A dynamic resource allocation strategy for collaborative constrained multi-objective optimization algorithm

Infeasible solutions are helpful for finding the feasible regions, but how many feasible and infeasible solutions should be invested to achieve the optimal search efficiency remains to be further studied. Combined with the recently proposed collaborative constrained multi-objective framework, the contributions of the helper population and original population in different types of CMOPs are discussed. It is unreasonable to assign equal resources to these two populations in different CMOPs and different searching stages. This paper aims to investigate resource allocation in a constraint environment to efficiently utilize the limited resources and obtain a better performance. Therefore, the concept of return on investment (ROI) is first introduced to measure the contributions of two populations, and then guide the population size allocation (APS). To prevent the ROI from continuously declining as the population size decreases, an evolutionary resource allocation strategy (AER) is proposed to adjust their evolutionary state according to the cooperative relationship, and to further increase their ROI and again compete for population size, to maximize the evolutionary efficiency of the two populations in competition and cooperation. The proposed CCMODRA is compared with seven popular algorithms that cover three types of CMOEAs and test them on three benchmarks that cover four types of CMOPs. The comprehensive performance of CCMODRA is better than the other seven CMOEAs on 71% of the 3-objective CDTLZs, 57% of the 5-objective CDTLZs and 46% of the MWs. The effectiveness of the APS and AER strategies are verified on generating contribution solutions and DOC test problems. In addition, the total profit obtained by CCMODRA in the knapsack problem with capacity constraints is improved by 0.2% to 216% compared with the other seven algorithms.

Automated summary

This paper investigates resource allocation in a constraint environment to efficiently utilize limited resources and improve performance. The concept of return on investment (ROI) is introduced to measure the contributions of two populations, and an evolutionary resource allocation strategy (AER) is proposed to increase their ROI.