RBF神经网络在卡伯值软测量中的应用研究

本文提出了一种新的RBF神经网络的在线学习算法,该算法能在线实时地调整RBF神经网络的隐层单元数目和网络参数,并且使用计算量小、运算速度快的基于逆QR分解的Givens递推最小二乘算法调整网络权值,克服了离线训练方式的不足,并将其用于制浆蒸煮过程中纸浆卡伯值的软测量.通过工厂蒸煮数据验证,表明此算法具有良好的性能,训练的网络具有学习速度快、精确度高、结构紧凑的优点,用于建立卡伯值数学模型,实现卡伯值的软测量是有效的.

Application Research on Radial Basis Function Neural Networks in the Kappa Number Soft Sensing

A new on-line learning algorithm based on the radial basis function neural networks was presented in this paper. The algorithm regulates the number of hidden layers and the parameters of the networks on line, and applied Givens recursive least square algorithm on the basis of the inverse QR decomposition to regulates the weights of the networks. The recursive algorithm has high efficiency. The new on-line algorithm avoided the shortage of the off-line algorithm, and was used in the soft sensing of pulp kappa number in pulping process. The simulation results using the on-site data in the pulp cooking process showed that the new algorithm had a superior properties. The advantages of the networks included quick learning, high accuracy and compact architecture. It was effective for the soft sensing and modeling of kappa number.