(687b) Large-Scale Dynamic and Multi-Objective Optimization for Power Plant Cycling | AIChE

(687b) Large-Scale Dynamic and Multi-Objective Optimization for Power Plant Cycling


Kim, R. - Presenter, West Virginia University
Lima, F. V., West Virginia University
Traditional electricity grids are currently facing challenges to manage high penetration levels of variable renewable energy (VRE), such as solar and wind[1]. Solar and wind power generation are known to have zero marginal costs and zero fuel emissions to dispatch, thus the VRE from these sources should be prioritized when available. However, the high variability introduced by the VRE leads to dispatchable energy sources such as baseload power plants to cycle their loads so that they can supply the net load in a reliable manner. The increased power plant cycling can introduce unexpected inefficiencies into the system that potentially incur increase in costs, emissions, and wear&tear, as the power plants are no longer operating at their optimal design points[2]. The development of approaches to address these challenges has become a significant topic of research in process systems engineering, from design to control and optimization[3-8].

In particular, a framework comprised of a Tchebycheff-based multi-objective and a hybrid of a Particle Swarm Optimization (PSO) and Sequential Quadratic Programming (SQP) dynamic real-time optimization was previously introduced for a carbon capture subsystem from a supercritical coal-fired (SCPC) power plant[7]. Due to the promising performance results for a low-dimensional case, the extension of the framework to plant-wide optimization was initiated[9]. Such plant-wide optimization had 10 decision variables per time step, including the slack variables, which leaded to a total number of decision variables up to 300 per time horizon. An additional customized Interior-Point-based (IP) step was designed to assess trajectory optimality of the plant-wide optimization. With that additional step, the PSO-SQP algorithm was used to generate an initial feasible trajectory, while the IP step was employed to search for trajectory optimality. Results indicated that the PSO-SQP component performed slower than real-time, while the IP component performed faster than real-time.

Based on the previous implementation results, in this presentation, the developed optimization framework is extended to address large-scale dynamic multi-objective optimization challenges for power plant cycling. The system dimensionality, the imposed load-following demand, and the computational time for real-time implementation are the main challenges for this framework. Assessments of subsections of the optimization algorithm are conducted to analyze large-scale suitability of the current methods as well as other IP algorithms to improve computational efficiency. Parallelization of the PSO component is also performed considering 4 and 64 cores clusters. The optimization is focused on economic and environmental objectives, considering overall efficiency and demand load as constraints. Electricity price market, renewable intermittency profiles, and policies are included as forcing functions/constraints to determine the optimal output trajectories. The obtained optimal dynamic trajectories for power generation can be ultimately sent to lower-level advanced model-based controllers[3][10][11].


[1] REN21 Renewable Energy Policy Network for 21st Century. Available at: http://www.ren21.net/status-of-renewables/global-status-report/. Accessed on April 16, 2018.

[2] Bentek Energy LLC., (2010). How Less Became More: Wind, Power and Unintended Consequences in the Colorado Energy Market. Prepared for Independent Petroleum Association of Mountain States.

[3] He, X., Lima, F. V., (2019). Development and Implementation of Advanced Control Strategies for Power Plant Cycling with Carbon Capture. Computers & Chemical Engineering, 121, pp. 497–509.

[4] Chen, C., Bollas, G. M., (2018). Dynamic Optimization of Subcritical Steam Power Plant Under Time-Varying Power Load. Processes, 6(8), pp. 114–114. doi: 10.3390/pr6080114.

[5] Bankole, T., Jones, D., Bhattacharyya, D., Turton, R., Zitney, S. E., (2018). Optimal Scheduling and Its Lyapunov Stability for Advanced Load-Following Energy Plants with CO2 Capture. Computers and Chemical Engineering, 109, pp. 30–47. doi: 10.1016/j.compchemeng.2017.10.025

[6] He, X, Wang, Y, Bhattacharyya, D, Lima, F.V, Turton, R., (2018). Dynamic Modeling and Advanced Control of Post-Combustion CO2 Capture Plants. Chemical Engineering Research and Design., 131, pp. 430-439. doi: 10.1016/j.cherd.2017.12.020.

[7] Kim, R., Lima, F. V., (2020). A Tchebycheff-Based Multi-Objective Combined with a PSO-SQP Dynamic Real-Time Optimization for Cycling Energy Systems. Chemical Engineering Research and Design, 156, pp. 180-194. doi: 10.1016/j.cherd.2020.01.020

[8] Wang, Y., Bhattacharyya, D., Turton, R., (2020). Evaluation of Novel Configurations of Natural Gas Combined Cycle (NGCC) Power Plants for Load-Following Operation using Dynamic Modeling and Optimization. Energy & Fuels, 34, pp. 1053-1070. doi: 10.1021/acs.energyfuels.9b03036

[9] Kim, R., Lima, F. V., (2019). Multi-objective and Dynamic Real-time Optimization of Supercritical Coal-fired Power Plant Cycling. AIChE Annual Meeting, Orlando, FL.

[10] Zhang, Q., Turton, R., Bhattacharyya, D., (2016). Development of Model and Model-Predictive Control of an MEA-Based Postcombustion CO2 Capture Process. Industrial and Engineering Chemistry Research, 55, pp. 1292-1308.

[11] He, X., Lima, F. V., (2019). A Modified SQP-based Model Predictive Control Algorithm: Application to Supercritical Coal-fired Power Plant Cycling. Submitted for publication.