Research On PV Power Prediction Considering Environmental Factors And Hydro-PV Complementary Operation

Although photovoltaic power generation has significant economic and environmental benefits,due to the influence of weather conditions,the output power of photovoltaic system shows strong randomness,intermittence and volatility,which not only has a certain impact on the large-scale deployment of photovoltaic system,but also brings challenges to the safe operation and dispatching of power grid.In this paper,starting with the analysis of the factors affecting photovoltaic power,a photovoltaic power prediction model based on hybrid deep learning is established,and The influence of photovoltaic power prediction accuracy on the operation of water-photovoltaic complementary system is analyzed.The main contents and research results of this paper are as follows:Firest,the influencing factors of PV power output are analyzed and studied.First of all,the results of measuring the module back temperature are corrected to ensure the validity and accuracy of the data.Secondly,the factors affecting module temperature and PV power are qualitatively and quantitatively analyzed by drawing curve,grey correlation analysis and multiple regression analysis based on R language.The grey correlation between solar irradiance,module temperature and PV power is more than 0.8,which together explain 91% of the variance in power.The grey correlation between solar irradiance,atmospheric temperature and module temperature is more than 0.8,which together explain 81% of the variance in module temperature.Finally,the historical power data are interpolated and extrapolated,the outliers are eliminated and a smoother power curve is obtained.Second,the application of prediction model is analyzed and studied.Based on the analysis of the second chapter,a PSO-ANFIS module temperature prediction model with solar irradiance and atmospheric temperature as input variables and a hybrid deep learning PV power prediction model with solar irradiance and module backplane temperature as input variables are established.The power prediction model is composed of CNN-LSTM and DBN integrated by GA-BPNN.The super parameters of neural network are determined by grid search method.K-fold cross verification is used in the model training process,which makes full use of all samples and improves the generalization ability of the prediction model.The parallel structure is introduced into CNN-LSTM,the input of CNN-LSTM is the component of historical power time series decomposition,the input of DBN is solar irradiance and module temperature,and the input variable of HDLM contains historical power information and meteorological data.The experimental results show that the module temperature prediction model and power prediction model proposed in this paper are better than other comparison models,and a significantly improve can be seen in the prediction accuracy of target variables.Third,The application and influence of different photovoltaic power prediction techniques in the operation of water-photovoltaic complementary system is compared.A short-term operation model of Lijiaxia-NJ Hydro-PV complementary system is established to minimize the fluctuation of residual load.The results show that in cloudy and rainy days,The net load fluctuation,system residual load fluctuation,system output fluctuation and reservoir water level fluctuation of the Hydro-PV complementary system based on CNN-LSTM are obviously smaller than those of BPNN,Lijiaxia is easier to complement CNN-LSTM-based photovoltaic output.Compared with BPNN,Lijiaxia units based on CNN-LSTM have less times of passing through the prohibited operation zone and fewer times of unit start-up and shutdown.