混合GP-GA用于信息系统建模预测的研究

该文克服了传统建模方法在模型选取及参数估计方面的困难与不足,提出了利用改进的遗传程序设计和改进的遗传算法相结合的混合GP-GA算法。一方面,遗传程序设计中加入了简约压力项,控制了代码过度增长,实现了不加先验知识的简洁非线性模型的自动获取。另一方面,遗传算法采用Gray编码,随机整群抽样选择,以优化模型中的参数,这在一定程度上补偿了遗传程序设计在演化过程中具有较好结构的模型可能因为其中的参数未能达到最优而被淘汰的损失。仿真实例和实际应用均表明混合GP-GA算法优于普通的回归分析及单纯的遗传程序设计方法,提高了拟合和预测精度,并且更适合反映问题的实际情况。

Study on Modeling and Forecasting of Information System with Hybrid GP-GA

This paper puts forward hybrid GP-GA algorithm by combining advanced Genetic Programming(GP)and advanced Genetic Algorithm(GA).It overcomes the difficulties of model selection and parameter optimization in traditional modeling methods.On the one hand,parsimony pressure is added to GP,which controls the code bloat.The compact non-linear model can be automatically achieved without any transcendent knowledge.On the other hand,Genetic Algorithm adopts Gray coding and stochastic universal sampling selection and optimizes parameters of the structure GP evolves.It makes up for the loss that results from the models that are washed out because of good structure and bad parameters.The simulation and the application show hybrid GP-GA is superior to simple GP and common regression analysis.The model it evolves is appropriate to reflect real world better.