MMFire-Train: A Multimodal Time-Series Dataset for Train Fire HRR Trend Prediction

The dataset is mainly generated using the Fire Dynamics Simulator (FDS) and SmokeView based on a full-scale metro train carriage model. A total of 125 parameterized fire scenarios are constructed by combining different fire growth coefficients, decay coefficients, and steady-state durations.

Each simulation covers the complete fire development process for up to 1,200 s.The dataset contains multiple synchronized modalities, including visual images, temperature measurements, smoke-related sensor responses, fire-process videos, spatial temperature-field data, and HRR records. The released dataset consists of 150,016 fire simulation images stored in 75 Parquet shards, 1,497,748 temperature records stored in 5 Parquet shards, 125 MP4 fire-process videos, NumPy-formatted image arrays provided as image_npy.zip, 249 virtual-sensor data files containing temperature, smoke, airflow velocity, carbon monoxide concentration and other fire-environment parameters, and 73,108 raw temperature-field text files organized by fire scenario and simulation time.All modalities originate from the same numerical fire scenarios and are aligned using a unified temporal index.