Physics-Informed Graph Neural Networks for Water Distribution Systems

Official Code for the paper "Physics-Informed Graph Neural Networks for Water Distribution Systems". \ All system and package requirements are listed in the document 'REQUIREMENTS.txt'.

Simulating scenarios to generate data

Data on heads, pressures, demands and flows for WDS can be generated for longer periods of time (Vrachimis et al. https://github.com/KIOS-Research/BattLeDIM) using:

python dataset_generator.py

A number of arguments can be passed to dataset generation parameters:

'--wdn'             "Specify the WDS for which you want to simulate the scenarios; default is l_town. Choices are ['hanoi', 'fossolo', 'pescara', 'l_town', 'zhijiang']" 
'--start_scenario'  "Specify the start scenario name, must be an integer; default is 1"
'--end_scenario'    "Specify the end scenario name, must be an integer; default is 50"
'--sigma_dem'       "Specify the standard deviation of the noise to be added to the demand patterns; default is 0.1."   
'--sigma_dia'       "Specify the standard deviation of the noise to be added to the diameters; default is 1/30"   
'--_seed'           "Specify the random seed for noise; default is None, where it will be set to the scenario name for every scenario."   
'--start_time'      "Specify the start time of the simulation; default is 2018-01-01 00:00, the simulation will be done every 30 minutes starting from this time."   
'--end_time'        "Specify the end time of the simulation; default is 2018-01-14 23:30."   

The simulation will produce an xlsx and multiple csv files in the folder 'Results-Clean' in the respective directories. The xlsx file will be used for training the models.

Training and Evaluation

Models can be trained using

python run.py

A number of arguments can be passed to specify model types and hyperparameters:

'--wdn'             "Specify the WDS for which you want to simulate the scenarios; default is l_town. Choices are ['hanoi', 'fossolo', 'pescara', 'l_town', 'zhijiang']" 
'--mode'            "train_test i.e. train and test a new model, or evaluate i.e. evaluate on an already trained model; default is train_test. "
'--warm_start'      "Specify True if you want to further train a partially trained model. model_path must also be specified; default is False."
'--model_path'      "Specify model path in case of re-training or evaluation; default is None."
'--start_scenario'  "Specify the start scenario name, must be an integer; default is 1"
'--end_scenario'    "Specify the end scenario name, must be an integer; default is 20"
'--n_days'          "Specify the number of days of data to be used for training; default is 2 days."
'--batch_size'      "Specify the mini-batch size; default is 48."
'--n_epochs'        "Specify the number of epochs of training; default is 3000."    
'--lr'              "Specify the learning rate; default is 1e-4."
'--decay_step'      "Specify the step size of the lr scheduler; default is 300."
'--decay_rate'      "Specify the decay rate of the lr scheduler; default is 0.75."
'--I'               "Specify the number of GCN layers; default is 5."
'--n_iter'          "Specify the minimum number of iterations; default is 10."
'--r_iter'          "Specify the maximum number of additional (random iterations; default is 5."
'--n_mlp'           "Specify the number of layers in the MLP; default is 2."
'--M_l'             "Specify the latent dimension; default is 128."

Trained models can be used for evaluation using run.py by specifying the 'evaluate' mode and 'model_path'.

Important Information

Every WDS is specified by an '.inp' file. We have included those files for all WDS (except Hanoi):

• Except for 'hanoi', we modify the '.inp' file to generate different scenarios.
• The '.inp' files for Hanoi WDS can be downloaded from: https://github.com/KIOS-Research/LeakDB (Vrachimis et al. 2018).