Deep Online Hierarchical Unsupervised Learning for Pattern Mining from Utility Usage Data

摘要

Machine learning approaches for non-intrusive load monitoring (NILM) have focused on supervised algorithms. Unsupervised approaches can be more interesting and of more practical use in real case scenarios. More specifically, they do not require labelled training data to be collected from individual appliances and the algorithm can be deployed to operate on the measured aggregate data directly. In this paper, we propose a fully unsupervised NILM framework based on Deep Belief network (DBN) and online Latent Dirichlet Allocation (LDA). Firstly, the raw signals of the house utilities are fed into DBN to extract low-level generic features in an unsupervised fashion, and then the hierarchical Bayesian model, LDA, learns high-level features that capture the correlations between the low-level ones. Thus, the proposed method (DBN-LDA) harnesses the DBN's ability of learning distributed hierarchies of features to extract sophisticated appliances-specific features without the need of precise human-crafted input representations. The clustering power of the hierarchical Bayesian models helps further summarise the input data by extracting higher-level information representing the residents' energy consumption patterns. Using Deep-Hierarchical models reduces the computational complexity since LDA is not directly applied to the raw data. The computational efficiency is crucial as our application involves massive data from different types of utility usages. Moreover, we develop a novel online inference algorithm to cope with this big data. Another novelty of this work is that the data is a combination of different utilities (e.g., electricity, water and gas) and some sensors measurements. Finally, we propose different methods to evaluate the results and preliminary experiments show that the DBN-LDA is promising to extract useful patterns.