基于模拟退火的基因改进型GEP算法

Revised gene-gene expression programming algorithms based on simulated annealing

基因表达式编程具有强大的函数挖掘能力,有助于在实验数据上提炼数学模型、揭示事物本质规律.尽管标准GEP算法通过改进遗传操作在一定程度上克服了早熟现象,但在解决实际问题中仍常表现出算法的不稳定;此外,标准GEP算法挖掘出的函数表达式往往冗长,可解释性差.针对这些问题本文做了如下工作:(1)对标准GEP算法的基因进行了新的定义,改进了标准GEP算法的基因构成,提高了GEP算法的通用性;(2)将模拟退火引入到标准GEP算法的选择算子中,提出了基于模拟退火的基因改进型基因表达式编程算法(RG-GEP-SA);(3)实验表明,RG-GEPSA算法比标准GEP算法具有更高的稳定性,RG-GEPSA算法比标准GEP算法成功率提高了11%,挖掘出的函数表达式更具有可解释性.

Mining functions from experimental data based on Gene Expression Programming （GEP） technique can help scientists to build mathematic model and discover the essential rules hidden in the objects. Traditional GEP avoids the problem of premature convergence to a certain extent by trying to use more genetic operations. However, it often represents instability when it is used to solve some practical problems. In addition, the target functions mined by traditional GEP are often very verbose, and are poorly explicable. To solve the problems mentioned above, this paper makes the following contributions： （1） Revises the Gene structure of the traditional GEP, improving its application domains. （2） Proposes Revised Gene-Gene Expression Pro- gramming Based on Simulated Annealing （RG-GEPSA） algorithm, which combines Gene Expression Pro- gramming and Simulated Annealing. （3） By extensive experiments demonstrates the effectiveness of RG- GEPSA. The results show that RG-GEPSA has higher stability than traditional GEP. RG-GEPSA increases the success-probability by 11% and the functions mined by using the new method are more explicable compared with the traditional GEP.