Increased renewable energy penetration into the grid has created new opportunities for baseload operating coal-fired power plants to participate in load-following [1]. This has led to operational challenges by driving the boilers to deviate from design specifications and to operate in a primarily transient condition, continually ramping up and down to follow the desired load. Although this has added flexibility into the supply-side of the electrical grid, this ramping frequently results in suboptimal performance, which is not desirable considering current efforts to minimize emissions [2]. In order to maintain usefulness of older fossil fuel plants, it is necessary to design better closed-loop control methods for maximizing performance under these new conditions. There have been previous works that have modeled a boiler at steady-state using feedforward artificial neural networks (FF-ANN) and support vector machines (SVM) [3]. There have also been works that use such models for closed-loop steady-state optimization [4, 5]. However, these works do not account for the process dynamics, which can be modeled using recurrent neural networks (RNN) such as long short-term memory (LSTM) [6]. In order to demonstrate the significance of the process dynamics and the need for real-time dynamic optimization, a simplified dynamic model of a coal-fired boiler is optimized using both steady-state and dynamic optimization. The same dataset is modeled using two types of machine learning: LSTM for the dynamic model and an FF-ANN for the steady-state model. Both models are in simulated closed-loop control with particle swarm optimization (PSO) to find the optimal operating parameters. The overall heat-rate of the simulated plant using dynamic optimization and steady-state optimization is compared. The results are not computed in real-time, but they demonstrate a significant improvement on overall boiler efficiency when process dynamics are included in the optimization. The steady-state and dynamic optimization methods are compared using equally-sized timesteps on the same load profile to demonstrate that accounting for the system dynamics yields significantly better performance. Successfully computing these results in real-time will allow more coal-fired boilers to efficiently perform load following and add additional flexibility to the electrical grid.
[1] Hannan, M. A., Tan, S. Y., Al-Shetwi, A. Q., Jern, K. P., & Begum, R. A. (2020). Optimised controller for renewable energy sources integration into microgrid: Functions, constraints and suggestions. Journal of Cleaner Production, 120419.
[2] Tuttle, J. F., & Powell, K. M. (2019). Analysis of a thermal generator’s participation in the Western Energy Imbalance Market and the resulting effects on overall performance and emissions. The Electricity Journal, 32(5), 38-46.
[3] Li, Q., & Yao, G. (2017). Improved coal combustion optimization model based on load balance and coal qualities. Energy, 132, 204-212.
[4] Tuttle, J. F., Vesel, R., Alagarsamy, S., Blackburn, L. D., & Powell, K. (2019). Sustainable NOx emission reduction at a coal-fired power station through the use of online neural network modeling and particle swarm optimization. Control Engineering Practice, 93, 104167.
[5] Tan, P., Xia, J., Zhang, C., Fang, Q., & Chen, G. (2016). Modeling and reduction of NOX emissions for a 700 MW coal-fired boiler with the advanced machine learning method. Energy, 94, 672-679.
[6] Tan, P., He, B., Zhang, C., Rao, D., Li, S., Fang, Q., & Chen, G. (2019). Dynamic modeling of NOX emission in a 660 MW coal-fired boiler with long short-term memory. Energy, 176, 429-436.
