基于支持向量机的无线电干扰预测算法

针对给定海拔高度、温湿度、风速风向等环境因素下的特高压直流线路无线电干扰分布无法仿真计算的问题,采用灰色关联度模型提取给定环境参数相似的现场测试样本数据,利用遗传算法优化惩罚系数和支持向量的核宽度,提出了特高压直流无线电干扰预测的最小二乘支持向量机法(lest squares support vector machine,LSSVM)。通过分析迭代步数与训练误差证明了灰色关联度的遗传LSSVM方法计算效率和计算精确度优于LSSVM方法和遗传LSSVM方法。对比本文预测方法的计算结果与实际测量值、同类算法计算结果表明:低海拔时0.5 MHz无线电干扰水平负极全压下平均偏差为10.1%,正极半压负极全压下平均偏差为6.75%,双极全压下平均偏差为4.64%;海拔1 900 m时,双极全压下0.5 MHz和10MHz无线电干扰水平平均偏差分别为4.63%和3.5%。

Radio interference prediction method based on support vector machine method

Great efforts have been made in this paper for the radio interference (RI) of ultera high voltage direct current(UHVDC)transmission lines for the given altitude,temperature,humidity,wind velocity,wind direction,etc.The degree of grey incidence and least squares support vector machine method,whose penalty coefficient and kernel functions can be optimized by genetic algorithm,has been presented to predict the RI of UHVDC transmission lines,where the degree of grey incidence model can extract the analogous sample data for the given environment.The numeric example has demonstrated that the better computational efficiency performance of least squares support vector machine(LSSVM)with the degree of grey incidence and genetic algorithm than that of the traditional LSSVM and the genetic LSSVM method.Furthermore,the maximum average relative error of the low altitude between the prediction results and test samples values for 0.5 MHz RI is 6.75％,4.64％ and 10.1％ for half-voltage of positive pole and full-voltage of negative pole,full-voltage of bipolar and full-voltage of the negative pole.Moreover,the average relative error for 10 MHz RI of full-voltage of bipolar with the 1 900 m altitude is less than that of the same measurement point for 0.5 MHz RI.