Wireless sensor networks (WSN) are battery-powered, low-resource systems. Node lifetime is influenced by the availability of battery power, device drive cycles, and environmental factors. The amount of energy in the batteries is largely dependent on their state of charge (SoC). For the prediction of device lifetime and safe device operation, accurate SoC estimation is essential. In this paper, we present a novel Gaussian Process Regression-based method for adaptive SoC estimation (GPR). The training data was gathered in a climate-controlled lab environment using IEEE 802.15.4-based drive loads for three different battery types, including lithium-ion, nickel-metal hydride, and lithium-polymer, at different temperatures.
The GPR model with hyper-tuned Radial Bias Filter (RBF) was trained at temperatures ranging from 5 °C to 45 °C for each battery parameter. The proposed method was compared to support vector machines and polynomial regression for model accuracy (SVM). Regarding this, the suggested model offered Mean Absolute Error (MAE) values of 2.53, 2.54, and 2%, respectively, and Root Mean Square Error (RMSE) values of 0.295, 0.292, and 0.35 for Nickel-Metal Hydride, Lithium-Polymer, and Lithium-Ion batteries at 25 °C. To the best of our knowledge, our suggested lightweight GPR scheme is the only one in use on embedded platforms for SoC estimation of WSN.