Search

This article aims to evaluate the performance of solar radiation forecasts produced by CMA-WSP v2.0 (version 2 of the China Meteorological Administration Wind and Solar Energy Prediction System) and to explore the application of machine learning algorithms from the scikit-learn Python library to improve the solar radiation prediction made by the CMA-WSP v2.0. It is found that the performance of the solar radiation forecasting from the CMA-WSP v2.0 is closely related to the weather conditions, with notable diurnal fluctuations. The mean absolute percentage error (MAPE) produced by the CMA-WSP v2.0 is approximately 74% between 11:00 and 13:00. However, the MAPE ranges from 193% to 242% at 07:00–08:00 and 17:00–18:00, which is greater than that observed at other daytime periods. The MAPE is relatively low (high) for both sunny and cloudy (overcast and rainy) conditions, with a high probability of an absolute percentage error below 25% (above 100%). The forecasts tend to underestimate (overestimate) the observed solar radiation in sunny and cloudy (overcast and rainy) conditions. By applying machine learning models (such as linear regression, decision trees, K-nearest neighbors, random forests regression, adaptive boosting, and gradient boosting regression) to revise the solar radiation forecasts, the MAPE produced by the CMA-WSP v2.0 is significantly reduced. The reduction in the MAPE is closely connected to the weather conditions. The models of K-nearest neighbors, random forests regression, and decision trees can reduce the MAPE in all weather conditions. The K-nearest neighbor model exhibits the most optimal performance among these models, particularly in rainy conditions. The random forest regression model demonstrates the second-best performance compared to that of the K-nearest neighbor model. The gradient boosting regression model has been observed to reduce the MAPE of the CMA-WSP v2.0 in all weather conditions except rainy. In contrast, the adaptive boosting (linear regression) model exhibited a diminished capacity to improve the CMA-WSP v2.0 solar radiation prediction, with a slight reduction in MAPE observed only in sunny (sunny and cloudy) conditions. In addition, the input feature selection has a considerable influence on the performance of the machine learning model. The incorporation of the time series data associated with the diurnal variation of solar radiation as an input feature can further improve the model's performance. [ABSTRACT FROM AUTHOR]

A sustainable energy sector and achieving carbon neutrality in microgrids require a firm commitment to renewable energy resources. A sharp focus on solar energy holds the most promising potential for a low-carbon energy pathway. Efficient and optimal energy management application in the case of such microgrid systems requires the development of an accurate forecasting technique. A significant obstacle in creating such an accurate prediction tool is the intermittent nature of solar energy. To address this challenge, this research proposes a novel threefold hybrid model that integrates empirical mode decomposition (EMD), convolutional neural networks (CNN), and long short-term memory (LSTM) neural networks to forecast solar radiation up to three steps ahead. We do this utilizing two distinct solar radiation datasets obtained from two different sites in Thailand: one at 30-min intervals and the second one at 15-min intervals. To assess the forecasting capabilities of the proposed model, this study has carried out a comprehensive analysis by comparing it to six alternative models considering hybrid and stand-alone options. The proposed model outperformed them all, establishing its supremacy with the lowest mean absolute error (MAE), root-mean-square error (RMSE), and mean absolute percentage error (MAPE), additionally, ensuring that it meets the industry standards for forecasting applications. The proposed model's effectiveness in accurately predicting solar radiation has been successfully tested for a microgrid case study site in Thailand which supplies power to a business complex and uses the innovative Green Energy Management System (GEMS) developed by Leonics Co. Ltd. [ABSTRACT FROM AUTHOR]

This article aims to evaluate the performance of solar radiation forecasts produced by CMA-WSP v2.0 (version 2 of the China Meteorological Administration Wind and Solar Energy Prediction System) and to explore the application of machine learning algorithms from the scikit-learn Python library to improve the solar radiation prediction made by the CMA-WSP v2.0. It is found that the performance of the solar radiation forecasting from the CMA-WSP v2.0 is closely related to the weather conditions, with notable diurnal fluctuations. The mean absolute percentage error (MAPE) produced by the CMA-WSP v2.0 is approximately 74% between 11:00 and 13:00. However, the MAPE ranges from 193% to 242% at 07:00–08:00 and 17:00–18:00, which is greater than that observed at other daytime periods. The MAPE is relatively low (high) for both sunny and cloudy (overcast and rainy) conditions, with a high probability of an absolute percentage error below 25% (above 100%). The forecasts tend to underestimate (overestimate) the observed solar radiation in sunny and cloudy (overcast and rainy) conditions. By applying machine learning models (such as linear regression, decision trees, K-nearest neighbors, random forests regression, adaptive boosting, and gradient boosting regression) to revise the solar radiation forecasts, the MAPE produced by the CMA-WSP v2.0 is significantly reduced. The reduction in the MAPE is closely connected to the weather conditions. The models of K-nearest neighbors, random forests regression, and decision trees can reduce the MAPE in all weather conditions. The K-nearest neighbor model exhibits the most optimal performance among these models, particularly in rainy conditions. The random forest regression model demonstrates the second-best performance compared to that of the K-nearest neighbor model. The gradient boosting regression model has been observed to reduce the MAPE of the CMA-WSP v2.0 in all weather conditions except rainy. In contrast, the adaptive boosting (linear regression) model exhibited a diminished capacity to improve the CMA-WSP v2.0 solar radiation prediction, with a slight reduction in MAPE observed only in sunny (sunny and cloudy) conditions. In addition, the input feature selection has a considerable influence on the performance of the machine learning model. The incorporation of the time series data associated with the diurnal variation of solar radiation as an input feature can further improve the model’s performance.

Solar energy systems which is one of the renewable energy sources takes more interest and gains prevalence day by day. A significant problem in solar energy systems as in other many renewable energy sources is the instability of the energy that the system will provide. Forecasting of the energy to be obtained is very important in this respect. In this study, solar radiation has been forecasted using meteorological data taken from the General Directorate of Meteorology for Isparta province. Random Forest (RF), k-Nearest Neighbor (k-NN), Artificial Neural Network (ANN) and Deep Learning (DL) methods have been used for forecasting. In addition, the results of dummy variable usage for time data have been examined with these different methods. According to the findings obtained, it is seen that the dummy variable usage increases performance for ANN and DL methods but decreases performance for RF and k-NN methods. Best results have been obtained with ANN and DL for the forecasting of the solar radiation. [ABSTRACT FROM AUTHOR]

Machine learning is routinely used to forecast solar radiation from inputs, which are forecasts of meteorological variables provided by numerical weather prediction (NWP) models, on a spatially distributed grid. However, the number of features resulting from these grids is usually large, especially if several vertical levels are included. Principal Components Analysis (PCA) is one of the simplest and most widely-used methods to extract features and reduce dimensionality in renewable energy forecasting, although this method has some limitations. First, it performs a global linear analysis, and second it is an unsupervised method. Locality Preserving Projection (LPP) overcomes the locality problem, and recently the Linear Optimal Low-Rank (LOL) method has extended Linear Discriminant Analysis (LDA) to be applicable when the number of features is larger than the number of samples. Supervised Nonnegative Matrix Factorization (SNMF) also achieves this goal extending the Nonnegative Matrix Factorization (NMF) framework to integrate the logistic regression loss function. In this article we try to overcome all these issues together by proposing a Supervised Local Maximum Variance Preserving (SLMVP) method, a supervised non-linear method for feature extraction and dimensionality reduction. PCA, LPP, LOL, SNMF and SLMVP have been compared on Global Horizontal Irradiance (GHI) and Direct Normal Irradiance (DNI) radiation data at two different Iberian locations: Seville and Lisbon. Results show that for both kinds of radiation (GHI and DNI) and the two locations, SLMVP produces smaller MAE errors than PCA, LPP, LOL, and SNMF, around 4.92% better for Seville and 3.12% for Lisbon. It has also been shown that, although SLMVP, PCA, and LPP benefit from using a non-linear regression method (Gradient Boosting in this work), this benefit is larger for PCA and LPP because SMLVP is able to perform non-linear transformations of inputs. [ABSTRACT FROM AUTHOR]

• Six types of solar radiation nowcasting models are explored. • UNet and DGMR-SO are utilized as deep learning models. • Satellite images are utilized for cloud properties and solar radiation forecast. • Deep learning models and blurred optical flow outperforms the other comparison methods. • Enable auxiliary decision-making for power systems. Reliable solar radiation and photovoltaic power prediction is essential for the safe and stable operation of electric power systems. Cloud cover is highly related with solar radiation, but existing extrapolation-based cloud forecast methods have difficulties in capturing cloud development. Therefore, we applied two deep learning models and a physical method for solar radiation forecast. First, for the first time we applied the novel Deep Generative Model of Radar (originally developed for radar precipitation nowcasting) to predict solar radiation (named as DGMR-SO). Second, the well-known UNet model was used for comparison. Third, we developed a physical method based on cloud physical properties forecasting. An optical flow model was used to predict the five cloud properties from satellite measurements, followed by a Cloud Physical Properties algorithm to compute solar radiation from the advected cloud properties. A spatial blurring strategy was also applied to the optical flow results in order to reduce the forecast errors. Finally, the smart persistence model and the HARMONIE numerical weather prediction model forecast were utilized as benchmark methods. The forecast horizon was 0–4 h with 15 min temporal resolution. All methods have been calibrated and tested using data from the Netherlands. In general, UNet shows the lowest errors, while DGMR-SO outperforms the competitors on qualitative performance after around 45 min. The forecast accuracy of each method also depends on sky conditions. The study findings are expected to encourage the inclusion of satellite data in solar radiation nowcasting, and can provide scientific guidance for power systems and solar power plants.Our code is open-sourced at: https://github.com/Yangcherry2024/SolarRadiation-nowcast-DGMR-SO/. [ABSTRACT FROM AUTHOR]

Due to the continuous increasing importance of renewable energy sources as an alternative to fossil fuels, to contrast air pollution and global warming, the prediction of Global Horizontal Irradiation (GHI), one of the main parameters determining solar energy production of photovoltaic systems, represents an attractive topic nowadays. Solar irradiance is determined by deterministic factors (i.e. the position of the sun) and stochastic factors (i.e. the presence of clouds). Since the stochastic element is difficult to model, this problem can benefit from machine learning techniques, like artificial neural networks. This work proposes a methodology to forecast GHI in short- (i.e. from 15 min to 60 min) and mid-term (i.e. from 60 to 120 min) time horizons. For this purpose, we designed, optimised and compared four neural network architectures for time-series forecasting, respectively based on: i) Non-Linear Autoregressive, ii) Feed-Forward, iii) Long Short-Term Memory and iv) Echo State Network. The original data-set, consisting of GHI values sampled every 15min, has been pre-processed by applying different filtering techniques. Our results analysis compares the performance of the proposed neural networks identifying the best in terms of error rate and forecast horizon. This analysis highlights that the clear-sky index results the preferred filtering technique by giving greatly improvements in data-set pre-processing, and Echo State Network gives best accuracy results.

West Africa is at the forefront of global environmental challenges with its commitment to reduce greenhouse gas emissions and harnessing the potential of renewable energy, especially the promising solar power. This study evaluates Global Horizontal Irradiance (GHI) operational forecast errors for the Zagtouli (Burkina-Faso) and Sococim (Senegal) solar plants, and investigates their links with local meteorological conditions, particularly clouds and dust aerosols. Firstly, the evaluation of aerosol products indicates that CAMS reanalysis is reliable for assessing aerosol optical depth. We then examine the accuracy of three operational GHI forecast products: the Global Forecast System (GFS, NCEP), the Integrated Forecast System (IFS, ECMWF), and SteadyMet (SM, French company Steadysun). The analysis reveals that IFS and SM outperform GFS, SM having a slight advantage due to its probabilistic nature. Closer examination reveals a significant relationship between GHI forecast errors and local meteorological characteristics. These errors are more pronounced during the wet season, primarily attributed to cloud occurrence. Dust events play a secondary influential role, especially during the dry season. Correlation analyses emphasize the forecast errors' major link with cloudiness, while co-occurrences highlight that dust aerosol is a secondary factor in forecast errors for the GHI directly or for cloud representation (aerosol-cloud interaction). [Display omitted] [ABSTRACT FROM AUTHOR]

Electrical power production by renewable energy sources is unpredictable in nature and this may cause imbalance between power generation and demand. Therefore, an accurate prediction of solar radiation is crucial for the stability and efficient management of electric grid. This study focuses on very short-term forecasts of solar radiation with a horizon in the range of 5–15 min. In this paper, a Convolutional Neural Network is proposed that uses sequences of infrared images captured by an All-Sky Imager to forecast the Global Horizontal Irradiance on different time horizon. A real case study, exploiting six months of high-resolution data, is analyzed. Additionally, an innovative technique, the Enhanced Convolutional Neural Network (ECNN), is proposed in which exogenous data, as the solar radiation measurement, is encoded in terms of colored pixels in the upper corner of the images. Considering the naïve persistence method as a baseline, a clear improvement across the key metrics has been noted with the proposed methodology. A deeper analysis of the results reveals that the proposed models are more accurate than persistence when high fluctuations of solar radiation are experienced. In that case, the ECNN achieves a forecast skill exceeding 19% for all the tested forecast horizons. [Display omitted] • Enhanced Convolutional Neural Network (ECNN) is proposed by exploiting auxiliary information (exogeneous parameters) which are encoded on images of the All-Sky Imager. • The performance of ECNN, CNN, and other baseline models are used for very short-term GHI forecast (nowcasting). • The ECNN outperforms naïve persistence models in case of partially cloudy days and highly variable weather conditions. [ABSTRACT FROM AUTHOR]

• Authors participated in the contest of net-zero energy house ENEMANE HOUSE 2017. • Renovated house realized net-zero energy and thermal comfort in operation phase. • HPWH and BT was schedule-controlled according to solar radiation forecast. • The control aimed for self-consumption and leveling of electrical supply and demand. • Renovation enabled residents' behavioural adaptation to the environment. In Japan, low thermal insulation performance in existing houses is a considerable issue that may lead to health problems. Therefore, it is necessary to acquire technologies and knowledge regarding thermal insulation renovation. On the other hand, with the spread of photovoltaic panels, urban scale energy management and self-consumption have become important. The purpose of this study is to introduce a net-zero energy house which was designed, built, and analyzed by the team of Waseda University and Shibaura Institute of Technology in the ENEMANE HOUSE 2017 competition. The team proposed a renovation method which insert highly insulated walls into the existing structure, creating not only highly insulated areas but also non-insulated areas called "loggias." By opening and closing the windows of the loggia according to the season, it functions as a passive system and enables residents' behavioural adaptation to the environment. On the other hand, to aim for self-consumption and leveling of reverse power flow, a heat pump water heater and a storage battery were operated according to solar radiation prediction. Actual measurements confirmed the implementation and effect of this operational method. Moreover, the energy consumption was greatly reduced, while thermal comfort was maintained by thermal insulation and efficient equipment introduction. [ABSTRACT FROM AUTHOR]

Given the wide applications of photovoltaic (PV) power generation, the volatility in generation caused by solar radiation, which limits the capacity of the power grid, cannot be ignored. Therefore, much research has aimed to address this issue through the development of methods for accurately predicting inter-hour solar radiation and then estimating PV power. However, most forecasting methods focus on adjusting the model structure or model parameters to achieve prediction accuracy. There is little research discussing how different factors influence solar radiation and, thereby, the effectiveness of these data-driven methods regarding their prediction accuracy. In this work, the effects of several potential factors on solar radiation are estimated using correlation analysis and a structural equation model; an ensemble model is developed for predicting inter-hour solar radiation based on the interaction of those key factors. Several experiments are carried out based on an open database provided by the National Renewable Energy Laboratory. The results show that solar zenith angle, cloud cover, aerosols, and airmass have great effects on solar radiation. It is also shown that the selection of the key factor is more important than the model structure construction for predicting solar radiation precisely. The proposed ensemble model proves to outperform all sub-models and achieves about a 12% improvement over the persistent model based on the normalized root mean squared error statistic.

The interest in solar radiation prediction has increased greatly in recent times among the scientific community. In this context, Machine Learning techniques have shown their ability to learn accurate prediction models. The aim of this paper is to go one step further and automatically achieve interpretability during the learning process by performing dimensionality reduction on the input variables. To this end, three non standard multivariate feature selection approaches are applied, based on the adaptation of strong learning algorithms to the feature selection task, as well as a battery of classic dimensionality reduction models. The goal is to obtain robust sets of features that not only improve prediction accuracy but also provide more interpretable and consistent results. Real data from the Weather Research and Forecasting model, which produces a very large number of variables, is used as the input. As is to be expected, the results prove that dimensionality reduction in general is a useful tool for improving performance, as well as easing the interpretability of the results. In fact, the proposed non standard methods offer important accuracy improvements and one of them provides with an intuitive and reduced selection of features and mesoscale nodes (around 10% of the initial variables centered on three specific nodes). [ABSTRACT FROM AUTHOR]

In electric power grids, generation must equal load at all times. Since wind and solar power are intermittent, system operators must predict renewable generation and allocate operating reserves to mitigate imbalances. If they overestimate the renewable generation during scheduling, insufficient generation will be available during operation, which can be very costly. However, if they underestimate the renewable generation, usually they will only face the cost of keeping some generation capacity online and idle. Therefore overestimation of renewable generation resources usually presents a more serious problem than underestimation. Many researchers train their solar radiation forecast algorithms using symmetric criteria like RMSE or MAE, and then a bias is applied to the forecast later to reflect the asymmetric cost faced by the system operator – a technique we call indirectly biased forecasting. We investigate solar radiation forecasts using asymmetric cost functions (convex piecewise linear (CPWL) and LinEx) and optimize directly in the forecast training stage. We use linear programming and a gradient descent algorithm to find a directly biased solution and compare it with the best indirectly biased solution. We also modify the LMS algorithm according to the cost functions to create an online forecast method. Simulation results show substantial cost savings using these methods. [ABSTRACT FROM AUTHOR]

The prediction of solar radiation allows estimating photovoltaic systems' power production in advance, guaranteeing a more reliable and stable energy supply. In this work, we present a novel approach for short-term solar radiation forecasting that leverages multi-channel images from the geostationary satellites of the Meteosat series, coupled with GHI values in clear-sky conditions. We propose two distinct deep learning models, a 3D-CNN and a ConvLSTM, to forecast solar radiation in terms of GHI values, up to 6-h ahead with a temporal granularity of 15 min, over a test study area, the city of Turin, Piedmont, Italy. The models have been validated with ground GHI measurements, and the results show that the ConvLSTM consistently outperforms the 3D-CNN for longer forecasting horizons, achieving a MAD of 27.18% and an nRMSE of 0.57 for 6-h ahead predictions. To motivate the use of satellite images, we compared the performance of our approach with a baseline Smart Persistence model and another benchmark model, which previously achieved state-of-the-art performance on the same data set by exploiting various kinds of meteorological inputs. The proposed models outperform the Smart Persistence for predictions farther than 15-min ahead, achieving a Forecast Skill of 0.56 for predictions 6-h ahead. Furthermore, the comparison shows that using raw satellite images overcomes the performance achievable by solely using meteorological variables, reducing the RMSD by more than 3% and the MAD by 1.37% for prediction horizons greater than 4-h ahead. [ABSTRACT FROM AUTHOR]

The use of renewable energy sources to supply electricity in customer grids is a current practise in developed countries and, particularly, in the case of distributed solar energy generation grids. Even though, the environmental benefits of the practise drive to economic incentives to increase the renewable energies. However, because of the unsteady behaviour of the renewable electric energy sources (Kaldellis et al., 2006), as wind energy and solar photovoltaic production, the increment in the use of these sources produces severe fluctuations in the electric grids. Then, their electric supply to the electric grid is limited and at the same time, an accurate renewable energy production forecast is required. In this work, the feasibility of a new approach to reduce this problem is tested by the application (on a yearly basis) of a simulator of a production-storage controlled system, namely FOTOV, to be applied in the design of a photovoltaic (PV) installation. [ABSTRACT FROM AUTHOR]

Nowadays, we are moving forward to more sustainable energy production systems based on renewable sources. Among all Photovoltaic (PV) systems are spreading in our cities. In this view, new models are needed to forecast Global Horizontal Solar Irradiance (GHI), which strongly influences PV production. For example, this forecast is crucial to develop novel control strategies for smart grid management. In this paper, we present a novel methodology to forecast GHI in short- and long-term time-horizons, i.e. from next 15 min up to next 24 h. It implements machine learning techniques to achieve this purpose. We start from the analysis of a real-world dataset with different meteorological information including GHI, in the form of time-series. Then, we combined Variational Mode Decomposition (VMD) and two Convolutional Neural Networks (CNN) together with Random Forest (RF) or Long Short Term Memory (LSTM). Finally, we present the experimental results and discuss their accuracy. • A combination of VMD and CNNs provide a novel methodology to forecast GHI. • GHI forecast aid the development of more accurate policies for energy management. • Methodologies to forecast GHI for short and long-term horizons are becoming crucial. [ABSTRACT FROM AUTHOR]

• The most relevant exogenous variables for predicting solar radiation are selected. • Different machine learning models for solar radiation forecasting are evaluated. • The Long Short-Term Memory neural network shows the best forecasting performance. • Multivariate models are compared with their univariate counterparts. • The adoption of exogenous inputs improves solar radiation forecasting performance. The prediction of Global Horizontal Irradiance (GHI) allows to estimate in advance the future energy production of photovoltaic systems, thus ensuring their full integration into the electricity grids. This paper investigates the effectiveness of using exogenous inputs in performing short-term GHI forecasting. To this aim, we identified a subset of relevant input variables for predicting GHI by applying different feature selection techniques. The results revealed that the most significant input variables for predicting GHI are ultraviolet index, cloud cover, air temperature, relative humidity, dew point, wind bearing, sunshine duration and hour-of-the-day. The predictive performance of the selected features was evaluated by feeding them into five different machine learning models based on Feedforward, Echo State, 1D-Convolutional, Long Short-Term Memory neural networks and Random Forest, respectively. Our Long Short-Term Memory solution presents the best prediction performance among the five models, predicting GHI up to 4 h ahead with a Mean Absolute Deviation (MAD) of 24.51%. Then, to demonstrate the effectiveness of using exogenous inputs for short-term GHI forecasting, we compare the multivariate models against their univariate counterparts. The results show that exogenous inputs significantly improve the forecasting performance for prediction horizons greater than 15 min, reducing errors by more than 22% in 4 h ahead predictions, while for very short prediction horizons (i.e. 15 min) the improvements are negligible. [ABSTRACT FROM AUTHOR]

Given the wide applications of photovoltaic (PV) power generation, the volatility in generation caused by solar radiation, which limits the capacity of the power grid, cannot be ignored. Therefore, much research has aimed to address this issue through the development of methods for accurately predicting inter-hour solar radiation and then estimating PV power. However, most forecasting methods focus on adjusting the model structure or model parameters to achieve prediction accuracy. There is little research discussing how different factors influence solar radiation and, thereby, the effectiveness of these data-driven methods regarding their prediction accuracy. In this work, the effects of several potential factors on solar radiation are estimated using correlation analysis and a structural equation model; an ensemble model is developed for predicting inter-hour solar radiation based on the interaction of those key factors. Several experiments are carried out based on an open database provided by the National Renewable Energy Laboratory. The results show that solar zenith angle, cloud cover, aerosols, and airmass have great effects on solar radiation. It is also shown that the selection of the key factor is more important than the model structure construction for predicting solar radiation precisely. The proposed ensemble model proves to outperform all sub-models and achieves about a 12% improvement over the persistent model based on the normalized root mean squared error statistic. [ABSTRACT FROM AUTHOR]