A tracking simulator that runs in parallel with the plant and exactly simulates dynamic plant behavior has been developed. This tracking simulator is currently in the validation phase at a commercial petrochemical plant site. The future prediction function of the tracking simulator is expected to have a major impact on plant operations, improvement, and control design. The tracking simulator is strongly dependent on the physical model based on physical and chemical laws, and cannot simulate unknown phenomena in the plant. A hybrid model consisting of a physical model and a statistical model based on historical plant operation data was proposed. The estimation accuracy of the statistical model was improved using the virtual input from the tracking simulator. Although the Kalman filter technique is applied to conventional software sensors for quality estimation as a model updating function, deterioration of estimation accuracy is often seen with marked changes in plant behavior and the analyzers shut down. To resolve these issues, we applied Just-In-Time modeling based on the historical data modeling technique with the tracking simulator, and the plant model coverage of the tracking simulator was expanded for various processes. (C) 2012 Elsevier Ltd. All rights reserved.