Python scripts for water quality simulators in WEST (MIKE DHI).

• Sensitivity analysis
• Model Inference/Calibration
• Automated simulation with control over inputs and model parameters

• Ironpython
• Python 2.7
• TornadoNet

• Install Ironpython and set environment variable (ipy).

• Install python 2.7 with the following libraries: numpy, pandas, os, time, timeit, json, shutil and sys

• Download Parallel_WESTRunningPyWrapper.py and RunModel.py and set them in a project folder under \AutomatedCall\

• RunModel.py Is an IronPython script which interfaces with Tornado.net in order to call a WEST experiment.

• Parallel_WESTRunningPyWrapper.py Python class which calls RunModel.py in parallel. Management of model resources. - Atributes:

            - self.InstanceNames = List the Model names
            - self.OccupationList = boolean list, 1 for occupied node, 0 for free
            - self.NumberofInstances = Number of total Model sessions (should not be higher than the number of local licenses or computer cores)
            - self.Paths = Project directory Path
            - self.threads_active = Number of threads initiated
  

• Instanciate n models with:

  ModelInstance = ModelInstances(['Model_1', 'Model_2', 'Model_n'], modelPath = "<ProjectFolder>", modelInternalName = 'Model_dummy_1.Dynamic.ObjEval.Exp')

  You can also run one model in parallel at different processors:

  ModelInstances(['Model_dummy_1'], modelPath = "<ProjectFolder>", modelInternalName = 'Model_dummy_1.Dynamic.ObjEval.Exp')

• Set Parameter values: e.g

  Timelist = {'StartTime':0,'StopTime':15}
  Parameterlist = {'n':0.07, 'KBODs':0.55, 'KLmin':0.1, 'KO2':1.6, 'SOD':1} 

  A Parameterlist is formed by a dictionary of parameter values. An entry of the form:

  • 'param' :value will set all parameters sharing the same name (even if it has been declared as a Top-level parameter) to the value specified.
  • 'param@submodel' :value will set the parameter 'param' only in the block model 'submodel'. This allows for local specifications.
  • 'param@Manipulated':value sets a manipulated variable value which doesn't carry any declared input. There is the option 'param@Manipulated#submodel':value to make the change only at submodel level.
  • 'param@Multiplier':value works by ofsetting its current saved value by a multiplyier value.
  • 'param@Multiplier':value works by ofsetting its current saved value by a multiplyier value.
  • 'inputname@Intput': Declare change of input file. Inputs should be added as: InputName@Input_DirectoryOfInputSamples Inputs in the directory should be labelled as Inputname_id (and all should be as txt)

• Run Model:

  ModelInstance.model_run(Parameterlist, Timelist)

  Note: each call to model_run will use the next available Model instance in an availability basis. This will open a call in the background. The execution of the program will be halted until all simulations are finished.

• Read outputs:

  output = ReadOutput('Model_dummy_1.Dynamic_water_quality.Simul.1.out.txt', ModelInstance.InstanceNames[0], TimeWindowStart = '01-01-2012 00:00:00', variable = '.River_5.DO')

  Note: Simulations produced by calling .model_run(Parameterlist, Timelist) store all output files in a new folder \SimulationOutput\ModelName\Sim_i. Each Sim_i contains: All output files from the WEST experiment which has .out. in their names. A txt file containing all iternal parameter values in the model <Internal_parameters.out.txt> and a file containing all changed parameter values from the Parameterlist in <External_parameters.txt>.

This script automatizes a parameter sampling scheme provided by:

SamplingLayout = pd.read_csv('\pyWEST\500SamplesUP.txt', index_col = [0])

An example for the format is found at pyWEST\500SamplesUP.txt

The script performs a simulation for each row in the layout. It will call in parallel as many simulations as resources allocated through the parameter TreadingPool at:

ModelInstance.model_run(Parameterlist, Timelist, ThreadingPool = 4, ParamDatabase = os.path.join(repPath, '500SamplesUPReduced.txt'))

The function Read_SA_outputVariable loads the selected variable from the output folder.

Included two jupyter notebooks containing:

• Bayesian_Inference_Example.ipynb : Examples and tutorial to perform bayesian parameter inference to a West model.
• Calibration_Example.ipynb : A tutorial for perform model calibration with scipy.minimize