Closed-loop Subspace Identification And Optimal Test Signal Design

A well-designed identification test and a proper identification algorithm play important roles in the successful process identification. It is the objective of the test signal design to excite maximum system information with the minimum test energy, which is a basic step in obtaining a more accurate process model. An effective identification method is critical in modeling the process when input and output data is obtained. Subspace identification method has been widely adopted due to its simple structure, computational efficiency and numerical robustness. This dissertation focuses on the closed-loop optimal test signal design and closed-loop subspace identification. The main contents are outlined as follows:1. The optimal test signal design for a closed-loop system with PID controller is proposed. The test signal is added at the set point to ensure the system safety and product quality. The corresponding SDP problem is constructed and solved in frequency domain and the cosine is superimposed to characterize the result. The design method is verified via Monte Carlo simulations which indicate that under the same level of test energy, the optimal signal can get a more accurate model than white noise signal and random binary signal.2. A novel closed-loop subspace identification method is proposed based on kernel partial least square method. Since this method can select latent eigenvectors to reduce the dimension of the problem, it can obtain accurate model without persistent excitation. Monte Carlo simulation examples are presented. The results show that the proposed method works well with the existence of correlations between the test signals.3. Performance analysis of the closed-loop identification based on kernel partial least square is presented, which proves that this method is asymptotic unbiased and consistent. Simulations indicate that performance of the identification is as good as PEM, and the previous optimal test signal can be utilized for identification.4. The closed-loop subspace identification method together with PEM is verified by industrial data. The application results prove that the identified model via this approach is better than PEM in data fitting. It can be utilized for the purpose of MPC control and the proposed closed-loop subspace identification is effective and applicable.Finally, a summary is given, and the perspective of future studies is presented in the end of the dissertation.