Cognitive data fusion for improving flexibility in smart homes

With the advancements in information and communication technology, accurate monitoring of energy consumption in smarts home has become feasible. This results in improved flexibility at household level by optimizing the appliances' scheduling. However, this is challenging at the same time due to uncertainties in energy consumption, and therefore demands a highly accurate forecasting of appliance-level energy consumption. In this context, this work presents a scheme where the energy consumption patterns of multiple appliances are explored jointly. The energy consumption of several appliances which are recorded as a time-series data, are clustered based on a well-established algorithm named k-shape. Utilizing k-shape time-series clustering algorithm, the consumption relation between different appliances is observed by computing the cross-correlation phenomenon which provides several clusters of appliances. All the appliances of a particular cluster show similarity in their energy consumption pattern. Further, the output from k-shape algorithm is utilized to forecast the consumption of a particular appliance in a cluster by utilizing the time-series energy consumption of other appliances of that cluster. We performed the clustering and forecasting experiments on a real dataset of 43 appliances in a household. We observed improved day-ahead forecasting performance utilizing the proposed scheme which motivates us to implement it in practical scenarios to enhance flexibility in smart homes.