Proc. of the 3rd International Conference on Smart Data (SmartData-2017)
The increase in electrical metering has created tremendous quantities of data and, as a result, possibilities for deep insights into energy usage, better energy management, and new ways of energy conservation. As buildings are responsible for a significant portion of overall energy consumption, conservation efforts targeting buildings can provide tremendous effect on energy savings. Building energy monitoring enables identification of anomalous or unexpected behaviors which, when corrected, can lead to energy savings. Although the available data is large, the limited availability of labels makes anomaly detection difficult. This research proposes a deep semi-supervised convolutional neural network with confidence sampling for electrical anomaly detection. To achieve semi-supervised learning, two sub-networks are used: the first performs reconstruction and uses unlabelled data, while the second performs classification with labelled data. The two sub-networks overlap: the encoder parameters are shared between the two. To quantify anomaly detection confidence, a valuable metric in anomaly detection, the network uses a dropout sampling method. The proposed approach has been evaluated with real-world electrical data from systems such as HVAC, lighting, and heat pumps. The results demonstrated the accuracy of the proposed anomaly detection solution.