融合外部属性的短时交通流预测研究

针对现有交通流量预测算法大多仅考虑常态下的预测，而未考虑天气属性、周围地理属性对预测结果的影响，提出一种融合外部属性的组合预测模型（A-STIGCN）。首先，将外部属性作为路网中路段的属性，同时对路段的属性和交通特征进行建模，得到增强的特征向量。其次，采用图小波变换和自适应矩阵分别提取交通流局部和全局空间特征信息，并借助门控循环单元（GRU）对时间信息的长时记忆能力以提取其时间特性。最后，通过注意力机制来捕获时空动态变化性进行交通流预测。采用深圳出租车轨迹数据、对应天气数据以及POI数据进行预测，研究结果表明：A-STIGCN组合模型预测效果优于传统线性模型及变体模型，与未引入注意力机制的ASTGCN模型相比，MAE降低了约0.131，精度提高了0.068，与未引入外部因素的TGCN模型对比分析，MAPE降低了约0.637%，精度提高了0.079，从而更好地为交通管理提供指导意见。

Short-term traffic flow prediction studies integrated with external properties

Most of the existing traffic flow prediction algorithms only consider the prediction under normal conditions, but not the influence of weather attributes and surrounding geographical attributes on the prediction results, this paper proposed a combined prediction model(A-STIGCN) integrating external attributes. Firstly, it took the external attributes as the attributes of the sections in the road network, and also put the attributes and traffic characteristics of the sections under modeling to obtain the enhanced feature vectors. Secondly, the method used graph wavelet transform and adaptive matrix to extract the local and global spatial feature information of the traffic flow respectively, with the help of the gating cycle unit(GRU) to extract the temporal information. Finally, it captured the temporal dynamic variability of the attention mechanism to predict the traffic flow. Shenzhen taxi trajectory data, corresponding weather data and POI data for prediction, the research results show that A-STIGCN combination model is better than the traditional linear model and variant model, compared with the ASTGCN model without introducing attention mechanism, MAE reduces about 0.131, accuracy improves 0.068, and compared with the TGCN model without introducing external factors, MAPE reduces 0.637% and the accuracy improves 0.079, it provides better guidance for traffic management.