Activity Prediction for Agent-based Home Energy Management

In this paper, we address the problem of predicting the usage of home appliances where a key challenge is to model the everyday routine of homeowners and the inter–dependency between the use of different appliances. In particular, given an efficient day–ahead prediction of electrical usage, home energy management systems can suggest homeowners when is the best time to run appliances in order to save cost, without violating their preferred everyday habits. To this end, we propose an agent based prediction algorithm that captures the everyday habits by exploiting their periodic features. In addition, our algorithm uses a episode generation hidden Markov model (EGH) to model the inter–dependency between appliances. We demonstrate that our approach outperforms existing methods by up to 40% in experiments based on real–world data from a prominent database of home energy usage. We also show that the computational cost of our algorithm is 100 times lower on average, compared to that of the benchmark algorithms.