On load disaggregation using discrete events

Abstract

Load disaggregation involves identification of the operating appliances and their respective power consumption using only total building power consumption data. Load disaggregation also called as non-intrusive load monitoring can offer consumers greater visibility about their individual appliance power consumption. This information will help customers (i) to optimize the use of appliances, resulting in energy savings and (ii) in knowing the health condition of the various appliances based on the power consumption patterns. Load disaggregation also provides utility companies the opportunity to gain new insights about their customers energy profile which helps them to come up with efficient demand response plans for peak load shaving. In this paper, discrete events based load disaggregation is discussed which is developed and implemented using apparent power data collected from actual residential building. Discrete events like switching ON/OFF of appliances, appliance operational time and some portion of the raw signal are considered for achieving load disaggregation. The proposed load disaggregation approach consists of event detection, pairing of the events, extraction of feature vector and learning & appliance identification using artificial neural network. The load disaggregation solution is developed and tested using the apparent power data sampled at 0.1 Hz, collected from a actual residential building in Bangalore, India using “Eyedro” power sensor and the results are very promising.