Power Management of Autonomous Drones using Machine Learning

The use of micro-autonomous drones are increasingly used for commercial and defense related applications. Importantly they play a critical role in search and rescue operations during a natural or human-made calamity. However, to operate in such a volatile environment, the sensor’s quality, prolonged flight-time, and resilience to the harsh operation conditions are vital characteristics of a micro-autonomous drone. For satisfying these characteristics, an adaptable, resilient and efficient power source is required. Compared to a battery or fuel cell-based power source, the type III-V based photovoltaics (PV) have shown a higher power-to-weight ratio. However, in a fixed configuration PV based micro-autonomous drones’ performance deteriorates due to partial or complete shading conditions. Therefore, instead of fixed topology PV module, we have used a complementary metal oxide semiconductor (CMOS) embedded PV module as a power source for micro-autonomous drone. In this work, we use machine learning techniques to determine the number of shaded PV cells present in CMOS embedded PV panel. We apply several machine learning techniques to enhance the performance of reconfigurable PV based power supply operating under different partial shading conditions. We present a comparative analysis of SVM, Naive Baiyes, Random Forest, Voting Classifier and Decision Trees as the machine learning techniques, verify their accuracy and present the classification results. The outcome of this work will lead to further usage of machine learning techniques in power management of micro-autonomous drone.