A Multi-index Control Performance Assessment Method Based on Historical Prediction Error Covariance

Abstract

Historical data based index has been widely studied in the field of control loop performance assessment in recent years. However, it has the limitation of subjectivity and cannot indicate the root causes of performance degradation. In addition, performance assessment based on a single historical data index can easily lead to erroneous results. In this paper, a multi-index control performance assessment method is proposed based on historical prediction error covariance to overcome the above limitations. In the proposed method, three covariance indices are firstly constructed to detect the changes in the covariance of output variables calculated from monitored data and the covariance of output variables' first τ-step prediction error calculated respectively from benchmark data and monitored data. Furthermore, the generalized eigenvalues of these built covariance matrices are analyzed to find the change directions of control performance. The effectiveness of the proposed method is illustrated through case studies on a two-variable numerical system and the Wood Berry distillation column system.