(494h) Closed-Loop Identification of Process and Disturbance Models for Multivariable Systems

Closed-loop identification plays an important role in improving existing underperforming controllers, also enabling the identification of systems that cannot operate in open-loop fashion. The methods based on the identification of a closed-loop system from which a process model can be obtained using knowledge of the operating controller have been termed as indirect closed-loop identification. The closed-loop identification using prediction error method (PEM) needs special parameterization of models. Therefore, subspace identification methods (SIM) have enjoyed tremendous development in the last 15 years and offered an attractive alternative to PEM due to simple and general parameterization for multivariable systems. However, the fundamental problem for closed-loop identification is the correlation between the unmeasured noise and the process input. As a result, the traditional SIM is biased under closed-loop condition, which required special treatment. To overcome the difficulties mentioned above, this paper presents a novel indirect closed-loop identification method for multivariable systems. Based on one closed-loop PRBS test, the complete models, including process and disturbance dynamics models, can be identified simultaneously. The finite impulse responses (FIR) of the closed-loop systems are first estimated by a recursive algorithm derived from subspace identification method. The use of SIM for indirect closed-loop identification can bypass the special treatment required for traditional direct closed-loop identification. Also, the recursive algorithm makes the estimation of FIR more efficiently. Then, the fast Fourier Transform (FFT) and inverse FFT (IFFT) techniques are used to construct the frequency responses and FIRs, respectively, of process and disturbance dynamics. The algorithm does not require any prior assumptions about the structure of the model. Simulation results have demonstrated the effectiveness of the proposed identification method.