Photovoltaic Power Output Forecasting Method Research For Weakly Related Meteorological Data Problem

With the growing pressure caused by the exhaustion of fossil fuel, the use of renewable energy sources are gaining more and more attention. With features of inexhaustion and free of pollution, solar energy is becoming one of the most valuable and potential energy type. At present, the photovoltaic (PV) industry in China has developed by leaps and bounds. However, the PV generation is severely affected by weather conditions, showing obvious randomness and fluctuation, which bring impact to the safe and stable operation of power system. So accurate forecasting of PV output is extremely important.The performance of forecasting methods which are based on traditional statistics can only be guaranteed by sufficient samples. But now in our country, weather measurement devices are not perfect enough, leading to the lack of comprehensive meteorological information which brings inconvenience to distributed PV output prediction. For centralized PV power plant, installing meteorological measurement device in every plant will cause large investment and complex operation. Also, if the same weather information are used by adjacent PV power plants, the test data will be inaccurate. So it is bound to face the problems such as the reduction of data correlation, large sample noise and unsatisfied data, causing the number of effective samples to decrease and the applicability of traditional statistical method to reduce.In this paper, PV output characteristics of periodicity, intermittence, fluctuation and randomness were analyzed. Through the analysis of correlation between PV output data and local historical meteorological data, the weak correlation was verified. PV output forecasting method aiming to solve the problem of weakly correlation between meteorological and generation data was studied, and two forecasting methods based on support vector machine were proposed in this paper. First of all, strong correlation factors were selected as the input factors of gray correlation analysis, and short-term PV output forecasting method based on similar days selection was constructed. Then volatility series of PV output was decomposed into a number of components by empirical mode decompositiom method, and ultra-short term PV output forecasting method based on empirical mode decomposition was constructed. Finally, tests were given to verify the validity and rationality of the proposed methods.