Research On The Strategy Of Model Predictive Control For Modular Multilevel Converter

The parameter design and basic control theory of modular multilevel converter(MMC)have been mature.The single-port STATCOM MMC and the dual-port AC/AC MMC have also been put into use in the traction power supply system of high-speed railway.However,with the development of transportation electrification,the traction power sup-ply system is facing higher and higher demand,which poses new challenges to the opera-tion control strategy of MMC.Taking MMC as the research object,this paper focuses on the strategy of model predictive control,in order to obtain the MMC control method with high universality,good dynamic characteristics,superior harmonic characteristics,low calculation requirements,strong parameter mismatch robustness and strong white noise robustness.Firstly,based on the mathematical model of dual-port AC/AC MMC,the dynamics of common-mode and differential-mode capacitor voltages are derived,and the decoupling of arm capacitor voltage balance control is realized in the frequency domain.By injecting circulating current components with different frequencies into the AC2 side,the proposed method realizes the balance control of the upper and lower arm capacitor voltages.Then,the basic operation control strategy of single-port STATCOM MMC is deduced,and the mapping relationship between single-port and dual-port MMC control logic is established by comparative study,which lays a foundation for the universality of model predictive control algorithm in the following chapters.The unification of control logic also means that controllers designed based on linear control theory face common shortcomings,such as complex structure caused by cascade scheme,setting difficulties caused by too many parameters,and poor dynamic performance,which limit the further development of MMC performance.Aiming at the above shortcomings,model predictive control scheme converts the control problem of MMC into an optimization problem.According to the difference of op-timization and control objectives,this paper designs model predictive current control strat-egy(MPCC),model predictive current and capacitor voltage control strategy(MPCVC)for the AC/AC MMC per-phase model.The model predictive control problem can be transformed into a convex optimization quadratic programming problem by designing ap-propriate optimization functions.In order to meet the requirements of real-time control,for the specific constrained quadratic programming problem in this paper,a special geo-metric solution method suitable for online operation is proposed,and its global optimality is mathematically proved.To overcome the problem that the performance of capacitor voltage balance is improved but the current tracking accuracy is reduced when MPCVC is used in per-phase model,an enhanced model predictive current and capacitor voltage control strategy(EMPCVC)for three-phase MMC system is proposed.The new opti-mization problem of capacitor voltage balance control is constructed with the minimum current tracking error as the constraint condition,and the tracking accuracy of current is guaranteed while the performance of capacitor voltage balance is improved.The strat-egy research and solution design of model predictive control lay the foundation for the follow-up robustness problem research.In order to solve the problem that conventional model predictive control is sensi-tive to model parameter errors,robust model predictive control algorithms for parameter mismatches are designed.Firstly,the influence path of parameter mismatches to system performance is analyzed.From the direct to the fundamental,this paper designs robust model predictive control strategy based on voltage compensation(VC based RMPC),ro-bust model predictive control strategy based on predictive current compensation(PCC based RMP)and robust model predictive control strategy based on parameter identifica-tion method(PIM based RMPC)from the different levels,output voltage,predictive state and model parameters.The VC based RMPC scheme adopts parallel controller to cor-rect the influence caused by parameter mismatches.As it does not need to modify the traditional model predictive controller itself,it has low deployment cost.The PCC based RMPC scheme improves the robustness by compensating the prediction error,and deduces the on-line learning algorithm of control parameters according to the optimization theory,and a designed Lyapunov function proves the stability of this algorithm.The PIM based RMPC strategy uses the results of parameter identification instead of nominal parameters to predict the state and obtain the optimal voltage.Finally,the above three RMPC strate-gies are compared and analyzed horizontally,and the experimental verification is carried out,which lays a foundation for the further fusion of white noise robustness.Different from the regular disturbance caused by parameter mismatches,both sam-pling noise and process noise are white noise with probability distribution,which cannot be accurately modeled or predicted.In view of the influence of system white noise on model predictive control,the optimization problem containing white noise is described mathe-matically,and the optimal solution of this problem is given based on Kalman filter theory(KF).Furthermore,the KF based RMPC is designed by fusing KF with model predictive control.By further modelling the regular disturbance caused by parameter mismatches,the robust model predictive control strategy based on extended state model-Kalman filter(ESM-KF based RMPC)is proposed.The algorithm uses KF to observe the system state and extended state,which not only enhances the robustness of white noise but also takes into account of parameter mismatch robustness.The RMPC strategy based on PIM-AKF is designed by combining parameter identification method and adaptive Kalman observer.In this strategy,parameter identification is introduced before observation and control to enhance the robustness of parameter mismatch.At the same time,the introduction of adaptive KF not only reduces the adjustment requirements of covariance matrix,but also achieves the better robust performance.Finally,three RMPC strategies integrating KF are compared and analyzed,and the RMPC strategy integrating KF theory is verified experi-mentally.