A backpropagation neural network-based hybrid energy recognition and management system

For several decades, small electronic devices like wireless sensor network nodes (WSNs) tend to be powered by ambient energy, and the multi-input energy platform attracts much attention because sensors are usually used in complicated surroundings. However, for multi-input energy platform energy management is complex and the demand of the consumers is stochastic. To solve the problems, this paper presents a backpropagation neural network (BPNN) based hybrid energy recognition and management System (ERMS). The design applies artificial intelligence algorithms to energy forecasting recognition. And it achieves energy-matching management according to recognition results. Besides, we implemented the energy recognition algorithm on an application specific integrated circuit (ASIC) innovatively, which is manufactured in a standard 180 nm CMOS technology. The energy recognition chip area is 1.45mm × 1.45 mm. The experimental data present that the system can identify different types of input energy and control the energy flows automatically. The current consumption of the ASIC is 65μA at 1 MHz and the recognition accuracy can reach 98 %. Moreover, the hybrid energy recognition and management system platform worked effectively. The measurement results show that the power conversion efficiency of the system to photovoltaic energy input is 85 %. Furthermore, when the input is piezoelectric energy, the power management system output power can achieve 7.4 mW.