摘要
支持向量机(S VM)模型的核心问题是惩罚因子c和核函数参数g的选取.通常支持向量机库工具箱(LIBSVM)采用传统网格搜索算法进行参数寻优,只能得到交叉验证意义下的全局最优解,在更大范围内进行参数寻优比较费时,且效率较低,针对这一问题,提出了基于遗传算法的启发式寻优,以交叉验证(CV)意义下的准确率为适应度,通过一系列的选择交叉变异操作,得到最优的c和g,将优化后的SVM模型应用于大坝扬压力的预测.通过某大坝扬压力监测的实例应用,将遗传算法优化的LIBSVM与传统的LIBSVM预测相对比,预测效果更好,精度更高.
The core issue of support vector machine(SVM)is the selection of penalty factor c and kernel parameter g.Generally,LIBSVM uses the traditional grid search algorithm for parameter optimization,which can only get the global optimal solution on the sense of cross-validation(CV)and is time-consuming and inefficient when carried out on a larger parameter scale.To solve this problem,the heuristic optimization is proposed based on genetic algorithm(GA),with the CV accuracy as adaption and through a certain series of selection,crossover and mutation operation,the best c and gare got.The optimized SVM model is applied to regress and predict the dam uplift pressure;through the practical application of a dam's uplift pressure monitoring,it turns out the prediction and accurate results of the new model is better compared with the traditional model.
出处
《三峡大学学报（自然科学版）》
CAS
2013年第6期24-28,共5页
Journal of China Three Gorges University：Natural Sciences
基金
国家自然科学基金资助项目(51139001)
新世纪优秀人才支持计划资助(NCET-11-0628)
高等学校博士学科点专项科研基金(20120094110005)
中央高校基本科研业务费项目(2012B07214)
