摘要
基于M-RAN算法的RBF神经网络是一种动态神经网络,适合于过程的在线建模。对M-RAN算法的删除策略进行了改进,不仅删除那些连续对网络输出贡献较小的隐层单元,同时还将相似的隐层单元合并,使网络结构更加紧凑。将基于这种算法的RBF神经网络用于电厂非性线模型热工过程的在线辨识,仿真研究表明了这种建模方法的有效性,且所得模型精度高,计算量小,可直接应用于基于模型的控制算法。
Minimal-resource allocating networks （M-RAN） are a kind of dynamical radial basis function （RBF） neural networks suitable for online process modeling. The pruning strategy of M-RAN＇ s learning algorithm is being improved, not only by deleting the hidden neurons that continuously contribute but little to the network＇ s output, but also by combining similar hidden neurons, herewith contributing to a more compact network structure. Online identification of power plant thermal processes with nonlinear models is carried out by applying this method. Simulation study results demonstrate the validity of this method, which is distinguished by a higher modeling precision and less entailing calculation work, directly applicable to model based control algorithm. Figs 4, table 1 and refs 18.
出处
《动力工程》
EI
CSCD
北大核心
2005年第6期844-848,共5页
Power Engineering
基金
国家自然科学基金资助项目(50576022)
江苏省高校自然科学研究计划项目(04KJB470036)
南京工程学院科研基金项目(KXJ04070)
关键词
自动控制技术
电厂
系统辨识
径向基函数
神经网络
热工过程
automatic control technique
power plant
system identification
radial basis function
neural network
thermal process
