The Dynamic Modelling And Flexible Operation Of Coal-fired Power Plant Integrated With Post-combustion Carbon Capture Process

With the increasing ecological concerns of global warming,post-combustion carbon capture（PCC）for coal-fired power plant（CFPP）is of significant importance to combat this trend.The operation of CFPP-PCC process differs from the standalone CFPP.CFPP-PCC process is required to participate in a frequent grid power regulation and to fulfill a strict carbon capture demand at the same time.However,CFPP-PCC process exhibits complex characteristics with significant nonlinearity,large time constant and strong system constraints.Besides,the tight coupling between CFPP and PCC process imposes intense influence on grid regulation and carbon capture.Under these circumstances,the conventional PID controllers are unable to provide with satisfactory control performance.This work aims to provide insights into the flexible operation of industrial-scale CFPP-PCC process through dynamic modelling,system optimization and advanced controller design.The major contributions of current work are proposed as follows:（1）A dynamic model of 660MW supercritical CFPP integrated with PCC process is developed and implemented in g CCS platform.To link CFPP and PCC process,flue gas flowrate and extracted steam from steam turbine are considered in CFPP modelling.Monoethanolamine（MEA）is used as absorption solvent in PCC process.The dynamic models of solvent-based PCC process are developed using rate-based approach and two-film theory.Then,a scale-up calculation for PCC process is carried out in order to process the flue gas from the 660MW CFPP.To minimize reboiler duty and to attain an optimal working condition,a steady-state optimization is executed in terms of absorber height,reboiler temperature and reboiler pressure.With the integration of CFPP and PCC model,an in-depth investigation on dynamic characteristics of CFPP-PCC process are undertaken in the presence of various working scenarios.This knowledge can be used for controller design.（2）A neural network inverse control scheme is presented to overcome the large time constant of CFPP-PCC process.The input and output values are used to train neural network model of CFPP-PCC process.The desired input values under given output setpoints can be therefore obtained.Then,these input values are introduced as feedforward signals.This enables neural network control to achieve a fast setpoints tracking ability.Afterwards,PID controllers are used to eliminate steady-state bias.Simulation indicates that the proposed neural network inverse control is able to realize a better control performance compared with conventional PID controller.（3）Model predictive controllers are utilized to deal with the strong connections between multi-variables of CFPP-PCC process.3 operational modes are presented according to the requirements of electricity generation and CO₂ capture,namely,normal operational model,fast power generation mode and strict carbon capture mode.（4）To enhance the robustness and the close-loop stability of CFPP-PCC process,an extended state observed based stable model predictive controller is proposed in this work.A stable model predictive controller is firstly designed to satisfy Lyapunov stability condition.Optimal manipulated variables can be acquired by solving a quasi-infinite horizon objective function.Afterwards,an extended state observer is obtained to estimate the lumped unknow disturbances.The effects of unknow disturbance can be eliminated by introducing a feed-forward compensation.（5）Considering the economic operation of CFPP-PCC process,this paper proposes a two-layer control structure and an economic model predictive control（EMPC）based on machine learning.In the two-layer controller,the deep belief network is used to build a steady-state economic objective function.Optimal setpoints can be calculated in the upper layer.The lower layer guarantees the setpoints tracking.However,the two-layer controller cannot guarantee a dynamic optimization.A stable EMPC is then proposed using LSTM network as predictive models.Particle swarm optimization technique is utilized to solve nonlinear optimization.Simulation shows that the EMPC can achieve an economic operation during the transient scenarios.