PhD opportunities

Methods for seamless multi-temporal forecasting of renewable production based on large amounts of heterogeneous data.

Thesis proposal

Area of expertiseEnergtique et gnie des procds
Doctoral SchoolSystems Engineering, Materials, Mechanics, Energy
SupervisorM. Georges KARINIOTAKIS
Research unitEnergy and Processes
Starting dateMarch 1st 2020
KeywordsRenewable Energies, Photovoltaics, Wind energy, Run-of-the river hydro, Data science, Smart-grids
AbstractContext and background:

Short-term forecasts of the power output of weather-dependent renewable energy (RES) power plants (wind farms, photovoltaic (PV) plants, run-of-the-river hydro,...) are required for their efficient integration into power systems and electricity markets. Research on RES forecasting dates back to the 80s. Today, the forecasting technology is quite mature thanks to a wealth of proposed approaches in the literature and to companies that develop, provide in the form of services, and use operationally forecasting solutions. The use of forecasts is generalised by transmission or distribution system operators, aggregators, virtual power plants operators and traders, RES plants operators and others. However, the performance of the actual forecasting models present limits in terms of accuracy due to the difficulty of the problem itself that relates to the complexity of the weather systems that drive RES generation, to the inherent uncertainties, the data quality, the computational limitations and other factors. As RES penetration in the electricity grids increases, the resulting challenges amplify. There is a consensus today that further multi-disciplinary research is required to significantly improve the accuracy and reliability of the forecasting models. Under this problematic, the EU supports the new H2020 research project Smart4RES launched in 2019 to address a number of promising directions for improving RES predictability. This PhD research is proposed in the frame of this project and is carried out at the Centre PERSEE of MINES ParisTech that disposes a long experience international visibility in the field.

Scientific objectives:

Although the type of research in RES forecasting models in the previous years was quite incremental, today there are several factors that make it possible to follow more disruptive directions. Firstly, the increased computational capacity and the development of artificial intelligence offer new possibilities. Secondly, with the emergence of digitalisation and other technologies there are much more data available than in the past. The aim of this research work is to propose RES forecasting models able to consider input from heterogeneous sources of data including measured data from geographically distributed RES plants and meteo stations, smart-meters, different types of satellite images, lidar, radar, sky-camera imagers, numerical weather predictions from different models and others. Each of these data sources contributes in improving predictability for specific prediction horizons (i.e. very short-term up to 5 min ahead, short-term up to 6 hours or longer-term up to a few days). For this reason, in the existing literature, RES forecasting models are built upon specific types of data and for specific prediction horizons. This results to a high data dependability of the models. For end-users that use forecasts for different time frames for decision making in complex applications (i.e. trading at multiple electricity markets, management of hybrid plants combined with storage), this translates to multiple forecasting models to tune and maintain. Here, the aim, and main scientific challenge, is to propose an approach that permits the simplification of the forecasting model chain by proposing a prredition approach that covers multiple prediction time frames while considering all the available sources of data as input. This model should be probabilistic, that is able to predict not only the most likely prediction but the associated uncertainty but also autoadaptive for continuous forecasting.


The research work will start with a literature review. The candidate will then familiarise with a portfolio of prediction models developed at MINES ParisTech and will carry out research on novel approaches i.e. based on AI. A number of test cases including heterogeneous types of data will be considered to evaluate the prediction models developed. The case of wind, solar, eventually run-of the river hydro, and aggregated production from virtual power plants will be considered to evaluate the proposed approaches.

Expected results:

A novel probabilistic forecasting approach for wind, PV and aggregated RES production. Evaluation results based on measured time-series from RES plants. A study on the incremental benefits in terms of performance brought by each source of data when considered as input to a model. A cost benefit analysis on the different sources of data (cost of data compared to the benefit

Engineer and / or Master of Science - Good level of general and scientific culture. Good analytical, synthesis, innovation and communication skills. Qualities of adaptability and creativity. Motivation for research activity. Coherent professional project. The desired profile should have a background in applied mathematics (statistics), data science, artificial intelligence and eventually electrical engineering. Skills in computer programming (eg MATLAB) are required. The candidate must be motivated to work in a team.


To apply please:
1) send your CV and motivation letter to using as subject of your email THESIS RESforecast 2020 PERSEE ,
2) AND fill in the on-line form:
FundingContrat de recherche
PartnershipEuropen H2020 Smart4RES Next Generation Modelling and Forecasting of Variable Renewable Generation for Large-scale Integration in Energy Systems and Markets