研究目的
To present a new algorithm tuning multiobjective particle swarm optimization (tMOPSO) for tuning the control parameter values (CPVs) of stochastic optimization algorithms under a range of objective function evaluation (OFE) budget constraints.
研究成果
The numerical experiments verify that tMOPSO is effective at tuning optimization algorithms under multiple OFE budgets. tMOPSO outperforms or is comparable to existing multiple OFE budget tuning algorithms and is more efficient than setting up multiple uncoupled tuning problems each focused on a different single OFE budget. The study concludes that tMOPSO is a viable alternative for tuning optimization algorithms under a range of OFE budgets.
研究不足
The study is limited to tuning stochastic optimization algorithms under multiple OFE budgets using problems from the CEC 2005 special session on real-parameter optimization. The effectiveness of tMOPSO on other types of optimization problems or with different computational constraints is not explored.
1：Experimental Design and Method Selection:
The study uses tMOPSO, a multiobjective particle swarm optimization algorithm, to tune the CPVs of stochastic optimization algorithms under multiple OFE budgets. The methodology includes formulating the control parameter tuning problem as a multiobjective optimization problem and employing a noise-handling strategy and CPV assessment procedure specialized for stochastic algorithms.
2：Sample Selection and Data Sources:
The study uses problems from the CEC 2005 special session on real-parameter optimization as the application layer for tuning. These problems are unconstrained, real-valued, and static noise-free single-objective minimization problems.
3：List of Experimental Equipment and Materials:
The study involves numerical experiments conducted using computational resources from the South African Center for High Performance Computing (CHPC) and the High Performance Computing Center (HPCC) of the Department of Electrical, Electronic, and Computer Engineering (EECE) at the University of Pretoria.
4：Experimental Procedures and Operational Workflow:
The study compares tMOPSO against other tuning algorithms focused on single and multiple OFE budgets. The performance of tMOPSO is gauged through numerical experiments, including parameter sweeps and statistical tests to determine the effectiveness of the proposed algorithm.
5：Data Analysis Methods:
The study uses statistical tests, including Friedman tests and Mann–Whitney U-tests (MWUTs), to analyze the performance of tMOPSO and compare it with other tuning algorithms. The hypervolume (HV) achieved on tuning problems is used as a performance measure for multiple OFE budget tuning algorithms.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
