Study On Short-term Power Forecast Methodology Of Photovoltaic System

As photovoltaic impact on the power structure increasing, a reliable prediction information on the expected power production is important and necessary as a basis for power-system management, scheduling, and dispatching operations. The prediction error will directly affect the security and economy of grid operation after photovoltaic system connected. The improvement of precision of short-term PV generation forecasting plays a key role in the development of PV generation system incorporating the existing power grid and the utilization of solar energy.In this paper, the prediction methods of short-term PV power generation are discussed and analysis. The short-term forecast technologies are summarized into two categories. One is the physical forecast technology based on solar radiation and PV generation models, the other is big data(historical data of meteorological, solar radiation and output power) driven method. The development trend, merits and drawbacks, application and prediction accuracy of irradiance prediction methods, physical model forecast methods based on solar radiation and big data driven prediction methods from historic data on solar radiation, output power and meteorological data are classified and summarized.The grey theory method and the BP neural network with double hidden layer method are chosen to make prediction. Grey theory prediction model don't need a large number of data input, but its limitations is that the accuracy is unstable with the weather types of the similar days. BP neural network as its own characteristics, play an important role in the prediction of nonlinear systems with an ideal result of accuracy.Afterwards, PV-generation is easily affected by weather conditions, geographical environment and other factors with high degree of randomness and big fluctuation. Based on the data instances of the PV plant, the factors affecting of PV generation are analyzed in this paper. The environmental factors- solar radiation, season, weather, temperature are analyzed emphatically. And then the influence of above factors on the PV output is validated by the data.Finally, based on the data instance of PV plant, the prediction of PV generation is simulated, and the accuracy of the prediction results are compared and analyzed. Some non-quantitative parameters such as weather, season, temperature and material are virtualized to quantitative ones and used as the explanatory variables taken account into developing the a complex-environment-factor model combined with the method of multiple linear regressions and software called SPSS. A new hybrid model for daily output power forecasting of a PV system is introduced, which is combined with PV-generated power, irradiation amplitude and efficiency models. The specificity of this hybrid model focuses on the single quantitative data source of temperature and the quantization of that non-quantitative weather information into dummy variables, which are all involved in forecasting the daily energy production. Experimental results show that the hybrid model can achieve a feasible prediction. The RMSE of the output power model is 46%, and the RMSE of the solar radiation model is 42.9%. The multiple linear regression model is not good enough to cope with the PV generation with strong random and nonlinear.