To save the electrical energy in a household, it is essential to monitor where and how the power is consumed. To maximize the efficiency of energy conservation, it is necessary to make the running power low in the power monitor system, which the tradition systems pay less attention to. This paper presents PowerAnalyzer, an energy-aware system for monitoring running states and power of each household appliance plugged into power line from a single point detection. PowerAnalyzer takes steady-state current waveforms as the appliances signature, and uses the deep neural network (DNN) models to infer the running states and running power of household appliances. We focus on the energy consumption of PowerAnalyzer itself. The energy efficiency of PowerAnalyzer is optimized from these aspects: Using dynamic time intervals to collect electric data, replacing a cloud server with an edge node to process data, and transmitting differential data over a low power wireless protocol. The evaluation results show that PowerAnalyzer offers 3.45% average power metering error and 98.38% average accuracy of inferring running states of appliances. PowerAnalyzer draws less than 247mW static power and 304mW peak power.