基于DKF-Bi-LSTM的阀控式铅酸电池SOC在线估计方法
The SOC online estimation of valve-regulated lead-acid battery based on DKF-Bi-LSTM
李练兵 1刘艳杰 2王海良 3李思佳 2李秉宇 4杜旭浩4
作者信息
- 1. 省部共建电工装备可靠性与智能化国家重点实验室(河北工业大学),天津 300130
- 2. 河北工业大学人工智能与数据科学学院,天津 300130
- 3. 新兴重工集团有限公司,北京 100070
- 4. 国网河北省电力有限公司电力科学研究院,河北石家庄 050021
- 折叠
摘要
精准估计阀控式铅酸蓄电池的荷电状态(SOC)对变电站直流系统的可靠性和安全性有着重要的作用,为提高SOC估算精度,提出一种基于DKF-Bi-LSTM的铅酸蓄电池SOC在线估计方法,基于二级结构的双卡尔曼滤波算法,分别进行模型估计和状态估计.通过卡尔曼滤波算法对模型参数进行动态跟踪,进而基于扩展卡尔曼滤波算法在线估算电池SOC值.将在线估算结果、电流、电压、温度值作为Bi-LSTM神经网络的输入,电池SOC预测值作为网络输出,实现对电池SOC的在线估计.经测试发现,与DKF和Bi-LSTM算法相比,DKF-Bi-LSTM算法的SOC预测均方根误差更小,其SOC在线估计方法具有更高的准确性.
Abstract
Accurate estimation of state of charge(SOC)of VRLA battery plays an important role in reliability and security of DC system in substation.In order to improve the accuracy of SOC estimation,a method for estimation of SOC of VRLA battery based on DKF-Bi-LSTM is proposed.Based on the double Kalman filter algorithm with secondary structure,the model estimation and state estimation are carried out respectively.Firstly,the model parameters are dynamically tracked by Kalman filter(KF)algorithm,and the battery SOC is estimated by extended Kalman filter(EKF)algorithm.Then,the online estimation result,current,voltage,temperature value are taken as the input of Bi-LSTM neural network,battery SOC real value as network output,and realize the estimation of battery SOC.Finally,compared with DKF and Bi-LSTM algorithm,The root mean square error of SOC prediction(RMSE)of DKF-Bi-LSTM algorithm is smaller,and its SOC online estimation method has higher accuracy.
关键词
阀控式铅酸电池/荷电状态/等效电路模型/卡尔曼滤波/扩展卡尔曼滤波/双向长短时记忆神经网络Key words
VRLA battery/SOC/equivalent circuit model/Kalman filter/extended Kalman filter/Bi-LSTM引用本文复制引用
基金项目
国家重点研发计划项目(2018YFC0810000)
出版年
2024
NETL
NSTL
万方数据