Research On Control Strategy Of PV Grid-connected Power Generation System

In order to alleviate the increasingly serious energy crisis and environmental problems caused by traditional energy consumption,the application of new energy has become the focus of development and research in the world.Solar energy has many advantages,such as clean,safe,renewable and so on.The conversion of solar energy into electric energy is an important form of its utilization.Therefore,the research of photovoltaic grid-connected power generation system is of great practical significance.Aiming at the shortcomings of traditional maximum power point tracking algorithm,this paper improves the algorithm of BP neural network,and proposes a new prediction model based on improved neural network.The algorithm adopts L-M(Levenberg-Marquardt)algorithm to improve the convergence speed greatly.It uses genetic algorithm to optimize BP neural network and avoid falling into local optimum.In view of the nonlinear and low precision of the predicted voltage,an approximate linear prediction model is proposed,which is more suitable for the online prediction of the neural network.Compared with the traditional MPPT algorithm,the proposed model has fast tracking speed and high precision,and can better track the maximum power point in real time.The traditional double closed loop control strategy has the disadvantages of slow response speed,high harmonic distortion rate(THD)and poor robustness,this paper designs a compound form of FADRC to improve the control effect and anti-interference capability of the grid-connected system.The controller combines the advantages of fuzzy control and ADRC.It not only makes use of the characteristics that the ADRC has strong anti-interference ability,but also uses fuzzy control algorithm to adjust the ADRC parameters online,improving the adaptive ability of ADRC.The simulation results show that the FADRC,which does not rely on the precise model of the system,has good control effect and anti-interference ability.In view of the shortcomings of traditional islanding detection,this paper proposes an islanding detection technology based on the energy characteristics of wavelet packet and neural network.The wavelet packet transform with better time-frequency characteristics is used to decompose the collected voltage and current signals to extract the energy eigenvalues respectively.Then,the information fusion technology is used to fuse the decomposed energy eigenvalue matrix and reduce dimension.An optimized BP neural network by genetic algorithm is used to identify the state and determine whether an islanding effect has occurred.Compared with the traditional method of islanding detection,this method has high accuracy,small detection area and no influence on power quality.When the islanding phenomenon occurs,the detection method can be detected quickly and accurately,with high reliability.