• edge intelligence
• embedded AI
• energy consumption forecasting
• energy efficient AI
• IoT
• LSTM
• resource-constrained intelligence
• resources analysis
• resources utilization
• TinyML

Machine Learning (ML) and Internet of Things (IoT) are two research fields that are progressing significantly in recent years and that are now complementary. Intelligent edge devices are an integral part of many real world applications where sensors are used to collect data that is sent to the Cloud to be processed and interpreted, with the help of ML. Bringing ML directly on embedded devices would pave the way for many new smart applications. TinyML is an emerging research field aiming at this goal. There are several open challenges in this area, including dealing with the very limited resources of the microcontrollers and with the fragmentation of the embedded world, that makes cross-platform interoperability very difficult. In this thesis, an approach has been proposed to implement energy consumption prediction on microcontrollers, using an open source TinyML inference framework, TensorFlow Lite for Microcontrollers (TFLM). An initial extensive study activity has been carried out in order to evaluate the current state-of-the-art of TinyML solutions. Then, the possibility to run a model for timeseries prediction on the Arduino Nano 33 BLE Sense has been evaluated. Once the model has been designed and implemented, an exhaustive analysis of resources utilization has been carried out, varying the number of the network’s parameters. At the end, the model has been trained and tested with real energy consumption data, in order to evaluate its effectiveness in a real context.