Code Repository for the Bachelor Thesis on "Train Schedule Optimization with focus on Robustness"

The src Folder contains all relevant components. It is structured in the following 6 submodules

• experiments: Jupyter notebook containing the implementation of the genetic algorithms as well as the experiment setups and results
• gym-train-env: A gym environment which was used to test reinfocement learning agent. But because the optimization method was changed to genetic optimization, those environments were not finished. They can be used as a basis for a future project involving RL Agents.
• log_viewer: The Jupyter Notebook of the Log viewer which was used to present the random number experiment in the intermediat presentation.
• model: The implementation of the components which were described in the documentation.
• simulation: Implementation of the behaviour for the components decribed in the model

The usage of the train network simulator is demonstrated in this section with a given example train network and two predefined timetables.

  +-----------+                    +-----------+ 
  |           |                    |           | 
  | Station A +--------------------+ Station C | 
  |           |                    |           | 
  +-----------+                    +-----------+ 

Import the simulator module and network architecture. If there is no predefined network architecture for your problem you can create on in the /src/simulation/networks file.

from src.simulation.simulator import Simulator
from src.simulation.networks import DebugNetwork

To evaluate timetables, you have to first instantiate the Simulator with the given network architecture

sim = Simulator(DebugNetwork)

The simulator can now be used to evaluate a timetable:

timetable = [[10,25,90,105],[5,15,70,85]]
sim.evaluate_timetable(timetable)

This automatically performs 12 simulations for the three scenarios (4 each):

• Train Increase
• Passenger Increase
• Train Breakdown

The returned results contain the performance of the network for each scenario, as well as the timetable which was provided.

{'passenger_increase': {'1': 2.37734693877551,
  '2': 2.37734693877551,
  '4': 2.37734693877551,
  '8': 2.37734693877551,
  'reward_total': 9.50938775510204},
 'trains_increase': {'1': 2.37734693877551,
  '2': 2.4054879075447912,
  '3': 2.3862420779295235,
  '4': 1.050675729560095,
  'reward_total': 8.21975265380992},
 'trains_delay': {'5': 1.6823481143668793,
  '30': 2.374356371936044,
  '60': 2.408574858259782,
  '120': 1.545365510890001,
  'reward_total': 8.010644855452707},
 'timetables': [[10, 25, 90, 105], [5, 15, 70, 85]]}

If you only want to conduct a single simulation on the network, you can use the simulate method:

env = sim.simulate(time_tables=timetable, 
            connections_per_h=1,
            passenger_increase_factor=1,
            force_train_stop_time=0,
            plot=False)

This command returns a train environment, which can then be used to evaluate the performance rewards of this network.

reward_early = env.get_reward_early_arrival()
reward_handled = env.get_reward_handled()
reward_delay = env.get_reward_delay(tolarance_per_line=5, num_lines=2)