导线或电缆不仅是高效的电磁干扰接收天线,也是高效的电磁干扰辐射天线.互连导线间的串扰是一种典型的电磁兼容问题.为了达到更好的电磁兼容预测效果,提出了基于支持向量机(SVM)的预测算法.算法以结构风险最小化为原则,比以经验风险最小化为原则的传统神经网络性能更优.利用遗传算法优化预测模型的关联参数,可以进一步提高计算效率和精度.实验结果表明利用SVM算法对导线串扰预测的误差最小.通过对导线串扰预测模型的影响因子进行相关性分析,降低了模型的自变量维度.简化后的模型预测具有更好的实用性.%Wire of cable is not only efficient receiver antenna, but also radiator antenna in Electromagnetic Interference (EMI). Crosstalk coupling between interconnecting wires is typical problem of Electromagnetic Compatibility (EMC). A methodology based on SVM was presented in order to get better prediction result. The proposed algorithm follows Structural Risk Minimization (SRM) principle is more generalized performance as compared to traditional Artificial Neural Network (ANN) which embodies the Embodies Risk Minimization (ERM) principle. By using the genetic algorithm to optimize inner parameters, the calculation efficiency and accuracy can be further improved. Experimental results show the effectiveness for prediction of EMC based on SVM algorithm which gets smaller calculation error than others. Through correlation analysis of the wires crosstalks impact factors, independent variables'dimension of the predictive model is reduced. Simplified model takes higher accuracy and greater practicality.
展开▼
