Photovoltaic (PV) systems have generated significant interest in recent years. There are many reasons for this: on the one hand, the costs of photovoltaic systems have fallen significantly, while on the other, the population’s environmental awareness is growing and government subsidies are supporting the expansion of renewable energies. In particular, the rising electricity prices compared to the falling prices of photovoltaic systems make them attractive. Photovoltaic systems already have a very high level of efficiency. For example, an inverter from Fronius has a maximum efficiency of 98.2 percent. The potential for improvement in these systems therefore no longer lies in the conversion of energy, but in its efficient utilisation. Energy management systems (EMS) can help to maximise self-consumption and increase energy efficiency. Photovoltaic systems generate little or no electricity at certain times, such as at night or when it is cloudy. In order to optimise the storage, use and distribution of the available energy, EMS have seen intensive development in recent years. Current research is focussing on the development of energy management systems that use symbolic regression. Mathematical models are used to optimise the energy flow in photovoltaic systems and increase overall efficiency. In order to ensure both efficient control and the protection of sensitive data, the system is to be executed directly on a microcontroller and use as few resources as possible. This thesis deals with the porting and optimisation of a heuristic energy management algorithm to an embedded device for load flow optimisation. The aim is to reduce the resource consumption of the calculation while maintaining almost the same accuracy. For this purpose, the existing algorithm is ported to resource-limited embedded hardware and evaluated in terms of computing time and accuracy. The work examines the limits and possibilities of using the heuristic calculation on microcontrollers and documents the results in order to create a basis for later integration into series products. This work shows that a significant improvement in performance can be achieved with targeted adjustments to the formula without a significant loss of accuracy. The findings provide a valuable foundation for the development of energy management systems on embedded hardware and can serve as a basis for future developments.
Optimierung eines KI-basierten Energiemanagementsystems für Embedded Hardware: Portierung symbolisch-regressiver Modelle, Performance-Steigerung und Retraining
Prielinger, A. (Author). 2025
Student thesis: Master's Thesis