Submission Master Thesis "Adversarial Attacks and Robustness in Water Distribution Systems"

This repository contains code, data and results from my Master Thesis. The contents of the folders are as follows:

• Data: networks used for experiments and results from the threshold validation of the leakage detector. Networks in this folder have been specifically adapted (e.g. their duration has been changed).
• Figures: All figures that also appeared in the submission
• Formalizations: Different formlizations of the least sensitive point problem whidh were initially developed before one was picked.
• Leakage Detector Plots: plots from the leakage detector comparison including some plots which were not shown in the thesis
• Max Residual Change and Mean Residual Change: results from additional experiments concerning the change of detector residuals based on the leak area. These are not directly linked to the results of the algorithms, so they do not appear in the thesis.
• Network Originals: the Net1 and Hanoi network before adaptation
• Results: results achieved by the algorithms on the one-week and two-week dataset. These were discussed in the thesis.
• Taxonomy Trees: taxonomies for adversarials in Water Distribution Systems and for robustness measures against them.
• src: source code

To run the code, I recommend that you set up a conda environment using the same packages as given in water.yml. In the best case, this works by running

conda env create -f water.yml

However, I cannot give a guarantee as this depends on system preferences and conda peculiarities. If the command above does not work, please install at least the following packages manually into a new environment:

• wntr
• pygad
• numpy
• scipy
• pandas
• matplotlib

If everything is set up correctly, you should be able to navigate to the src folder and run

python find_lsp.py

to run the Enhanced Genetic Algorithm on the two-week dataset. All classes and class methods in the src-folder contain explanatory doc-strings. In case of questions, feel free to open an issue.