我以为我理解了一月份关于在优化中使用元模型的讨论。但到目前为止，我的尝试失败了。我有点曲折。我想使用几个学科之一的元模型，其中一月份讨论的示例仅针对元模型的结果进行了优化。我基于抛物面优化教程示例构建了一个示例问题。我添加了第二个组件，将两个组件耦合在一个组中并对其进行了优化。然后我尝试将其中一个组件替换为该组件的元模型。以下是带有真实模型的版本。
from __future__ import print_function
from openmdao.api import IndepVarComp, Component, Problem, Group, ExecComp, NLGaussSeidel, ScipyOptimizer, UniformDriver, DumpRecorder, InMemoryRecorder
import math
import numpy as np
class Paraboloid(Component):
""" Evaluates the equation f(x,y) = (x-3)^2 + xy + (y+4)^2 - 3 """
def __init__(self):
super(Paraboloid, self).__init__()
self.add_param('z', val=0.0)
self.add_param('y', val=0.0)
self.add_output('f_xy', val=0.0)
def solve_nonlinear(self, params, unknowns, resids):
"""f(x,y) = (x-3)^2 + xy + (y+4)^2 - 3
"""
x = params['z']
y = params['y']
f_xy = (x-3.0)**2 + x*y + (y+4.0)**2 - 3.0
unknowns['f_xy'] = f_xy
def linearize(self, params, unknowns, resids):
""" Jacobian for our paraboloid."""
x = params['z']
y = params['y']
J = {}
J['f_xy', 'z'] = 2.0*x - 6.0 + y
J['f_xy', 'y'] = 2.0*y + 8.0 + x
return J
class Sinx(Component):
""" Evaluates the equation f(x) = sin(x) """
def __init__(self):
super(Sinx, self).__init__()
self.C = 50.
self.add_param('x', val=0.0)
self.add_output('z', val=0.0)
def solve_nonlinear(self, params, unknowns, resids):
"""f(x) = 10.* sin(x/C)
"""
x = params['x']
sinx = math.sin(x/self.C) *10.
unknowns['z'] = sinx
def linearize(self, params, unknowns, resids):
""" Jacobian for our paraboloid."""
x = params['x']
J = {}
J['z', 'x'] = math.cos(x/self.C)/self.C *10.
return J
class Mda(Group):
def __init__(self):
super(Mda, self).__init__()
self.f1in=['x']
self.f1out=['z']
self.f2in=['z', 'y']
self.f2out=['f_xy']
self.add('x', IndepVarComp('x', 10.), promotes=['*'])
self.add('y', IndepVarComp('y', 0.), promotes=['*'])
self.add('f1', Sinx(), promotes=self.f1in+self.f1out)
self.add('f2', Paraboloid(), promotes=self.f2in+self.f2out)
self.add('obj_cmp', ExecComp('obj = f_xy'), promotes=['f_xy', 'obj'])
self.nl_solver = NLGaussSeidel()
if __name__ == "__main__":
top = Problem()
root = top.root = Mda()
top.driver = ScipyOptimizer()
top.driver.options['optimizer'] = 'SLSQP'
top.driver.add_desvar('x', lower=-50, upper=50)
top.driver.add_desvar('y', lower=-50, upper=50)
top.driver.add_objective('obj')
top.setup()
top.run()
print('obj =%f at (%f,%f)'%(top['obj'],top['x'],top['y']))
以下是我认为可以用作元模型的课程。
class SinxSurr(MetaModel):
''' Evaluate surrogate of Sinx '''
def __init__(self):
super(SinxSurr, self).__init__()
self.add_param('x', val=0.)
self.add_output('z', val=0., surrogate=FloatKrigingSurrogate())
以下是运行它的部分：
if __name__ == "__main__":
# build training data by sampling
top = Problem()
root = top.root = Group()
root.add('sinx', Sinx())
root.add('p1', IndepVarComp('x', 50.0))
root.connect('p1.x', 'sinx.x')
top.driver = UniformDriver(num_samples=500)
top.driver.add_desvar('p1.x', lower=-50, upper=50)
top.driver.add_objective('sinx.z')
recorder = InMemoryRecorder()
top.driver.add_recorder(recorder)
top.setup()
top.run()
top.cleanup()
# retrieve sampling results from recorder
xlist=[]
slist=[]
for line in recorder.iters:
xlist.append(line['unknowns']['p1.x'])
slist.append(line['unknowns']['sinx.z'])
recorder.close()
if len(xlist) != len(slist):
print('Recorder parse failure')
sys.exit(1)
x=10.
top = Problem()
root = top.root = Mda()
top.driver = ScipyOptimizer()
top.driver.options['optimizer'] = 'SLSQP'
top.driver.add_desvar('x', lower=-50, upper=50)
top.driver.add_desvar('y', lower=-50, upper=50)
top.driver.add_objective('obj')
top.setup()
top.root['f1.train:x'] = xlist
top.root['f1.train:z'] = slist
top.run()
print('obj =%f at (%f,%f)'%(top['obj'],top['x'],top['y']))
当我运行它时，我收到错误：
Setup: Checking for potential issues...
No recorders have been specified, so no data will be saved.
The following parameters have no associated unknowns:
f1.train:x
f1.train:z
Setup: Check complete.
##############################################
Traceback (most recent call last):
File "surrDisc.py", line 164, in <module>
top.root['f1.train:x'] = xlist
TypeError: 'Mda' object does not support item assignment
我不明白如何（至少）将元模型中的数据与 Group 交互。