基于正则化随机扰动优化LSTM的动力电池SOC估计研究  

Research on Power Battery SOC Estimation Based on Regularized Stochastic Perturbation Method Optimized LSTM

作　　者：邢云祥 王玥琦 易吉良 刘路 Xing Yunxiang;Wang Yueqi;Yi Jiliang;Liu Lu(School of Vehicle Engineering,Guilin University of Aerospace Technology,Guilin,Guangxi 541004,China)

机构地区：[1]桂林航天工业学院汽车工程学院,广西桂林541004

出　　处：《机电工程技术》2025年第5期148-151,共4页Mechanical & Electrical Engineering Technology

基　　金：广西自然科学基金项目(2024JJA160324);桂林航天工业学院2023年度本科教改项目(2023JA02)。

摘　　要：全球汽车市场的变化对新能源汽车产业升级提出新的要求和挑战,精确监测SOC状态对于优化BMS性能具有重要意义。针对现有SOC估计方法无法同时满足实时监测和精度需求的问题,提出一种利用正则化随机扰动法优化LSTM模型的方法,通过随机扰动对模型注入高斯噪声作为动态扰动,以正则化为模型参数的设置构建空间界限,将多维参数空间构建与自适应噪声衰减机制结合实现门控单元协同优化,使改良后的LSTM模型能够在正则化约束的安全边界内开展参数空间定向搜索。实验结果表明,该模型在复杂的动态工况下MAE指标稳定维持在1.1%附近,相比传统SOC估计方法误差显著降低,对电流突变等干扰因素展现出较强的鲁棒性和监测精度,为优化BMS性能提供新路径。The evolving global automotive market imposes new requirements and challenges for the industrial upgrading of new energy vehicles.Precise SOC monitoring holds significant importance for optimizing BMS performance.Targeting at the limitations of existing SOC estimation methods in simultaneously meeting real-time monitoring and precision demands,a regularized stochastic perturbation method is proposed to optimize LSTM models.The methodology introduces Gaussian noise through stochastic perturbations as dynamic disturbances,establishes spatial boundaries via regularization-based parameter configuration,and integrates multidimensional parameter space construction with an adaptive noise attenuation mechanism.This synergistic approach enables collaborative optimization of gated units,allowing the enhanced LSTM model to conduct targeted parameter space exploration within regularization-constrained safety boundaries.Experimental results demonstrate that the proposed model maintains a stable MAE metric of approximately 1.1%under complex dynamic operating conditions.Compared with conventional SOC estimation methods,the model exhibits significantly reduced error margins and enhanced robustness against interference factors such as current transients,providing a novel pathway for advancing BMS performance optimization.

关 键 词：SOC LSTM网络 随机扰动 

分 类 号：U463[机械工程—车辆工程]