INTELLIGENT APPROACHES FOR INTRUSIVE MONITORING OF APPLIANCE LOADS

Power utilities must understand how customers use their products and monitoring/survey analysis of customer usage patterns is the best way to understand how customers use electricity. This is essential to formulate any long term planning, forecasting or marketing strategy. However, only customer's service level loads (total loads) are available. As the collection and analysis of customer usage data is very costly and time consuming, most power utilities do not allocate the necessary resources required to properly address this issue. That's why such an important necessity has been left unattended. If we could decompose the service level loads into end-use load profiles, then we are able to analyze them. In this study, a multi-agent system is applied to analyze the residential customer usage patterns in a cost-effective way. For this study, a power utility has recorded the service level load of residential customers in 15-minute intervals over 3 months. A residential survey collected by the same utility is also used. The seven-channel end-use meters are installed as a part of a pilot program to study the validity of end-use load research. In this research, every agent represents an appliance. Also, an artificial neural network (ANN) is assigned for each agent. Back-propagation (BP) learning algorithm is used to learn about the environment and communication among agents. We analyzed the manner in which the BP learning algorithm can be used in such a system. We also discovered several problems discussed in the paper in applying BP learning to multi-agent systems. Here for the sake of simplicity, only one household is taken into account for simulation. The simulation shows that the present study can provide detail information that currently does not exist. This project is currently being implemented.