基于代理模型可行规则法的优化算法

Optimization Algorithm Based on Surrogate Model Feasible Rule Method

针对约束优化问题中约束函数计算费时的情况,提出使用代理模型拟合待优化约束函数的方法。该方法先求解个体所有约束违反度并求和,然后建立代理模型预测个体的约束违反度之和,并根据可行规则法筛选最优解。最后采用CEC2017约束测试函数,与不加代理模型的优化算法对比,验证了该方法的有效性,提高了优化结果和进化算法效率。

In many constrained optimization problems, the evolutionary algorithm is inefficient due to the time-consuming constraints.A method of matching the constraint condition function of the constraint function to be optimized by proxy model is proposed to accelerate the efficiency of population evolution.At the same time, in order to solve the problem of overlapping error rates caused by multiple surrogate model matching and surrogate model time-consuming problems when there are many constraints, this article firstly solves all constraint violations of the individual and sums them, and then establishes a surrogate model to predict the individual’s sum of constraint violations.In order to verify the effectiveness of the proposed method, the CEC2017 constraint test function is used to compare with the original optimization algorithm without proxy model.It is found through experiments that the method proposed in this paper can effectively improve the optimization results with a limited number of evaluations.