Performance Monitoring Of Multivariable Model Predictive Controller

Performance monitoring of model predictive controller (MPC) is not only valuable to the theory research, but also to the engineering application of the industrial facilities, with the aspect of the control quality improvement, maintaining of efficient operation and increasing the application level of advanced process control strategies. Several performance monitoring methods for multivariable MPC controller are proposed. The main research works are as follows:A survey of current controller performance monitoring on both theory and application is introduced, including the basic concept, flow and algorithms of performance assessment and diagnosis as well as future needs for development. The performance assessment based on covariance benchmark is demonstrated in detail.To locate the performance deterioration root causes, a distance clustering based approach for model predictive controller performance diagnosis is proposed. After concept of eigenvector subspace which can describe the characteristic of various subspaces is presented, subspace distance is introduced to demonstrate the similarity of two performance data. Classification could be made by calculating the distance between current subspace and the predefined ones, and then the causes contributing to the performance deterioration can correctly be located. Simulation results on the Wood-Berry distillation column system validate the efficiency of the novel method.A coherent statistical 2-norm based approach is proposed as an extended method for monitoring the performance of multivariate MPC controller. One 2-norm based covariance benchmark is established through the SVD on the extended monitored variable set. Characteristic direction information is mined from the periods of operating data and the corresponding classes are built. The eigenvector angle is defined to describe the similarity between the current data set and the established classes, and an angle based classifier is introduced to identify the root cause of MPC performance degradation when a poor performance is detected. The proposed methodology is successfully demonstrated in a detailed case study of the Wood-Berry distillation column system.Using nonlinear scaled covariance to maximize the relationship of process data, a correspondence analysis (CA) based approach for MPC controller performance monitoring is proposed. The performance benchmark based on CA eigenvalues is deduced, and MPC controller performance index is established. To identify the root cause of performance deterioration, principal axis based CA subspace method is proposed. Correspondence analysis is performed on different performance pattern datasets so that the subspace is available to describe the major data directions, and the subspace classifier is constructed to recognize performance patterns. Simulations on Wood-Berry process demonstrate that the proposed method could carry out performance assessment and diagnosis on MPC controller effectively.By means of DPCA modeling of the extended monitoring variable set, an approach based on dynamic principal component analysis (DPCA) is proposed to monitor the performance of multivariable MPC controller. The Hotelling statistics based benchmark can be established, which can be used to carry out the performance assessment procedure. After classes which stand for different causes of the performance deterioration can be obtained, the united weighted DPCA similarity is introduced to identify the root cause of the current controller poor performance according to the classification analysis. Simulations on the Wood-Berry process validate the efficiency of the novel method.