LAUSR

The key advantage of using power-utility-owned smart meters is the ability to transmit electrical energy consumption data to power utilities’ remote data centers for various purposes, such as billing. Several useful consumer-centric use cases can also be identified for the collection and further analysis of consumers’ electrical energy consumption data from smart meters. One of the use cases is home automation. Recent related solutions for home automation involving home security and healthcare depend on the installation of sensors and/or other devices such as video cameras, which have high costs for installation and annual maintenance. Because the electrical energy consumption patterns mined from smart meter data are indicative of residents’ daily life, it is possible to develop a new home automation approach based on energy decomposition for smart home automation. Accordingly, in this work, a smart home energy management system (SHEMS) utilizing a parallel-processing-implemented, GPU-accelerated neurocomputing-based time-series load modeling and forecasting mechanism is proposed for smart home automation. Energy decomposition is used to facilitate the time-series load modeling and forecasting mechanism, which tracks appliance-level electrical energy consumption to be quantitatively modeled from circuit-level consumption, with no intrusive deployment of networked plug-level power meters for individual electrical home appliances. For the neurocomputing approach applied in this mechanism, an autoregressive multilayer perceptron methodology is compared against a stacked long short-term memory methodology. The presented neurocomputing-based time-series load modeling and forecasting mechanism facilitated by energy decomposition is capable of predicting residents’ daily behavioral patterns by nonintrusively analyzing and modeling relevant electrical home appliances based on their past trends for smart home automation.