论文部分内容阅读
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。
Artificial neural networks usually need to prepare a large number of data samples for modeling, meanwhile, the network structure is generally complex. When using support vector machine (SVM) for modeling, different kernel functions have different effects, and each has advantages and disadvantages The theoretical support for choosing SVM model parameters is not yet complete. In order to solve these problems, a new hybrid SVM kernel function based on Particle Swarm Optimization (PSO) algorithm is proposed in this paper. The hybrid kernel function combines the Cauchy kernel function in the local kernel function and the polynomial kernel in the global kernel function The functions are linearly combined, and the combination coefficients and parameters in each kernel function are optimized using the PSO algorithm. The mixed kernel function is verified by using the wine-red dataset in the UCI database. The simulation results show that the hybrid kernel function can improve the learning ability and generalization ability of the model. Finally, the hybrid kernel-based PSO-SVM method is used to model the L-shaped microstrip antenna resonant frequency, which further proves that this method is feasible and effective.