This project has a relevant contribution to the following CERENA strategic areas:
- Machine Learning and Artificial Intelligence;
- Geomodelling, Geophysics and Geochemistry;
- Physical, Chemical and Biological Processes
Supervisor: Leonardo Azevedo
Co-Supervisor: Ana Catarina Braz
The PhD project’s core is to evaluate the production capacity of renewable hydrogen for Portugal from excess of renewable energy up to 2050. With this aim, the plan includes the following goals:
i) Model the spatiotemporal evolution of relevant climate variables for renewable energy generation for the upcoming 30 years at the country level using geostatistics and geo-spatial data science techniques and applied to existing satellite data and direct observations.
ii) Build spatiotemporal wind turbine (WT) and photovoltaic (PV) power generation models to produce temporal WT/PV power density maps and statistical-based forecasting models of the total load demand considering socio-economic, calendar, and climatic factors.
iii) Optimal design and sizing, and hybrid model control of alkaline (ALKEL) and proton exchange membrane (PEMEL) electrolyzer plants, using systematic methods to boost energy and cost efficiencies, keeping a safe operation. A comparison between ALKEL and PEMEL will be performed.
The research plan comprises:
Task 1 - Spatiotemporal modeling of relevant climate variables (12M). Statistical-based models of the spatial-temporal distribution of climate variables (e.g., solar irradiance, wind speed, etc.) will be generated for the next 30 years using legacy data available at IPMA and from LandSaF satellite. The spatial description will be modeled through geostatistical simulation and co-simulation methods, integrating data with distinct quality, and assessing the uncertainty about predictions. The spatial models will be evolved in time with the application of innovative geo-spatial data science tools (e.g., deep neural networks; DNN) to model the temporal evolution of the system.
Task 2 - Forecasting of renewable energy power generation and load demand (8M). Temporal WT/PV power density maps will be built from climate models (T1). A hybrid mechanistic-ML (e.g., DNN) modeling approach will be developed by combining operating power curves of WTs and PV modules with historical data available at REN, the location and characteristics of installed WT/PV farms in Portugal. Load demand time series will be gathered from legacy data and linked with calendar, climate, and social-economic data. Stochastic optimization algorithms (e.g., particle swarm techniques) will be considered to train the models.
Task 3 - Evaluation of electrolyzer plants in terms of economic performance and dynamic response (16M). ALKEL/PEMEL plants (electrolysis cells, gas separation, H2 purification) will be modeled and designed in detail using specialized software. Heat/power integration methods and flowsheet optimization will be applied to reduce levelized cost of energy (LCOE) and enhance energy efficiency for various H2 production rates. Optimal sizing of WT/PV/electrolyzer/H2 storage systems satisfying the forecasted load demand within the timeframe will be conducted. These systems’ dynamic response will be analyzed, and mechanistic-ML hybrid control schemes developed.
We expect to obtain the following results:
i) a model of the spatiotemporal evolution of climate variables combining geostatistics with ML;
ii) a set of maps describing alternative scenarios of spatiotemporal evolution of climate variables produced with 1);
iii) The total amount and the geographical distribution of wind/solar energy, load demand time series, average WT/PV power density, and temporal WT/PV power density maps forecasting models;
iv) High-fidelity models of electrolyzers and respective plants optimized design.;
v) Optimal sizing of electrolyzers and scheduling of hydrogen production.
Comparative evaluation between WT/PV/electrolyzer/H2 storage systems using ALKEL or PEMEL electrolyzers in terms of LCOE and dynamic response to several dispatch scenarios, particularly start-up and shutdown phases.