The European Union (EU) policy for climate and energy imposes significant targets for a high integration of renewable energy sources (RES) in the period from 2020 to 2030. More flexibility of the power system is required to respond to variability and uncertainty of the renewable generation without compromise the network reliability and security. The European promotion for a fast deployment of Smart Grids will allow accommodating high intermittent generation and providing the flexibility requirements. Distribution System Operators (DSOs) are facing numerous challenges of the existing infrastructures and they need to know in advance the amount of variable resources available within operations. Solar power generation forecasts are important tools for power grid operations. The solar power generation forecasting has also a significant influence on grid operation costs, considering that forecast uncertainty increases further the overall costs for utilities, customers, system operators and electricity market participants. Therefore the knowledge in advance of the solar generation and high forecast accuracy can lead to more efficient management of distribution systems in a cost-effective way, enabling a large-scale deployment of variable distributed resources.
The state-of-the-art for photovoltaic (PV) power forecasts deals with various forecasting techniques, several temporal and spatial horizons and different approaches. The solar power forecasting is traditionally performed by using statistical methods which use past observations of weather variables and solar power as input. Recently, Artificial Intelligence techniques are introduced for predicting the PV power output, extracting relations on past data to predict the PV power output without any information from system.
Forecasts can be classified is according to the time horizons, impacting on different aspects of grid operation. The intraday forecasts ensure the grid quality and stability, the one-day forecasts are normally used for planning, unit commitment and energy market. PV forecast can be performed for a single plant or at regional scale (spatial horizon of forecasts). Considering that grid operators deal with balancing between demand and supply in the electric system, the regional forecasts are more attractive than point forecasts. Depending on the availability of data regional forecasts can be approached by several ways. Generally, only PV power from some plants is known, so the up-scaling method represents the most suitable solution, where a set of PV plants, selected by correlation parameters and smoothing effect, became representative of the power output of the whole ensemble.
The overall objectives of this project is to improve the forecast accuracy for PV generation by the development of a novel spatial-temporal forecasting method. A forecasting method on a regional scale and at multi-time horizon for
PV generation has been developed using machine learning techniques applied to huge amount of power generation data collected from solar PV plants in the smart grids environment. Such data collection has supported the investigation of the exiting correlations between the different PV plants in order to improve the predictions accuracy with respect to time-domain forecasting models.
Further goals of the project were:
- To address reliability issues of PV systems in operational condition by using Artificial intelligence techniques, to identify and characterize failures and successively support the predictive maintenance strategies.
- To transfer of knowledge on Artificial intelligence techniques for PV generation. The findings of the project were disseminated in research, academia and the industry sectors.
- To foster the development of the individual researcher.
