Hybrid genetic algorithm method for efficient and robust evaluation of remaining useful life of supercapacitors. (15th February 2020)

• Record Type: Journal Article
• Authors: Zhou, Yanting; Wang, Yanan; Wang, Kai; Kang, Le; Peng, Fei; Wang, Licheng; Pang, Jinbo
• Is Part Of: Applied energy. Volume 260(2020)
• Publication Date: 2020-02-15
• Subjects: Supercapacitor -- Remaining useful life -- Device degradation -- Electric vehicle -- Hybrid genetic algorithm -- Long short-term memory; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2019.114169
• Languages: English
• ISSNs: 0306-2619

Abstract

Graphical abstract: Highlights: Efficient HGA-LSTM method proposed for predicting life of energy storage devices. Precise and robust lifetime estimation for Supercapacitors with error low to 1.61% Low time cost of 59 min for one remaining life predication, a 60% reduction. Adaptability for predicting life of supercapacitor at real-time dynamic cycling. High versatility of such method to deal with both online and offline untrained data. Abstract: Supercapacitor as a clean energy storage device has been widely adopted in powering electric motors of vehicles. Precise evaluation of aging state of supercapacitors, i.e., the remaining useful life provides a feedback to replace damaged cells to sustain the comfort and safety of electric vehicle. Currently reported evaluation methods for such aim are data or model-based predications, which are either time consuming or of low precision. To achieve efficient and robust evaluation of the remaining lifetime, this work proposes a general strategy based on the combination between a recurrent neutral network method, i.e., long short-term memory, and hybrid genetic algorithm. The sequential quadratic programming as a local search operator of the genetic algorithm, enhances its global search ability, which allows quickly search for the local optimal solution in the means of the dropout probability and the number of hidden layer units. Eventually we apply this predication method to supercapacitors charging at steady state mode and succeed in estimating their remaining useful life. Such life prediction approach also gains validity in supercapacitors with dynamic operative cycle. Indeed, high accuracy has been achieved at both the online trained supercapacitors with root mean square errors ranging from 0.0161 and 0.0214, and offline historical data with 0.0264 error. Moreover, the estimation time is shortened to 3550 s, which is shortened by 60%. This evaluation model may pave the way in predicting the remaining useful lifetime of supercapacitors as well as secondary ion batteries in a precise and robust fashion.