一种参数优化旋转广义回归神经网络模型

Parameter-optimized rotated general regression neural network model

针对传统广义回归神经网络的模型结构与数据分布失配问题和模型参数难以确定问题,提出了一种参数优化旋转广义回归神经网络模型的设计方法。在传统广义回归神经网络模型的基础上,通过坐标旋转,增加了一个模型结构参数,并采用粒子群算法对旋转广义回归神经网络的模型参数寻找最优值,从而改进了广义回归神经网络模型精确度。两个工业实例的实验结果表明该方法的有效性。

To resolve the problem of the mismatching of model structure and data distribution as well as the problem of determining model parameters difficultly in the traditional general regression neural network （GRNN）, a scheme is proposed to design a parameter-optimized rotated network. Through the coordinate rotation, an additional parameter of model structure is introduced to the traditional general regression neural network. Moreover, the particle swarm optimization algorithm is adopted to find the best values of parameters of the rotated GRNN ; hence the model precision is improved. The experimental results of two industrial applications have shown the effectiveness of the method.