Predicting solar radiation at high resolutions: A comparison of time series forecasts

摘要: Abstract The increasing use of solar power as a source electricity has led to increased interest in forecasting radiation over short time horizons. relevant horizons for generation and transmission can range from little 5 minutes long several hours. Forecasting experiments are run using six data sets, at resolutions 5, 15, 30, 60 min, the global horizontal component. exhibits nonlinear variability, due variations weather cloud cover. Nevertheless, dominance 24-h cycle makes it straightforward build predictive models. tests regressions logs, Autoregressive Integrated Moving Average (ARIMA), Unobserved Components Transfer functions, neural networks, hybrid models also evaluated. All true out-of-sample forecasts: estimated history prior start forecast horizon, is forecasted, predicted values compared with actuals. In nearly all tests, best results obtained ARIMA time-varying coefficients. There some exceptions. At high resolutions, transfer function cover found improve ARIMA. few cases, net or very on order 5 min. success attributable mainly its ability capture diurnal more effectively than other methods.