Incorporating Appliance Usage Patterns for Non-Intrusive Load Monitoring and Load Forecasting

摘要

This paper proposes a novel Non-Intrusive LoadudMonitoring (NILM) method which incorporates appliance usageudpatterns (AUPs) to improve performance of active load identi-udfication and forecasting. In the first stage, the AUPs of a givenudresidence were learnt using a spectral decomposition based standardudNILM algorithm. Then, learnt AUPs were utilized to biasudthe priori probabilities of the appliances through a specificallyudconstructed fuzzy system. The AUPs contain likelihood measuresudfor each appliance to be active at the present instant based onudthe recent activity/inactivity of appliances and the time of day.udHence, the priori probabilities determined through the AUPsudincrease the active load identification accuracy of the NILMudalgorithm. The proposed method was successfully tested forudtwo standard databases containing real household measurementsudin USA and Germany. The proposed method demonstrates anudimprovement in active load estimation when applied to theudaforementioned databases as the proposed method augments theudsmart meter readings with the behavioral trends obtained fromudAUPs. Furthermore, a residential power consumption forecastingudmechanism, which can predict the total active power demand ofudan aggregated set of houses, five minutes ahead of real time, wasudsuccessfully formulated and implemented utilizing the proposedudAUP based technique.