Is there a way to make this routine to work on OpenLCA 2.1.0? It used to work on previous version, but it is not working anymore.
Any clue could help is appreciated.
Best regards,
import csv
from org.openlca.app import App
from org.openlca.app.components import FileChooser, ModelSelectionDialog
from org.openlca.app.db import Cache
from org.openlca.app.util import Labels
from org.openlca.core.math import CalculationSetup, CalculationType, SystemCalculator
from org.openlca.core.matrix import ProductSystemBuilder, LinkingConfig
from org.openlca.core.model import ProcessType, ProductSystem
from org.openlca.core.database import ImpactMethodDao, ProcessDao
from org.eclipse.swt.widgets import Display
def main():
global db
# select the processes
processes = ModelSelectionDialog.multiSelect(ModelType.PROCESS)
if processes is None or len(processes) == 0:
print("No processes were selected")
return
print("Selected %i processes" % len(processes))
# select the LCIA method
method = ModelSelectionDialog.select(ModelType.IMPACT_METHOD)
if method is None:
print("No LCIA method was selected")
return
indicators = ImpactMethodDao(db).getCategoryDescriptors(method.id)
print("Selected LCIA method '%s' with %i indicators" %
(method.name, len(indicators)))
file_name=method.name + ".csv";
# select the CSV file where the results should be written to
f = FileChooser.forExport('*.csv', 'export.csv')
#f = FileChooser.forExport('*.csv', file_name)
if f is None:
print("No CSV file selected")
return
print("Selected CSV file: %s" % f.absolutePath)
# init the CSV file, run calculations, and write results
with open(f.getAbsolutePath(), 'wb') as stream:
# configure the CSV writer
# see
https://docs.python.org/2/library/csv.html
writer = csv.writer(stream, delimiter=';')
# write the indicators as column headers
header = ['Process', 'Type', 'Product', 'Amount', 'Unit']
for i in indicators:
header.append(enc('%s (%s)' % (i.name, i.referenceUnit)))
#header.append(enc('%s (%s)' % (i.name, i.id)))
#header.append(enc('%s' % (i.name)))
writer.writerow(header)
for d in processes:
# load the process
process = ProcessDao(db).getForId(d.id)
ptype = 'Unit process'
if process.processType != ProcessType.UNIT_PROCESS:
ptype = 'LCI result'
qref = process.quantitativeReference
# we can only create a product system from a process
# when it has a quantitative reference flow which is
# a product output or waste input
if qref is None:
print('Cannot calculate %s -> no quant.ref.' % d.name)
continue
# prepare the CSV row; we will calculate the results
# related to 1.0 unit of the reference flow
row = [
enc(Labels.getDisplayName(process)),
ptype,
enc(Labels.getDisplayName(qref.flow)),
1.0, # qref.amount,
enc(Labels.getDisplayName(qref.unit))
]
# build the product system with a configuration
print('Build product system for: %s' % d.name)
config = LinkingConfig()
# set ProcessType.UNIT_PROCESS to prefer unit processes
config.preferredType = ProcessType.LCI_RESULT
# provider linking: the other options are IGNORE and PREFER
config.providerLinking = LinkingConfig.DefaultProviders.ONLY
builder = ProductSystemBuilder(
Cache.getMatrixCache(), config)
system = builder.build(process)
system.targetAmount = 1.0 # the reference amount
# run the calculation
print('Calculate process: %s' % d.name)
calculator = SystemCalculator(
Cache.getMatrixCache(), App.getSolver())
setup = CalculationSetup(
CalculationType.SIMPLE_CALCULATION, system)
setup.impactMethod = method
result = calculator.calculateSimple(setup)
# write results
print('Write results for: %s' % d.name)
for i in indicators:
value = result.getTotalImpactResult(i)
row.append(value)
writer.writerow(row)
def enc(s):
return unicode(s).encode("utf-8")
if __name__ == "__main__":
Display.getDefault().asyncExec(main)