Research On Control Performance Assessment Based On Subspace Identification

Process control system is the basis of industrial production automation which guarantees the safety, stability and optimization, low consumption and high benefit of modern enterprises. Control loops generally have good control performance in the early stage of operation. However, lack of regular maintenance over time will result in control performance degradation. Poor control performance will reduce the effectiveness of control system and operating costs increase, ect. Only those control systems which are good designed and regularly maintenaned can guarantee high product quality, low costs, and consequently create economic benefits.The research on control performance assessment based on subspace identification is carried out in this paper. The main work of this thesis as follows:1) A brief review of subspace identification theory is presented with computation of orthogonal and oblique projection which is widely used in subspace identification. Basic steps of subspace identification algorithm are described under the uniform framework.2) A method of time delay estimation is presented using open-loop subspace identification. Using the estimated time delay and stochastic subspace identification algorithm, minimum variance control benchmark and improved benchmark based on desired closed-loop dynamics are developed for control performance assessment.3) A new closed-loop subspace identification method is explored through orthogonal decomposition of signals. First, intermediate variables which are independent with process noise are directly estimated through orthogonal decomposition of controller and process output signals. Secondly, subspace matrices can be estimated through subspace transformation of intermediated variables. Identified subspace matrices are used for the computation of LQG benchmark rather than explicit model parameters to avoid bias of nonlinear least squares fitting. 4) A subspace-based Model-Plant-Mismatch approach for control systems is proposed. First, the residual is constructed through subspace transformation. Then the residual is disposed using statistical local approach, based on which an indicator is constructed to detect the mismatch. The effectiveness of the proposed approach is illustrated by both simulation and industrial applications.