An improved feedforward-long short-term memory modeling method for the whole-life-cycle state of charge prediction of lithium-ion batteries considering current-voltage-temperature variation

• Published in: Energy (Oxford) Vol. 254; p. 124224
• Main Authors: Wang, Shunli, Takyi-Aninakwa, Paul, Jin, Siyu, Yu, Chunmei, Fernandez, Carlos, Stroe, Daniel-Ioan
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01-09-2022
• Subjects: Capacity fading; Feedforward-long short-term memory; Lithium-ion battery; Sliding balance window; Steady-state screening model; Whole-life-cycle state of charge; Lithium-ion battery; Feedforward-long short-term memory; Capacity fading; Steady-state screening model; Whole-life-cycle state of charge; Sliding balance window
• ISSN: 0360-5442
• DOI: 10.1016/j.energy.2022.124224

The whole-life-cycle state of charge (SOC) prediction plays a significant role in various applications of lithium-ion batteries, but with great difficulties due to their internal capacity, working temperature, and current-rate variations. In this paper, an improved feedforward-long short-term memory (FF-LSTM) modeling method is proposed to realize an accurate whole-life-cycle SOC prediction by effectively considering the current, voltage, and temperature variations. An optimized sliding balance window is constructed for the measured current filtering to establish a new three-dimensional vector as the input matrix for the filtered current and voltage. Then, an improved steady-state screening model is constructed for the predicted SOC redundancy reduction that is obtained by the Ampere-hour integral method and taken as a one-dimensional output vector. The long-term charging capacity decay tests are conducted on two batteries, C7 and C8. The results show that the battery charging capacity reduces significantly with increasing time, and the capacity decreases by 21.30% and 22.61%, respectively, after 200 cycles. The maximum whole-life-cycle SOC prediction error is 3.53% with RMSE, MAE, and MAPE values of 3.451%, 2.541%, and 0.074%, respectively, under the complex DST working condition. The improved FF-LSTM modeling method provides an effective reference for the whole-life-cycle SOC prediction in battery system applications.
•Improved feedforward-long short-term memory (FF-LSTM) modeling for SOC prediction.•Sliding balance window of dimensional current-voltage-temperature variation vectors.•Optimized steady-state screening model built with Ah integration and output vector.•Whole-life-cycle feature analysis for current, voltage, temperature, and capacity.