基于GRNN与BPNN的二维向量模式分类对比研究

Comparative study of two dimensional vectors pattern classification based on GRNN and BPNN

为了研究广义回归神经网络(GRNN)和标准BP神经网络(BPNN)在解决二维向量的模式分类问题时的性能差异,分别构建了GRNN分类模型和标准BPNN分类模型,详细阐述了2种分类模型的建立方法,并对所建立的2种分类模型进行训练和泛化能力测试。仿真结果表明,GRNN模型的人为调节参数少,构建方法简单,不易陷入局部极小值,在解决相同的二维向量模式分类问题时,GRNN模型比BPNN模型具有更高的分类精度、更快的收敛速度、更适合于解决二维向量的模式分类问题。

To study the differences between GRNN and standard BPNN in pattern classification of two dimensional vectors, classification models based on GRNN and standard BPNN are established respectively. The establishment methods of the two models are illustrated in detail. The two models are trained and their generalization abilities are tested. The simulation result shows that GRNN has less artificial adjustment parameters, simpler establishment method and it is not easy to fall into local minimum. When applied to the same pattern classification problem of two dimensional vectors, GRNN has higher classification accuracy and faster convergence speed than BPNN and it is more suitable for solving the problem of pattern classification of two dimensional vectors.