Data-driven Lithium-ion Battery Health Management Research

Lithium-ion batteries are widely used in electric vehicles and energy storage systems,and battery health management is essential for operation and maintenance.Although datadriven methods are more suitable for large-scale engineering applications than modelbased methods,there are still difficulties in data pre-processing,applicability to actual engineering data,and employment of data analysis tools.Based on the existing research and literature,four main challenges in applying datadriven methods are summarized.First,the direct application of multi-source heterogeneous data is quite difficult.Second,it is hard to screen abnormal batteries quickly in large-scale application scenarios.Third,achieving accurate health status estimation based on engineering data can be challenging.Fourth,there is a lack of convenient and general data analysis software.Research work has been carried out with respect to the aforementioned four aspects.First,for the massive data generated during the full life cycle of lithium-ion batteries,a multi-source heterogeneous data fusion method is proposed,and an extensible database is designed to import and merge multi-source heterogeneous data to resolve field inconsistencies,data conflicts,and different degrees of particles.At the same time,the pre-processing flow including deduplication,filling,interpolation smoothing,outlier removal,and data dimensionality reduction is proposed to provide high-quality data support for the implementation of data-driven algorithms.Second,for large-scale lithium-ion battery operation scenarios,most fault detection algorithms are difficult to implement due to the limitation of hardware acquisition and software calculation capabilities at present.Therefore,feature extraction schemes including variance,extreme range,and inconsistency parameters of power variation are proposed,which have good adaptability to poor quality engineering data.Further,a comprehensive health scoring scheme based on a hybrid clustering algorithm was developed,which can predict the health level of each battery in the system to achieve rapid screening of abnormal batteries.The experimental data set is used for verification,and the accuracy of sorting is higher than 92%.This health scoring scheme has been implemented in more than one year in many large-scale energy storage power plants.Then,for scenarios with high user demand,in addition to rapid faulty battery detection,the battery health status needs to be accurately calculated.For scenarios with high data quality,such as in-vehicle applications,a feature extraction and combination scheme based on incremental capacity analysis is proposed,which is not limited by the depth of charge and discharge.Based on the Gaussian process regression training,the health state estimation model is established,which is applied to the online vehicle estimation scenario.The estimation error is less than 1%,which indicates good accuracy and robustness.Finally,the development process of lithium-ion battery data analysis software is proposed,and the application requirements of university users and enterprise users are analyzed.Based on APP Designer,lithium-ion battery data analysis software-AILi On is developed,which provides multi-source heterogeneous data import and fusion,data display,operation statistics,safety and health analysis,and integrates the proposed algorithm.After application testing,the developed software can save at least half of the time and effort for lithium-ion battery data analysts.