(541h) A Flexible, Open-Source Framework for Data Reconciliation, Parameter Estimation, and Optimizing Operation: Application to Power Plants

Large-scale process models based on first principles have the potential to enable the optimization of a wide variety of processes. Their use in improving plant operations, however, remains scarce due to challenges associated with quality of the process data, model construction and validation, and computational tractability. For example, raw plant data is noisy, may contain gross measurement errors, may not satisfy mass and energy balances, and is not always comprehensive. Furthermore, a mix of data-driven, surrogate, and first-principles models is often required to strike the appropriate balance between the desired model accuracy and computational performance (i.e., speed and robustness). In addition, identification of optimal operations frequently requires solving large-scale optimization problems involving thousands of decision variables and highly nonlinear equations, which is challenging. Finally, developing useful model-driven recommendations requires significant effort to gain confidence that the model reflects true plant operations and to understand which knobs a plant can practically turn to achieve ‘optimal’ performance.

To help address these challenges, the Institute for the Design of Advanced Energy Systems (IDAES) [1] has developed a flexible, open, and extensive process modeling library and framework [2] that can leverage state-of-the-art nonlinear programming solvers throughout a comprehensive workflow that includes data reconciliation, parameter estimation, and system-wide optimization. In this work, a step-by-step procedure is introduced that reconciles data from an existing process, uses the reconciled data to identify parameters for first-principles models that ensure their accuracy over a broad range of conditions, and applies the models in large-scale optimization problems to identify new, and potentially non-intuitive, operating paradigms.

We demonstrate this general framework to address challenges faced by the existing fleet of coal-fired power plants, who have been under immense economic pressures in recent years due largely to low cost natural gas and the increased adoption of renewable energy sources. Specifically, the capabilities are demonstrated through a partnership with Tri-State Generation and Transmission Association, Inc., and their Escalante Generating Station in Prewitt, New Mexico. The 245 MW subcritical coal-fired power plant, originally designed to operate continuously at 100% capacity (i.e., baseload operation), is now frequently required to cycle its power output throughout the day in response to variable demand (i.e., load following operation) or maintain operations at minimal loads for extended periods of time when demand for electricity is low (i.e., low load operation).

The plant model includes: 1) a hybrid boiler fire-side model that captures flow characteristics and reaction kinetics in 1-D zones and radiative heat transfer on a 3-D mesh [3] that is integrated into the flowsheet using highly accurate algebraic surrogate models generated using ALAMO [4], 2) fully equation-oriented models of all other unit operations including the boiler heat exchanger network (water wall, economizer, superheaters, reheater), turbine, feedwater heaters, condenser, deaerator, and pumps, and 3) an equation-oriented implementation of the IAPWS steam properties [5] that can handle vapor-liquid phase transitions and returns exact 1st and 2nd derivatives.

In order to identify process improvement strategies, several months of operational data were reconciled to eliminate gross measurement errors, identify and fill knowledge and data gaps, and quantify the variability in both measured and calculated flow rates, temperatures and pressures. The data reconciliation approach enabled calculation of several unmeasured flow rates throughout the steam cycle and boiler leading to testable hypotheses regarding what is limiting the current minimum load at which the plant can operate. Lowering this limit can lead to significant cost savings by keeping the plant warm (i.e., avoiding cold starts that significantly reduce equipment life), safe, and stable using the minimum amount of fuel when demand for electricity is low.

The ensuing parameter estimation step generated a plant-wide model with prediction errors of less than 3% in all key quantities across the load range (e.g., coal flow, flue gas oxygen composition, power output, main steam temperature) and in most cases the errors were far lower. The final optimization revealed that significant improvements to plant’s heat rate (i.e., efficiency) are potentially achievable upon taking a steeper sliding pressure approach to load following that results in lower throttling losses at the inlet to the high-pressure turbine and less power consumption by the main boiler feed pump.

References

1. D. Miller, J. Siirola, D. Agarwal, A. Burgard, A. Lee, J. Eslick, B. Nicholson, C. Laird, L. Biegler, D. Bhattacharyya, N. Sahinidis, I. Grossmann, C. Gounaris, and D. Gunter, “Next Generation Multi-Scale Process Systems Engineering Framework”, Computer Aided Chemical Engineering, 44, 2209-2214 (2018).
2. https://idaes-pse.readthedocs.io/en/stable/
3. J. Ma, J.P. Eason, A.W. Dowling, L.T. Biegler, and D. Miller, “Development of a First-Principles Hybrid Boiler Model for Oxycombustion Power Generation”, International Journal of Greenhouse Gas Control, 46, 136-157 (2016).
4. A. Cozad, N.V. Sahinidis, and D. Miller, “Learning Surrogate Models for Simulation-Based Optimization”, AICHE Journal, 60(6), 2211-2227 (2014).
5. IAPWS, “International Association for the Properties of Water and Steam”, IAPWS R6-95 (2016).