首页|基于注意力机制和Mogrifier LSTNet的道路交通占有率预测

基于注意力机制和Mogrifier LSTNet的道路交通占有率预测

扫码查看
提出一种改进的LSTNet深度学习框架用于交通占有率数据预测.采用不同大小的卷积核来捕捉时间序列数据中不同时间范围内的模式和趋势,并融合CBAM注意力机制可以在通道维度和空间维度上自适应地调整特征的权重.通过引入 Mogrifier 机制多次迭代交替更新LSTM的输入门和遗忘门的权重,以更好地捕捉序列数据中的长期依赖关系.而 AR 模型充分考虑了数据集的自相关性帮助模型更好地理解历史信息.实验结果表明,提出的模型相对绝对值误差为 0.349 7,明显优于其他模型,能够有效提高交通占有率的准确预测.
Prediction of road traffic share based on attention mechanism and Mogrifier-LSTNet
This paper proposes an improved LSTNet deep learning framework for traffic occupancy data prediction.Convolution kernels of different sizes are used to capture patterns and trends in different time ranges in time series data,and CBAM attention mechanism can be integrated to adaptively adjust the weight of features in channel dimension and spatial dimension.By introducing the Mogrifier mechanism,the weights of the LSTM's input and forget gates are updated alternately over several iterations to better capture long-term dependencies in the sequence data.The AR model fully considers the autocorrelation of the data set to help the model better understand the historical information.The experimental results show that the relative absolute error of our proposed model is 0.349 7,which is obviously better than other models and can effectively improve the accurate prediction of traffic occupancy.

LSTNet modelCNN(Convolutional Neural Network)CBAM attention mechanismMogrifier LSTMtraffic occupancy forecast

秦喜文、潘星宇、张斯琪、石红玉、董小刚

展开 >

长春工业大学 大数据科学研究院,吉林 长春 130012

长春工业大学 数学与统计学院,吉林 长春 130012

LSTNet模型 卷积神经网络 CBAM注意力机制 Mogrifier LSTM 交通占有率预测

吉林省科技厅项目吉林省科技厅项目吉林省教育厅项目

20200403182SF20210101149JCJJKH20210716KJ

2024

长春工业大学学报
长春工业大学

长春工业大学学报

影响因子:0.282
ISSN:1674-1374
年,卷(期):2024.45(3)