人工神经网络进行建模时通常需要准备大量的数据样本，同时，网络结构一般都比较复杂；而采用支持向量机（SVM）进行建模时，不同核函数有不同的效果，各有利弊，且选取SVM模型参数的理论支撑尚不完整。为了解决这些问题，文中提出了一种基于粒子群优化（PSO）算法的新的SVM混合核函数，这种混合核函数是将局部核函数中的柯西核函数和全局核函数中的多项式核函数进行线性组合，且组合系数和各个核函数中的参数采用PSO算法来优化选取。采用UCI数据库中的wine-red数据集对该混合核函数进行了验证，仿真结果表明，该混合核函数可以提高模型的学习能力和泛化能力。最后，将基于混合核函数的PSO SVM方法用于L形微带天线谐振频率建模，进一步证明了这种方法是可行的和有效的。

A large number of samples and complex structure are needed when predicting the resonant frequency of L-shaped microstrip antenna with artificial neural network. And each ordinary kernel function of support vector machine (SVM) has advantages and disadvantages. Moreover, the theory selecting the parameters of SVM model is incomplete when using SVM. In order to solve these problems, a hybrid kernel function is proposed, in which the parameters of SVM and weight coefficient are optimized by means of particle swarm optimization (PSO) algorithm. Two kernel functions of SVM, namely global kernel function (polynomial kernel function) and local kernel function (cauchy kernel function), and combination kernel function of SVM is presented. This method is validated by using wine red data in UCI database, and result shows that the hybrid kernel function can improve the model's learning ability and generalization ability. Also, the resonant frequency of L shaped microstrip antenna is modeled based on the method, and experimental analysis shows it is feasible and effective.

TN820;O242.1

国家自然科学基金资助项目