A novel automatic underwater vehicle SOC estimator based on BPNN-AUKF at different temperatures
李青 1张劭玮 2罗斯伦 2李炬晨 2成海超 1卢丞一2
扫码查看
点击上方二维码区域，可以放大扫码查看
作者信息
1. 中国电子科技集团公司第十八研究所,天津 300384
2. 西北工业大学,陕西西安 710072
折叠
摘要
本研究提出了一种基于反向传播神经网络(BPNN)和自适应无迹卡尔曼滤波器(AUKF)的SOC(state of charge)估计方法.首先针对电池SOC与端电压之间在不同温度下的关系,研究设计了一系列温度补偿策略,以提高在低温、低SOC条件下的估计精度.其次,利用反向传播神经网络(BPNN)建立了一个耦合了温度补偿策略的电池模型.这个模型能更好地适应低温和低SOC条件下的电池状态变化,提高了SOC估计的准确性.最后,基于BPNN电池模型建立了BPNN-AUKF的SOC估计框架,通过利用测量值与测量预测值之间的信息和残差序列,对系统过程和测量噪声协方差进行估计修正.通过实验验证,发现该方法在低温环境下具有明显优势,相比传统方法能够更准确地估计电池的SOC,且具备较好的泛化能力.这种基于BPNN-AUKF方法的SOC估计器不仅适用于自主无人潜水器(AUV),而且对于其他在复杂环境中工作的车辆也具有广泛的应用价值.
Abstract
This study proposes a state of charge (SOC) estimation method based on backpropagation neural network (BPNN) and adaptive unscented Kalman filter (AUKF). Firstly, a series of temperature compensation strategies were studied and designed to improve the estimation accuracy under low temperature and low SOC conditions, focusing on the relationship between battery SOC and terminal voltage at different temperatures. Secondly, a battery model coupled with temperature compensation strategy was established using backpropagation neural network (BPNN). This model can better adapt to battery state changes under low temperature and low SOC conditions, improving the accuracy of SOC estimation. Finally, a SOC estimation framework for BPNN-AUKF was established based on the BPNN battery model. By utilizing the information and residual sequences between measured and predicted values, the system process and measurement noise covariance were estimated and corrected. Through experimental verification, it was found that this method has significant advantages in low-temperature environments. Compared with traditional methods, it can more accurately estimate the SOC of batteries and has good generalization ability. This SOC estimator based on BPNN-AUKF method is not only suitable for autonomous unmanned underwater vehicles (AUV), but also has broad application value for other vehicles working in complex environments.
维普
万方数据