Adel Mellit, Professor of Electronics at the Faculty of Sciences and TechnologyJijel University, Algeria. He received his M.S. Degree and Phd inElectronics from the University of Sciences Technologies (USTHB)Algiers in 2002 and 2006 respectively. Research interests of Dr. Adel Mellit focus on the application of the artificial intelligence techniques in photovoltaic systems and micro-grids, control, fault diagnosis, optimization and real time applications. He has authored and co-authored more than 100 papers in international peer reviewed journals (mostly in Elsevier and papers in conference proceedings (Mostly in IEEE) mainly on photovoltaic systems and nine book chapters. He is the Director of the Renewable Energy Laboratory at the Jijel University and is an Associate Member at the ICTP Trieste (Italy). He is actually serving on the Editorial Board of the Renewable Energy and of the Energy Journals (Elsevier Ltd).
Abstract:Advanced methods in photovoltaic output power forecasting: State of the art
In 2018, the global photovoltaic capacity reached about 500 GWp corresponding to several millions of photovoltaic (PV) systems installed worldwide. Thus, the operation and maintenance activities of such plants are today important for a great number of professionals working in this solar sector. Forecasting of photovoltaic output power play very important role in power planning and dispatching, optimal management and grid quality and stability.Designing of an accurate PV output power forecasting models stay a challenging issue and a crucial task, as the PV output power is extremely uncertain due mainly to solar irradiance variation.Forecasting methods can be classified mainly into three groups: Physical (numerical weather prediction models), statistical methods (eg. AR, ARMA, ARIMA, etc), method-based artificial intelligence techniques (eg. Neural networks, machine learning and deep learning) and the last group is a hybrid methods (eg. Combining of two methods). Different timescales forecasting are important for PV plants, for exampleintra-hourforecasts (up to 1h) are useful for grid quality and stability. Intra-day forecasts (up to 6 hours) are essential and could be used for optimal integration.Forecasts up to one-day mainly used for unit commitment planning and dispatching power. Up to one-week forecasts could be used for trading, management and maintenance. The main aim of this talk is to give a detailed overview on the available forecasting methods, special attention will be paid to methods that can accurately forecast the output PV power in different timescales, including artificial intelligence techniques, machine learning, deep learning and numerical weather forecasting models.Advantages and limits ofreviewed methods in terms of accuracy, complexity, cost-effectiveness will be discussed in this presentation.