پیوندهای مفید
آمار
| تعداد نشریات | 8 |
| تعداد شمارهها | 435 |
| تعداد مقالات | 5,636 |
| تعداد مشاهده مقاله | 7,444,364 |
| تعداد دریافت فایل اصل مقاله | 6,220,589 |
Advanced State of Health Estimation for Lithium-Ion Batteries Using Deep Learning and Feature Engineering | ||
| AUT Journal of Electrical Engineering | ||
| مقالات آماده انتشار، پذیرفته شده، انتشار آنلاین از تاریخ 15 آذر 1404 اصل مقاله (1.8 M) | ||
| نوع مقاله: Research Article | ||
| شناسه دیجیتال (DOI): 10.22060/eej.2025.24863.5779 | ||
| نویسندگان | ||
| Negar Khalili1؛ Saeed Khankalantary* 2؛ Ali Najafi Ardekany3 | ||
| 1Master degree, Electrical Engineering Department, K. N. Toosi University of Technology, Tehran, Iran | ||
| 2Assistant Professor, Electrical Engineering Department, K. N. Toosi University of Technology, Tehran, Iran | ||
| 3Assistant Professor, Mechanical Engineering Department, K. N. Toosi University of Technology, Tehran, Iran | ||
| چکیده | ||
| Obtaining an accurate estimate of the state of health of a lithium-ion battery is important for its efficiency and stability, but it's hard because the aging processes are so complicated and non-linear. Deep neural networks and long short-term memory networks are powerful tools, but their potential is often not realized if the raw operating features fail to capture synergistic aging mechanisms. This article proposes a novel two-stage hybrid feature engineering methodology to address this constraint. In the first stage, the method uses a binary particle swarm optimization algorithm to look for a small set of important predictive features. In the second stage, the parsimonious subset is enhanced with a physics-constrained Electro-Thermal Interaction Feature that incorporates terminal voltage and temperature interaction stresses. The resulting feature set was subsequently utilized for the training and evaluation of both deep neural networks and long short-term memory networks models. Adding the electro-thermal interaction feature significantly improves the predictability of both models on the primary B05 cell, raising the R² value from about 0.93 to over 0.99. To assess generalizability, the framework was rigorously validated using a cross-battery approach on two additional cells (B07 and B055), where the models maintained high performance with an average R² > 0.97. The findings indicate that domain-knowledge-intensive feature engineering significantly influences performance more than the architectural decision between deep neural networks and long short-term memory networks, facilitating highly accurate and robust state of health predictions, which are crucial in advanced battery management systems. | ||
| کلیدواژهها | ||
| Deep Learning Algorithms؛ State of Health (SOH)؛ Feature Engineering؛ Binary Particle Swarm Optimization (BPSO)؛ Optimization | ||
|
آمار تعداد مشاهده مقاله: 1 |
||