摘要
电池分选方法能够有效降低电池间的不一致性,显著提升储能系统的性能、降低安全隐患。由于液态金属电池具有容量大、内阻小等特点,现有分选方法难以满足其分选的要求。针对现有分选方法耗时长、精度较低的问题,该文提出了一种基于集成机器学习的液态金属电池快速分选方法。该方法利用活化期的电池数据作为输入,通过特征选择和集成学习方法结合不同模型优势,并根据集成模型对预测得到的电池容量进行分选。与现有分选方法相比,该方法避免了额外的电池测试从而实现了快速分选,同时具有更低的预测误差与更好的可靠性。研究结果表明:相较于主流的分选方法,该分选方法的预测误差降低了52.16%,可靠性提升了9.10%,在电池分选上实现了高准确率和召回率,分别为96.62%和93.18%。在规模储能的电池分选中,该方法具有显著的潜在应用价值。
Liquid metal battery(LMB)is a newly emerged battery technology for large scale energy storage,which requires massive batteries in group operation.Battery sorting,which could effectively reduce the inconsistency among batteries,is a useful method to significantly improve the performance of energy storage and reduce potential safety risks.However,because of the characteristics of LMB,including large capacity and small internal resistance,existing sorting methods,which mainly focus on lithium ion batteries,cannot meet the precision requirements of liquid metal battery sorting.In addition,these methods also require additional battery testing to obtain input features,thus there is room for further optimization of test time and cost.In order to address these issues,a rapid sorting method for liquid metal batteries based on ensemble learning is proposed.Combining the advantages of different models by adopting feature selection and ensemble learning methods,it precisely predicts the sorting index and conducts battery sorting with satisfactory accuracy.Firstly,a dataset is constructed based on LMB’s cycling data during activation,which comprises samples in 1D vector form.Secondly,features of the dataset are selected on the basis of a comprehensive method,which takes various aspects into consideration.Thirdly,three different base models are trained and optimized with the help of three-fold validation and grid search optimization,and then two ensemble model,which adopts voting and stacked ensemble method respectively,are trained based on these base models.Fourthly,all the models,including base models and ensemble ones,and a contrast model(neural network)are evaluated on the test set,which demonstrates the superiority of the stacked ensemble model.Finally,battery sorting is conducted on the basis of capacity predictions made by the stacked ensemble model.In this sorting method,the feature selection method reduces the dimension of data and selects the most effective features and ensemble methods integrate advantages of diff
作者
夏珺羿
石琼林
蒋凯
何亚玲
王康丽
Xia Junyi;Shi Qionglin;Jiang Kai;He Yaling;Wang Kangli(State Key Laboratory of Advanced Electromagnetic Engineering and Technology School of Electrical and Electronic Engineering Huazhong University of Science and Technology,Wuhan 430074 China;Engineering Research Center of Power Safety and Efficiency Ministry of Education,Wuhan 430074 China)
出处
《电工技术学报》
EI
CSCD
北大核心
2023年第21期5900-5912,共13页
Transactions of China Electrotechnical Society
基金
国家自然科学基金项目(52277217)
国家重点研发计划项目(2018YFB0905600)资助。
关键词
电池分选
液态金属电池
机器学习
集成机器学习
特征选择
Battery sorting
liquid metal battery
machine learning
ensemble learning
feature selection
