Machine Learning Algorithm Integrated in a Residential Home Automation Software

The growing adoption of residential monitoring and automation systems due to the rapid development of LVAC and LVDC microgrids now requires SCADA (Supervisory Control And Data Acquisition) systems that can manage a huge and varied amount of data. In this sense, the recent development of artificial intelligence algorithms allows us to build predictive scenarios that help the system even more, improving energy efficiency, comfort, and building useful automation cycles. However, the irregular temporal resolution, limited datasets, and anomalous values typically found in real data sets pose significant challenges for the direct application of machine learning (ML) algorithms. This paper presents a methodology for the pre-processing and modeling of energy data collected by the LAMBDA Laboratory of Sapienza University of Rome, where the Home Assistant (HA) platform acts as a central data hub. Two ML architectures were studied: An LSTM (Long Short-Term Memory) recurrent neural network (RNN) and a FCNN (Fully Connected Neural Network). Finally, the two networks were tested and compared, making predictions on the power generated by the PV and also on the internal temperature of the laboratory.