Flexible operation of large-scale coal-fired power plant integrated with solvent-based post-combustion CO2 capture based on neural network inverse control

Abstract

Post-combustion carbon capture (PCC) with chemical absorption has strong interactions with coal-fired power plant (CFPP). It is necessary to investigate dynamic characteristics of the integrated CFPP-PCC system to gain knowledge for flexible operation. It has been demonstrated that the integrated system exhibits large time inertial and this will incur additional challenge for controller design. Conventional PID controller cannot effectively control CFPP-PCC process. To overcome these barriers, this paper presents an improved neural network inverse control (NNIC) which can quickly operate the integrated system and handle with large time constant. Neural network (NN) is used to approximate inverse dynamic relationships of integrated CFPP-PCC system. The NN inverse model uses setpoints as model inputs and gets predictions of manipulated variables. The predicted manipulated variables are then introduced as feed-forward signals. In order to eliminate steady-state bias and to operate the integrated CFPP-PCC under different working conditions, improvements have been achieved with the addition of PID compensator. The improved NNIC is evaluated in a large-scale supercritical CFPP-PCC plant which is implemented in gCCS toolkit. Case studies are carried out considering variations in power setpoint and capture level setpoint. Simulation results reveal that proposed NNIC can track setpoints quickly and exhibit satisfactory control performances.

Optimization of a 660 MWe Supercritical Power Plant Performance—A Case of Industry 4.0 in the Data-Driven Operational Management Part 1. Thermal Efficiency

“…The new generation of artificial intelligence, called AI 2.0, had seen a rapid development worldwide, especially in smart energy and electric power systems (Smart EEPS) and fueling its enormous applications in operation, optimization, control, and management of Smart EEPS [41,42]. Moreover, a comprehensive review of machine learning tools for energy efficiency objectives was presented based on the published forty-two research articles.…”

Flexible operation of supercritical coal-fired power plant integrated with solvent-based CO2 capture through collaborative predictive control

“…2. The CFPP model was developed based on a hybrid modelling approach, which fuses the physics-based understanding of the process and the information contained in the real-time operation data to develop a satisfactory process model [30]. The hybrid modelling approach can be divided into two steps.…”

“…Particle swarm optimization (PSO) algorithm, which is a powerful machine leaning technique is explored for the parameter identification of this model. The CFPP model has been validated via steady-state and dynamic operational data archived from a 660MWe supercritical CFPP located in Anhui, China and has been shown to reflect the key dynamics of the CFPP satisfactorily [30].…”

“…The PCC dynamic model in this study was developed based on a pilot-scale reference model [32], which has been validated through experimental data collected from the Separations Research Program (SRP) pilot plant in the University of Texas at Austin. This model was then scaled up in our previous works [30,33] using the generalized pressure drop principle to deal with the flue gas from a 660MW CFPP and provide convincible simulations of the industrial-scale PCC process. Three absorbers are used in the PCC design.…”

“…We assumed that the flue gas and lean solvent are equally distributed to the three absorbers. Detailed information of the CFPP-PCC model can be found in [30], thus it is not repeated here. in Table 1.…”