Computer Science & Artificial Intelligence
[1], [4]
Real-time forecasting of solar radiation posses several benefits and has huge potential for industrial applications. However, the intermittent nature of solar radiation makes it difficult to forecast accurately. Cloud cover is one of the major influencing factors of solar radiation. Thus, forecasting cloud motion effectively can help to improve the accuracy of short-term solar radiation forecasts. In this study, a novel Multi-Layer Cloud Motion Vector (referred as 3D-CMV) forecasting technique was introduced, which combined with the fast radiative transfer model (FRTM) produces forecasts up to 3 h ahead at 15 min intervals over 5km 5km grids across Europe and North Africa. The cloud microphysics obtained from the Support to Nowcasting and Very Short Range Forecasting (SAFNWC) of the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) was used as input to the forecasting system. The results obtained improvements in forecasts as compared to the conventional cloud motion vector techniques, across all seasons and sky conditions. Comparisons against ground-based measurements from the Baseline Surface Radiation Network (BSRN) revealed an overall maximum percentage difference of less than 12%, bias under -20 Wm−2 and a root mean square error (RMSE) under 80 Wm−2. Performance evaluations of Multi-Layer Cloud Motion Vector has been performed against several state-of-the-art techniques and presented in this study. Short-term solar energy forecasting has an established market and rising demand. Accurate forecasts from Multi-Layer CMV hence pose a high potential for real world applications.
[2]
This study introduces an Earth observation (EO)-based system which is capable of operationally estimating and continuously monitoring the ultraviolet index (UVI) in Europe. UVIOS (i.e., UV-Index Operating System) exploits a synergy of radiative transfer models with high-performance computing and EO data from satellites (Meteosat Second Generation and Meteorological Operational Satellite-B) and retrieval processes (Tropospheric Emission Monitoring Internet Service, Copernicus Atmosphere Monitoring Service and the Global Land Service). It provides a near-real-time nowcasting and short-term forecasting service for UV radiation over Europe. The main atmospheric inputs for the UVI simulations include ozone, clouds and aerosols, while the impacts of ground elevation and surface albedo are also taken into account. The UVIOS output is the UVI at high spatial and temporal resolution (5 km and 15 min, respectively) for Europe (i.e., 1.5 million pixels) in real time. The UVI is empirically related to biologically important UV dose rates, and the reliability of this EO-based solution was verified against ground-based measurements from 17 stations across Europe. Stations are equipped with spectral, broadband or multi-filter instruments and cover a range of topographic and atmospheric conditions. A period of over 1 year of forecasted 15 min retrievals under all-sky conditions was compared with the ground-based measurements. UVIOS forecasts were within ±0.5 of the measured UVI for at least 70 % of the data compared at all stations. For clear-sky conditions the agreement was better than 0.5 UVI for 80 % of the data. A sensitivity analysis of EO inputs and UVIOS outputs was performed in order to quantify the level of uncertainty in the derived products and to identify the covariance between the accuracy of the output and the spatial and temporal resolution and the quality of the inputs. Overall, UVIOS slightly overestimated the UVI due to observational uncertainties in inputs of cloud and aerosol. This service will hopefully contribute to EO capabilities and will assist the provision of operational early warning systems that will help raise awareness among European Union citizens of the health implications of high UVI doses.
[5]
This study focuses on the use of cloud motion vectors (CMV) and fast radiative transfer models (FRTM) in the prospect of forecasting downwelling surface solar irradiation (DSSI). Using near-real-time cloud optical thickness (COT) data derived from multispectral images from the spinning enhanced visible and infrared imager (SEVIRI) onboard the Meteosat second generation (MSG) satellite, we introduce a novel short-term forecasting system (3 h ahead) that is capable of calculating solar energy in large-scale (1.5 million-pixel area covering Europe and North Africa) and in high spatial (5 km over nadir) and temporal resolution (15 min intervals). For the operational implementation of such a big data computing architecture (20 million simulations in less than a minute), we exploit a synergy of high-performance computing and deterministic image processing technologies (dense optical flow estimation). The impact of clouds forecasting uncertainty on DSSI was quantified in terms of cloud modification factor (CMF), for all-sky and clear sky conditions, for more generalized results. The forecast accuracy was evaluated against the real COT and CMF images under different cloud movement patterns, and the correlation was found to range from 0.9 to 0.5 for 15 min and 3 h ahead, respectively. The CMV forecast variability revealed an overall DSSI uncertainty in the range 18–34% under consecutive alternations of cloud presence, highlighting the ability of the proposed system to follow the cloud movement in opposition to the baseline persistent forecasting, which considers the effects of topocentric sun path on DSSI but keeps the clouds in “fixed” positions, and which presented an overall uncertainty of 30–43%. The proposed system aims to support the distributed solar plant energy production management, as well as electricity handling entities and smart grid operations.
[6]
The rapid expansion of solar industries presents unknown technological challenges. A dedicated and suitable energy forecast is an effective solution for the daily dispatching and production of the electricity grid. The traditional forecast technique uses weather and plant parameters as the model information. Nevertheless, these are insufficient to consider problematic weather variability and the various plant characteristics in the actual field. Considering the above facts and inspired by the excellent implementation of the multi-column convolutional neural network (MCNN) in image processing, we developed a novel approach for forecasting solar energy by transforming multipoint time series (MT) into images for the MCNN to examine. We first processed the data to convert the time series solar energy into image matrices. We observed that the MCNN showed a preeminent response under a ground-based high-resolution spatial–temporal image matrix with a 0.2826% and 0.5826% RMSE for 15 min-ahead forecast under clear (CR) and cloudy (CD) conditions, respectively. Our process was performed on the MATLAB deep learning platform and tested on CR and CD solar energy conditions. The excellent execution of the suggested technique was compared with state-of-the-art deep neural network solar forecasting techniques.
[7]
The rapid growth in the use of Solar Energy for sustaining energy demand around the world requires accurate forecasts of Solar Irradiance to estimate the contribution of solar power to the power grid. Accurate forecasts for higher time horizons help to balance the power grid effectively and efficiently. Traditional forecasting techniques rely on physical weather parameters and complex mathematical models. However, these techniques are time-consuming and produce accurate results only for short forecast horizons. Deep Learning Techniques like Long Short Term Memory (LSTM) networks are employed to learn and predict complex varying time series data. However, LSTM networks are susceptible to poor performance due to improper configuration of hyperparameters. This work introduces two new algorithms for hyperparameter tuning of LSTM networks and a Fast Fourier Transform (FFT) based data decomposition technique. This work also proposes an optimised workflow for training LSTM networks based on the above techniques. The results show a significant fitness increase from 81.20% to 95.23% and a 53.42% reduction in RMSE for 90 min ahead forecast after using the optimised training workflow. The results were compared to several other techniques for forecasting solar energy for multiple forecast horizons.
[8]
This paper reports on the development of a neural network (NN) model for instantaneous and accurate estimation of solar radiation spectra and budgets geared toward satellite cloud data using a ≈2.4 M record, high-spectral resolution look up table (LUT) generated with the radiative transfer model libRadtran. Two NN solvers, one for clear sky conditions dominated by aerosol and one for cloudy skies, were trained on a normally-distributed and multiparametric subset of the LUT that spans a very broad class of atmospheric and meteorological conditions as inputs with corresponding high resolution solar irradiance target spectra as outputs. The NN solvers were tested by feeding them with a large (10 K record) “off-grid” random subset of the LUT spanning the training data space, and then comparing simulated outputs with target values provided by the LUT. The NN solvers demonstrated a capability to interpolate accurately over the entire multiparametric space. Once trained, the NN solvers allow for high-speed estimation of solar radiation spectra with high spectral resolution (1 nm) and for a quantification of the effect of aerosol and cloud optical parameters on the solar radiation budget without the need for a massive database. The cloudy sky NN solver was applied to high spatial resolution (54 K pixel) cloud data extracted from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) onboard the geostationary Meteosat Second Generation 3 (MSG3) satellite and demonstrated that coherent maps of spectrally-integrated global horizontal irradiance at this resolution can be produced on the order of 1 min.