Skip to content
Snippets Groups Projects
Commit 32b67bf1 authored by arnetfi's avatar arnetfi
Browse files

Upload New File

parent d607c741
No related branches found
No related tags found
No related merge requests found
# Time Series Analysis for Train Delay Prediction
## Files
There are three files each of which trains one of the three models.
##### Non-Interacting Model
_Parameter value that need to be chosen_
- Train service: S2, S8 or S24
- Direction: Down or Up (North-South or South-North direction)
- p: how many preceding stations should be included (1, 2 or 3)
- k: k-parameter for k-fold cross validation
_Input_
Depending on the parameter values that are chosen, the input is given as a set of files containing the necessary information to train the model for every station of the respective train and direction.
_Output_
Calculated:
- Parameters (a_1, ..., a_p and c_k) for each station k
- Delay predictions for test set
Stored: several dictionaries containing information about MAEs and parameters values trained:
- MAE of non-interacting model at every station of respective train service and direction
- MAE of baseline approach at every station
- Trained parameters of non-interacting model for every station of respective train service and direction
- Standard deviations of trained parameters at every station of respective train service and direction
##### Intercating Model
_Parameter values that need to be chosen_
- k (int): k-parameter for k-fold cross validation
- headway (int): maximum allowed scheduled headway between train j and train i [min]
- threshold (int or float): minimal delay that train j needs to have for it to be considered to have a knock-on influence
- train_on_knockOn (boolean): if set true, the non-interacting model is trained on runs where a knock-on delay was detected only. If set false, it is traind on all data.
- predict_S24_delay (boolean): if set true, S24 delay in Enge is predicted, else the actual delay value is taken
_Input_
- Dataframe with arrival events of S24 and S8 service in Enge
- Trained parameters (a and c) of S24 on section Zürich Wollishofen - Zürich Enge
- Dataframe with information about S2 on section Thalwil-Zürich Enge
- Dataframe with information about S24 on section Thalwil-Zürich Enge
_Output_
Calculated:
- Parameters a_1 and c_k for S2 service in South-North direction at station Enge for non-interacting model
- Parameters a_1, b and c_k for S2 service in South-North direction at station Enge for interacting model
- Delay predictions with both models for test set
Stored: Two dataframes, one containing operating figures to analyse and assess the performance of the interacting model and one containing operating figures to analyse and assess the performance of the non-interacting model:
- MAE of the model
- MAE of the baseline approach
- Trained parameters
- Standard deviations of trained parameters
##### Binary Decision Model
_Parameters that need to be chosen_
- k (int): k-parameter for k-fold cross validation
- headway (int): maximum allowed scheduled headway between train j and train i [min]
- threshold (int or float): minimal delay that train j needs to have for it to be considered to have a knock-on influence
- predict_S24_delay (boolean): if set true, S24 delay in Enge is predicted, else the actual delay value is taken
_Input_
- Dataframe with arrival events of S24 and S8 service in Enge
- Trained parameters (a and c) of S24 on section Zürich Wollishofen - Zürich Enge
- Dataframe with information about S2 on section Thalwil-Zürich Enge
- Dataframe with information about S24 on section Thalwil-Zürich Enge
_Output_
Calculated:
- Parameters a, b and c for S2 service in South-North direction at station Enge
- Delay predictions for test set
Stored: Dataframe containing operating figures to analyse the model and assess its performance:
- MAE of the model
- MAE of the baseline approach
- Trained parameters
- Standard deviations of trained parameters
\ No newline at end of file
0% Loading or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment