多源数据驱动CNN-GRU模型的公交客流量分类预测

为精准分析公交线路与站点不同客流的出行特征及时变差异性，结合深度学习理论，提出了一种基于卷积神经网络(CNN)与门控制循环单元(GRU)组合的公交客流分类预测模型；融合匹配公交一卡通刷卡、公交车GPS轨迹、线路和站点基础信息、气象等多源数据，实现公交客流数据重构；采用K-Medians算法将乘客分为通勤类和非通勤类；以乘客类型、历史客流量、时段、高/平峰、星期、降水量、重大活动等因素为输入向量，分别建立CNN与GRU单一模型，并利用均方误差、均方根误差、平均绝对误差为评价指标，开展预测；针对单一模型不适用多特征时间序列预测等问题，分别构建了由CNN和GRU组合的线路客流和断面客流预测模型；以北京市特15路公交为例，预测工作日与非工作日场景下的线路及断面的分类客流。分析结果表明：对于通勤类和非通勤类线路及断面客流，组合模型的均方误差相比单一模型平均降低了57.932、13.106和33.987，均方根误差平均降低了1.862、1.058和1.538，平均绝对误差平均降低了1.399、0.487和0.613，可见，多源数据驱动下的CNN-GRU组合模型具有良好的预测性能。

Bus passenger flow classification prediction driven by CNN-GRU model and multi-source data

To accurately analyze the trip characteristics and time-varying differences of different passenger flows of bus routes and stops, combined with deep learning theory, a bus passenger flow classification prediction model based on a combination of a convolutional neural network (CNN) and gated recurrent unit (GRU) was proposed. By integrating and matching multi-source data, such as bus card swiping, bus global positioning system (GPS) trajectory, route and station basic information, and weather data, bus passenger flow data was reconstructed. The K-medians algorithm was used to divide passengers into commuter and non-commuter categories. Taking the factors of passenger type, historical passenger flow, time period, high/flat peak, week, precipitation, and major events as input vectors, a single model of CNN and GRU was established, and forecasts were conducted using mean squared error (MSE), root mean squared error (RMSE), and mean absolute error (MAE) as evaluation indicators. As a single model is not suitable for multi-feature time series forecasting, line passenger flow and cross-section passenger flow prediction models combined with a CNN and GRU were constructed. Taking Beijing Special 15 Bus as an example, the classified passenger flows of routes and cross-sections under the scenarios of working days and non-working days were predicted. Analysis results show that for commuter and non-commuter routes and cross-section passenger flows, the MSEs of the combined model reduce by 57.932, 13.106, and 33.987 on average, the RMSEs reduce by 1.862, 1.058, and 1.538 on average, and the MAEs reduce by 1.399, 0.487, and 0.613 on average, respectively. Thus, the CNN-GRU combined model driven by multi-source data has a good prediction performance. 3 tabs, 7 figs, 36 refs.