摘要
建立热工过程的全局非线性模型是热工控制系统全局优化的基础,而静态神经网络难于对非线性动态过程进行建模。资源分配网络(RAN)可以动态调整网络参数,而扩展卡尔曼滤波器(EKF)算法可加快收敛速度。将上述方法有机地结合起来,在此基础上加入剪枝策略和滑动窗口RMS准则,形成了改进的最小资源分配网络(MRAN)。将改进方法应用于典型热工过程的非线性动态建模中,仿真结果表明MRAN网络结构紧凑,建模精度高,适合于在线应用。最后分析了网络初参数对其性能的影响。
The establishment of a comprehensive nonlinear model for a thermodynamic process serves as a basis for the overall optimization of a thermodynamic control system.However,it is difficult for a static neural network to establish a model for nonlinear dynamic processes.A resource allocation network(RAN) lends itself to dynamically adjust the network parameters while an extension Kalman filter(EKF) algorithm can accelerate the converging speed.By organically combining the above-mentioned methods and adding on this basis pruning tactics and a slidingwindow root-mean-square criterion,an improved minimum resource allocation network(MRAN) can be formed.The improved MRAN has been applied to the nonlinear dynamic modeling of a typical thermodynamic process.The simulation results show that the MRAN features a compact network structure and high modeling accuracy,thus making it suitable for on-line applications.Finally,analyzed is the impact of network initial parameters on its performance.
出处
《热能动力工程》
EI
CAS
CSCD
北大核心
2007年第1期91-95,共5页
Journal of Engineering for Thermal Energy and Power
基金
高校博士点基金资助项目(20050286041)
关键词
神经网络
最小资源分配网络
建模
热工过程
neural network, minimum resource allocation network,modeling,thermodynamic process