Online Implementable Open Circuit Voltage Prediction for Battery-Powered Internet of Things Devices

Accurate open circuit voltage (OCV) prediction is crucial for battery modeling and state evaluation. However, existing OCV prediction methods are not online implementable for battery powered Internet-of-Things (IoT) devices, as these methods typically require disconnecting battery from load and have high computational complexity. To solve this problem, this paper proposes a lightweight OCV prediction method that can be implemented online. Firstly, by obtaining discharge data within the limited current range of IoT devices and fully utilizing the intrinsic relationship between OCV, terminal voltage and current, the dataset used for training is significantly reduced. Then, utilizing the advantages of support vector regression in handling nonlinear data with small size, a nonlinear model for OCV with terminal voltage and current is established. The inference model obtained in this way requires very few support vectors while achieving relatively accurate OCV prediction. Experimental results show that the probability that the prediction error is less than 6mV exceeds 80% for two commonly used primary batteries under both constant and variable current conditions. Meanwhile, the inference time overhead is less than 20ms. Compared with existing fast OCV prediction methods, the proposed method can achieve prediction errors of the same order of magnitude while reducing the time overhead by at least two orders of magnitude. More importantly, the proposed method does not require disconnecting battery from load, which makes it online implementable for IoT devices.