| 引用本文: | 岳家辉,夏向阳,吕崇耿,等.计及健康特征信息量的锂离子电池健康状态与剩余寿命预测研究[J].电力系统保护与控制,2023,51(22):74-87.[点击复制] |
| YUE Jiahui,XIA Xiangyang,LÜ Chonggeng,et al.Research on the prediction of state of health and remaining useful life of lithium-ion batteriesconsidering the amount of health factors information[J].Power System Protection and Control,2023,51(22):74-87[点击复制] |
|
| 摘要: |
| 电池状态有效评估过程中数据驱动法的模型输入虽与容量呈现相关性，但并没有考虑其信息量及信息质量，低质量的数据输入会造成一定程度的预测偏差。针对上述问题，提出一种计及健康特征信息量的加权神经网络电池健康状态(state of health, SOH)预测与剩余寿命(remaining useful life, RUL)估计模型。该模型在GA-BP神经网络的基础上，通过确定有效健康特征数据集，利用数据信息度构建动量因子来保证神经网络迭代收敛速度。并基于熵权思想过滤出低信息量健康特征的预测结果，将过滤后的预测结果作为电池老化模型的输入，进一步实现剩余寿命的估计。通过公开电池老化数据集与实验平台进行验证，得到该模型健康状态预测结果MAE、RMSE分别控制在0.63%、0.81%之下，剩余寿命估计结果MAE、RMSE分别控制在0.0031 mA·h、0.0042 mA·h之下，具有良好的可行性与有效性。 |
| 关键词: 锂离子电池 数据驱动技术 健康状态 剩余使用寿命 神经网络 熵权法 |
| DOI：10.19783/j.cnki.pspc.230606 |
| 投稿时间：2023-05-23修订日期：2023-06-30 |
| 基金项目:国家自然科学基金项目资助(51977014)；湖南省研究生科研创新项目资助(CX20220917) |
|
| Research on the prediction of state of health and remaining useful life of lithium-ion batteriesconsidering the amount of health factors information |
| YUE Jiahui1,XIA Xiangyang1,LÜ Chonggeng1,WU Xiaozhong2,KONG Lin3,KONG Lin1,CHEN Laien1 |
| (1.School of Electrical and Information Engineering, Changsha University of Science and Technology, Changsha 410114,
China; 2. State Grid Hunan Electric Power Co., Ltd., Changsha 410004, China; 3. China Energy Construction
Group Hunan Electric Power Design Institute Co., Ltd., Changsha 410004, China) |
| Abstract: |
| The model input of the data-driven method in the effective evaluation process of battery state, although related to capacity, does not consider its information content and quality. Low-quality data input can cause a certain degree of prediction bias. To address this issue, this paper proposes a weighted neural network battery SOH prediction and RUL estimation model that takes into account the degree of health factor information. Based on the GA-BP neural network, this model identifies effective health feature data sets and uses data information to generate momentum factors to ensure neural network iteration convergence speed. And this paper filters out low information health feature prediction findings using the entropy weight concept and then uses the filtered prediction results as the input to the battery aging model to further achieve the RUL estimation. It is discovered through the publicly available battery aging datasets and experimental platforms that the model's SOH prediction results have a MAE and RMSE range controlled within 0.63% and 0.81%, and the remaining useful life estimation results have a MAE and RMSE range controlled within 0.0031 mA·h and 0.0042 mA·h, indicating good feasibility and effectiveness. |
| Key words: lithium-ion battery data-driven technology state of health remaining useful life neural network entropy weight method |
