利用手机信令数据、高德实时交通态势数据、POI数据等多源大数据，以宁波市拥堵核心区域为研究单元，对比了普通最小二乘法回归模型、经典地理加权回归模型、多尺度地理加权回归（MGWR）模型的精度；并选用最优模型定量探究了宁波市拥堵核心区不同时段的交通拥堵驱动力。结果表明：(1)MGWR模型对交通拥堵驱动力的拟合效果较好，且能反映不同驱动力间的空间异质性；(2)实时人口密度、路网密度、土地利用混合度、职住平衡指数对交通拥堵影响存在时空分异，公司分布密度对交通拥堵影响存在圈层性。研究结果可为大数据在交通拥堵领域的应用提供一定参考，为管理者分析交通拥堵根源提供依据。

By using multi-source big data such as mobile phone signaling data, real-time traffic data of Amap and POI data, taking the congestion core area in Ningbo as the research unit, we compared the accuracy of ordinary least squares regression model, classical geo-weighted regression model and multi-scale geo-weighted regression(MGWR) model, and selected the optimal model to quantitatively explore the traffic congestion driving force of different time periods in Ningbo's congestion core area. The results show that(1)the MGWR model fits the traffic congestion driving force better and can reflect the spatial heterogeneity among different drivers.(2)There are spatio-temporal differences in the effects of real-time population density, road network density, land-use mix and job-occupancy balance index on traffic congestion. There are circles in the effects of company distribution density on traffic congestion. This study can provide some references for the future application of big data in the traffic congestion field, and a basis for managers to analyze the root causes of traffic congestion.