基于注意力-长短期记忆模型的偶发性交通流异常检测

Traffic Flow Occasional Anomaly Detection Based on Attention-LSTM Model

摘要:

提出了基于注意力（attention）机制的LSTM（long short-term memory）交通异常检测模型，并利用整体交通网格点数据对交通流量进行异常检测。模拟数据集的验证结果表明，Attention-LSTM预测模型具有较好的检测效果；SKAB（skoltech anomaly benchmark）公开数据集进一步检验了模型的检测能力。以上海市出租车GPS实时数据代表实时交通流状况，对交通流异常进行检测，并对检测结果进行了分析，验证了Attention-LSTM模型的有效性。

Abstract:

This paper proposes an long short-term memory （LSTM） traffic anomaly detection model based on the attention mechanism， which uses the overall traffic grid point data to predict the future traffic flow at various points in order to achieve the purpose of anomaly detection. By evaluating the simulation data， the Attention-LSTM model has a better detection effect. Furthermore， this paper uses the SKAB dataset to test the anomaly detection ability of the model， and obtains good results. Finally， this paper conducts experiments on the actual GPS signal data of Shanghai taxis， empirically analyzes some of the actual traffic anomalies detected， and proves the effectiveness of the model in detecting actual traffic anomalies.