Recurrent Neural Network Models of Human Mobility

Abstract

Locational data generated by mobile devices present an opportunity to substantially simplify methodologies and reduce analysis latencies in short-term transportation planning applications. Short-term transportation planning, such as traffic flow management or traffic demand management, requires accurate prediction of daily network congestion levels and the congestion contributors. The existing human mobility models using locational data have fo- cused on predicting next activities, and many models limited the prediction to only temporal features or only spatial features that they cannot be directly applied to such applications. In this dissertation, we propose Long Short Term Memory (LSTM) models for learning and predicting human mobility sequences using mobile locational data. The major contributions of this dissertation include the following: first, we developed the LSTM mobility models that are capable of learning and predicting the entire mobility sequences within a time window of interest; second, we developed the LSTM mobility models that are able to predict activity sequences with activity type choices and explicit spatial-temporal choices; third, the LSTM mobility models are able to capture long-term activity dependencies. The LSTM models can be applied for transportation demand forecasting problems, including typical-day activity prediction, medium-term activity prediction, and activity prediction with social-demographic information. We performed validation through micro-simulation and compared the simula- tion results to real-world traffic counts. The results showed high similarities between gen- erated traffic volumes and observed traffic volumes. The performance of LSTM models was also compared against baseline sequence models including Hidden Markov Models and near- est neighbour models. Using daily activity structure and daily travel distance as metrics, we observed better performance of LSTM models due to the capability of learning long-term activity dependencies. Lastly, we extended the LSTM mobility models for learning activ- ity sequences with contextual information. We demonstrated the capability of the LSTM models to handle both discrete and continuous contextual information.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 4 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.