纯电动汽车磷酸铁锂电池组的建模及优化

Modeling and Optimization of Pure Electric Vehicle's LiFePO4 Battery Pack

摘要:

鉴于传统神经网络和支持向量机机理复杂、计算量大的缺陷,很难实时跟踪磷酸铁锂电池组复杂快速的内部反应,影响电池荷电状态的估算精度,提出应用一种简单、有效的极限学习机对一额定容量为100 Ah、额定电压为72 V的纯电动汽车磷酸铁锂电池组建模,并分别与BP神经网络、RBF神经网络、支持向量机进行对比；随后,以学习时间和泛化性能为优化目标,应用粒子群方法寻找最佳隐层节点个数；结果表明,基于极限学习机的磷酸铁锂电池组模型的学习时间、泛化性能优于BP神经网络、RBF神经网络、支持向量机；隐层节点优化后,模型的学习时间和泛化性能达到最优。

Abstract:

The traditional neural networks and support vector machine have the weakness of complex mechanism and large amount of computation. It is difficult to track the complex and fast inner reaction of LiFePO₄ battery pack in real time, affecting the estimation accuracy of the battery state of charge. A simple and effective extreme learning machine is proposed for the modeling of pure electric vehicle’s LiFePO₄ battery pack,whose rated capacity is 100 Ah and nominal voltage is 72 V, then compared with the back-propagation neural networks-based, radical basis function neural networks-based and support vector machines-based. Subsequently, taking the learning time and generalization performance as the optimization goal and using the particle swarm to find the optimal hidden node. The results show that the model of LiFePO₄ battery pack based on extreme learning machine has shorter learning time and higher generalization performance compared with the model based on BP neural networks, RBF neural networks and support vector machines. After optimization of hidden nodes, learning time and generalization performance of the model is optimal.