DeepTrack – Digital Twins and Deep Learning and Their Implementation in PV Tracker Algorithms

How should PV trackers optimally track? This question is being researched in the project »DeepTrack« together with PV Zimmermann Tracker GmbH. In DeepTrack, the latest AI and power plant simulation methods are applied to maximize the yield of complex power plants, such as Agri-PV power plants. The methods are implemented and investigated in a pilot plant at the solar test field in Merdingen.

Modern artificial intelligence approaches such as digital twinning and deep learning show great potential for the renewable energy sector. Especially for tracked photovoltaic plants with bifacial solar modules, there is the possibility to extend control algorithms. In doing so, the entire complexity of the power plant environment can be taken into account to increase the electrical yield in such tracked systems. These novel AI concepts are being developed and implemented in a prototype system in Merdingen, Germany, with a focus on system enhancement and diffuse light optimization, as well as system adaptation to novel application areas such as agri-photovoltaics. To this end, state-of-the-art PV monitoring and modeling tools (digital twins) are coupled with weather forecasts to achieve optimal tracking positions for the overall photovoltaic system yield. For example, microclimatic requirements and plant-level needs are considered within an agri-photovoltaic system. Neural networks are then trained based on the digital twin results to create a less computationally-intensive and predictive metamodel. The computational cost of the approach will be reduced to the point where it can be implemented on standard industrial computers in the power plant field, thus showing prospects for marketable product applications.