一种变步长CMAC的沉降NARMAX模型

摘要

To improve quality and reduce energy consumption of alumina production, the paper analyzed the various factors of alumina settlement process. It used the system identification method to establish the auto-regressive moving average exogenous (ARMAX) model of settlement systems based on cerebella model articulation controller (CMAC). Considered the convergence performance problem of CMAC neural networks, presented the changed step method to solve the problems of standard algurithm, such as convergence speed and accuracy, which adopted hyperbolic secant function to optimize learning step of CMAC. Optimized the ARMAX model of settlement density based on the changed step CMAC. Simulation results show that the density of the settlement process is accurately identified by ARMAX model based on the presented algorithm and the settlement of alumina production operations can be guided.%为了提高氧化铝生产质量和降低能耗,分析了氧化铝沉降工艺中影响沉降过程的各种因素,采用小脑模型神经网络(CMAC)系统辨识的方法建立沉降系统的带外部输入的自回归滑移模型(ARMAX).针对CMAC收敛性存在的问题,提出了基于变步长小脑模型神经网络(CMAC)算法,通过双曲正割函数优化学习步长,提高了小脑模型神经网络算法的收敛速度和计算精度,进而优化了沉降槽密度ARMAX模型.仿真实验表明,该算法的ARMAX模型可以对沉降过程中的槽内密度进行准确识别,指导氧化铝的沉降生产操作.