(184p) Machine Learning Techniques for Model Identification from Historical Data for Control

Model predictive controllers (MPC) rely on system model to identify optimal input profile. Data collection for identifying such models is the most time consuming step in implementation of MPC in industries [1]. Storing operational data using process historian is a common practice in industries. In general usage of such data is restricted to performance assessment, fault detection etc. Historical data is not used for model identification due to the presence of multiple time scales, presence of long time disturbance effects and correlated inputs etc. Within the time period for which the historical data is collected, there will be intervals of time where the data segment is informative enough to be used for modeling exercise. Identifying and marking these segments is a complex problem. Peretzki et al [2], proposed a method for data segmentation based on variance of the variables, condition number of the information matrix and the estimated parameters using Laguerre’s filter being non-zero. They also provided a quality measure for the data. In this paper we propose a new data segmentation technique that can identify regions of informative data from historical data. An interval-halving technique that can handle multivariate data is used for this purpose [3]. Machine learning techniques are used to identify regions of data with similar features to address the presence of disturbance effects. Self-learning and model update after online implementation of such models will also be discussed.

References

[1] H. Genceli and M. Nikolaou, “New Approach to Constrained Predictive Control with Simultaneous Model Identification,” vol. 42, no. 10, pp. 2857–2868, 1996.

[2] D. Peretzki, A. J. Isaksson, A. C. Bittencourt, and K. Forsman, “Data mining of historic data for process identification,” AIChE Annu. Meet., 2011.

[3] L. Das, B. Srinivasan, and R. Rengaswamy, “Multivariate Control Loop Performance Assessment With Hurst Exponent and Mahalanobis Distance,” IEEE Trans. Control Syst. Technol., vol. 24, no. 3, pp. 1067–1074, 2016.