def setScenarioData(self):
"""
Method to setup the scenario data for scenario tree construction
@ In, None
@ Out, None
"""
MultipleKnapsack.setScenarioData(self)
def __init__(self):
"""
Constructor
@ In, None
@ Out, None
"""
MultipleKnapsack.__init__(self)
def createModel(self):
"""
This method is used to create pyomo model.
@ In, None
@ Out, model, pyomo.AbstractModel, abstract pyomo model
"""
model = MultipleKnapsack.createModel(self)
return model
def addAdditionalSets(self, model):
"""
Add specific Sets for DROMKP problems
@ In, model, pyomo model instance, pyomo abstract model
@ Out, model, pyomo model instance, pyomo abstract model
"""
model = MultipleKnapsack.addAdditionalSets(self, model)
return model
def initialize(self, initDict):
"""
Mehod to initialize
@ In, initDict, dict, dictionary of preprocessed input data
{
'Sets':{setName: list of setValues},
'Parameters':{paramName:{setsIndex:paramValue}} or {paramName:{'None':paramValue}},
'Settings':{xmlTag:xmlVal},
'Meta':{paramName:{setIndexName:indexDim}} or {paramName:None},
'Uncertainties':{paramName:{'scenarios':{scenarioName:{setIndex:uncertaintyVal}}, 'probabilities': [ProbVals]}}
}
@ Out, None
"""
MultipleKnapsack.initialize(self, initDict)
if self.uncertainties is None:
raise IOError(
'Uncertainty data is required by "{}", but not provided!'.
format(self.name))
def addAdditionalParams(self, model):
"""
Add specific Params for DROMKP problems
@ In, model, pyomo model instance, pyomo abstract model
@ Out, model, pyomo model instance, pyomo abstract model
"""
model = MultipleKnapsack.addAdditionalParams(self, model)
model._lambda = pyomo.Param(within=pyomo.UnitInterval, mutable=True)
model.alpha = pyomo.Param(within=pyomo.UnitInterval, mutable=True)
return model
def addVariables(self, model):
"""
Add variables for DROMKP problems
@ In, model, pyomo model instance, pyomo abstract model
@ Out, model, pyomo model instance, pyomo abstract model
"""
model = MultipleKnapsack.addVariables(self, model)
# variables for robust optimization
model.gamma = pyomo.Var(within=pyomo.NonNegativeReals)
model.nu = pyomo.Var(model.sigma, within=pyomo.NonNegativeReals)
return model
def pysp_scenario_tree_model_callback(self):
"""
scenario tree instance creation callback
@ In, None
@ Out, treeModel, Instance, pyomo scenario tree model for two stage stochastic programming,
extra variables 'y[*,*]' is used to define the priorities of investments,
'gamma' and 'nu[*]' is used for the distributionally robust optimization counter-part.
"""
treeModel = MultipleKnapsack.pysp_scenario_tree_model_callback(self)
# additional variables:
firstStage = treeModel.Stages.first()
treeModel.StageVariables[firstStage].add('gamma')
treeModel.StageVariables[firstStage].add('nu[*]')
return treeModel
def addVariables(self, model):
"""
Add variables for DROMKP problems
@ In, model, pyomo model instance, pyomo abstract model
@ Out, model, pyomo model instance, pyomo abstract model
"""
model = MultipleKnapsack.addVariables(self, model)
# variables for CVaR optimization
model.nu = pyomo.Var(domain=pyomo.NonNegativeReals)
model.u = pyomo.Var(domain=pyomo.Reals)
# variables to retrieve additional information
model.cvar = pyomo.Var(domain=pyomo.Reals)
model.expectProfit = pyomo.Var(domain=pyomo.Reals)
return model
def addAdditionalParams(self, model):
"""
Add specific Params for DROMKP problems
@ In, model, pyomo model instance, pyomo abstract model
@ Out, model, pyomo model instance, pyomo abstract model
"""
model = MultipleKnapsack.addAdditionalParams(self, model)
model.epsilon = pyomo.Param(within=pyomo.NonNegativeReals,
mutable=True)
model.prob = pyomo.Param(model.sigma,
within=pyomo.UnitInterval,
mutable=True)
# model.dist will be changed on the fly via scenario callback functions
model.dist = pyomo.Param(model.sigma, mutable=True)
return model
def addAdditionalConstraints(self, model):
"""
Add specific constraints for DROMKP problems
@ In, model, pyomo model instance, pyomo abstract model
@ Out, model, pyomo model instance, pyomo abstract model
"""
model = MultipleKnapsack.addAdditionalConstraints(self, model)
## model.dist will be changed on the fly
def constraintWasserstein(model, i):
expr = sum(
sum(model.net_present_values[i] * model.x[i, m]
for i in model.investments) for m in model.capitals)
return -model.gamma * model.dist[i] + model.nu[i] <= expr
model.constraintWassersteinDistance = pyomo.Constraint(
model.sigma, rule=constraintWasserstein)
return model
def pysp_scenario_tree_model_callback(self):
"""
scenario tree instance creation callback
@ In, None
@ Out, treeModel, Instance, pyomo scenario tree model for two stage stochastic programming,
extra variables 'y[*,*]' is used to define the priorities of investments,
'u' and 'nu' is used for CVaR optimization counter-part.
'cvar' is used to retrieve the calculated CVaR value
'expectProfit' is used to retrieve the expect profit under CVaR optimization
"""
treeModel = MultipleKnapsack.pysp_scenario_tree_model_callback(self)
# additional variables:
firstStage = treeModel.Stages.first()
secondStage = treeModel.Stages.last()
treeModel.StageVariables[firstStage].add('u')
treeModel.StageVariables[secondStage].add('nu')
treeModel.StageVariables[secondStage].add('cvar')
treeModel.StageVariables[secondStage].add('expectProfit')
return treeModel
def addAdditionalConstraints(self, model):
"""
Add specific constraints for DROMKP problems
@ In, model, pyomo model instance, pyomo abstract model
@ Out, model, pyomo model instance, pyomo abstract model
"""
model = MultipleKnapsack.addAdditionalConstraints(self, model)
def cvarConstraint(model):
return model.nu + sum(
sum(model.net_present_values[i] * model.x[i, m]
for i in model.investments)
for m in model.capitals) + model.u >= 0
model.cvarConstraint = pyomo.Constraint(rule=cvarConstraint)
## used to retrieve additional information from the optimization
def expectProfit(model):
"""
Method to compute the expect profit of stochastic programming
@ In, model, instance, pyomo abstract model instance
@ Out, expr, pyomo.expression, constraint to compute the expect profit
"""
return model.expectProfit - sum(
sum(model.net_present_values[i] * model.x[i, m]
for i in model.investments) for m in model.capitals) == 0
model.computeExpectProfit = pyomo.Constraint(rule=expectProfit)
def cvar(model):
"""
Method to compute the conditional value at risk
@ In, model, instance, pyomo abstract model instance
@ Out, expr, pyomo.expression, constraint to compute the CVaR
"""
return model.cvar - model.u + 1. / (1. -
model.alpha) * model.nu == 0
model.computeCVAR = pyomo.Constraint(rule=cvar)
return model