加权支持向量机回归算法,几乎都是以样本输入空间中的一个重要特征量的函数来确定权值,造成了在高维特征空间中作回归可能存在较大误差。针对这一问题,提出利用高维特征空间中的欧基里德距离来确定权值的方法,构造了一种改进的加权支持向量机回归算法,并将其应用到电子器件高功率微波易损性评估中。仿真结果表明: 该方法具有比模糊神经网络法、标准支持向量机回归算法和一般的加权支持向量机回归算法更高的预测精度。由于增加了权值的计算过程,相对于标准支持向量机回归和模糊神经网络方法,该方法的效率较低,但与一般的加权支持向量机回归算法相当。

At present weighted support vector regression(WSVR)algorithms almost select a function of an important eigen quantity to calculate the weights, which leads to higher errors in doing regression in high dimension eigen space. Aiming at above problem, a method of certaining weights by the Euclidean distance in high dimension eigen space is presented, therefore an improved weighted support vector regression algorithm is built up and applied to the vulnerability assessment of electronic devices illuminated or injected by high power microwave(HPM). The simulation results show that our method is more accurate than fuzzy neural network(FNN), standard support vector regression and common weighted support vector regression. Because of the additional process of calculating weights, the presented method’s efficiency which is as high as common WSVR is a little lower than standard support vector regression and FNN.