Search

With the fast increase of solar energy plants, a high-quality short-term forecast is required to smoothly integrate their production in the electricity grids. Usually, forecasting systems predict the future solar energy as a continuous variable. But for particular applications, such as concentrated solar plants with tracking devices, the operator needs to anticipate the achievement of a solar irradiance threshold to start or stop their system. In this case, binary forecasts are more relevant. Moreover, while most forecasting systems are deterministic, the probabilistic approach provides additional information about their inherent uncertainty that is essential for decision-making. The objective of this work is to propose a methodology to generate probabilistic solar forecasts as a binary event for very short-term horizons between 1 and 30 min. Among the various techniques developed to predict the solar potential for the next few minutes, sky imagery is one of the most promising. Therefore, we propose in this work to combine a state-of-the-art model based on a sky camera and a discrete choice model to predict the probability of an irradiance threshold suitable for plant operators. Two well-known parametric discrete choice models, logit and probit models, and a machine learning technique, random forest, were tested to post-process the deterministic forecast derived from sky images. All three models significantly improve the quality of the original deterministic forecast. However, random forest gives the best results and especially provides reliable probability predictions. [ABSTRACT FROM AUTHOR]

In the realm of solar forecasting, it is common to use a clear sky model output to deseasonalise the solar irradiance time series needed to build the forecasting models. However, most of these clear sky models require the setting of atmospheric parameters for which accurate values may not be available for the site under study. This can hamper the accuracy of the prediction models. Normalisation of the irradiance data with a clear sky model leads to the construction of forecasting models with the so-called clear sky index. Another way to normalize the irradiance data is to rely on the extraterrestrial irradiance, which is the irradiance at the top of the atmosphere. Extraterrestrial irradiance is defined by a simple equation that is related to the geometric course of the sun. Normalisation with the extraterrestrial irradiance leads to the building of models with the clearness index. In the solar forecasting domain, most models are built using time series based on the clear sky index. However, there is no empirical evidence thus far that the clear sky index approach outperforms the clearness index approach. Therefore the goal of this preliminary study is to evaluate and compare the two approaches. The numerical experimental setup for evaluating the two approaches is based on three forecasting methods, namely, a simple persistence model, a linear AutoRegressive (AR) model, and a non-linear neural network (NN) model, all of which are applied at six sites with different sky conditions. It is shown that normalization of the solar irradiance with the help of a clear sky model produces better forecasts irrespective of the type of model used. However, it is demonstrated that a nonlinear forecasting technique such as a neural network built with clearness time series can beat simple linear models constructed with the clear sky index. [ABSTRACT FROM AUTHOR]

In this work, we assess the performance of three probabilistic models for intra-day solar forecasting. More precisely, a linear quantile regression method is used to build three models for generating 1 h-6 h-ahead probabilistic forecasts. Our approach is applied to forecasting solar irradiance at a site experiencing highly variable sky conditions using the historical ground observations of solar irradiance as endogenous inputs and day-ahead forecasts as exogenous inputs. Day-ahead irradiance forecasts are obtained from the Integrated Forecast System (IFS), a Numerical Weather Prediction (NWP) model maintained by the European Center for Medium-RangeWeather Forecast (ECMWF). Several metrics, mainly originated from the weather forecasting community, are used to evaluate the performance of the probabilistic forecasts. The results demonstrated that the NWP exogenous inputs improve the quality of the intra-day probabilistic forecasts. The analysis considered two locations with very dissimilar solar variability. Comparison between the two locations highlighted that the statistical performance of the probabilistic models depends on the local sky conditions. [ABSTRACT FROM AUTHOR]

This paper aims at assessing the accuracy of different solar forecasting methods in the case of an insular context. Two sites of La Réunion Island, Le Tampon and Saint-Pierre, are chosen to do the benchmarking exercise. Réunion Island is a tropical island with a complex orography where cloud processes are mainly governed by local dynamics. As a consequence, Réunion Island exhibits numerous micro-climates. The two aforementioned sites are quite representative of the challenging character of solar forecasting in the case of a tropical island with complex orography. Hence, although distant from only 10 km, these two sites exhibit very different sky conditions. This work focuses on day-ahead and intra-day solar forecasting. Day-ahead solar forecasts are provided by the European Center for Medium-Range Weather Forecast (ECMWF). This organization maintains and runs the Numerical Weather Prediction (NWP) model named Integrated Forecast System (IFS). In this work, post-processing techniques are applied to refine the output of the IFS model for day-ahead forecasting. Statistical models like a recursive linear model or a nonlinear model such as an artificial neural network are used to produce the intra-day solar forecasts. It is shown that a combination of the IFS model and the neural network model further improves the accuracy of the forecasts. [ABSTRACT FROM AUTHOR]

In this paper, the performances of two approaches for solar probabilistic are evaluated using a set of metrics previously tested by the meteorology verification community. A particular focus is put on several scores and the decomposition of a specific probabilistic metric: the continuous rank probability score (CRPS) as they give extensive information to compare the forecasting performance of both methodologies. The two solar probabilistic forecasting methodologies are used to produce intra-day solar forecasts with time horizons ranging from 1 h to 6 h. The first methodology is based on two steps. In the first step, we generated a point forecast for each horizon and in a second step, we use quantile regression methods to estimate the prediction intervals. The second methodology directly estimates the prediction intervals of the forecasted clear sky index distribution using past data as inputs. With this second methodology we also propose to add solar geometric angles as inputs. Overall, nine probabilistic forecasting performances are compared at six measurements stations with different climatic conditions. This paper shows a detailed picture of the overall performance of the models and consequently may help in selecting the best methodology. [ABSTRACT FROM AUTHOR]