Three-Level Active Power Filter Based On Model Predictive Control And Its Parameter Identification

As a large number of nonlinear devices have been put into industrial production and daily life,power quality management has become the focus of today’s research.As a typical harmonic source,nonlinear equipment will not only reduce the power quality of the power grid,bring harmonic pollution,but also affect the normal operation of the power system,resulting in security risks and property losses.Active Power filter(APF),as a new type of power electronic device,can compensate the harmonics of varying size and frequency.In this thesis,three-level APF with diode clamping is taken as the research object.First,the current research status of APF,current control and parameter identification is analyzed,the power topology structure of three-level APF is given,and the working mechanism is deeply analyzed,the relevant mathematical model is established,and the point potential balance of the key technology is deeply studied.Secondly,the control strategy of APF is analyzed.Based on the three-level mathematical model,the prediction model and value function of model predictive control(MPC)are established.In view of the problems of traditional model predictive control(MPC)algorithm,such as large computation amount,difficult balance of midpoint potential and difficult setting of weight factor of value function,an MPC algorithm based on sector vector judgment is proposed.Among th em,the algorithm based on the small vector judgment balances the capacitor voltage by means of the opposite effect of the small vector on the midpoint potential,so that the value function only needs to consider the current tracking effect,without compli cated multiobjective weight renormalization,and reduces the number of rolling optimization from 27 times to 21 times.In the algorithm based on sector judgment,the optimization times are further reduced to 8 times by introducing sector judgment.After t he redundant vector is omitted,the algorithm only needs to perform 7 iterations,which greatly reduces the computation amount of the algorithm.Thirdly,in order to improve the real-time performance of the algorithm,the gradient descent algorithm is added to identify the parameters in real time.Considering the disadvantages of the traditional gradient descent algorithm such as low identification accuracy,unstable identification process and easy to fall into local extreme points,the momentum gradient al gorithm and Nesterov gradient algorithm(NAG)are introduced.The momentum gradient is optimized in real time by integrating the historical gradient and the current gradient,so that the algorithm can break out of the extreme point,while the NAG gradient algorithm is updated in real time by integrating the historical gradient and the future gradient,so as to improve the stability and accuracy of the algorithm.Fourthly,a three-level APF system simulation platform is built through MATLAB to verify the identification accuracy of three gradient descent methods.The results show that NAG algorithm has higher stability and accuracy in parameter identification;At the same time,the simulation results of the three model prediction methods show that the model pr ediction method based on sector vector judgment can adapt to a variety of working conditions,and has higher compensation performance and capacitance midpoint potential balance capability.Finally,the three-level APF system was designed and the experimen tal platform was built.The correctness and feasibility of the three-level APF system and its parameter identification system based on the sector vector judgment model were proved through the experimental verification.