Ultra-short-term Prediction Model For Solar Irradiance Of Photovoltaic Power Generation Based On Complex Network Theory

The output power of the solar photovoltaic(PV)plant is mainly determined by the received solar irradiance of the solar PV panels.The step-wise PV power forecast methods need to predict the irradiance,temperature,humidity,and other influencing factors that affect the photovoltaic power first,and then predict the power according to the influencing factors.Among them,solar irradiance is the most important factor affecting the PV power.Therefore,to improve the prediction accuracy of solar irradiance is of significance.However,the diversity of irradiance fluctuation laws poses a huge challenge for PV power prediction.Thus,it is necessary to mine the law of irradiance fluctuation to improve the prediction accuracy.At present,most of the research is based on the distance similarity of time series to match the fluctuation law but neglects the correlation of its fluctuation morphology,so it is difficult to guarantee the matching accuracy.Complex networks can map the non-linear time series into networks,and extract the fluctuation characteristics of data through the network's topol ogical structure.The intrinsic fluctuation patterns of irradiance series can be deeply excavated from the aspect of topological morphology.Therefore,this paper proposes a prediction method of ultra-short-term solar irradiance based on complex network th eory and machine learning,which uses complex network theory to match irradiance fluctuation patterns and constructs a prediction model through the BP neural network.This paper first studies the method of constructing complex networks.According to the current application of complex networks in time series analysis,the applicability of different complex network construction methods to the analysis of solar irradiance data is analyzed,and the coarse-graining algorithm is used to construct complex networks and excavate the fluctuation characteristics of solar irradiance.The fluctuation network of solar irradiance data is further constructed.By using the coarse-graining algorithm,the solar irradiance sequence is transformed into the corresponding fluctuation network,and the process of extracting the solar irradiance fluctuation mode by using the complex network is described.At the same time,the index of relative matching rate is defined,and the performance of this method and traditional data matching methods such as Euclidean distance in solar irradiance feature extraction is compared and analyzed.Finally,the ultra-short-term prediction model of solar irradiance is established.Under three different types of weather,using actual solar irradiance data,a direct prediction model,a prediction model based on traditional data matching methods,and a prediction model constructed in this paper were established through BP neural networks.The root mean square error(RMSE)and mean absolute error(MAE)are used to analyze the prediction results of each model,and the prediction results obtained by different complex network construction methods are compared.Simulation results under three weather types show that the prediction result of the ultra-short-term solar radiation prediction model constructed in this paper is better than the direct prediction,and compared with the two traditional data screening methods,Euclidean distance and Pearson correlation coefficient,the prediction result of the model in this paper is still better than the prediction result of the two methods.In terms of the construction method of the complex network,for the irradiance sequence,the coarse-grained algorithm can better explore the inherent fluctuation characteristics of irradiance compared with the visibility algorithm,so it has better prediction performance,thus verifying the validity and applicability of the model in this paper.