2004 年 2 巻 2 号 p. 627-632
Usable remnant energy of a rechargeable battery is proportional to its state of charge, but the values of these two parameters are not exactly the same. A circuit based on neural network is developed for available capacity estimation of lithium-ion battery. To ensure the network consisting of optimal numbers of hidden layers and neurons, intensive experiments have been performed and various training algorithms have been evaluated. The experimental results reveal that the circuit is able to estimate the available capacity of a battery with arbitrary discharging profiles as well as continuous constant current discharging profiles. This paper presents the roadmap for the design and the practical constraints for hardware implementation of a neural network based lithium-ion battery capacity indicator.