基于K近邻算法和混合模型的短期交通流预测

Short-Term Traffic Flow Forecast Based on K near Neighbor Algorithm and Hybrid Model

道路拥堵情况由于车辆数量的不断增加而一直存在,及时、准确地进行交通流预测仍是研究的重点。由于交通流数据是庞大的、复杂的,本研究使用KNN算法对数据进行挑选,选择与目标监测点相关性更高的数据,将其输入到CNN-GRU-ATT模型中,对交通流数据进行预测。模型中的CNN层提取特征,GRU层描述时间趋势,ATT层实现对关键信息的关注。实验发现:该模型与其他基线模型相比,模型精度更高,MAPE最高降低了28.33%;与未引入KNN算法相比,模型拟合优度有所提升,达到了97.79%。

Road congestion has always existed due to the increasing number of vehicles, and traffic flow fore-casting in time and accurately is still the focus of research. Because traffic flow data is huge and complex, this study uses the KNN algorithm to select the data, select data with higher correlation with the target monitoring point, and enter it into the CNN-GRU-ATT model. Perform predictions. The CNN layer extracts feature in the model, the GRU layer describes time trends, and the ATT layer achieves attention to key information. The experiment found that compared with other baseline models, the model has higher accuracy, and MAPE has reduced up to 28.33%;compared with the KNN algorithm, the model fitting superiority has improved, reaching 97.79%.