本数据集以洛杉矶县高速公路为研究区域，以Caltrans PeMS系统采集的交通速度数据为基础，经Li等人（2018）在DCRNN项目中筛选和预处理后形成METR-LA交通速度基准数据集。数据集包含207个环路检测器，时间范围为2012年3月1日至2012年6月30日，时间分辨率为5分钟，共34,272个时间步长，约6,519,002个观测记录。数据仅包含交通速度指标（mph），采用Z-score标准化处理。按照70%、10%、20%的比例划分为训练集、验证集和测试集。基于路网距离使用阈值化高斯核构建传感器间的加权邻接矩阵。该数据集被广泛用于交通预测领域的深度学习模型评估，是时空交通预测的标准基准数据集之一。
", "ds_source": "交通状态数据：源自加州交通局Caltrans PeMS系统（https://pems.dot.gov/），由Li等人（2018）在DCRNN项目中从原始数据中筛选207个传感器并进行预处理后公开。数据代码地址：https://github.com/liyaguang/DCRNN
", "ds_process_way": "一、数据采集
\n从Caltrans PeMS业务运营系统获取原始环路检测器数据。PeMS系统实时采集加州高速公路全网的车流量、平均速度、占用率等交通参数，数据由加州交通局运营维护。
\n二、数据筛选与预处理
\nLi等人（2018）从原始PeMS数据中筛选出207个可靠传感器，覆盖洛杉矶县主要高速公路路段（I-10、I-105、I-110、I-405、I-210等），剔除异常率过高的传感器。原始数据采样频率为20秒至5分钟不等，统一聚合为5分钟时间窗口的平均速度值（mph）。缺失值保留以检验模型鲁棒性。
\n三、数据集划分
\n全部时间序列数据按70%、10%、20%比例划分为训练集、验证集和测试集。
\n四、数据标准化
\n采用Z-score标准化方法，在训练集上计算均值和标准差后，统一应用于验证集和测试集，避免数据泄露。
\n五、传感器图构建
\n计算各传感器之间的路网距离（采用道路网络距离而非欧氏距离），使用阈值化高斯核构建加权邻接矩阵：
\nW_ij = exp(-dist(v_i, v_j)² / σ²) 若 dist(v_i, v_j) ≤ κ，否则为0
\n其中σ为距离标准差，κ为阈值。构建的有向加权图的邻接矩阵与速度数据一起提供，用于建模交通流在路网上的扩散过程。
", "ds_quality": "一、原始数据来源可靠性
\n数据源自Caltrans PeMS业务运营系统，加州交通局官方交通监测平台，数据采集和维护有标准化的质量保障流程。
\nLi等人（2018）进一步对传感器进行质量筛选，保留207个高质量传感器。
\n二、数据标准化
\nZ-score标准化参数仅在训练集上计算，确保验证集和测试集的独立性。
\n三、缺失值标注
\nMETR-LA数据集中保留原始缺失值位置，不进行人为填充，可用于评估模型对缺失数据的鲁棒性。
\n四、图结构验证
\n邻接矩阵为方阵且维度与传感器数量一致（207x207），确保图结构与节点数匹配。
", "ds_acq_start_time": "2012-01-01 00:00:00", "ds_acq_end_time": null, "ds_acq_place": "美国加利福尼亚州洛杉矶县高速公路网", "ds_acq_lon_east": -116.17777777777778, "ds_acq_lat_south": 34.04277777777778, "ds_acq_lon_west": -116.53666666666666, "ds_acq_lat_north": 34.22138888888889, "ds_acq_alt_low": null, "ds_acq_alt_high": null, "ds_share_type": "open-access", "ds_total_size": 63437745, "ds_files_count": 0, "ds_format": "HDF5、CSV", "ds_space_res": "", "ds_time_res": "", "ds_coordinate": "无", "ds_projection": "", "ds_thumbnail": "2af8c989-9763-49fd-a824-a5401fab9100.png", "ds_thumb_from": 0, "ds_ref_way": "", "paper_ref_way": "", "ds_ref_instruction": "本数据集为DCRNN项目（Li et al., ICLR 2018）公开的METR-LA交通速度基准数据集。原始数据来源于Caltrans PeMS系统（https://pems.dot.gov/）。数据集通过项目页面（https://github.com/liyaguang/DCRNN）以开源形式获取。使用时请引用Li et al. (2018)。", "ds_from_station": "", "organization_id": "9ecaaa78-39e9-411e-9f24-274e12aa643f", "ds_serv_man": "张新民", "ds_serv_phone": "15869040017", "ds_serv_mail": "xinminzhang@zju.edu.cn", "doi_value": "", "subject_codes": [ "410" ], "quality_level": 0, "publish_time": "2026-07-09 10:58:16", "last_updated": "2026-07-09 10:58:16", "protected": false, "protected_to": "2028-06-30 00:00:00", "lang": "zh", "cstr": "", "i18n": { "en": { "title": "Los Angeles County Highway Traffic Speed Benchmark Dataset (METR-LA)", "ds_format": "HDF5、CSV", "ds_source": "Traffic data: from Caltrans PeMS, curated by Li et al. (2018) for the DCRNN project. Publicly available at https://github.com/liyaguang/DCRNN", "ds_quality": "1. Source Reliability\r\nData from Caltrans PeMS, California official traffic monitoring platform with standardized quality assurance. Li et al. (2018) further filtered 207 high-quality sensors.\r\n\r\n2. Normalization Integrity\r\nZ-score computed on training set only, preventing data leakage.\r\n\r\n3. Missing Value Preservation\r\nOriginal missing values preserved to reflect real-world conditions.\r\n\r\n4. Graph Validation\r\nAdjacency matrix verified as square (207x207) matching sensor count.", "ds_ref_way": "", "ds_abstract": "This dataset covers the highway network of Los Angeles County, collected from loop detectors in the Caltrans Performance Measurement System (PeMS). Curated by Li et al. (2018) for the DCRNN project, METR-LA contains traffic speed readings from 207 sensors over 4 months (March 1, 2012 to June 30, 2012), aggregated at 5-minute intervals, yielding 34,272 timestamps and approximately 6,519,002 observations. Traffic speed (mph) is the sole variable, normalized via Z-score standardization. The dataset is split into training (70%), validation (10%), and testing (20%). A weighted adjacency matrix based on road network distances is constructed using a thresholded Gaussian kernel. METR-LA is widely adopted as a standard benchmark for evaluating spatiotemporal traffic forecasting models including DCRNN, Graph WaveNet, and STGCN.", "ds_time_res": "", "ds_acq_place": "Los Angeles County freeway network, California, USA", "ds_space_res": "", "ds_projection": "", "ds_process_way": "1. Data Collection\r\nRaw loop detector data acquired from Caltrans PeMS, California official traffic monitoring platform, covering major highways in Los Angeles County.\r\n\r\n2. Data Curation\r\nLi et al. (2018) selected 207 reliable sensors from raw PeMS data, filtering out sensors with excessive missing rates. Raw readings (20s-5min intervals) aggregated to 5-minute average speed (mph). Missing values preserved for robustness evaluation.\r\n\r\n3. Data Split\r\nTime series split: training (70%), validation (10%), testing (20%).\r\n\r\n4. Normalization\r\nZ-score standardization fitted on training set, applied to all splits without data leakage.\r\n\r\n5. Sensor Graph Construction\r\nPairwise road network distances computed between sensors. Weighted adjacency matrix using thresholded Gaussian kernel: W_ij = exp(-dist(v_i, v_j)^2 / sigma^2) if dist(v_i, v_j) <= kappa, else 0. The directed graph captures traffic flow diffusion on the highway network.", "ds_ref_instruction": "This is the METR-LA traffic speed benchmark from DCRNN (Li et al., ICLR 2018). Raw data from Caltrans PeMS (https://pems.dot.gov/). Dataset available at https://github.com/liyaguang/DCRNN. Please cite Li et al. (2018) when using." } }, "submit_center_id": "ncdc", "data_level": 0, "recommendation_value": 0, "license_type": "https://creativecommons.org/licenses/by/4.0/", "doi_reg_from": "reg_local", "cstr_reg_from": "reg_local", "doi_not_reg_reason": null, "cstr_not_reg_reason": null, "is_paper_in_submitting": false, "belong_to_nieer": false, "ds_topic_tags": [ "交通速度预测", "交通流预测", "时空预测", "图神经网络", "深度学习", "基准数据集" ], "ds_subject_tags": [ "工程与技术科学基础学科" ], "ds_class_tags": [], "ds_locus_tags": [], "ds_time_tags": [], "ds_contributors": [ { "true_name": "张新民", "email": "xinminzhang@zju.edu.cn", "work_for": "浙江大学", "country": "中国" } ], "ds_meta_authors": [ { "true_name": "Yaguang Li", "email": "yaguang.li@usc.edu", "work_for": "University of Southern California", "country": "中国" } ], "ds_managers": [ { "true_name": "张新民", "email": "xinminzhang@zju.edu.cn", "work_for": "浙江大学", "country": "中国" } ], "category": "其他" }