Dynamic analysis of variance methods for monitoring control system performance.

摘要

Efficient control performance monitoring requires large amounts of process data to be compressed into a few statistics that can be monitored over time to alert plant personnel of deteriorating controller performance, changes in disturbance behavior, or other anomalies in process operation. Control performance monitoring is sometimes considered to consist of two stages: a hands-off monitoring stage where problems are reported by exception, and an analysis stage for detailed investigation of potential problems. This thesis presents research results on four control performance monitoring topics that are relevant in both the monitoring and analysis stages: (1) analysis methods for variable setpoint single-input, single-output (SISO) control loops, (2) output variance decompositions for multi-input, single-output (MISO) control loops with cross-correlated disturbances, (3) multi-output dynamic modeling and analysis using vector autoregressive models, and (4) an investigation into the role of cointegration in the analysis of control system data.