Variability of solar resource poses difficulties in grid management as solar penetration rates rise continuously. Thus, the task of solar power forecasting becomes crucial to ensure grid stability and to enable an optimal unit commitment and economical dispatch. Several forecast horizons can be identified, spanning from a few seconds to days or weeks ahead, as well as spatial horizons, from single site to regional forecasts. New techniques and approaches arise worldwide each year to improve accuracy of models with the ultimate goal of reducing uncertainty in the predictions. This paper appears with the aim of compiling a large part of the knowledge about solar power forecasting, focusing on the latest advancements and future trends. Firstly, the motivation to achieve an accurate forecast is presented with the analysis of the economic implications it may have. It is followed by a summary of the main techniques used to issue the predictions. Then, the benefits of point/regional forecasts and deterministic/probabilistic forecasts are discussed. It has been observed that most recent papers highlight the importance of probabilistic predictions and they incorporate an economic assessment of the impact of the accuracy of the forecasts on the grid. Later on, a classification of authors according to forecast horizons and origin of inputs is presented, which represents the most up-to-date compilation of solar power forecasting studies. Finally, all the different metrics used by the researchers have been collected and some remarks for enabling a fair comparison among studies have been stated.
Recently, the 2015 United Nations Climate Change Conference (COP21), now known as the Paris Agreement, has become a milestone in fighting global warming. The 196 countries that signed the document agreed to make efforts to limit the global warming to less than 2 °C with respect to pre-industrial levels, which implies reducing the anthropogenic greenhouse emissions to zero during the second half of the 21st century. Reaching those goals involves an electrification of many current thermal systems, among many other actions. This Agreement stresses the necessity of generating energy via renewable sources and motivates the research on how to manage and integrate into the grid these variable generation systems.
Focusing on solar technology, photovoltaics have experienced enormous growth over the last years, amounting to a total installed capacity of around 177 GW worldwide by the end of 2014 (IEA, 2015) and growth is projected to continue at a similar rate in the future. Moreover, photovoltaic (PV) prices have seen a strong reduction, bottoming below $1.5/Wp for fixed-tilt systems, boosting more installations (GTM). PV has already become a key agent in some electricity markets, reaching an annual 8% of solar share in Italy or close to 7% in Germany, and the number of countries where that percentage is greater than 1% is about 20 (IEA, 2015). In this context, the high penetration of PV in electric systems poses many economic benefits, but may also threaten the stability of the power grid without accurate forecasts.
PV production mainly depends on the amount of solar global irradiation incident on the panels, but that irradiation is not uniform over time. Solar resource variability and the uncertainty associated to forecasts are behind most of the problems that must be handled to maintain the stability of the power grid. A part of the fluctuations are deterministic and explained by the rotational and translational movements of the Earth with respect to the Sun, which are accurately described by physical equations. However, there also exists unexpected changes in the amount of solar irradiance arriving at the Earth’s surface, mainly derived from the presence of clouds, which stochastically block the Sun’s rays and grant PV power forecasting a certain level of uncertainty.
The ability of precisely forecasting the energy produced by PV systems is of great importance and has been identified as one of the key challenges for massive PV integration (EPIA, 2012, PV GRID, 2014). It is decisive for grid operators, since deviations between forecasted and produced energy must be supplied by the rest of technologies that form the energy portfolio. Some of the units that build the electric system act as operating reserve generators. Thus, a proper PV forecast would be able to lower the number of units in hot standby and, consequently, reduce the operation costs. Table 1 depicts the flexibility of conventional power plants and the ability to respond to such deviations.
An accurate forecast is not only beneficial for system operators (and, eventually, for all customers from the grid) since it reduces costs and uncertainties, but also for PV plant managers, as they avoid possible penalties that are incurred due to deviations between forecasted and produced energy.
The importance of the issue has boosted the development of many studies worldwide to obtain accurate forecasts. Two main approaches can be found in the forecasting of PV plant production: indirect and direct. Indirect forecasts firstly predict solar irradiation and then, using a PV performance model of the plant, obtain the power produced. On the other hand, direct forecasts directly calculate the power output of the plant. Also, many other studies only focus on the prediction of solar irradiation, since it is the most difficult element to model and have other applications apart from solar power forecasting. Both forecasts (power and irradiation) are approached via similar techniques. This review paper was based on those articles that have as the output the power produced by the plants, to establish a boundary in the scope and since that variable can be directly used by grid operators and plant managers. This work is limited to the study of scientific articles; the analysis of commercial forecasting tools is out of the scope of this review.
This paper presents a complete review of the state-of-the-art techniques to produce power forecasts for photovoltaics. There are some previous review articles with also a wide scope (forecasting techniques, sources of inputs, performance metrics, temporal and spatial coverage, …), such as the work developed by Inman et al., 2013, IEA, 2013, but the rate at which new studies are developed requires that a new review showing current trends is conducted. Some of these new trends are the focus on the economic impact of forecasting, the importance of probabilistic forecasting and the necessity of agreement for a common suite of performance metrics. Other more recent reviews are only focused on a specific aspect of forecasting, such as ensemble forecasting (Ren et al., 2015) or different forecasting techniques (Wan et al., 2015).
The article is structured in such a way that it tackles some of the issues that arise when planning a forecast, such as the necessity to issue and improve solar power forecasts, the different techniques that can be used, spatial and temporal coverage, information that should be provided, measurement of accuracy and previous work developed by other researchers.
Thus, the paper is structured as follows: Section 2 explains some basic concepts that are used throughout the paper. Section 3 sets the foundations and main motivations of the study as it talks about the importance of forecasting, showing possible economic consequences of improved forecasts. Then, Section 4 shows the main approaches to forecasting power output: physical, statistical or hybrid. Section 5 discusses the benefits and characteristics of forecasting for either a single PV plant or for an ensemble of them. Section 6 talks about the different options to present the forecast: a single value or a probabilistic term. It also discusses about the implications it may have for grid operation. Section 7 discusses about the different time horizons that are necessary to be taken into account for a proper grid operation. In contrast to most of review papers about solar forecasting, we have classified the studies according to the forecast horizon instead of the techniques used. Here are collated and summarized all the articles found about solar power forecasting. Finally, a review of the metrics that are used to evaluate forecasts and the convenience of each of them is given in Section 8, along with some recommendations for a better comparability of studies. Moreover, at the end of certain sections and subsections a short summary is presented, which depicts the main findings and conclusions about each topic.
In this section some basic concepts about solar irradiation and solar power generation are explained, which will ease the comprehension of the remaining parts of the text.
The economics of forecasting
The main purpose of improving the accuracy of solar power forecasts is to reduce the uncertainties related to this type of variable energy source, which would directly result in a safer and easier grid management. Moreover, curtailment applied to photovoltaics could be reduced (Bird et al., 2014). Plant managers also find motivation in issuing better predictions as they can better plan maintenance stops and generate more precise bids. As solar penetration increases in the energy portfolio, the
As presented in the Introduction, there are two main approaches for solar power forecasting. The first option consists in using analytical equations to model the PV system. Normally, most efforts are dedicated to obtain accurate irradiance forecasts, which is the main factor related to the power production. This approach is denoted as PV performance, physical, parametric or “white box” method. Contrarily, the second option consists in directly predicting the power output using statistical and
Spatial horizon: single plant and regional forecasts
Forecasts can be made for a single PV system or for an ensemble of them. Normally, grid operators prefer regional forecasts since they are more useful to keep the balance between demand and supply in the electric system. To better understand the differences between point and regional forecasts, first we study the short term power output variability.
As detailed in Mills and Wiser (2010), variability of solar resource at different time scales poses several problems to the integration of solar
Deterministic vs. probabilistic
Energy forecasts have been applied for a long time, not only predicting production (wind, solar) but also forecasting load. Each domain has its own peculiarities and differences in accuracy can be found between them. As seen in Fig. 4, solar power forecasts are the least mature of the energy forecasts analyzed by Hong et al. (in press), due to the relatively low solar penetration in electricity markets so far. However, wind forecasting shows a high level of maturity for its similarity to
The main way in which forecasts can be classified is according to the time horizon. As will be discussed later on, predictions made for the diverse time horizons are important for different aspects of grid operation, such as maintenance of grid stability, scheduling of spinning reserves, load following or unit commitment.
What follows is a complete classification of studies regarding the time horizon. A general description detailing the main characteristics of each study and their most relevant
The performance and accuracy of a certain model can be assessed via several metrics. Metrics permit the comparison between different models and locations. Each one focus on a certain aspect of a point distribution. Thus, there is not a unique metric valid for all situations; instead, each one adds some information about the accuracy of the model. In the bibliography, several metrics can be found, although there are a group of them that are more commonly used. In the last years, some authors
Solar power forecasting becomes a crucial task as solar energy starts to play a key role in electricity markets. The complexity of issuing reliable forecasts is mainly caused by the uncertainty in the solar resource assessment. Moreover, energy markets work within different time frames and, thus, specific forecasts are needed for each time horizon. Several models appear to issue forecasts as accurate as possible.
From the collection of studies shown in this paper, the following trends and
Conflict of interest
The authors declare that they do not have any confict of interests.
J. Antonanzas and R. Urraca acknowledge the fellowships FPI-UR-2014 and ATUR grant 15/03 granted by the University of La Rioja. Also, F. Antonanzas-Torres expresses his gratitude for the fellowship FPI-UR-2012 and ATUR Grant No. 03061402 at the University of La Rioja. R. Escobar acknowledges the generous financial support provided by CORFO (Corporación de Fomento de la Producción) under the Project 13CEI2-21803.
We would also like to thank the reviewers for their dedicate reading of the paper
- A. Mellit et al.Short-term forecasting of power production in a large-scale photovoltaic plantSol. Energy(2014)
- A. Mellit et al.A 24-h forecast of solar irradiance using artificial neural network: application for performance prediction of a grid-connected PV plant at Trieste, ItalySol. Energy(2010)
- D. Masa-Bote et al.Improving photovoltaics grid integration through short time forecasting and self-consumptionAppl. Energy(2014)
- P. Mandal et al.Forecasting power output of solar photovoltaic system using wavelet transform and artificial intelligence techniquesProc. Comput. Sci.(2012)
- V. Lonij et al.Intra-hour forecasts of solar power production using measurements from a network of irradiance sensorsSol. Energy(2013)
- H. Long et al.Analysis of daily solar power prediction with data-driven approachesAppl. Energy(2014)
- M. Lipperheide et al.Embedded nowcasting method using cloud speed persistence for a photovoltaic power plantSol. Energy(2015)
- Y. Li et al.An ARMAX model for forecasting the power output of a grid connected photovoltaic systemRenew. Energy(2014)
- Z. Li et al.A hierarchical approach using machine learning methods in solar photovoltaic energy production forecastingEnergies(2016)
- D. Larson et al.Day-ahead forecasting of solar power output from photovoltaic plants in the American SouthwestRenew. Energy(2016)