粗糙集约简的WNN隐层节点优化方法

Optimization method for hidden layer nodes of WNN based on rough set reduction

在确保网络性能的前提下,如何确定最佳隐层节点,获得最简网络结构是小波神经网络(WNN)应用推广的关键.对此,引入粗糙集理论,提出了基于信息熵的卡方离散化算法和启发式的属性约简递归算法,利用粗糙集约简过程对WNN隐层节点进行精简,并将其应用于飞行器气动力建模.仿真结果表明,采用改进的粗糙集方法设计WNN,不仅能够简化网络结构,而且与未经结构优化的WNN相比,其模型精度和训练速度都得到了实质性改善.

Under the premise of ensuring network performance, the key of wavelet neural network（WNN） application and promotion is how to get the most simple network structure by determining the optimal hidden layer nodes. Therefore, a Chi-Square discretization algorithm based on the information entropy and heuristic attribute reduction recursive algorithm is proposed, reduction process of the rough set theory is used to optimize wavelet neural network hidden layer nodes without changing network performance, and an aircraft aerodynamic model is built by modifying wavelet neural network. Simulation results show that WNN optimized by the proposed improved rough set method the can not only simplify the network structure, but also improve model accuracy and training speed.