EU PVSEC Programme Online
35th EU PVSEC 2018, 24 - 28 September 2018, Brussels
Presentation: 5BV.1.9 Wide-Range Solar Resource Forecasting Using Radiation Measurements, All-Sky Camera Imagery and High-Resolution Large-Eddy Simulations on a GPU
Type: Visual
Date: Tuesday, 23rd September 2014
08:30 - 10:00
Location / Room: Hall 2 / Poster Area
Author(s): A. Los, H. Jonker, J. Schalkwijk, S. de Roode, T. Zinner
Presenter / Speaker: A. Los, EKO Instruments, The Hague, Netherlands et al.
Event: Conference Conference
Session: 5BV.1 Operation of PV Systems and Plants
Type(s) of Access:  Conference Registration
Topic: 5. 1 Operation of PV Systems and Plants
Summary / Abstract: With the current tendency to larger photovoltaic power plants the demand for accurate forecasts of solar radiation increases steadily. In contrast to the traditional, controllable power generation that can handle the power grid variability, renewables must cope with natural power resource fluctuations. To maintain grid stability with an increasing penetration of renewables, the variability of natural power sources has to be accounted for in the operation of the renewables by accurate forecasts of natural energy resources, minutes to days ahead. ! We present two forecasting methods based on novel techniques. For short time-scale predictions up to 15-30 minutes we use atmospheric observations performed with an all-sky camera and solar radiometers, and for longer time scales we use a highresolution Large-Eddy Simulation (LES) model. These methods allow us to cover forecasting time horizons from one minute to several days ahead. !F irst we discuss results of solar radiation forecasts based on atmospheric observations. Clouds are observed with an all-sky camera and their effects on the solar radiation are monitored by means of ground-based broadband radiometers. These measurements are processed by a sophisticated cloud tracking algorithm to derive cloud motion vector fields. The cloud motion vector fields are used in turn to predict solar radiative fluxes at the surface every minute up to 15 minutes ahead. The evolution of clouds scenes is analysed in order to differentiate between advective and convective clouds so that we are able to choose the appropriate forecasting method. In the convective cloud case we apply a so-called morphing method which analyses the optical flow of cloudy pixels on several spatial scales. In the advective case cloud motion is based on the area-averaged cloud displacement. In both cases the motion vector field is then applied to the latest cloud mask allowing cloud position and radiation forecasts up to 15 minutes. Statistical analysis reveals that cloud position errors for 10 minutes forecasts range between 10% and 35% on a 1-sigma confidence level. Similarly, errors for cloud cover range between +/- 15%. Although radiation forecast errors manifest itself in the same way as cloud position errors, the radiation forecast accuracy depends strongly on cloud type, cloud layers, cloud displacement velocity, etc. Therefore, the presentation will elaborate on various aspects of radiation forecasting reliability by means of cloud classification. ! The second forecasting technique fully relies on intensive numerical calculations with the Graphics Processing Unit ( GPU) resident Atmospheric Large-Eddy Simulation (GALES) model. For this application the GPU offers superior computational speed compared to the more traditional CPU. The lateral boundary conditions as well as the large-scale forcing conditions are obtained from the KNMI Regional Atmospheric Climate Model (RACMO). In contrast to most regional weather forecast models GALES operates at a horizontal grid resolution that is sufficiently fine to explicitly resolve boundary-layer clouds such as shallow cumulus and stratocumulus. Currently GALES runs on a continuous basis in real-time forecasting mode and for re-analysis studies. Because the horizontal domain of GALES includes the Cabauw meteorological measurement platform operated by the KNMI, the GALES results can be critically compared against observations.