Smart metering data analytics for non-intrusive load monitoring

Advanced metering infrastructure (AMI) exerts an enormous function on the smart grid. It has a great effect on implementing remote load control and improving the utilization of energy. Load decomposition and data analysis of energy could be realized based on the data provided by AMI. Users can manage their power consumption with the help of data analysis of energy or can control the load in real time according to the current load status in the system. Non-intrusive load monitoring (NILM) is able to obtain the electricity information of different components in the system by installing a smart meter at the home end and then using load decomposition algorithm. This method could get the real-time status of different electrical components without changing the existing circuit structure. It has the advantages of low implementation cost and small interference to users. Furthermore, NILM is easy to be popularized in various fields. In this report, a long short-term memory (LSTM) based framework is proposed and used in the load disaggregation algorithm. Two different network structures are constructed and then three types of look back time-steps as well as 2 types of input features are investigated in the two network structures respectively. A dataset from a real chiller plant system is used for case study and performances are evaluated based on two metrics. The simulation results show that the proposed LSTM method could realize a relatively high accuracy and efficiency in load disaggregation.Master of Science (Power Engineering