Predicting trajectory destinations based on diffusion model integrating spatiotemporal features and urban contexts

Real world driving trajectory dataset in Nanshan and Futian districts, Shenzhen, China, collected in October, 2017 by Amap platform. The dataset is used to predict trajectory destinations. Road network and POI statistics are utilized in this dataset, serving as urban contexts. The geometries are in Gauss-Kruger zone 38 (epsg:4526) with GCJ-02 latitude-longitude coordinate confusion.The published article is available (Hu et al., 2024) on International Journal of Digital Earth.The latest version of our code is available on GithubFile description:<code>code.zip</code>: code for model structure, data pipeline and training, testing procedure.<code>data.zip</code>: dataset and code for this study, including:<code>data.zip/embedding/</code>: the trained embeddings of road topology by LINE method.<code>data.zip/predict_model/</code>: the trained parameters of our model and baselines, with *.pth suffix for pytorch framework.<code>data.zip/roads/</code>: the shp file of road network. POI statistics are contained in <code>road_input.csv</code><code>data.zip/trajectories/</code>: driving trajectories of each day. <code>metadata.csv</code> contains the departure time, destination and other statistics.

本数据集为2017年10月由高德（Amap）平台于中国深圳市南山区与福田区采集的真实驾驶轨迹数据集，旨在用于轨迹目的地预测任务。数据集采用道路网络与POI（兴趣点）统计信息作为城市上下文特征。其几何数据采用高斯-克吕格（Gauss-Kruger）3度带第38带坐标系（epsg:4526），并使用GCJ-02加密经纬度坐标。相关研究成果以Hu等人（2024）的名义发表于《国际数字地球学报》(International Journal of Digital Earth)。本研究的最新代码已开源至GitHub，文件说明如下：
- `code.zip`：包含模型结构、数据流水线、训练与测试流程的实现代码。
- `data.zip`：本研究所用数据集与配套代码，具体包含以下子内容：
- `data.zip/embedding/`：通过LINE（Large-scale Information Network Embedding）方法训练得到的道路拓扑嵌入向量。
- `data.zip/predict_model/`：本研究模型及基线模型的训练参数，文件后缀为`.pth`，适配PyTorch框架。
- `data.zip/roads/`：道路网络的SHP格式文件，POI统计信息存储于`road_input.csv`文件中。
- `data.zip/trajectories/`：每日的驾驶轨迹数据。
- `metadata.csv`：包含出发时间、目的地及其他统计信息的元数据文件。