DC Component Suppression Strategy For Grid-connected Inverter Based On Neural Network

In order to alleviate the current energy shortage,the microgrid that can exchange renewable energy into the grid for energy exchange is gradually being promoted.The grid-connected inverter is an important component to ensure the stable operation of the microgrid.It is imperative to study the grid-connected inverter.In this paper,the DC(direct current)component injection problem caused by transformerless three-phase LCL grid-connected inverter is taken as the research object,and the causes of the DC component and its suppression method in grid-connected inverter are studied.In this paper,the mechanisms of the DC component in the grid-connected inverter are analyzed.According to the circuit topology of the three-phase grid-connected inverter,the corresponding grid-connected current mathematical models are established,and the generation mechanisms of the DC component of the non-isolated grid-connected inverter under different reasons are analyzed,including: power switch drive signal delay,current sensor signal acquisition error,power switching device asymmetry,grid voltage imbalance,and so on.In this paper,a sliding window quadratic integration method is proposed to detect the DC component in the grid-connected current in real time.Since in actual applications,the grid frequency will be slightly offset at a power frequency point.Therefore,this paper expands on the basis of the sliding window one-time integration method,and integrates the result of one integration again to obtain the result of the corresponding second integral.The detection result of the secondary integration of the sliding window is delayed by one power frequency period compared with the one-time integration of the sliding window,but the steady-state error of the extracted DC component waveform is greatly reduced.Moreover,the method for detecting the DC component of the grid-connected current proposed in this paper does not need to add an additional hardware detection circuit,which is very easy to implement.This paper proposes a DC component suppression method based on neural network.This paper draws on the basic idea of the DC component compensation method in single-phase grid-connected inverters,and extends it to three-phase grid-connected inverters.Firstly,the difference method is used to eliminate the zero drift error in the three-phase grid-connected current.Secondly,the sliding window secondary integration method is used to detect the DC component in the grid-connected current in real time.Then,the DC component is sent to the DC component compensation control.The compensation amount of the corresponding DC component is obtained in the device,and the compensation amount is superimposed on the output of the grid-connected current controller to adjust the driving signal of the grid-connected inverter power switch tube,and finally the DC component is suppressed.The neural network is introduced into the suppression of DC components,and a BP(Back Propagation)-PID controller is proposed as the compensation component of the DC component.The strong nonlinear approximation ability and learning ability of the neural network are used to dynamically adjust the PID parameters,which greatly optimizes the problem that the DC component dynamic adjustment time and overshoot are large in the traditional PID controller with fixed parameters.Finally,the neural network-based DC component suppression method is verified by MATLAB/Simulink simulation.This paper presents DC component suppression method based on an improved neural network.In the original BP-PID algorithm,the learning rate is a fixed constant.The choice of the learning rate parameter will affect the convergence speed of the system.This paper proposes an adaptive learning rate adjustment algorithm,which increases or decreases the learning rate by observing the increase or decrease of the DC component evaluation function value reflecting the current time value and the changing trend of the DC component.The adjustment step size of the neural network weight increment is correspondingly increased or decreased,and finally the performance of the algorithm is improved,so that the improved neural network-based DC component suppression method has a shorter DC component dynamic adjustment time and smaller overshoot.Finally,the improved neural network-based DC component suppression method is verified by MATLAB/Simulink simulation.Finally,based on the hardware circuit design and software program design,the 2kW three-phase LCL grid-connected inverter experimental prototype is built.By comparing the grid-connected current waveforms before and after the DC suppression strategy using BP-PID controller,the experimental results are obtained,and the feasibility of the proposed DC suppression scheme is verified.