基于渐进黑箱理论的多变量辨识方法

Multivariable Process Identification Method Based on Asymptotic Theory

为克服传统单变量开环阶跃测试法测试时间长和误差大的缺点,提出一种基于渐进黑箱理论的多变量辨识方法。针对辨识的几个基本问题:测试信号的设计、模型结构的选择、模型阶次的判别和参数估计,进行了全新的设计。采用平移的方法,把一个周期较长的伪随机二进制序列平移若干次,从而得到若干个近似两两互不相关的伪随机二进制序列作为多变量测试信号。选取高ARX模型作为参数模型,并用输出误差(OE)模型进行降阶模型的参数估计,降阶模型的阶次由最小描述长度(MDL)准则来判别。实例仿真的结果表明,用该方法解决多变量辨识问题,能减少测试时间,降低测试期间对设备产生的干扰,辨识的结果也优于常规辨识法。

To overcome the long test time and big error in single variable open loop step test method, a novel identification method based on asymptotic theory （ASYM） is presented. Fundamental problems, such as test signal design for control, model structure selection, model order and parameter estimation, are solved in a systematic manner. Using horizontally shift method, the longer period pseudo random binary sequence （PRBS） is horizontally shifted several times in order to get the PRBS signals that any of the two PRBS signals are non-correlation for multivariable test signal. The high order ARX model is choosed as parameter model, output error model is used to estimate the parameter of reduced model. The order selection for reduced model is decided by MDL （ minimum description length） criterion. The simulation results show that the proposed method can reduce test time, and decrease disturbance to the equipment during the test period. It can get better results comparing with conventional identification method.