機械学習を用いた輸送障害時の旅客流動予測モデルの開発

抄録

When a train service disruption happens, dispatchers make a train rescheduling plan to control train delay and congestion. To support dispatcher's work, we developed a passenger flow prediction model during disruption using machine learning techniques. The proposed model consists of time-series waveform prediction using LightGBM and rule-based correction. We defined 5 time series waveform clusters using unsupervised learning with dynamic time warping(DTW). As a result, we verified prediction average accuracy 75% about 4 cases of unlearned disruptions.