基于载荷谱分析和混合深度学习车载储氢气瓶路况载荷模式识别

Road Load Pattern Recognition for Vehicle-loaded Hydrogen Storage Cylinders Based on Load Spectrum Analysis and Hybrid Deep Learning

氢燃料电池汽车在行驶过程中受路况影响产生振动,引起的振动载荷可能导致车载气瓶产生表面损伤,直接影响气瓶的使用安全和效率。针对储氢气瓶路况载荷数据分布不平衡导致载荷识别效果不佳的问题,提出一种改进深度卷积生成模型(Deep Convolutional Generative Adversarial Networks,DCGAN)结合卷积神经网络(Convolutional Neural Network,CNN)的路况振动载荷识别方法。DCGAN可以实现样本扩充,提高模型的识别性能。同时,针对DCGAN的卷积计算只能处理相邻数据特征的问题,将自注意力机制(Self Attention,SA)引入DCGAN中,自注意力机制可以计算样本的特征点之间的关系,帮助DGCAN的生成器充分学习样本的全局特征,增强模型泛化性。最后通过CNN实现载荷识别。通过实验对提出模型进行测试,并与多种模型比较;实验结果表明,提出的模型对路况振动载荷识别准确率达到96.3%,与其他模型相比,该模型表现出更好的性能。

Vibration of hydrogen fuel cell vehicles is greatly influenced by road conditions during driving,and the resulting vibration load may cause surface damage of the on-board gas cylinders,which directly affects their safety and efficiency.To solve the issue of poor load identification due to the unbalanced distribution of road load data for hydrogen storage gas cylinders,an improved deep convolutional generative adversarial networks(DCGAN)model combined with convolutional neural networks(CNN)is proposed for road condition vibration load identification.The DCGAN can realize sample expansion and improve the recognition performance of the model.Meanwhile,aiming at the problem that the convolutional calculation of DCGAN can only deal with the characteristics of adjacent data,the self-attention mechanism(SA)is introduced into the DCGAN to calculate the relationship between feature points of the sample and help the generator of DGCAN to fully learn the global features of the sample and enhance the generalization of the model.Finally,load recognition is realized through CNN.The proposed model is experimentally tested and the results are compared with those of various models.It is found that the accuracy rate of the proposed model for road condition vibration load recognition can reach 96.3%,and the proposed model exhibits better performance than the other models.