←Back
247 – Advances in Spatio-Temporal Modeling
Cyclostationary Models of Solar Irradiance
Sponsor: Section on Statistics and the Environment
Keywords: Time Series Analysis, Cyclostationarity, Prediction, Spectrum Analysis
Charlotte Haley
Argonne National Laboratory
Mihai Anitescu
Argonne National Laboratory
In this study we model solar irradiance at ground based sites in the continental US in order to predict into the future. The obvious nonstationary, non-Gaussian data have nightly zeros, or slightly negative values, producing dual frequency spectra with large values on the diagonals with offsets of 24 hours. Analysis of this particular covariance structure implies a a non-Gaussian process model with cyclostationary noise structure. Fitting stationary vs cyclostationary models reveals simpler and more accurate results are obtained using the cyclostationary framework.