Research On Power Quality Of Photovoltaic Grid-Connected System Based On Deep Learning

Under the support of carbon peaking and carbon neutrality policy,China has increased the development and utilization of photovoltaic energy,built a large number of photovoltaic power stations,and increased the penetration rate of photovoltaic in the power grid,but at present,there are still many shortcomings in photovoltaic power generation.First,photovoltaic power generation is greatly affected by external environmental factors,and has strong random fluctuations and control complexity.Second,after being integrated into the power grid,the harmonics of the system show the characteristics of wide frequency domain and high frequency,which makes the power quality problem very serious.Third,the exponential increase in the number of a large number of nonlinear loads leads to a more prominent problem of harmonic pollution,which will not only affect the normal operation of each power equipment,but also lead to an increase in the probability of series and parallel resonance,endangering the safe operation of the power system.Therefore,it is of great significance to accurately and quickly control the power quality of photovoltaic gridconnected systems.This paper focuses on the power quality problem of photovoltaic grid-connected,combines the power quality analysis method with deep learning,and focuses on the study of its data law.The study is conducted on two core issues.The first is to explore the power quality change law of harmonics and voltage of photovoltaic grid-connected system under the actual temperature and light intensity changes,and improve the traditional harmonic detection ip-iq method.The second is to accurately predict power quality based on the improved deep learning model,and use parallel active power filter for harmonic compensation verification.The main research results are as follows:Firstly,based on the existing ip-iq harmonic detection method,an interpolated differential harmonic detection method is designed,which aims to reduce the data redundancy of power quality and retain its intrinsic characteristics,and improve the characteristic analysis ability of deep learning on power quality data.A two-stage threephase photovoltaic grid-connected system was built on the Matlab/Simulink platform,and the Longdong region with abundant sunshine resources and great development potential was selected as the sample basis,and the measured light intensity and temperature data of its photovoltaic power station were selected for experimental simulation,which simulated the grid-connected photovoltaic system under three weather conditions: sunny,cloudy and cloudy,explored the characteristics of system power quality in different environments,and specifically analyzed the influence of light intensity and temperature on current distortion and voltage overrun.Secondly,an improved Bi-Long-Short-Term Memory network(Bi-LSTM)is proposed,which is applied to the analysis and prediction of harmonic current data,which provides a scheme for power quality monitoring and harmonic governance.Firstly,the deep learning methods with different structures are analyzed,and the Recurrent Neural Network(RNN)with good analysis ability for time-series harmonics is selected as the basic deep learning model.Deep learning models such as BP,LSTM,GRU,BI-LSTM are built on the Matlab platform,and deep learning analysis is performed on the optimized harmonic data.Secondly,the grid search method introduced into the Bi-LSTM algorithm is improved,and four sets of Batch size and three sets of Learnrate parameters are set for grid search automatic parameter optimization.Then,the magnitude and iteration speed of the loss function value verify the rapidity and accuracy of the improved Bi-LSTM method,which effectively improves the monitoring of power quality and can accurately predict the change of future harmonics,which plays an early warning role.Finally,the shunt active power filter is used to compensate the predicted harmonic data,and the effectiveness of the scheme is demonstrated according to the harmonic Fourier analysis before and after compensation.