考虑采样失准的卡尔曼滤波改进神经网络锌溴液流电池状态估计策略

Kalman Filter Improved Neural Network State of Charge Estimation Strategy of Zinc Bromine Flow Battery Considering Sampling Inaccuracy

针对锌溴液流电池自身特性导致的电池状态估计困难问题,该文提出一种适用于锌溴液流电池的卡尔曼滤波改进神经网络荷电状态估计策略。核心工作在于通过神经网络模型解决锌溴液流电池泵转速变化带来的开路电压非线性问题,并进一步将自放电电流加入电池状态方程中,使用考虑自放电电流的卡尔曼滤波方法解决锌溴液流电池的自放电以及采样噪声问题。现场实测数据的验证表明,所提方法的状态估计平均绝对误差0.98%,最大误差3.73%,优于单独使用的安时积分、无迹卡尔曼滤波以及神经网络等方法。同时采样失准情况下的数据验证表明,该方法在单传感器采样失准的情况下也能保持较好的效果,能够满足长时储能应用场景的需求。

Aiming at the difficulty in state of charge estimation of Zinc Bromine flow battery caused by its own characteristics,a Kalman Filter improved neural network state of charge estimation strategy is proposed.The core of this work is to solve the nonlinear characteristic caused by the variable pump speed through the neural network model.Meanwhile,the self-discharge current is incorporated into the state equation of Kalman filter to solve the self-discharge and sampling noise problems of Zinc Bromine flow battery.Validated by site operation data,the mean absolute error of proposed strategy is 0.98%and the max absolute error is 3.73%,which is superior to the individual use of Ah integral,the unscented Kalman filter and the neural network methods.Furthermore,the proposed strategy can maintain good result when single sensor sampling is inaccurate,which can meet the needs of long-term energy storage applications.