基于Koopman算子的超超临界火电机组模型预测控制OA

[Objective]In the context of achieving carbon peaking and carbon neutrality goals,it is crucial for the thermal power plants to improve the load response capability to accelerate their transition from primary power sources to auxiliary units that provide peak shaving and frequency regulation services.Traditional control methods often face challenges such as delayed responses,poor steady-state accuracy,and high computational complexity during wide-range load variation operations..[Methods]To address the above issues,this paper takes the coordinated control system of a 1 000 MW ultra-supercritical coal-fired power unit as the research object,and proposes a Koopman operator-based model predictive control(KMPC)method.First,the original nonlinear system is discretized using the fourth-order Runge-Kutta method to generate a dataset.Based on extended dynamic mode decomposition,a finite-dimensional approximation of the Koopman operator is constructed to obtain a high-dimensional linear approximation model of the power unit.Based on this model,the system's future dynamics are predicted,and a receding horizon optimization strategy is introduced,in which control constraints,control objectives,and performance metrics are comprehensively considered.A Koopman model predictive control algorithm is then designed for the ultra-supercritical thermal power units.The effectiveness of the proposed KMPC method is validated through simulation experiments,with the local linear MPC(LMPC)serving as the benchmark.[Results]Corresponding to the unit's main steam pressure,separator steam enthalpy,steam turbine power generation of these three quantities,the results of this study are summarized as follows:(1)In the load-ramping simulation with successive step increases,the relative root mean square errors between the high-dimensional approximate model and the original nonlinear system outputs were 1.00%,0.40%,and 0.36%,respectively.(2)Under nominal operating conditions,the KMPC algorithm reduced the integral of time-weighted absolute error(ITAE)of the outputs by 46.67%,48.66%,and 21.46%compared with the LMPC algorithm in the load-increasing experiment.(3)Under model mismatch conditions,the KMPC algorithm reduced the ITAE of the outputs by 19.57%,22.45%,and 30.94%compared with the LMPC algorithm in the load-increasing experiment.[Conclusion]The above experimental results demonstrate that the high-dimensional linear approximation model constructed using the Koopman operator can accurately capture the nonlinear dynamic characteristics of the original system.Compared with LMPC algorithm,the KMPC algorithm provides a faster response and smaller steady-state errors during wide-range load variations.Moreover,it demonstrates enhanced robustness,which is beneficial for practical implementation in thermal power plant operations.