Mean absolute log-return and solar radiation forecastabity
Cyril Voyant, Ted Soubdhan, Philippe Lauret, Mathieu David, Marc Muselli


Papier accepté pour la conférence ISES Solar World Congress 2015
Lors de cette conférence nous exposerons les principaux résultats de nos travaux publiés dans l'article Statistical parameters as a means to a priori assess the accuracy of solar forecasting models. Le résumé est:
"In this paper we propose to determinate and to test a set of statistical parameters (20) to estimate the predictability of the global horizontal irradiation time series and thereby propose a new prospective tool indicating the expected error regardless the forecasting methods a modeller can possibly implement. The mean absolute log return, which is a tool usually used in econometry, proves to be a very good estimator. Some examples of the use of this tool are exposed, showing the interest of this statistical parameter in concrete cases of predictions or optimizations."

ises.pdf ISES.pdf  (483.2 Ko)


Rédigé le Vendredi 24 Juillet 2015 à 08:28

Statistical parameters as a means to a priori assess the accuracy of solar forecasting models
Cyril Voyant, Ted Soubdhan, Philippe Lauret, Mathieu David, Marc Muselli


Nouveau papier accepté dans la revue Energy Journal
Dans ce papier , nous démontrons qu'il existe des paramètres statistiques (tel le rendement logarithmique) qui permettent d'estimer l'erreur de prédiction relative à une série temporelle spécifique, et ce quelque soit la méthode de machine learning utilisée. A travers 3 exemples simples, nous proposons une méthodologie d'utilisation de ces paramètres. Le résumé est:

"In this paper we propose to determinate and to test a set of statistical parameters (20) to estimate the predictability of the global horizontal irradiation time series and thereby propose a new prospective tool indicating the expected error regardless the forecasting methods a modeller can possibly implement. The mean absolute log return, which is a tool usually used in econometry, proves to be a very good estimator. Some examples of the use of this tool are exposed, showing the interest of this statistical parameter in concrete cases of predictions or optimizations."

discriminent_param.pdf discriminent_param.pdf  (836.37 Ko)


Rédigé le Vendredi 24 Juillet 2015 à 08:18

Energétique, génie des procédés (62° CNU)


Nous recherchons des candidats pour une thèse débutant en septembre 2015. Cette thèse sera rattachée à l'UMR CNRS SPE 6134 (Projet Structurant ENR) de l'université de Corse. L'étudiant sera basé sur Ajaccio.

Modélisation stochastique et prévisions numériques du rayonnement solaire pour le pilotage de la plateforme hybride de R&D MYRTE.

Résumé: Lors de cette thèse, l’étudiant aura la tâche de manipuler des séries temporelles, des outils de modélisation stochastique (ANN, SVM, etc.), des sorties de modèles globaux (ECMWF, AROME, etc.), et des modèles numériques régionaux haute résolution (WRF ou Méso-NH) dans le but de développer un outil de prévision du rayonnement solaire à différents horizons, intégré au système de contrôle-commande de la plateforme PV MYRTE, pour la production d’énergie électrique sur un réseau non-interconnecté.

Contact:
-MUSELLI Marc, PRU, directeur de thèse
marc.muselli@univ-corse.fr / 06 08 07 05 01
-VOYANT Cyril, Dr, (PAST – adossement Projet ENR), co-directeur de thèse
voyant@univ-corse.fr

1_fiche_these_2015_16.doc 1 Fiche_Thèse 2015-16.doc  (63 Ko)


Rédigé le Dimanche 19 Avril 2015 à 21:18

A benchmarking of machine learning techniques for solar radiation forecasting in an insular context
Philippe Lauret, Cyril Voyant, Ted Soubdhan, Mathieu David, Philippe Poggi


Ce papier est le fruit de la collaboration de 3 universités Françaises présentant une caractéristique commune : l'insularité. En effet, que ce soit à La Réunion, la Guadeloupe ou encore la Corse, la problématique liée à l'approvisionnement énergétique est dominante. Dans ce papier une large revue des méthodes de prévision dites "Machine Learning" est proposée et des tests de prévision de rayonnement global solaire sont réalisés sur les trois iles. Le but étant de déterminer les modèles les plus performants et surtout d'essayer d'établir des règles de sélection de modèle en fonction du site et des caractéristiques de l'étude. Le résumé de cet article est :
"In this paper, we propose a benchmarking of supervised machine learning techniques (neural networks, Gaussian processes and support vector machines) in order to forecast the Global Horizontal solar Irradiance (GHI). We also include in this benchmark a simple linear autoregressive (AR) model as well as two naive models based on persistence of the GHI and persistence of the clear sky index (denoted herein scaled persistence model). The models are calibrated and validated with data from three French islands: Corsica (41.91°N; 8.73°E), Guadeloupe (16.26°N; 61.51°W) and Reunion (21.34°S ; 55.49°E). The main findings of this work are, that for hour ahead solar forecasting, the machine learning techniques slightly improve the performances exhibited by the linear AR and the scaled persistence model. However, the improvement appears to be more pronounced in case of unstable sky conditions. These nonlinear techniques start to outperform their simple counterparts for forecasting horizons greater than one hour."

https://www.researchgate.net/publication/270214354_A_benchmarking_of_machine_learning_techniques_for_solar_radiation_forecasting_in_an_insular_context https://www.researchgate.net/publication/270214354_A_benchmarking_of_machine_learning_techniques_for_solar_radiation_forecasting_in_an_insular_context

Nouveau papier accepté dans la revue Solar Energy

Rédigé le Jeudi 1 Janvier 2015 à 16:07

Heterogeneous transfer functions MultiLayer Perceptron (MLP) for meteorological time series forecasting International Journal of Modeling, Simulation, and Scientific Computing
C. Voyant, ML. Nivet, C. Paoli, M. Muselli,G. Notton


Article accepté dans la revue "International Journal of Modeling, Simulation, and Scientific Computing"
Cet article fait suite à ce qui avait été exposé lors de conférence ICMSQUARE de Madrid: étude d'un MLP particulier pour la prédiction de séries temporelles météorologiques. Ces MLP sont dit à fonction de transfert hétérogène car il cumule sur leur couche cachée des fonctions de type excitatrice pure mais aussi des fonctions non bijectives de type inhibitrice/excitatrice. Le résumé de cet article est :
"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."

Rédigé le Mercredi 31 Décembre 2014 à 16:53

On meteorological forecasts for energy management and large historical data: A first look
Cyril Voyant, Cédric Join, Michel Fliess, Marie-Laure Nivet, Marc Muselli et Christophe Paoli


papier accepté pour l International Conference on Renewable Energies and Power Quality (ICREPQ’14) à la Corogne (Mars 2015)
Dans ce papier nous proposons une comparaison des deux types de méthode de prédiction de séries temporelles météorologiques:
-les méthode dites "with large historical data"
-et celles qualifiées de "without large historical data"
Le résumé de ce papier est :
"This communication is devoted to a comparison between various meteorological forecasts, for the purpose of energy management, via different time series techniques. The first group of methods necessitates a large number of historical data. The second one does not and is much easier to implement, although its performances are today only slightly inferior. Theoretical justifications are related to methods stemming from a new approach to time series, artificial neural networks, computational intelligence and machine learning. Several numerical simulations are provided and discussed."

https://hal.archives-ouvertes.fr/hal-01093635v1 https://hal.archives-ouvertes.fr/hal-01093635v1


Rédigé le Vendredi 5 Décembre 2014 à 17:35

Influence of Global Solar Radiation Typical Days on Error Forecasting Models

Ted Soubdhan, Cyril Voyant, Philippe Lauret


Papier accepté à la conférence "The Third Southern African Solar Energy Conference (SASEC2015)"
La collaboration sur la thématique de la prévision de la ressource solaire en milieu insulaire (Réunion, Guadeloupe et Corse) nous a permis de proposer un nouveau papier lors de cette conférence en Afrique du sud. Le résumé est:
"Large and frequent variations of solar radiation can be observed in tropical climates with amplitudes reaching 800 W/m² and occurring within a short time interval, from few seconds to few minutes, according to the geographical location. Such fluctuations can be due for example to the dynamic of clouds which can be very complex and depend on cloud type, size, speed and spatial distribution and, more generally, due to some specific local meteorological conditions. In this work, we have led an analysis of the error of different global solar radiation prediction models. Different predictions models where performed such as machine learning techniques (Neural Networks, Gaussian processes and support vector machines) in order to forecast the Global Horizontal solar Irradiance (GHI). We also include in this study a simple linear autoregressive (AR) model as well as two naive models based on persistence of the GHI and persistence of the clear sky index (denoted herein scaled persistence model). The models are calibrated and tested with data from three 3 French islands: Corsica (42.15°N ; 9.08°E), Guadeloupe (16.25°N ; 61.58°W) and Reunion (21.15°S ; 55.5°E). Guadeloupe and Reunion are located in a subtropical climatic zone whereas Corsica is in a tempered climatic zone. Hence the global solar radiation variation differs significantly. The output error of the different models was quantified by the nRSME. In order to quantify the influence of the global solar radiation variation on the error forecasting models we performed a classification of typical days according to different typical days. Each class of typical day is defined by a variation of global solar radiation rate. For each class and for each location, the selected forecasting models where performed and the error was quantified. With this analysis a global solar radiation forecasting models can be selected according to the location, the global solar radiation fluctuations, hence the meteorological conditions."

https://www.researchgate.net/publication/270369794_INFLUENCE_OF_GLOBAL_SOLAR_RADIATION_TYPICAL_DAYS_ON_FORECASTING_MODELS_ERROR https://www.researchgate.net/publication/270369794_INFLUENCE_OF_GLOBAL_SOLAR_RADIATION_TYPICAL_DAYS_ON_FORECASTING_MODELS_ERROR


Rédigé le Mardi 21 Octobre 2014 à 13:04

Short-term solar irradiance and irradiation forecasts via different time series techniques: A preliminary study

C. Join, C. Voyant, M. Fliess, M. Muselli, ML. Nivet, C. Paoli, F. Chaxel.


Papier accepté lors de " the 3rd International Symposium on Environment-Friendly Energies and Applications (EFEA 2014)"
Avec ce papier, nous proposons une comparaison des méthodes de prédictions du rayonnement global nécessitant beaucoup d'historique de mesure (ANN) et d'autre plus simples à mettre en oeuvre et ne nécessitant que de très peu de de données (pas de phase d'apprentissage, i.e. persistance, model-free control). Le résumé de ce papier est:
"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 who need them, especially when those data are quite noisy."

http://hal.archives-ouvertes.fr/index.php?halsid=j888b60cn1eica5seponkg6pr3&view_this_doc=hal-01068569&version=1 http://hal.archives-ouvertes.fr/index.php?halsid=j888b60cn1eica5seponkg6pr3&view_this_doc=hal-01068569&version=1


Rédigé le Jeudi 25 Septembre 2014 à 18:39

Meteorological time series forecasting with pruned multi-layer perceptron and 2-stage Levenberg-Marquardt method

cyril voyant, Wani Tamas, Marie-Laure Nivet, Gilles Notton, Christophe Paoli, Aurélia Balu, Marc Muselli


Nouvel article accepté dans un journal indersceince: "International Journal of Modelling, Identification and Control"
Cet article fait suite au travail mené et valorisé lors de l'ICM² 2014 de Prague. La nouvelle méthodologie de pruning exposée semble donner de bon résultats, il convient maintenant de la comparer avec les méthodes existantes afin de d'établir une hiérarchisation des algorithmes sur le cas réel de séries temporelles de rayonnement global. L'abstract :

"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."

https://www.researchgate.net/publication/263717805_Meteorological_time_series_forecasting_with_pruned_multi-layer_perceptron_and_2-stage_Levenberg-Marquardt_method?ev=prf_pub https://www.researchgate.net/publication/263717805_Meteorological_time_series_forecasting_with_pruned_multi-layer_perceptron_and_2-stage_Levenberg-Marquardt_method?ev=prf_pub


Rédigé le Mardi 8 Juillet 2014 à 15:29

Meteorological time series forecasting based on MLP modelling using heterogeneous transfer functions
C. Voyant, ML. Nivet, C. Paoli, M. Muselli and G. Notton


Papier accepté à la conférence "International Conference on Mathematical Modeling in Physical Sciences"
Dans ce papier nous proposons une nouvelle variété de perceptrons multicouches avec plusieurs fonctions de transferts. L'apport de ce nouvel outil est testé sur différentes séries temporelles météorologiques, montrant ainsi un léger gain dés qu'un indice temporelle est positionné en entrée (variable auto-indicatice) .
Le résumé de ce papier est :
"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."

icmsquare2014.pdf ICMSQUARE2014.pdf  (439.5 Ko)


Rédigé le Mardi 29 Avril 2014 à 06:27
1 2 3 4 5 » ... 6