Algorithms and computational processes implementation for improving the solar radiation forecast reliability

Supervised by:
  1. Joaquín Tovar Pescador Director
  2. Rafael Jesús Segura Sánchez Co-director
  3. Antonio David Pozo Vázquez Co-director

Defence university: Universidad de Jaén

Fecha de defensa: 19 January 2015

  1. Francisco José Olmo Reyes Chair
  2. María José del Jesús Díaz Secretary
  3. Andreas Kazantzidis Committee member

Type: Thesis

Teseo: 415384 DIALNET


This thesis aims at the development of computational algorithms and tools for solar radiation prediction in the very short-term, ranging from few minutes up to 6 hours ahead. In the context of solar radiation, this forecasting period is referred to as nowcasting. The use of high-quality data from state-of-the-art measuring techniques is essential for this purpose. Reliable solar radiation nowcasts are seminal to address some of the challenges posed by the current state-of-the-art technologies for solar power production. A major underlying problem is that the high spatio-temporal variability of weather patterns make solar power production highly fluctuating in both space and time, what may cause disturbances in electrical networks with high solar shares. Solar radiation forecasts contribute to mitigate these problems by anticipating the fluctuations of solar power production and allowing the schedule of counter-measurements. Clouds are the main source of solar radiation variability. The current state-of-the-art techniques for cloud physical properties and cloud coverage retrieval are based on satellite and sky-camera imagery. The approaches for solar radiation nowcasting are largely based on these datasets. Other cloud-retrieval techniques based on active remote sensing from ground stations using instruments, such as lidars and ceilometers, are starting to surface. However, the reliability of these measuring techniques and the forecasting methods have not coped yet with the solar industry requirements. One of the main purposes of this thesis is thus the development of improved cloudcoverage estimations and forecasting methods by processing both sky-camera and satellite images in a computationally efficient manner. This efficiency is a required condition to provide near-real time forecasts. The potential improvement achieved in the reliability of the solar forecasts by incorporating additional information from ceilometers and numerical weather prediction (NWP) models is also explored. A new computationally-efficient cloud-tracking method -hereinafter referred to as Sector Method (SM)- for intra-hourly DNI forecasting based on the processing of total-sky imager (TSI) images is presented and evaluated. After a revision and enhancement of the standard pixel-to-irradiance methods for MSG (Meteosat Second Generation) imagery, an improved and adapted version of SM is presented for its use with MSG images for both GHI and DNI forecasting (up to 6 hours ahead). Later, a new hybrid forecasting method -referred to as the WRF (Weather Research and Forecasting) Advection methodthat combines the use of MSG images and wind speed forecasts derived from the WRF NWP model is presented and evaluated. The SM and WRF-Advection methods are benchmarked against the WRF model radiation forecasts and smart persistence. Finally, an optimized method to derive GHI from satellite images based on an ANN ensemble model that exploits the MSG channels multi-spectral information is proposed and tested.