Photovoltaic Power Generation Output Forecasting Based On Improved Neutral Network

With the enhancement of people’s awareness of environmental protection besides the great support of the national policy, and the maturing of the photovoltaic power generation technology besides the gradually reducing of the cost of photovoltaic power generation, photovoltaic power generation has become one of the most commercial development potential renewable energy industry currently. But the output power of photovoltaic system is easily affected by environmental factors, which leads to highly intermittence, volatility and uncertainty. The large-scale connect of photovoltaic and grid will impact on the power grid, causing bad influence on the security, stability and economic operation of power grid. Predicting the output of photovoltaic power accurately can achieve friendly interconnection between photovoltaic system and grid, improve the power grids’ ability of accepting photovoltaic power, realize the optimal scheduling and stable operation, promote the construction of photovoltaic system on large-scale.This paper is to study the short-term photovoltaic power generation output forecasting technology. This paper briefly introduces the background of photovoltaic power generation and the domestic and foreign research status of the photovoltaic output forecasting technology. Based on the example of a photovoltaic system, a detailed analysis of the relationship between the photovoltaic output and season, weather type, temperature and other meteorological elements is presented. The paper elaborates the revision method of the original photovoltaic output data and meteorological data; and presents two methods for sample selection, one is the sample selection method based on the temperature of the similar days, the other one is the sample selection method based on phase space reconstruction. The paper details the grey GM forecasting method and BP artificial neural network forecasting method, combination forecasting method and puts forward an improved neural network which based on Legendre orthogonal basis function and derivative algorithm. Combined with the actual photovoltaic power generation system, 4 typical weather types are put forward for example analysis. By comparing the forecasting results and evaluated by MAPE and RMSE, this paper proves that the improved neural network model and the combination model in the short-term photovoltaic output prediction is practical and effective.