This paper proposes a novel Non-Intrusive Load Monitoring (NILM) method which incorporates appliance usage patterns (AUPs) to improve performance of active load identification and forecasting. In the first stage, the AUPs of a given residence were learnt using a spectral decomposition based standard NILM algorithm. Then, learnt AUPs were utilized to bias the priori probabilities of the appliances through a specifically constructed fuzzy system. The AUPs contain likelihood measures for each appliance to be active at the present instant based on the recent activity/inactivity of appliances and the time of day. Hence, the priori probabilities determined through the AUPs increase the active load identification accuracy of the NILM algorithm. The proposed method was successfully tested for two standard databases containing real household measurements in USA and Germany. The proposed method demonstrates an improvement in active load estimation when applied to the aforementioned databases as the proposed method augments the smart meter readings with the behavioral trends obtained from AUPs. Furthermore, a residential power consumption forecasting mechanism, which can predict the total active power demand of an aggregated set of houses, five minutes ahead of real time, was successfully formulated and implemented utilizing the proposed AUP based technique.

No abstract

This paper proposes a novel Non-Intrusive Load Monitoring (NILM) method for a consumer premises with a residentially installed solar plant. This method simultaneously identifies the amount of solar power influx as well as the turned ON appliances, their operating modes, and power consumption levels. Further, it works effectively with a single active power measurement taken at the total power entry point with a sampling rate of 1 Hz. First, a unique set of appliance and solar signatures were constructed using a high-resolution implementation of Karhunen Loéve expansion (KLE). Then, different operating modes of multi-state appliances were automatically classified utilizing a spectral clustering based method. Finally, using the total power demand profile, through a subspace component power level matching algorithm, the turned ON appliances along with their operating modes and power levels as well as the solar influx amount were found at each time point. The proposed NILM method was first successfully validated on six synthetically generated houses (with solar units) using real household data taken from the Reference Energy Disaggregation Dataset (REDD) -USA. Then, in order to demonstrate the scalability of the proposed NILM method, it was employed on a set of 400 individual households. From that, reliable estimations were obtained for the total residential solar generation and for the total load that can be shed to provide reserve services. Finally, through a developed prediction technique, NILM results observed from 400 households during four days in the recent past were utilized to predict the next day's total load that can be shed. * Corresponding AuthorEmail addresses: hchintha@sfu.ca; dineshhgcp@gmail.com (Chinthaka Dinesh), shirantha.welikala.2015@IEEE.org; shirantha27@gmail.com (Shirantha Welikala), yasithashehanliyanage@gmail.com; wlysliyanage@IEEE.org (Yasitha Liyanage), mpb.ekanayake@ee.pdn.ac.lk (Mervyn Parakrama B. Ekanayake), roshangodd@ee.pdn.ac.lk (Roshan Indika Godaliyadda), jbe@ee.pdn.ac.lk; EkanayakeJ@cardiff.ac.uk (Janaka Ekanayake) Proposed

No abstract

No abstract