Gosch, KonstantinKonstantinGoschGöpfert, JohannesJohannesGöpfertRenner, Bernd-ChristianBernd-ChristianRenner2025-10-202025-10-202025-09-1122. GI/ITG KuVS Fachgespräch Sensornetze, FGSN 2025https://hdl.handle.net/11420/58030The Internet of Things (IoT) is steadily gaining traction, but its reliance on battery power raises significant challenges. To address this, researchers are developing energyharvesting IoT devices that operate on intermittent power, eliminating the need for batteries but complicating accurate timekeeping due to unreliable energy supply. Traditional Real-Time Clocks (RTCs) and synchronization methods are ineffective under such conditions, hindering tasks that require precise timing. We propose timekeeping using TinyML to predict sunrise and sunset based on power usage patterns, enabling autonomous time inference without continuous power or external synchronization. We trained and evaluated multiple lightweight models through simulation and with real hardware, and we demonstrate their potential and shortcomings compared to conventional methods.enhttps://creativecommons.org/licenses/by/4.0/Technology::621: Applied Physics::621.3: Electrical Engineering, Electronic EngineeringConference Paperhttps://doi.org/10.15480/882.1600710.21268/20250822-110.15480/882.16007Conference Paper