Towards the intrahour forecasting of direct normal irradiance using sky-imaging data

Abstract

Increasing power plant efficiency through improved operation is key in the development of Concentrating Solar Power (CSP) technologies. To this end, one of the most challenging topics remains accurately forecasting the solar resource at a short-term horizon. Indeed, in CSP plants, production is directly impacted by both the availability and variability of the solar resource and, more specifically, by Direct Normal Irradiance (DNI). The present paper deals with a new approach to the intrahour forecasting (the forecast horizon normalΔtf is up to 300.25emmin ahead) of DNI, taking advantage of the fact that this quantity can be split into two terms, i.e. clear-sky DNI and the clear sky index. Clear-sky DNI is forecasted from DNI measurements, using an empirical model (Ineichen and Perez, 2002) combined with a persistence of atmospheric turbidity. Moreover, in the framework of the CSPIMP (Concentrating Solar Power plant efficiency IMProvement) research project, PROMES-CNRS has developed a sky imager able to provide High Dynamic Range (HDR) images. So, regarding the clear-sky index, it is forecasted from sky-imaging data, using an Adaptive Network-based Fuzzy Inference System (ANFIS). A hybrid algorithm that takes inspiration from the classification algorithm proposed by Ghonima et al. (2012) when clear-sky anisotropy is known and from the hybrid thresholding algorithm proposed by Li et al. (2011) in the opposite case has been developed to the detection of clouds. Performance is evaluated via a comparative study in which persistence models – either a persistence of DNI or a persistence of the clear-sky index – are included. Preliminary results highlight that the proposed approach has the potential to outperform these models (both persistence models achieve similar performance) in terms of forecasting accuracy: over the test data used, RMSE (the Root Mean Square Error) is reduced of about 200.25emW0.2emm−2, with normalΔtf=150.25emmin, and 400.25emW0.2emm−2, with normalΔtf=300.25emmin.

Benchmarking of solar irradiance nowcast performance derived from all-sky imagers

“…Similarly, Chu et al [57] used as inputs to ANN the time-lagged GHI, total sky cloud coverage and cloud indices. ANNs have been widely used to nowcast either GHI [58][59][60][61] or DNI [62]. Recently, Kamadinata et al [59] directly utilized the color information captured by an ASI providing adequately good nowcasts (up to 5 min) with less computational effort.…”

Benchmarking of Deep Learning Irradiance Forecasting Models from Sky Images -- an in-depth Analysis

“…An in-depth analysis of individual components of the above methods and a review of the current literature on the topic are presented in Kuhn (2019) and Yang et al (2018), respectively. However, challenges in irradiance forecasting remain: our perception of the complexity of cloud physical properties from sky images, but also their spatial and temporal dynamics, is still limited (Brad and Letia, 2002;Nou et al, 2018).…”

Comparison between Time- and Observation-Based Gaussian Process Regression Models for Global Horizontal Irradiance Forecasting

“…Depending on the available data and the forecast horizons, different forecasting methods are adequate. Intrahour solar irradiance forecasts with a high spatial and temporal resolution may be provided by models using ground-based sky images [7][8][9]. For forecast horizons ranging from a few minutes to 6 h, statistical models with on-site solar irradiance observations as input are more suitable [10,11].…”