Generation Forecasting For Photovoltaic System Based On Artificial Neural Networks

With the energy crisis and environmental situation is increasingly serious,Photovoltaic power generation which is clean, green, renewable, etc. has muchnational attention. In recent years, driven by policy, photovoltaic industry has a rapiddevelopment, photovoltaic power plants has gradually increased in the proportion ofpower system. However, due to the power generation affected by many factors, theoutput has a significant cyclical output fluctuations and randomness. Its paralleloperation make it is an uncontrollable source relative to power system, it will affectthe safe and stable operation of the power system when large-scale photovoltaicpower plants parallel operation. Therefore, Research on generation forecasting forphotovoltaic system has great theoretical and practical significance.Firstly, the working principle of the photovoltaic cell and classification ofphotovoltaic power generation system were discussed, detailed composition ofgrid-connected photovoltaic power generation system, given the basic structure andthe basic parameters of State Key Laboratory of Alternate Electrical Power Systemwith Renewable Energy Sources in North China Electric Power University. Then, thecharacteristics of photovoltaic power generation systems was analyzed, consideringradiation intensity, ambient temperature, solar cell temperature and relative humidityas the main factors which affect the power generation of photovoltaic systems,making the theoretical support for the establishment of power generation forecastingmodel.Artificial neural networks as a widely used mathematical tool was widespread inthe application of photovoltaic power forecast. This paper has described the basicprinciples of BP neural network, wavelet neural network, echo state networks andempirical mode decomposition-echo state networks, and established four kinds ofthe corresponding photovoltaic power generation forecasting model.Finally, using the measured data of State Key Laboratory of Alternate ElectricalPower System with Renewable Energy Sources in North China Electric PowerUniversity, the experimental verifications on four kinds of forecasting models weretaken. The results show that four kinds of forecasting models established can predictaccurately, and the empirical mode decomposition-echo state networks whichcombines the advantages of the two algorithms, has higher prediction accuracy andbetter performance and stability.