As a soft sensor, the state-of-power (SoP) estimator reveals critical information on battery-based energy storage systems. A set of reliable ‘referenced values’ is the key to evaluate the precision of such soft sensors at their designing stage and could influence the overall reliability of the battery systems. However, experimentally obtaining the ‘referenced SoP’ is non-trivial since high-current pulse tests (>10C) are required to charge/discharge the batteries to their cut-off conditions. The associated high-power experimental platforms could be expensive, while frequently applying large current at boundary conditions may leave potential safety issues. Aiming at these problems, this paper focuses on obtaining referenced SoP, rather than onboard SoP estimations. A novel equivalent discharging test is designed to accurately recover the voltage response of high-current pulses from a set of low-current tests, resulting in a 33% reduction of the peak discharging current. In addition, a flexible softmax neural network is further proposed to generate SoP values for the intervals between pulse tests. With these tools, reliable SoP values with errors lower than 0.5% can be readily obtained. The SoP obtained from our approach can be further utilised as a highly accurate benchmark to evaluate the accuracy of other onboard battery SoP estimators.
Bibliographical noteFunding Information:
The first author would like to thank the Guangzhou HKUST Fok Ying Tung Research Institute for the continuing support during the Hong Kong's unrest and 2019-nCoV's outbreak.
This work was supported partly by the Guangdong scientific and technological project (Grant No. 2017B010120002), Guangzhou Scientific and Technological Project (Grant No. 202002030323), Hong Kong Research Grants Council CERG project (Grant No. 16207717, 16208520), High Value Manufacturing Catapult project (Grant No. 8248CORE), and Shenzhen Science and Technology Innovation Commission under the grant Shenzhen-Hong Kong Innovation Circle Category D Project: SGDX 2019081623240948.
© 2021 Elsevier B.V.
- Battery management system
- Discharging test
- Electric vehicles
- Lithium-ion battery
- Machine learning