A Blind Modeling Tool for Standardized Evaluation of Battery State of Charge Estimation Algorithms

Abstract

There are hundreds of approaches to estimating battery state of charge (SOC). It is difficult to compare results reported in different papers because each typically uses a different dataset. While some papers compare multiple SOC estimation algorithms, the author's bias, skill, or effort towards each algorithm may unintentionally skew the results. A standardized way to test and compare methodologies between authors is necessary to allow the best algorithms to stand out. An example in another application area is the National Institute of Standards (NIST) Face Recognition Vendor Test, which compares facial recognition software using a standardized dataset. A similar approach is proposed here for batteries, where data is provided for users to parameterize and train their algorithms. An online tool is provided to subject the algorithms to a wide range of blinded test cases. A high-quality dataset is prepared using battery cells from a prevalent electric vehicle. A total of sixty-four drive cycles are performed at each of six temperatures ranging from -20 °C to 40 °C. The blind modelling tool is demonstrated for one SOC estimation algorithm. It will be made available for researchers to benchmark and compare their algorithms.