基于距离和时序的层次粒度支持向量回归机

Hierarchical Granular Support Vector Regression Based on Distance and Temporal

针对目前粒度支持向量回归机的粒划算法只考虑了距离因素,引入时序因素,提出适用于金融时间序列的基于距离和时序的层次粒度支持向量回归机(DTHGSVR).该方法首先将训练样本通过核函数映射到高维空间,并在该特征空间中进行初始粒划.然后,通过衡量样本粒与当前回归超平面的距离以及当前样本粒时序的综合因素,找到含有较多回归信息的粒,并通过计算其半径、密度及时序信息进行深层次的动态粒划.如此循环迭代,直到没有粒需要进行深层划分为止.最后,对不同层次的粒进行回归训练.采用提出的基于距离和时序因素的层次粒度支持向量回归机对基金净值进行预测,实验结果表明回归的泛化性有所提高.

Only distance factor is considered in the granular algorithm of granular support vectorregression. Temporal factor was introduced simultaneously in granular algorithm. Hierarchicalgranular support vector regression based on distance and temporal factors （ DTHGSVR） wasproposed which is applicable for financial time series. The training samples were mapped into thehigh-dimensional space by mercer kernel, and the samples were divided into some granulesinitially. Then, the granules which have more regression information was found by measuring thedistances between the granules and regression hyperplane and the granule？ s temporal factor. Bycomputing the radius, density of granules and the temporal factor, the deeper hierarchicalgranulation process was executed until no granules was needed to be granulated. Finally,thosegranules in different granulation levels were trained by SVR. Fund net was forecast by thehierarchical granular support vector regression based on distance and temporal factors.Experimental results showed the generalization performance of regression had been improved.