Machine Learning and Clustering techniques for Anomalies Detection in Household Appliances

The incredible growth of the Smart Grids technology during the last 5 to 10 years has seen in parallel the need to have technologies capable of improving the user experience. The user has gained new control over his devices, but he has yet to obtain the right awareness, for example of his consumption. One of the leading sectors in which the use of technology can improve the user experience is, in fact, in the appliances consumption and behaviours. This project aims at creating a framework placed in this environment, specifically focused on the task of detecting and analysing anomalies in the household appliances. An anomaly is an abnormal event which differs significantly from the normal behaviour of the data set in which it occurs. The detection of these phenomena is crucial, to find problems and trying to prevent them in the future. The project is aimed at studying the household appliances, but it will not focus only on a specific use case: the novelty of the project is the fact that the developed framework will analyse different types of anomalies. In particular, the framework is composed of three main sections: i) Single-Point Outliers, ii) Past and Current Anomalous Trends and iii) Bad User Behaviours. The first two sections will handle a disaggregated consumption data set describing different appliances, while the last section will be focused on a data set composed of Meta-Data of the household general characteristics like yearly consumption, squared meters, or family components. The methodologies used derive from the branch of Machine Learning dedicated to the unsupervised algorithms: the use case discourages the use of supervised approaches because the search for anomalies denies the possibility of having labelled data. In particular, the project will focus on studying Change Point Detection and Anomaly Detection algorithms for the first two sections, while the last part will handle Clustering Algorithms.