Weighted minimum variance based on-line performance assessment of multivariate processes: A data-driven approach

In multivariate processes, the previous minimum variance benchmarking techniques still lack an on-line applicability for industrial implementation under routine operating conditions due to the fact that the minimum required process knowledge cannot be obtained from the routine operating data. A data-driven approach is proposed for the on-line performance assessment without relying on prior process knowledge, where the weighted minimum variance benchmark is directly estimated from the closed-loop output data under routine operating conditions. Moreover, in the case of that the knowledge of the multivariate process is available, the practical scheme is proposed for on-line benchmark estimation by using the routine operating data, in which the prior process knowledge is employed by means of the proposed normalizability of first non-zero impulse response coefficient of the plant. In the case study of "Shell" heavy oil fractionator, the effectiveness of the proposed approach is verified from the benchmark/index on-line estimations, their relative estimation errors and time cost for each update.