def unpack(self, df=False):
"""
Unpack stored data in `_xy` and `_z` into arrays `t`, `xy`, and `z`.
Parameters
----------
df : bool
True to construct DataFrames `self.df` and `self.df_z` (time-consuming).
"""
if df is True:
self.df = pd.DataFrame.from_dict(self._xy,
orient='index',
columns=self.dae.xy_name)
self.t = self.df.index.to_numpy()
self.xy = self.df.to_numpy()
self.df_z = pd.DataFrame.from_dict(self._z,
orient='index',
columns=self.dae.z_name)
self.z = self.df_z.to_numpy()
else:
n_steps = len(self._xy)
self.t = np.array(list(self._xy.keys()))
self.xy = np.zeros((n_steps, self.dae.m + self.dae.n))
self.z = np.zeros((n_steps, self.dae.o))
for idx, xy in enumerate(self._xy.values()):
self.xy[idx, :] = xy
for idx, z in enumerate(self._z.values()):
self.z[idx, :] = z
def get(self,
src: str,
idx,
attr: str = 'v',
allow_none=False,
default=0.0):
"""
Based on the indexer, get the `attr` field of the `src` parameter or variable.
Parameters
----------
src : str
param or var name
idx : array-like
device idx
attr
The attribute of the param or var to retrieve
allow_none : bool
True to allow None values in the indexer
default : float
If `allow_none` is true, the default value to use for None indexer.
Returns
-------
The requested param or variable attribute
"""
self._check_src(src)
self._check_idx(idx)
n = len(idx)
if n == 0:
return np.zeros(0)
ret = [''] * n
_type_set = False
models = self.idx2model(idx, allow_none=allow_none)
for i, idx in enumerate(idx):
if models[i] is not None:
uid = models[i].idx2uid(idx)
instance = models[i].__dict__[src]
val = instance.__dict__[attr][uid]
else:
val = default
# deduce the type for ret
if not _type_set:
if isinstance(val, str):
ret = [''] * n
else:
ret = np.zeros(n)
_type_set = True
ret[i] = val
return ret
def list2array(self, n):
"""
Allocate memory for storage arrays.
"""
super().list2array(n)
if self.mode == 'step':
self._v_mem = np.zeros((n, self.delay + 1))
self.t = np.zeros(self.delay + 1)
else:
self._v_mem = np.zeros((n, 1))
def link_external(self, ext_model):
"""
Update variable addresses provided by external models
This method sets attributes including `parent_model`,
`parent_instance`, `uid`, `a`, `n`, `e_code` and
`v_code`. It initializes the `e` and `v` to zero.
Returns
-------
None
Parameters
----------
ext_model : Model
Instance of the parent model
"""
self.parent = ext_model
if isinstance(ext_model, GroupBase):
if self.indexer.n > 0 and isinstance(self.indexer.v[0],
(list, np.ndarray)):
self._n = [len(i) for i in self.indexer.v
] # number of elements in each sublist
self._idx = np.concatenate(
[np.array(i) for i in self.indexer.v])
else:
self._n = [len(self.indexer.v)]
self._idx = self.indexer.v
self.a = ext_model.get(src=self.src, idx=self._idx,
attr='a').astype(int)
self.n = len(self.a)
self.v = np.zeros(self.n)
self.e = np.zeros(self.n)
else:
original_var = ext_model.__dict__[self.src]
if self.indexer is not None:
uid = ext_model.idx2uid(self.indexer.v)
else:
uid = np.arange(ext_model.n, dtype=int)
self._n = [len(uid)]
if len(uid) > 0:
self.a = original_var.a[uid]
else:
self.a = np.array([], dtype=int)
# set initial v and e values to zero
self.n = len(self.a)
self.v = np.zeros(self.n)
self.e = np.zeros(self.n)
def _set_arrays_alloc(self):
"""
Allocate for internal v and e arrays that cannot
share memory with dae arrays.
"""
if not self.v_inplace:
self.v = np.zeros(self.n)
if not self.e_inplace:
self.e = np.zeros(self.n)
def set_address(self, addr):
"""
Set the address of this variables
Parameters
----------
addr : array-like
The assigned address for this variable
"""
self.a = addr
self.n = len(self.a)
self.v = np.zeros(self.n)
self.e = np.zeros(self.n)
def store_sparse_pattern(self,
models: Optional[Union[str, List,
OrderedDict]] = None):
models = self._get_models(models)
self._call_models_method('store_sparse_pattern', models)
# add variable jacobian values
for j_name in self.dae.jac_name:
ii, jj, vv = list(), list(), list()
# for `gy` matrix, always make sure the diagonal is reserved
# It is a safeguard if the modeling user omitted the diagonal
# term in the equations
if j_name == 'gy':
ii.extend(np.arange(self.dae.m))
jj.extend(np.arange(self.dae.m))
vv.extend(np.zeros(self.dae.m))
# logger.debug(f'Jac <{j_name}>, row={ii}')
for mdl in models.values():
row_idx = mdl.row_of(f'{j_name}')
col_idx = mdl.col_of(f'{j_name}')
# logger.debug(f'Model <{name}>, row={row_idx}')
ii.extend(row_idx)
jj.extend(col_idx)
vv.extend(np.zeros(len(np.array(row_idx))))
# add the constant jacobian values
for row, col, val in mdl.zip_ijv(f'{j_name}c'):
ii.extend(row)
jj.extend(col)
if isinstance(val, (float, int)):
vv.extend(val * np.ones(len(row)))
elif isinstance(val, (list, np.ndarray)):
vv.extend(val)
else:
raise TypeError(
f'Unknown type {type(val)} in constant jacobian {j_name}'
)
if len(ii) > 0:
ii = np.array(ii).astype(int)
jj = np.array(jj).astype(int)
vv = np.array(vv).astype(float)
self.dae.store_sparse_ijv(j_name, ii, jj, vv)
self.dae.build_pattern(j_name)
def resize_array(self):
"""
Resize arrays to the new `m` and `n`
Returns
-------
"""
self.x = np.append(self.x, np.zeros(self.n - len(self.x)))
self.y = np.append(self.y, np.zeros(self.m - len(self.y)))
self.z = np.append(self.z, np.zeros(self.o - len(self.z)))
self.f = np.append(self.f, np.zeros(self.n - len(self.f)))
self.g = np.append(self.g, np.zeros(self.m - len(self.g)))
def __init__(self,
u,
mode='step',
delay=0,
name=None,
tex_name=None,
info=None):
Discrete.__init__(self, name=name, tex_name=tex_name, info=info)
if mode not in ('step', 'time'):
raise ValueError(
f'mode {mode} is invalid. Must be in "step" or "time"')
self.u = u
self.mode = mode
self.delay = delay
self.export_flags = ['v']
self.export_flags_tex = ['v']
self.t = np.array([0])
self.v = np.array([0])
self._v_mem = np.zeros((0, 1))
self.rewind = False
self.has_check_var = True
def setUp(self) -> None:
self.n = 5 # number of input values to delay
self.step = 2 # steps of delay
self.time = 1.0 # delay period in second
self.data = DummyValue(0)
self.data.v = np.zeros(self.n)
self.dstep = Delay(u=self.data, mode='step', delay=self.step)
self.dstep.list2array(self.n)
self.dtime = Delay(u=self.data, mode='time', delay=self.time)
self.dtime.list2array(self.n)
self.avg = Average(u=self.data, mode='step', delay=1)
self.avg.list2array(self.n)
self.v = self.dstep.v
self.vt = self.dtime.v
self.va = self.avg.v
self.n_forward = 5
self.tstep = 0.2
self.dae_t = 0
self.k = 0.2
def init(self):
"""
Initialize the status, storage and values for TDS.
Returns
-------
array-like
The initial values of xy.
"""
t0, _ = elapsed()
system = self.system
if self.initialized:
return system.dae.xy
self._reset()
self._load_pert()
# Note:
# calling `set_address` on `system.exist.pflow_tds` will point all variables
# to the new array after extending `dae.y`
system.set_address(models=system.exist.pflow_tds)
system.set_dae_names(models=system.exist.tds)
system.dae.clear_ts()
system.store_sparse_pattern(models=system.exist.pflow_tds)
system.store_adder_setter(models=system.exist.pflow_tds)
system.vars_to_models()
# Initialize `system.exist.tds` only to avoid Bus overwriting power flow solutions
system.init(system.exist.tds)
system.store_switch_times(system.exist.tds)
# Build mass matrix into `self.Teye`
self.Teye = spdiag(system.dae.Tf.tolist())
self.qg = np.zeros(system.dae.n + system.dae.m)
self.initialized = self.test_init()
# if `dae.n == 1`, `calc_h_first` depends on new `dae.gy`
self.calc_h()
_, s1 = elapsed(t0)
if self.initialized is True:
logger.info(f"Initialization was successful in {s1}.")
else:
logger.error(f"Initialization failed in {s1}.")
if system.dae.n == 0:
tqdm.write('No dynamic component loaded.')
return system.dae.xy
def set_arrays(self, dae):
"""
Set the equation and values arrays.
It slicing into DAE (when contiguous) or allocating new memory (when not contiguous).
Parameters
----------
dae : DAE
Reference to System.dae
"""
slice_idx = slice(self.a[0], self.a[-1] + 1)
if self.v_inplace:
self.v = dae.__dict__[self.v_code][slice_idx]
else:
self.v = np.zeros(self.n)
if self.e_inplace:
self.e = dae.__dict__[self.e_code][slice_idx]
else:
self.e = np.zeros(self.n)
def get(self, src: str, idx, attr: str = 'v'):
"""
Based on the indexer, get the `attr` field of the `src` parameter or variable.
Parameters
----------
src : str
param or var name
idx : array-like
attr
The attribute of the param or var to retrieve
Returns
-------
The requested param or variable attribute
"""
self._check_src(src)
self._check_idx(idx)
n = len(idx)
if n == 0:
return np.zeros(0)
ret = None
models = self.idx2model(idx)
for i, idx in enumerate(idx):
uid = models[i].idx2uid(idx)
instance = models[i].__dict__[src]
val = instance.__dict__[attr][uid]
# deduce the type for ret
if ret is None:
if isinstance(val, str):
ret = [''] * n
else:
ret = np.zeros(n)
ret[i] = val
return ret
def setUp(self) -> None:
self.n = 5
self.data = DummyValue(0)
self.data.v = np.zeros(self.n)
self.der = Derivative(u=self.data)
self.der.list2array(self.n)
self.v = self.der.v
self.n_forward = 10
self.k = 0.2
self.t_step = 0.1
self.dae_t = 0
def link_external(self, ext_model):
"""
Method to be called by ``System`` for getting values from the external model or group.
Parameters
----------
ext_model
An instance of a model or group provided by System
"""
# set initial v values to zero
self.v = np.zeros(self.n)
if self.n == 0:
return
# the same `get` api for Group and Model
self.v = ext_model.get(src=self.src, idx=self.indexer.v, attr='v')
def v(self):
"""
Return the reduced values from the reduction function in an array
Returns
-------
The array ``self._v`` storing the reduced values
"""
if self._v is None:
self._v = np.zeros(len(self.ref.v))
idx = 0
for i, v in enumerate(self.ref.v):
self._v[i] = self.fun(self.u.v[idx:idx + len(v)])
idx += len(v)
return self._v
else:
return self._v
def v(self):
"""
Return the values of the repeated values in a sequantial 1-D array
Returns
-------
The array, ``self._v`` storing the repeated values
"""
if self._v is None:
self._v = np.zeros(len(list_flatten(self.ref.v)))
idx = 0
for i, v in enumerate(self.ref.v):
self._v[idx:idx + len(v)] = self.u.v[i]
idx += len(v)
return self._v
else:
return self._v
def init(self):
"""
Initialize the status, storage and values for TDS.
Returns
-------
array-like
The initial values of xy.
"""
t0, _ = elapsed()
system = self.system
if self.initialized:
return system.dae.xy
self._reset()
self._load_pert()
system.set_address(models=system.exist.tds)
system.set_dae_names(models=system.exist.tds)
system.dae.clear_ts()
system.store_sparse_pattern(models=system.exist.pflow_tds)
system.store_adder_setter(models=system.exist.pflow_tds)
system.vars_to_models()
system.init(system.exist.tds)
system.store_switch_times(system.exist.tds)
self.eye = spdiag([1] * system.dae.n)
self.Teye = spdiag(system.dae.Tf.tolist()) * self.eye
self.qg = np.zeros(system.dae.n + system.dae.m)
self.calc_h()
self.initialized = self.test_init()
_, s1 = elapsed(t0)
if self.initialized is True:
logger.info(f"Initialization was successful in {s1}.")
else:
logger.error(f"Initialization failed in {s1}.")
if system.dae.n == 0:
tqdm.write('No dynamic component loaded.')
return system.dae.xy
def store_sparse_pattern(self, models: Optional[Union[str, List, OrderedDict]] = None):
"""
Collect and store the sparsity pattern of Jacobian matrices.
This is a runtime function specific to cases.
Notes
-----
For `gy` matrix, always make sure the diagonal is reserved.
It is a safeguard if the modeling user omitted the diagonal
term in the equations.
"""
models = self._get_models(models)
self._call_models_method('store_sparse_pattern', models)
# add variable jacobian values
for jname in jac_names:
ii, jj, vv = list(), list(), list()
if jname == 'gy':
ii.extend(np.arange(self.dae.m))
jj.extend(np.arange(self.dae.m))
vv.extend(np.zeros(self.dae.m))
for mdl in models.values():
for row, col, val in mdl.triplets.zip_ijv(jname):
ii.extend(row)
jj.extend(col)
vv.extend(np.zeros_like(row))
for row, col, val in mdl.triplets.zip_ijv(jname + 'c'):
# process constant Jacobians separately
ii.extend(row)
jj.extend(col)
vv.extend(val * np.ones_like(row))
if len(ii) > 0:
ii = np.array(ii, dtype=int)
jj = np.array(jj, dtype=int)
vv = np.array(vv, dtype=float)
self.dae.store_sparse_ijv(jname, ii, jj, vv)
self.dae.build_pattern(jname)
def link_external(self, ext_model):
"""
Update variable addresses provided by external models
This method sets attributes including `parent_model`,
`parent_instance`, `uid`, `a`, `n`, `e_code` and
`v_code`. It initializes the `e` and `v` to zero.
Returns
-------
None
Parameters
----------
ext_model : Model
Instance of the parent model
Warnings
--------
`link_external` does not check if the ExtVar type is the same
as the original variable to reduce performance overhead.
It will be a silent error (a dimension too small error from `dae.build_pattern`)
if a model uses `ExtAlgeb` to access a `State`, or vice versa.
"""
self.parent = ext_model
if isinstance(ext_model, GroupBase):
# determine the number of elements based on `indexer.v`
if self.indexer.n > 0 and isinstance(self.indexer.v[0],
(list, np.ndarray)):
self._n = [len(i) for i in self.indexer.v
] # number of elements in each sublist
self._idx = np.concatenate(
[np.array(i) for i in self.indexer.v])
else:
self._n = [len(self.indexer.v)]
self._idx = self.indexer.v
# use `0` for non-existent addresses (corr. to None in indexer)
self.a = ext_model.get(
src=self.src,
idx=self._idx,
attr='a',
allow_none=self.allow_none,
default=0,
).astype(int)
# check if source var type is the same as this ExtVar
vcodes = np.array(
ext_model.get_field(src=self.src,
idx=self._idx,
field='v_code'))
vcodes = vcodes[vcodes != np.array(None)].astype(str)
if not all(vcodes == np.array(self.v_code)):
raise TypeError(
"ExtVar <%s.%s> is of type <%s>, but source Vars <%s.%s> may not."
% (self.owner.class_name, self.name, self.v_code,
ext_model.class_name, self.src))
self.n = len(self.a)
else:
original_var = ext_model.__dict__[self.src]
if self.allow_none:
raise NotImplementedError(
f"{self.name}: allow_none not implemented for Model")
if original_var.v_code != self.v_code:
raise TypeError("Linking %s of %s to %s of %s is not allowed" %
(self.name, self.class_name, original_var.name,
original_var.class_name))
if self.indexer is not None:
uid = ext_model.idx2uid(self.indexer.v)
else:
uid = np.arange(ext_model.n, dtype=int)
self._n = [len(uid)]
if len(uid) > 0:
self.a = original_var.a[uid]
else:
self.a = np.array([], dtype=int)
# set initial v and e values to zero
self.n = len(self.a)
self.v = np.zeros(self.n)
def clear_xy(self):
"""Reset variable arrays to empty
"""
self.x = np.zeros(self.n)
self.y = np.zeros(self.m)
def list2array(self, n):
super().list2array(n)
self._last_v = np.zeros(n)
def assign_memory(self, n):
"""Assign memory for ``self.v`` and set the array to zero."""
self.v = np.zeros(n, dtype=self.vtype)
def clear_z(self):
"""
Reset status arrays to empty
"""
self.z = np.zeros(self.n)
def init(self):
"""
Initialize the status, storage and values for TDS.
Returns
-------
array-like
The initial values of xy.
"""
t0, _ = elapsed()
system = self.system
if self.initialized:
return system.dae.xy
self.reset()
self._load_pert()
# restore power flow solutions
system.dae.x[:len(system.PFlow.x_sol)] = system.PFlow.x_sol
system.dae.y[:len(system.PFlow.y_sol)] = system.PFlow.y_sol
# Note:
# calling `set_address` on `system.exist.pflow_tds` will point all variables
# to the new array after extending `dae.y`
system.set_address(models=system.exist.pflow_tds)
system.set_dae_names(models=system.exist.tds)
system.dae.clear_ts()
system.store_sparse_pattern(models=system.exist.pflow_tds)
system.store_adder_setter(models=system.exist.pflow_tds)
system.vars_to_models()
system.init(system.exist.tds, routine='tds')
system.store_switch_times(system.exist.tds)
# Build mass matrix into `self.Teye`
self.Teye = spdiag(system.dae.Tf.tolist())
self.qg = np.zeros(system.dae.n + system.dae.m)
self.initialized = self.test_init()
# connect to dime server
if system.config.dime_enabled:
if system.streaming.dimec is None:
system.streaming.connect()
# send out system data using DiME
self.streaming_init()
self.streaming_step()
# if `dae.n == 1`, `calc_h_first` depends on new `dae.gy`
self.calc_h()
_, s1 = elapsed(t0)
if self.initialized is True:
logger.info(f"Initialization for dynamics was successful in {s1}.")
else:
logger.error(f"Initialization for dynamics failed in {s1}.")
if system.dae.n == 0:
tqdm.write('No dynamic component loaded.')
return system.dae.xy
def init(self):
"""
Initialize the status, storage and values for TDS.
Returns
-------
array-like
The initial values of xy.
"""
t0, _ = elapsed()
system = self.system
if self.initialized:
return system.dae.xy
self.reset()
self._load_pert()
# restore power flow solutions
system.dae.x[:len(system.PFlow.x_sol)] = system.PFlow.x_sol
system.dae.y[:len(system.PFlow.y_sol)] = system.PFlow.y_sol
# Note:
# calling `set_address` on `system.exist.pflow_tds` will point all variables
# to the new array after extending `dae.y`.
system.set_address(models=system.exist.pflow_tds)
system.set_dae_names(models=system.exist.tds)
system.dae.clear_ts()
system.store_sparse_pattern(models=system.exist.pflow_tds)
system.store_adder_setter(models=system.exist.pflow_tds)
system.store_no_check_init(models=system.exist.pflow_tds)
system.vars_to_models()
system.init(system.exist.tds, routine='tds')
# only store switch times when not replaying CSV data
if self.data_csv is None:
system.store_switch_times(system.exist.tds)
# Build mass matrix into `self.Teye`
self.Teye = spdiag(system.dae.Tf.tolist())
self.qg = np.zeros(system.dae.n + system.dae.m)
self.initialized = True
# test if residuals are close enough to zero
if self.config.test_init:
self.test_ok = self.test_init()
# discard initialized values and use that from CSV if provided
if self.data_csv is not None:
system.dae.x[:] = self.data_csv[0, 1:system.dae.n + 1]
system.dae.y[:] = self.data_csv[0, system.dae.n + 1:system.dae.n + system.dae.m + 1]
system.vars_to_models()
# connect to data streaming server
if system.streaming.dimec is None:
system.streaming.connect()
if system.config.dime_enabled:
# send out system data using DiME
self.streaming_init()
self.streaming_step()
# if `dae.n == 1`, `calc_h_first` depends on new `dae.gy`
self.calc_h()
# allocate for internal variables
self.x0 = np.zeros_like(system.dae.x)
self.y0 = np.zeros_like(system.dae.y)
self.f0 = np.zeros_like(system.dae.f)
_, s1 = elapsed(t0)
logger.info("Initialization for dynamics completed in %s.", s1)
if self.test_ok is True:
logger.info("Initialization was successful.")
elif self.test_ok is False:
logger.error("Initialization failed!!")
else:
logger.warning("Initialization results were not verified.")
if system.dae.n == 0:
tqdm.write('No differential equation detected.')
return system.dae.xy
def clear_fg(self):
"""Resets equation arrays to empty
"""
self.f = np.zeros(self.n)
self.g = np.zeros(self.m)
