(610f) Advanced Model Predictive Control for Reducing Equipment Damage in a Supercritical Pulverized Coal Fired Power Plant during Load-Following Operation

Supercritical pulverized coal (SCPC) power plants are being forced to cycle their load rapidly and frequently due to the increasing penetration of intermittent renewable energy sources to the grid. Traditionally, maintaining the load and main steam temperature and pressure have been some of the key challenges of power plant control [1]. However, SCPC plants are facing increasing equipment damage – especially in high-temperature boiler components – due to rapid load changes that lead to temperature and pressure cycling and significant temperature excursions. Large temperature excursions will lead to increased thermal stresses and in turn creep damage, while repetitive pressure cycling will lead to fatigue damage. Thus, a control system that can also take equipment damage into account is desired. However, the real-time and in-situ measurement of damage in boiler components is practically impossible due to the harsh operating conditions. Furthermore, the damage is typically highly localized at the locations of stress concentration and can vary with time as a function of the operational profile. Additionally, as the boiler pressure is changed during load-following operation, the boiler can transition from supercritical to subcritical steam conditions and vice versa. During this phase transition there are considerable nonlinear changes in the steam properties and heat transfer characteristics that make control challenging.

While commercial power plants are typically operated using feedback-augmented feedforward control (FBAFF), known as the coordinated control system (CCS), there are few works in the open literature that have investigated the use of model predictive control (MPC). These works have primarily considered linear MPC (LMPC) such as dynamic matrix control (DMC) [2]–[4]. However, these works do not consider transitioning between supercritical and subcritical operation, nor do they account for equipment health. Furthermore, due to the considerable changes in the steam properties and plant characteristics in SCPC plants, it is difficult for traditional LMPC using a single linear model to obtain satisfactory closed-loop performance.

In previous work, a detailed dynamic model for an SCPC power plant was developed [1]. This model includes a first-principles, distributed model for all sections of the boiler, custom models of the steam turbine and feed water heaters, and detailed modeling of the balance of the plant. In this work, the model is validated using load-following data from an industrial SCPC plant. The operating data capture plant dynamics and performance characteristics over a wide load range, including transitions through the critical point in either direction. As the plant data do not satisfy mass and energy balances, a dynamic data reconciliation problem is solved. The dynamic model results compare well with a large number of measured plant variables over a long time span and when considering multiple ramp rates. The validated dynamic model is used for developing and evaluating predictive control strategies with an emphasis on reducing equipment damage during load-following operation.

Due to considerable nonlinearity of the plant and equipment health models along with the spatial and temporal variability of the location of stress concentration, a highly complicated nonlinear predictive control model is required for representing the system over the entire operating range of interest. However, the resulting computational expense and nonlinearity makes it difficult to execute MPC at the desired time intervals. To address this issue, two strategies are developed. First, we note that there is considerable time-scale separation for the underlying control problems. While the load dynamics are very fast, temperature transients are much slower. Further, equipment damage occurs over a considerably slower time scale. Therefore we have developed a distributed MPC (DMPC) where multiple MPCs operate under both hierarchical and cooperative paradigms [5]. Each MPC can use a linear or nonlinear model as appropriate. The second strategy is to use a bank of linear models leading to a multi-model predictive control (MMPC) strategy [6]. The contribution of each model is computed recursively using Bayesian statistics based on the model residuals.

Both the DMPC and MMPC algorithms are evaluated alone and in combination and are compared with the industry standard FBAFF-based CCS. When these algorithms are used in combination, only linear or a bank of linear models are used in the DMPC strategy. A number of load-following scenarios are evaluated to evaluate control performance under difference ramp rates and for different initial and final loads. It is observed that while DMPC and MMPC strategies lead to superior control performance compared to the CCS, there is considerable difference in their execution times. The hybrid DMPC-MMPC strategy offers a tradeoff between control performance and execution time. All MPC strategies that are evaluated in this work are found to reduce equipment damage considerably compared to the CCS.

[1] P. Sarda, E. Hedrick, K. Reynolds, D. Bhattacharyya, S. E. Zitney, and B. Omell, “Development of a dynamic model and control system for load-following studies of supercritical pulverized coal power plants,” Processes, vol. 6, no. 11, Nov. 2018, doi: 10.3390/pr6110226.

[2] Z.-J. Li, Z.-X. Li, W. Tan, and J.-Z. Liu, “CONSTRAINED DYNAMIC MATRIX CONTROL FOR A BOILER-TURBINE UNIT,” Proc. Fifth Int. Conf. Mach. Learn. Cybern., pp. 665–670, Aug. 2006.

[3] U. C. Moon and K. Y. Lee, “Step-response model development for dynamic matrix control of a drum-type boiler-turbine system,” IEEE Trans. Energy Convers., vol. 24, no. 2, pp. 423–430, 2009, doi: 10.1109/TEC.2009.2015986.

[4] T. Lee, E. Han, U. C. Moon, and K. Y. Lee, “Supplementary control of air-fuel ratio using dynamic matrix control for thermal power plant emission,” Energies, vol. 13, no. 1, Jan. 2020, doi: 10.3390/en13010226.

[5] P. D. Christofides, R. Scattolini, D. Muñoz de la Peña, and J. Liu, “Distributed model predictive control: A tutorial review and future research directions,” Comput. Chem. Eng., vol. 51, pp. 21–41, Apr. 2013, doi: 10.1016/j.compchemeng.2012.05.011.

[6] M. Thoma, F. Allgöwer, and M. Morari, Nonlinear Model Predictive Control, vol. 384. Springer, 2009.