(646h) Adaptive ARX Models for Non-Linear Chemical Processes: An Industrial Application

The ever-increasing capability of data collection and storage in chemical industry has provided us access to an immense amount of historical data. With open-loop or close-loop identification algorithms^1,2, dynamic predictive models are being developed to address both process and business related questions. However, a larger dataset does not always lead to a higher model accuracy. This is partly due to the fact that as the chemical processes usually evolve over time, the data collected over past years may not accurately reflect the current situation³. In addition to this, the intrinsic nonlinear behaviors of many chemical processes make the data analytics even more challenging.

In this paper, we discuss the utilization of a recursive identification algorithm⁴ to analyze the historical data collected from time-evolving nonlinear industrial processes at Air Products. The identified dynamic models with properly selected forgetting factors were developed to be used for monitoring, optimizing and controlling Air Products’ assets⁵. To verify the advantages of the proposed method, we compare the prediction accuracy of three models: an off-line ARX model, an adaptive ARX model and a set of two ARX models estimated at two main operating conditions. The set of two ARX models were found to be capable of accounting for the nonlinearity of the process, but not the time-evolving dynamics. Among the three models, the adaptive ARX model was found to achieve the smallest prediction error, which verifies that the proposed method is an effective option for modeling time-evolving nonlinear real industrial processes using manufacturing data.

Reference

1. Esmaili A, MacGregor JF, Taylor PA. Direct and two-step methods for closed-loop identification: a comparison of asymptotic and finite data set performance. Journal of Process Control. 2000;10(6):525-537.

2. Ljung L. System identification.Wiley Online Library; 1999.

3. Reis MS, Braatz RD, Chiang LH. Big Data: Challenges and Future Research Directions. Chemical Engineering Progress. 2016;112(3):46-50.

4. Ljung L, Söderström T. Theory and practice of recursive identification.Vol 5: JSTOR; 1983.

5. Kumar D, Chen Y, Esmaili A. Inclusion of Long-term Production Planning/Scheduling into Real-time Optimization. Ifac Papersonline. 2015;48(8):229-233.