In early January, a group of researchers affiliated with the EnerSHelF project published the article “Day-Ahead Electric Load Forecast for a Ghanaian Health Facility Using Different Algorithms” in the peer-reviewed open access journal “energies”
Summary by Samer Chaaraoui
The main scope of the research article is to identify a forecasting algorithm, which is most suitable for electric load-forecasting purposes. To address the peculiarities of the Ghanaian health sector, real load data from one of the project sites – the St. Dominic’s hospital in Akwatia – are used to conduct this comparison. The main idea of performing such forecasts is the possibility to apply a so-called model predictive control for PV-hybrid-systems, which uses predictions to optimize the dispatch of the PV-hybrid-system. It enables a higher efficiency and reliability compared to the widely used rule-based control.
The main finding of the research article is that forecast algorithms based on artificial intelligence, in particular long-short-term-memory neural networks, show the most promising results with regards to plasticity, robustness, and accuracy. However, the authors emphasize that they need to conduct further analysis with data from the field measurements and from the national utility provider. This will help to make a statement regarding the potential to generalize such forecasting methods.
Also, the exceptionally high measurement frequency of the electric load at the measurement sites is unique in this field, which enables the researchers to run simulations close to real conditions. By doing so, the gap between the theoretical development and the practical implementation of such algorithms becomes much smaller.