Abstract: To solve the hyper-parameter adaptive problem of Support Vector Regression (SVR) in prediction problems,especially in the big data prediction problem,a hyper-parametric adaptive support vector regression approach was proposed using K-means clustering and Chaotic Harris Hawks Optimization (CHHO).SVR Hyper-Parameter Cluster (HPC) was randomly generated within the range of super parameter values.K-means clustering was used to obtain the Cluster Centers (CCs) of the training set,and the prediction results of the verification set were obtained according to CCs and HPC.Taking the mean square error as the target,CHHO was used to continuously update HPC for searching the best hyper-parameter.Compared with other well-established methods,the accuracy,stability and computational complexity of prediction were evaluated based on five sets of data sets.The proposed approach ranked first with an overall performance score of 0.226.Finally,the approach was applied to the field of hobbing parameters prediction,and the application effect was compared with other methods.The feasibility and effectiveness of the proposed approach were verified.

Key words: prediction problem, support vector regression, hyper-parameter adaptation, K-means clustering, chaotic harris hawks optimization, hobbing parameters prediction

摘要： 针对预测问题中,特别是在大数据预测问题中的支持向量回归(SVR)超参数自适应调整问题,提出一种基于K-means聚类和混沌哈里斯鹰算法的超参数自适应支持向量回归方法。在超参数数值范围内,随机生成SVR超参数集群。使用K-means聚类获取训练集聚类中心,根据聚类中心和SVR超参数集群获得验证集的预测结果。以均方误差为目标,使用混沌哈里斯鹰算法不断更新超参数集群,找出最佳超参数,并对测试集进行预测,获得最终的均方误差,以测试方法的泛化能力。与其他知名方法进行对比,基于5组数据集,对预测的准确性、稳定性和时间复杂度进行了评估,本方法以0-226的整体性能评分排名第一。最后将本文方法应用到滚齿工艺参数预测领域,与其他方法的应用效果进行比较,验证了本方法的可行性和有效性。

关键词: 预测问题, 支持向量回归, 超参数自适应, K-means聚类, 混沌哈里斯鹰算法, 滚齿工艺参数预测