFKH-C algorithm takes into account the constraints of the computational cost budget and assigns higher computational costs to higher fidelity problem models as the algorithm optimization progresses.The MFKH-D algorithm increases the problem fidelity level based on the status dominance relationship in the optimization phase.When the optimization of the algorithm goes into a degenerated status,the algorithm updates the fidelity level of the problem to a higher level.Furthermore,in order to make the MFKH algorithm generalizable in solving multi-fidelity problems,a general multi-fidelity problem framework for modeling multi-fidelity problems is also proposed in this paper.The framework is the reference and basis for modeling high computational complexity problems.As a problem base class for the MFKH algorithm,it provides an interface for model control of the fidelity adjustment strategy of the MFKH algorithm optimization process.In experimental tests of a multi-fidelity test suite and an engineering application,the three proposed MFKH algorithms show a substantial reduction in computational cost in most cases compared to the original KH algorithm.At the same time,the MFKH algorithm optimization achieved at least the same performance compared to the KH algorithm in terms of accuracy and stability.