During our long-term research collaboration with SFÉRA, we have addressed several research-industrial problems in the energy domain. We’ve worked on predicting electricity production from renewable sources, detecting anomalous electricity consumption among consumers, and estimating energy system deviation.
One of the most interesting problems was a day-ahead and intraday photovoltaic prediction. We’ve utilized several approaches and machine learning methods with the goal of providing the most accurate estimates of energy production by photovoltaic panels for the upcoming time period, considering the weather and current season.
Among the most successful approaches in solving this problem was transfer learning. Transfer learning is a machine learning technique where a model developed for one task is reused as the starting point for a model on a second task. It leverages the knowledge gained from a previously learned task to improve generalization in a new task. This approach is particularly useful when the new task has limited data, as it allows the new model to benefit from the patterns and representations learned from a larger, related dataset.
The limited amount of data is almost always the case for newly installed photovoltaic panels. Therefore, we’ve utilized data from multiple photovoltaic panel locations in the Central European region to create a large model that can be easily calibrated for new photovoltaic panel installations, ensuring more precise prediction outputs.
Another impactful and practical task was the design of a method and development of an application for estimating the optimal size of photovoltaic panels and battery storage for households and businesses. The calculation takes into account the consumption pattern of the selected household/business as well as the energy price. In addition to the size of the photovoltaic panels and battery storage, the output of the optimization also contains the information about the potential financial savings resulting from effective utilization of the generated energy.
Currently, we are collaborating on solving a very challenging research problem, which is the short-term prediction of power outages and faults in the energy infrastructure. To correctly predict the upcoming power outage, machine learning models utilize statistical information about historical fault events and weather forecasts.
How does Dr. Krbaťa, the Innovation Department Director at SFÉRA, perceive our collaboration?
“Working with a research organisation gives us access to the latest technologies and algorithms in the field of artificial intelligence, which keeps us at the forefront of innovation and trends. Working together on projects such as renewable electricity generation prediction and photovoltaic panel sizing optimization gives us real outputs that can be used further in our solutions. Long-term collaboration also increases our ability to adapt quickly to change and solve complex problems, making the company more competitive. At the same time, together we help to increase Slovakia’s competitiveness internationally.”
Ing. Rastislav Krbaťa, PhD., MBA
SFÉRA, Innovation Department Director
