S4S project – H2elios Neurocams

“H2elios Neurocams” is part of the EPFL Solutions4Sustainability projects.

Photo-voltaic (PV) generation is experiencing a significant growth thanks to the decreasing costs of installations and reduced carbon footprint. The progressive displacement of conventional electrical power generation in favor of intermittent renewables like PV, requires energy storage systems to store and restore electrical energy to ensure reliable operation of interconnected power systems.

The need to accurately forecast solar irradiance with local precision and global reach, in different time horizons from minutes – hours – days – months- is paramount to better schedule energy storage, dispatch and trading.

By utilizing a dataset comprising of publicly available images captured by webcams and combining with forecasts produced by conventional weather prediction methods, we aim to design and train specific neural networks to forecast the solar irradiance for intra-day time horizons ranging from 2 to 4 hours – a notably difficult time horizon to predict.

The intra-day solar irradiance forecast will empower EPFL to accurately predict on-campus photovoltaic (PV) power production with the ultimate goal of improving the tracking performance of day-ahead computed dispatch plan of energy resources connected to the EPFL power grid, avoiding imbalances that result in penalty charges. These charges typically range from 500K to 2 Million CHF per GWh for deviations of 2% above or below the day-ahead computed dispatch plan.

The enhancement in dispatch plan adherence will be ensured through the use of dedicated energy storage technologies composed by Li-based batteries and power-to-gas systems.

The forecasting method is conceived to be scalable from city to regional and country levels due to the widespread availability of solar irradiation data and public webcams. An economic model will be developed to incorporate technological constraints and costs associated with short- and long-term storage, prevailing wholesale energy prices, and an estimation of penalty charges. This model will determine the most suitable strategy to minimize the operational cost of the system.

The performance assessment of the developed solutions will be carried out on the EPFL low voltage microgrid as part of the EPFL Smart Grid demonstrator.

Data