Cyril Voyant

Cyril Voyant, Milan Despotovic, Gilles Notton et al.

Clearsky models are widely used in solar energy for many applications such as quality control, resource assessment, satellite-base irradiance estimation and forecasting. However, their use in forecasting and nowcasting is associated with a number of challenges. Synchronization errors, reliance on the Clearsky index (ratio of the global horizontal irradiance to its cloud-free counterpart) and high sensitivity of the clearsky model to errors in aerosol optical depth at low solar elevation limit their added value in real-time applications. This paper explores the feasibility of short-term forecasting without relying on a clearsky model. We propose a Clearsky-Free forecasting approach using Extreme Learning Machine (ELM) models. ELM learns daily periodicity and local variability directly from raw Global Horizontal Irradiance (GHI) data. It eliminates the need for Clearsky normalization, simplifying the forecasting process and improving scalability. Our approach is a non-linear adaptative statistical method that implicitely learns the irradiance in cloud-free conditions removing the need for an clear-sky model and the related operational issues. Deterministic and probabilistic results are compared to traditional benchmarks, including ARMA with McClear-generated Clearsky data and quantile regression for probabilistic forecasts. ELM matches or outperforms these methods, providing accurate predictions and robust uncertainty quantification. This approach offers a simple, efficient solution for real-time solar forecasting. By overcoming the stationarization process limitations based on usual multiplicative scheme Clearsky models, it provides a flexible and reliable framework for modern energy systems.

Cyril Voyant, Milan Despotovic, Luis Garcia-Gutierrez et al.

A novel methodology for short-term energy forecasting using an Extreme Learning Machine ($\mathtt{ELM}$) is proposed. Using six years of hourly data collected in Corsica (France) from multiple energy sources (solar, wind, hydro, thermal, bioenergy, and imported electricity), our approach predicts both individual energy outputs and total production (including imports, which closely follow energy demand, modulo losses) through a Multi-Input Multi-Output ($\mathtt{MIMO}$) architecture. To address non-stationarity and seasonal variability, sliding window techniques and cyclic time encoding are incorporated, enabling dynamic adaptation to fluctuations. The $\mathtt{ELM}$ model significantly outperforms persistence-based forecasting, particularly for solar and thermal energy, achieving an $\mathtt{nRMSE}$ of $17.9\%$ and $5.1\%$, respectively, with $\mathtt{R^2} > 0.98$ (1-hour horizon). The model maintains high accuracy up to five hours ahead, beyond which renewable energy sources become increasingly volatile. While $\mathtt{MIMO}$ provides marginal gains over Single-Input Single-Output ($\mathtt{SISO}$) architectures and offers key advantages over deep learning methods such as $\mathtt{LSTM}$, it provides a closed-form solution with lower computational demands, making it well-suited for real-time applications, including online learning. Beyond predictive accuracy, the proposed methodology is adaptable to various contexts and datasets, as it can be tuned to local constraints such as resource availability, grid characteristics, and market structures.

Cédric Join, Cyril Voyant, Michel Fliess et al.

This communication is devoted to solar irradiance and irradiation short-term forecasts, which are useful for electricity production. Several different time series approaches are employed. Our results and the corresponding numerical simulations show that techniques which do not need a large amount of historical data behave better than those which need them, especially when those data are quite noisy.

Cyril Voyant, Wani W. Tamas, Marie Laure Nivet et al.

A Multi-Layer Perceptron (MLP) defines a family of artificial neural networks often used in TS modeling and forecasting. Because of its "black box" aspect, many researchers refuse to use it. Moreover, the optimization (often based on the exhaustive approach where "all" configurations are tested) and learning phases of this artificial intelligence tool (often based on the Levenberg-Marquardt algorithm; LMA) are weaknesses of this approach (exhaustively and local minima). These two tasks must be repeated depending on the knowledge of each new problem studied, making the process, long, laborious and not systematically robust. In this paper a pruning process is proposed. This method allows, during the training phase, to carry out an inputs selecting method activating (or not) inter-nodes connections in order to verify if forecasting is improved. We propose to use iteratively the popular damped least-squares method to activate inputs and neurons. A first pass is applied to 10% of the learning sample to determine weights significantly different from 0 and delete other. Then a classical batch process based on LMA is used with the new MLP. The validation is done using 25 measured meteorological TS and cross-comparing the prediction results of the classical LMA and the 2-stage LMA.

Cyril Voyant, Marie Laure Nivet, Christophe Paoli et al.

In this paper, we propose to study four meteorological and seasonal time series coupled with a multi-layer perceptron (MLP) modeling. We chose to combine two transfer functions for the nodes of the hidden layer, and to use a temporal indicator (time index as input) in order to take into account the seasonal aspect of the studied time series. The results of the prediction concern two years of measurements and the learning step, eight independent years. We show that this methodology can improve the accuracy of meteorological data estimation compared to a classical MLP modelling with a homogenous transfer function.

Cyril Voyant, Gilles Notton, Christophe Paoli et al.

Numerous methods exist and were developed for global radiation forecasting. The two most popular types are the numerical weather predictions (NWP) and the predictions using stochastic approaches. We propose to compute a parameter noted constructed in part from the mutual information which is a quantity that measures the mutual dependence of two variables. Both of these are calculated with the objective to establish the more relevant method between NWP and stochastic models concerning the current problem.

Cyril Voyant, C. Darras, Marc Muselli et al.

It is essential to find solar predictive methods to massively insert renewable energies on the electrical distribution grid. The goal of this study is to find the best methodology allowing predicting with high accuracy the hourly global radiation. The knowledge of this quantity is essential for the grid manager or the private PV producer in order to anticipate fluctuations related to clouds occurrences and to stabilize the injected PV power. In this paper, we test both methodologies: single and hybrid predictors. In the first class, we include the multi-layer perceptron (MLP), auto-regressive and moving average (ARMA), and persistence models. In the second class, we mix these predictors with Bayesian rules to obtain ad-hoc models selections, and Bayesian averages of outputs related to single models. If MLP and ARMA are equivalent (nRMSE close to 40.5% for the both), this hybridization allows a nRMSE gain upper than 14 percentage points compared to the persistence estimation (nRMSE=37% versus 51%).

Cyril Voyant, Wani W. Tamas, Christophe Paoli et al.

A Multi-Layer Perceptron (MLP) defines a family of artificial neural networks often used in TS modeling and forecasting. Because of its "black box" aspect, many researchers refuse to use it. Moreover, the optimization (often based on the exhaustive approach where "all" configurations are tested) and learning phases of this artificial intelligence tool (often based on the Levenberg-Marquardt algorithm; LMA) are weaknesses of this approach (exhaustively and local minima). These two tasks must be repeated depending on the knowledge of each new problem studied, making the process, long, laborious and not systematically robust. In this paper a pruning process is proposed. This method allows, during the training phase, to carry out an inputs selecting method activating (or not) inter-nodes connections in order to verify if forecasting is improved. We propose to use iteratively the popular damped least-squares method to activate inputs and neurons. A first pass is applied to 10% of the learning sample to determine weights significantly different from 0 and delete other. Then a classical batch process based on LMA is used with the new MLP. The validation is done using 25 measured meteorological TS and cross-comparing the prediction results of the classical LMA and the 2-stage LMA.

Cyril Voyant, Christophe Paoli, Marc Muselli et al.

Considering the grid manager's point of view, needs in terms of prediction of intermittent energy like the photovoltaic resource can be distinguished according to the considered horizon: following days (d+1, d+2 and d+3), next day by hourly step (h+24), next hour (h+1) and next few minutes (m+5 e.g.). Through this work, we have identified methodologies using time series models for the prediction horizon of global radiation and photovoltaic power. What we present here is a comparison of different predictors developed and tested to propose a hierarchy. For horizons d+1 and h+1, without advanced ad hoc time series pre-processing (stationarity) we find it is not easy to differentiate between autoregressive moving average (ARMA) and multilayer perceptron (MLP). However we observed that using exogenous variables improves significantly the results for MLP . We have shown that the MLP were more adapted for horizons h+24 and m+5. In summary, our results are complementary and improve the existing prediction techniques with innovative tools: stationarity, numerical weather prediction combination, MLP and ARMA hybridization, multivariate analysis, time index, etc.

Wani W. Tamas, Gilles Notton, Christophe Paoli et al.

Atmospheric pollutants concentration forecasting is an important issue in air quality monitoring. Qualitair Corse, the organization responsible for monitoring air quality in Corsica (France) region, needs to develop a short-term prediction model to lead its mission of information towards the public. Various deterministic models exist for meso-scale or local forecasting, but need powerful large variable sets, a good knowledge of atmospheric processes, and can be inaccurate because of local climatical or geographical particularities, as observed in Corsica, a mountainous island located in a Mediterranean Sea. As a result, we focus in this study on statistical models, and particularly Artificial Neural Networks (ANN) that have shown good results in the prediction of ozone concentration at horizon h+1 with data measured locally. The purpose of this study is to build a predictor to realize predictions of ozone and PM10 at horizon d+1 in Corsica in order to be able to anticipate pollution peak formation and to take appropriated prevention measures. Specific meteorological conditions are known to lead to particular pollution event in Corsica (e.g. Saharan dust event). Therefore, several ANN models will be used, for meteorological conditions clustering and for operational forecasting.

Philippe Lauret, Auline Rodler, Marc Muselli et al.

This paper proposes to use a rather new modelling approach in the realm of solar radiation forecasting. In this work, two forecasting models: Autoregressive Moving Average (ARMA) and Neural Network (NN) models are combined to form a model committee. The Bayesian inference is used to affect a probability to each model in the committee. Hence, each model's predictions are weighted by their respective probability. The models are fitted to one year of hourly Global Horizontal Irradiance (GHI) measurements. Another year (the test set) is used for making genuine one hour ahead (h+1) out-of-sample forecast comparisons. The proposed approach is benchmarked against the persistence model. The very first results show an improvement brought by this approach.

Cyril Voyant, Marc Muselli, Christophe Paoli et al.

We propose in this paper an original technique to predict global radiation using a hybrid ARMA/ANN model and data issued from a numerical weather prediction model (ALADIN). We particularly look at the Multi-Layer Perceptron. After optimizing our architecture with ALADIN and endogenous data previously made stationary and using an innovative pre-input layer selection method, we combined it to an ARMA model from a rule based on the analysis of hourly data series. This model has been used to forecast the hourly global radiation for five places in Mediterranean area. Our technique outperforms classical models for all the places. The nRMSE for our hybrid model ANN/ARMA is 14.9% compared to 26.2% for the naïve persistence predictor. Note that in the stand alone ANN case the nRMSE is 18.4%. Finally, in order to discuss the reliability of the forecaster outputs, a complementary study concerning the confidence interval of each prediction is proposed

Christophe Paoli, Cyril Voyant, Marc Muselli et al.

In this paper, we present an application of neural networks in the renewable energy domain. We have developed a methodology for the daily prediction of global solar radiation on a horizontal surface. We use an ad-hoc time series preprocessing and a Multi-Layer Perceptron (MLP) in order to predict solar radiation at daily horizon. First results are promising with nRMSE < 21% and RMSE < 998 Wh/m2. Our optimized MLP presents prediction similar to or even better than conventional methods such as ARIMA techniques, Bayesian inference, Markov chains and k-Nearest-Neighbors approximators. Moreover we found that our data preprocessing approach can reduce significantly forecasting errors.