基于遗传神经网络的电动汽车锂电池SOC预测

针对传统方法在电动汽车锂电池荷电状态（State of Charge,SOC）预测中的局限和不足,提出了一种基于遗传神经网络的电池SOC预测算法.该算法的整体方案首先给出了车载锂电池状态监测系统的软硬件实现,在该系统上以不同的放电倍率对磷酸铁锂电池进行了放电实验,获取了其放电过程中电压、电流和SOC的样本数据,然后利用遗传算法全局寻优能力对神经网络中的连接权值和阈值进行了优化,用实验所得的样本数据训练BP神经网络,根据训练好的神经网络对锂电池SOC进行了预测并将其与真实SOC进行对比,以验证算法的可行性.研究结果表明,该方案可通过电压、电流的实时测量值获知锂电池的剩余电量,具有收敛速度快、预测误差小、适应范围广的特点,有效解决了电动汽车锂电池的SOC预测问题.

Prediction of lithium-ion battery SOC in EV based on genetic neural network

Aiming at limits and insufficiency of orthodox methods in state of charge（SOC） estimation for lithium-ion battery of electric vehicle （EV）,an algorithm based on genetic neural network was presented.Firstly,the realization of a lithium-ion battery condition monitoring system was introduced; the sample data of voltage,current and SOC of a lithium-ion battery were obtained by different rate discharge.Secondly,the genetic algorithm was used to train the weight values and threshold values of BP network considering its ability of global optimization;the BP network was trained with the data collected from experiments,and then the SOC was distinguished from the well-trained neural network; the feasibility of the algorithm was demonstrated by comparing real SOC with predicted SOC.The results indicate that this scheme not only obtains the residual capacity through the voltage and current but also has a quick convergent velocity,less error,wide adaptation range,and can estimate the SOC of a lithium-ion battery in EV effectively.