交通流动态预测LSTM模型设计

Design of LSTM Model for Traffic Flow Dynamic Prediction

摘要: 针对交通流具有周期性动态性的特点，为提高交通流预测精度，在充分分析交通流日波动和周波动特征规律的基础上，构建了交通流预测LSTM模型；以均方根误差和平均绝对误差为评价指标，构建了LSTM模型参数寻优设计逻辑，提出了适用于交通流动态变化的层数、批处理大小、隐藏层节点数、序列长度、学习率参数的选定准则，并以PyTorch智能框架搭建了测试环境。研究结果表明，优化后的LSTM模型在交通流量预测任务上表现出显著的优越性，预测结果与实际值之间的差距明显缩小，同时也为LSTM模型超参数优化提供了一种可行有效的方法。

Abstract: To address the periodic and dynamic characteristics of traffic flow and improve the accuracy of traf-fic flow prediction, a traffic flow prediction LSTM model was constructed on the basis of fully analyz-ing the daily fluctuation and weekly fluctuation characteristics of traffic flow. Using the root mean square error and the mean absolute error as evaluation indicators, an LSTM model parameter search was constructed. Optimize the design logic, propose selection criteria for the number of lay-ers, batch size, number of hidden layer nodes, sequence length, and learning rate parameters that are suitable for dynamic changes in traffic flow, and build a test environment with the PyTorch in-telligent framework. The research results show that the optimized LSTM model shows significant superiority in traffic flow prediction tasks, and the gap between the prediction results and the actu-al values is significantly narrowed. It also provides a feasible and effective method for hyperparam-eter optimization of the LSTM model.