Prediction of PV output based on local mean decomposition under limited information

Photovoltaic (PV) power generation's advantage of mitigating fossil energy consumption and environment friendly results in its widespread utilization. However, inaccurate PV output prediction will cause problems of dispatching, stability and reliability in power system. PV output is directly affected by the solar irradiance. Without solar irradiance monitoring facilities and only relying on the basic weather information and historical output, we develop a short-term PV output prediction model to achieve high-precision prediction through the data mining. Impacts of external factors on power generation, such as weather type, light duration, date and temperature, are considered and analyzed, and classification of historical data is accomplished according to weather type. The PV daily output can be considered as a nonlinear random time series. Therefore, the output series could be decomposed into relatively stable product function (PF) components based on Local Mean Decomposition (LMD). With weather information added into each PF, the Least Squares Support Vector Machine (LSSVM) prediction model is established. In this paper, as an additional factor, the daily maximum temperature is used in PV output prediction at noon. Results of the simulations show that without solar radiation data, the prediction based on LMD-LSSVM is practical and can better preserve the original physical characteristics of PV output series.