摘要
提出了一种基于最优分段函数逼近的过程神经网络学习算法。将网络时变输入信号和连接权在一定精度下表示为分段函数的拟合形式,根据最小均方误差准则,构建PNN基于函数基展开的训练算法。选择低阶分段函数作为基函数,利用其良好的柔韧逼近和光滑可导性质,快速实现网络待定参数对函数样本的自适应学习。网络训练中,只需迭代调整分段函数的连接系数,可有效减少模型中的参数冗余、提高PNN对实际问题的建模能力。
A learning algorithm of process neural network based on the optimal approximating piecewise function is pro-posed .the time-varying input signals and connection weights in the network is represented as a fitting form of the piecewise function under a certain accuracy .According to minimum mean square error approach ,learning algorithm of PNN based on function basis expansion is built .It chooses low-order piecewise function as the basis function ,uses its good flexible approxi-mation and smooth nature to rapidly implement implementation adaptive learning of the network undetermined parameters on the function sample .In the network training ,it can reduce redundancy in the model parameters and improve the modeling a-bility of the PNN to the actual problems effectively by just iterative adjusting the connection coefficient of the piecewise func-tion .
出处
《计算机与数字工程》
2014年第6期919-923,共5页
Computer & Digital Engineering
基金
国家自然科学基金项目(编号:61170132)
中国石油科技创新基金项目(编号:2010D-5006-0302)资助
关键词
过程神经网络
训练算法
分段函数
最优逼近
process neural network
training algorithm
piecewise function
optimal approximating
