A Mobility Model for Synthetic Travel Demand from Sparse Individual Trajectories

Yuan Liao, Kristoffer Ek, Eric Wennerberg, Sonia Yeh, Jorge Gil

1. Model (Section 2)

The code for the proposed individual mobility model can be found in lib/models.py. See lib/gs_model.py for how it's configured.

There are other scripts under lib/ defined to be called by lib/models.py, lib/gs_model.py, and the below scripts and notebooks.

2. Data pre-processing (Section 3.1)

In this part, the sparse mobility trajectories from the geolocations of Twitter data are extracted and preprocessed.

1) Extract sparse data from their raw databases

This is done in the script src/py/sqlite3-converter.py.

2) Preprocess the sparse geolocations of Twitter data

This corresponds to Section 3.1. Sparse trajectories: geotagged tweets. This is done in the script src/py/tweets-filter.py.

3. Experiment: model validation (Section 3.2)

1) Split the processed Twitter data into validation and calibration sets

This is done in the notebook src/py/1-split-input-for-validation.ipynb.

2) Bayesian optimisation

This is done in the script src/py/parasearch-bayesian.py.

3) Model output on validation and calibration datasets

This is done in the script src/py/parameters-validation.py. The selection of parameter is tested in the script src/py/parameter-D-KL-relationship.py.

4) A test of model parameter transferability (Section 4.3)

This is done in the script src/py/parameters-sensitivity-test.py.

5) Summarise the optimal parameters and model performance

This is done in the notebook src/py/2-parasearch-summary.ipynb.

4. Application: characterising trip distance (Section 3.3)

1) Generate visits using the calibrated model

This is done in the script src/py/multi-region-visits-generation.py.

2) Create statistics and trips from the generated visits

This is done in the script src/py/multi-region-visits-describe.py.

3) Summarise the statistics of the generated visits of multiple regions

This is done in the script src/py/3-multi-region-summary.ipynb.

4) Preprocess the model-generated trips into domestic trips

This is done in the script src/py/multi-region-visits-trips-filtering.py.

5) Characterising trip distance

This is done in the notebook src/py/4-characterising-trip-distance.ipynb and the R script src/r/correlation-multi-region-agg.R. The 10-fold theoretical distributions fitting is done in the script src/py/distance-models-sensitivity-test.py.

5. Value difference between Euclidean trip distance and travel distance (Appendix B.2)

This is to quantify to what extent we underestimate the travel distance by using Haversine distance. This is done by comparing Haversine distance with the reported & the simulated travel distance.

1) Prepare survey data

This is done in the notebook src/py/7-distance-error-data-based.ipynb.

2) Prepare simulation data

2-1) Download drive road networks

This is done in the script src/py/multi-region-drive-network-downloader.py.

2-2) Calculate network distances

This is done in the script src/py/multi-region-distance-error-computation.py.

2-3) Summarise simulated network distances

This is done in the notebook src/py/8-distance-error-simulation-based.ipynb.

6. Descriptive analysis and results illustration

Description of the applied datasets and regional properties is found in the notebook src/py/5-descriptive-analysis.ipynb. The data used for an illustration example of model input and output can be found in the notebook src/py/6-model-input-output-illustration.ipynb.

The below shows a summary of figures.

Script	Objective	Output	No. in the manuscript
`src/r/plot-regions.R`	Visualise the regions for the experiment	`figures/gt-zones.png`	3
`src/r/plot-para-search.R`	The results of parameters search	`figures/para-search.png`	4
`src/r/plot-input-output-chord.R`	Individual mobility ODMs based on the benchmark model and the proposed model	`figures/chord_input_output.png`	5
`src/r/plot-odm-r1-calibration.R`	Comparison of trip frequency rate between the ground truth data and model output - calibration data	`figures/od_pairs_calibration.png`	6
`src/r/plot-odm-r1-validation.R`	Comparison of trip frequency rate between the ground truth data and model output - validation data	`figures/od_pairs_validation.png`	B.1
`src/r/plot-distance.R`	Comparison of trip distance distribution between the ground truth data and the model output	`figures/distance.png`	7
`src/r/plot-sensitivity-r2.R`	Relative performance of each region using the model parameters calibrated from the other regions	`figures/sensitivity.png`	8
`src/r/plot-multi-region-pdf.R`	Distance distributions of model-synthesised domestic trips (PDF)	`figures/trip_distance.png`	9
`src/r/plot-parameters-M-D.R`	Value settings of the model parameters and .	`figures/M_day_D.png`	A.1
`src/r/plot-distance-error-data.R`	Trip distance, Euclidean vs. reported travel distance (survey data)	`figures/distance_error_data.png`	B.2
`src/r/plot-distance-error-simulation.R`	Distance ratio of selected urban areas (simulated)	`figures/distance_error_simulation.png`	B.3

HaoCong1221/individual_mobility_model