def __init__(self, system, rc=None):
self.system = system
self.config = Tds(rc=rc)
self.solver = Solver(system.config.sparselib)
# internal states
self.F = None
self.initialized = False
self.switch = False
self.next_pc = 0.1
self.step = 0
self.t = self.config.t0
self.h = 0
self.headroom = 0
self.t_jac = 0
self.inc = None
self.callpert = None
self.solved = False
self.fixed_times = []
self.convergence = True
self.niter = 0
self.err = 1
self.x0 = None
self.y0 = None
self.f0 = None
self.success = False
def __init__(self, system, rc=None):
self.system = system
self.solver = Solver(system.config.sparselib)
self.config = Eig(rc=rc)
# internal flags and storages
self.As = None
self.eigs = None
self.mu = None
self.part_fact = None
def __init__(self, system, rc=None):
self.system = system
self.config = Pflow(rc=rc)
self.solver = Solver(system.config.sparselib)
# store status and internal flags
self.solved = False
self.niter = 0
self.iter_mis = []
self.F = None
def __init__(self, system, name):
super(AGCMPC, self).__init__(system, name)
if platform.system() == 'Darwin':
logger.error(
"** AGCMPC optimization does not work correctly on macOS!!!")
self._group = "AGCGroup"
self._name = "AGCMPC"
self.param_remove('Vn')
self.param_remove('Sn')
self._data.update({
'tg': None,
'avr': None,
'vsc': None,
'qw': 15000,
'qu': 10,
})
self._params.extend(['qw', 'qu'])
self._descr.update({
'tg': 'idx for turbine governors',
'vsc': 'idx for VSC dynamic models',
'qw': 'the coeff for minimizing frequency deviation',
'qu': 'the coeff for minimizing input deviation'
})
self._units.update({'tg': 'list', 'vsc': 'list'})
self._mandatory.extend(['tg', 'avr'])
self.calls.update({
'init1': True,
'gcall': True,
'jac0': True,
'fxcall': True
})
self._service.extend([
'xg10', 'pin0', 'delta0', 'omega0', 't', 'dpin0', 'x0', 'xlast'
'xidx', 'uidx', 'yxidx', 'sfx', 'sfu', 'sfy', 'sgx', 'sgu', 'sgy',
'A', 'B', 'Aa', 'Ba', 'obj', 'domega', 'du', 'dx', 'x', 'xpred'
'xa'
])
self._algebs.extend(['dpin'])
self._fnamey.extend(r'\Delta P_{in}')
self.solver = Solver(system.config.sparselib)
self.H = 6
self.uvar = None
self.op = None
self._linearized = False
self._interval = 0 # AGC apply interval in seconds. 0 - continuous
self._init()
class TDS(RoutineBase):
"""
Time domain simulation (TDS) routine
"""
def __init__(self, system, rc=None):
self.system = system
self.config = Tds(rc=rc)
self.solver = Solver(system.config.sparselib)
# internal states
self.F = None
self.initialized = False
self.switch = False
self.next_pc = 0.1
self.step = 0
self.t = self.config.t0
self.h = 0
self.headroom = 0
self.t_jac = 0
self.inc = None
self.callpert = None
self.solved = False
self.fixed_times = []
self.convergence = True
self.niter = 0
self.err = 1
self.x0 = None
self.y0 = None
self.f0 = None
self.success = False
def reset(self):
self.F = None
self.initialized = False
self.switch = False
self.next_pc = 0.1
self.step = 0
self.t = self.config.t0
self.h = 0
self.headroom = 0
self.t_jac = 0
self.inc = None
self.callpert = None
self.solved = False
self.fixed_times = []
self.convergence = True
self.niter = 0
self.err = 1
self.x0 = None
self.y0 = None
self.f0 = None
self.success = False
def _calc_time_step_first(self):
"""
Compute the first time step and save to ``self.h``
Returns
-------
None
"""
system = self.system
config = self.config
if not system.dae.n:
freq = 1.0
elif system.dae.n == 1:
B = matrix(system.dae.Gx)
self.solver.linsolve(system.dae.Gy, B)
As = system.dae.Fx - system.dae.Fy * B
freq = abs(As[0, 0])
else:
freq = 20.0
if freq > system.freq:
freq = float(system.freq)
tspan = abs(config.tf - config.t0)
tcycle = 1 / freq
config.deltatmax = min(5 * tcycle, tspan / 100.0)
config.deltat = min(tcycle, tspan / 100.0)
config.deltatmin = min(tcycle / 64, config.deltatmax / 20)
if config.fixt:
if config.tstep <= 0:
logger.warning('Fixed time step is negative or zero')
logger.warning('Switching to automatic time step')
config.fixt = False
else:
config.deltat = config.tstep
if config.tstep < config.deltatmin:
logger.warning(
'Fixed time step is below the estimated minimum')
self.h = config.deltat
def calc_time_step(self):
"""
Set the time step during time domain simulations
Parameters
----------
convergence: bool
truth value of the convergence of the last step
niter: int
current iteration count
t: float
current simulation time
Returns
-------
float
computed time step size
"""
system = self.system
config = self.config
convergence = self.convergence
niter = self.niter
t = self.t
if t == 0:
self._calc_time_step_first()
return
if convergence:
if niter >= 15:
config.deltat = max(config.deltat * 0.5, config.deltatmin)
elif niter <= 6:
config.deltat = min(config.deltat * 1.1, config.deltatmax)
else:
config.deltat = max(config.deltat * 0.95, config.deltatmin)
# adjust fixed time step if niter is high
if config.fixt:
config.deltat = min(config.tstep, config.deltat)
else:
config.deltat *= 0.9
if config.deltat < config.deltatmin:
config.deltat = 0
if system.Fault.is_time(t) or system.Breaker.is_time(t):
config.deltat = min(config.deltat, 0.002778)
elif system.check_event(t):
config.deltat = min(config.deltat, 0.002778)
if config.method == 'fwdeuler':
config.deltat = min(config.deltat, config.tstep)
# last step size
if self.t + config.deltat > config.tf:
config.deltat = config.tf - self.t
# reduce time step for fixed_times events
for fixed_t in self.fixed_times:
if (fixed_t > self.t) and (fixed_t <= self.t + config.deltat):
config.deltat = fixed_t - self.t
self.switch = True
break
self.h = config.deltat
def init(self):
"""
Initialize time domain simulation
Returns
-------
None
"""
system = self.system
if self.t != self.config.t0:
logger.warning(
'TDS has been previously run and cannot be re-initialized. Please reload the system.'
)
return False
if system.pflow.solved is False:
logger.info('Attempting to solve power flow before TDS.')
if not system.pflow.run():
return False
t, s = elapsed()
# Assign indices for post-powerflow device variables
system.xy_addr1()
# Assign variable names for bus injections and line flows if enabled
system.varname.resize_for_flows()
system.varname.bus_line_names()
# Reshape dae to retain power flow solutions
system.dae.init1()
# Initialize post-powerflow device variables
for device, init1 in zip(system.devman.devices, system.call.init1):
if init1:
system.__dict__[device].init1(system.dae)
t, s = elapsed(t)
if system.dae.n:
logger.debug('Dynamic models initialized in {:s}.'.format(s))
else:
logger.debug('No dynamic model loaded.')
# system.dae flags initialize
system.dae.factorize = True
system.dae.mu = 1.0
system.dae.kg = 0.0
self.initialized = True
return True
def run(self, **kwargs):
"""
Run time domain simulation
Returns
-------
bool
Success flag
"""
# check if initialized
if not self.initialized:
if self.init() is False:
logger.info(
'Call to TDS initialization failed in `tds.run()`.')
ret = False
system = self.system
config = self.config
dae = self.system.dae
# maxit = config.maxit
# tol = config.tol
if system.pflow.solved is False:
logger.warning(
'Power flow not solved. Simulation cannot continue.')
return ret
t0, _ = elapsed()
t1 = t0
self.streaming_init()
logger.info('')
logger.info('-> Time Domain Simulation: {} method, t={} s'.format(
self.config.method, self.config.tf))
self.load_pert()
self.run_step0()
config.qrtstart = time()
while self.t < config.tf:
self.check_fixed_times()
self.calc_time_step()
if self.callpert is not None:
self.callpert(self.t, self.system)
if self.h == 0:
break
# progress time and set time in dae
self.t += self.h
dae.t = self.t
# backup actual variables
self.x0 = matrix(dae.x)
self.y0 = matrix(dae.y)
self.f0 = matrix(dae.f)
# apply fixed_time interventions and perturbations
self.event_actions()
# reset flags used in each step
self.err = 1
self.niter = 0
self.convergence = False
self.implicit_step()
if self.convergence is False:
try:
self.restore_values()
continue
except ValueError:
self.t = config.tf
ret = False
break
self.step += 1
self.compute_flows()
system.varout.store(self.t, self.step)
self.streaming_step()
# plot variables and display iteration status
perc = max(
min((self.t - config.t0) / (config.tf - config.t0) * 100, 100),
0)
# show iteration info every 30 seconds or every 20%
t2, _ = elapsed(t1)
if t2 - t1 >= 30:
t1 = t2
logger.info(
' ({:.0f}%) time = {:.4f}s, step = {}, niter = {}'.format(
100 * self.t / config.tf, self.t, self.step,
self.niter))
if perc > self.next_pc or self.t == config.tf:
self.next_pc += 20
logger.info(
' ({:.0f}%) time = {:.4f}s, step = {}, niter = {}'.format(
100 * self.t / config.tf, self.t, self.step,
self.niter))
# compute max rotor angle difference
# diff_max = anglediff()
# quasi-real-time check and wait
rt_end = config.qrtstart + (self.t - config.t0) * config.kqrt
if config.qrt:
# the ending time has passed
if time() - rt_end > 0:
# simulation is too slow
if time() - rt_end > config.kqrt:
logger.debug(
'Simulation over-run at t={:4.4g} s.'.format(
self.t))
# wait to finish
else:
self.headroom += (rt_end - time())
while time() - rt_end < 0:
sleep(1e-5)
if config.qrt:
logger.debug('RT headroom time: {} s.'.format(str(self.headroom)))
if self.t != config.tf:
logger.error(
'Reached minimum time step. Convergence is not likely.')
ret = False
else:
ret = True
if system.config.dime_enable:
system.streaming.finalize()
_, s = elapsed(t0)
if ret is True:
logger.info(' Time domain simulation finished in {:s}.'.format(s))
else:
logger.info(' Time domain simulation failed in {:s}.'.format(s))
self.success = ret
self.dump_results(success=self.success)
return ret
def restore_values(self):
"""
Restore x, y, and f values if not converged
Returns
-------
None
"""
if self.convergence is True:
return
dae = self.system.dae
system = self.system
inc_g = self.inc[dae.n:dae.m + dae.n]
max_g_err_sign = 1 if abs(max(inc_g)) > abs(min(inc_g)) else -1
if max_g_err_sign == 1:
max_g_err_idx = list(inc_g).index(max(inc_g))
else:
max_g_err_idx = list(inc_g).index(min(inc_g))
logger.debug('Maximum mismatch = {:.4g} at equation <{}>'.format(
max(abs(inc_g)), system.varname.unamey[max_g_err_idx]))
logger.debug('Reducing time step h={:.4g}s for t={:.4g}'.format(
self.h, self.t))
# restore initial variable data
dae.x = matrix(self.x0)
dae.y = matrix(self.y0)
dae.f = matrix(self.f0)
def implicit_step(self):
"""
Integrate one step using trapezoidal method. Sets convergence and niter flags.
Returns
-------
None
"""
config = self.config
system = self.system
dae = self.system.dae
# constant short names
In = spdiag([1] * dae.n)
h = self.h
while self.err > config.tol and self.niter < config.maxit:
if self.t - self.t_jac >= 5:
dae.rebuild = True
self.t_jac = self.t
elif self.niter > 4:
dae.rebuild = True
elif dae.factorize:
dae.rebuild = True
# rebuild Jacobian
if dae.rebuild:
system.call.int()
dae.rebuild = False
else:
system.call.int_fg()
# complete Jacobian matrix dae.Ac
if config.method == 'euler':
dae.Ac = sparse(
[[In - h * dae.Fx, dae.Gx], [-h * dae.Fy, dae.Gy]], 'd')
dae.q = dae.x - self.x0 - h * dae.f
elif config.method == 'trapezoidal':
dae.Ac = sparse([[In - h * 0.5 * dae.Fx, dae.Gx],
[-h * 0.5 * dae.Fy, dae.Gy]], 'd')
dae.q = dae.x - self.x0 - h * 0.5 * (dae.f + self.f0)
# windup limiters
dae.reset_Ac()
if dae.factorize:
try:
self.F = self.solver.symbolic(dae.Ac)
dae.factorize = False
except NotImplementedError:
pass
self.inc = -matrix([dae.q, dae.g])
try:
N = self.solver.numeric(dae.Ac, self.F)
self.solver.solve(dae.Ac, self.F, N, self.inc)
except ArithmeticError:
logger.error('Singular matrix')
dae.check_diag(dae.Gy, 'unamey')
dae.check_diag(dae.Fx, 'unamex')
# force quit
self.niter = config.maxit + 1
break
except ValueError:
logger.warning('Unexpected symbolic factorization')
dae.factorize = True
continue
except NotImplementedError:
self.inc = self.solver.linsolve(dae.Ac, self.inc)
inc_x = self.inc[:dae.n]
inc_y = self.inc[dae.n:dae.m + dae.n]
dae.x += inc_x
dae.y += inc_y
self.err = max(abs(self.inc))
if np.isnan(self.inc).any():
logger.error('Iteration error: NaN detected.')
self.niter = config.maxit + 1
break
self.niter += 1
if self.niter <= config.maxit:
self.convergence = True
def event_actions(self):
"""
Take actions for timed events
Returns
-------
None
"""
system = self.system
dae = system.dae
if self.switch:
system.Breaker.apply(self.t)
for item in system.check_event(self.t):
system.__dict__[item].apply(self.t)
dae.rebuild = True
self.switch = False
def check_fixed_times(self):
"""
Check for fixed times and store in ``self.fixed_times``.
Returns
-------
None
"""
self.fixed_times = self.system.get_event_times()
def load_pert(self):
"""
Load perturbation files to ``self.callpert``
Returns
-------
None
"""
system = self.system
if system.files.pert:
try:
sys.path.append(system.files.case_path)
module = importlib.import_module(system.files.pert[:-3])
self.callpert = getattr(module, 'pert')
except ImportError:
logger.warning('Pert file is discarded due to import errors.')
self.callpert = None
def run_step0(self):
"""
For the 0th step, store the data and stream data
Returns
-------
None
"""
dae = self.system.dae
system = self.system
config = self.config
# compute line and area flow
if config.compute_flows:
dae.init_fg()
self.compute_flows() # TODO: move to PowerSystem
self.inc = zeros(dae.m + dae.n, 1)
system.varout.store(self.t, self.step)
self.streaming_step()
def streaming_step(self):
"""
Sync, handle and streaming for each integration step
Returns
-------
None
"""
system = self.system
if system.config.dime_enable:
system.streaming.sync_and_handle()
system.streaming.vars_to_modules()
system.streaming.vars_to_pmu()
def streaming_init(self):
"""
Send out initialization variables and process init from modules
Returns
-------
None
"""
system = self.system
config = self.config
if system.config.dime_enable:
config.compute_flows = True
system.streaming.send_init(recepient='all')
logger.info('Waiting for modules to send init info...')
sleep(0.5)
system.streaming.sync_and_handle()
def angle_diff(self):
"""
Compute the maximum angle difference between generators
Returns
-------
float
maximum angular difference
"""
return 0
def compute_flows(self):
"""
If enabled, compute the line flows after each step
Returns
-------
None
"""
system = self.system
config = self.config
dae = system.dae
if config.compute_flows:
# compute and append series injections on buses
system.call.bus_injection()
bus_inj = dae.g[:2 * system.Bus.n]
system.call.seriesflow()
system.Area.seriesflow(system.dae)
system.Area.interchange_varout()
dae.y = matrix([
dae.y[:system.dae.m], bus_inj, system.Line._line_flows,
system.Area.inter_varout
])
def dump_results(self, success):
"""
Dump simulation results to ``dat`` and ``lst`` files
Returns
-------
None
"""
system = self.system
t, _ = elapsed()
if success and (not system.files.no_output):
# system.varout.dump()
system.varout.dump_np_vars()
_, s = elapsed(t)
logger.info('Simulation data dumped in {:s}.'.format(s))
class AGCMPC(ModelBase):
"""MPC based AGC using TG and VSC"""
def __init__(self, system, name):
super(AGCMPC, self).__init__(system, name)
if platform.system() == 'Darwin':
logger.error(
"** AGCMPC optimization does not work correctly on macOS!!!")
self._group = "AGCGroup"
self._name = "AGCMPC"
self.param_remove('Vn')
self.param_remove('Sn')
self._data.update({
'tg': None,
'avr': None,
'vsc': None,
'qw': 15000,
'qu': 10,
})
self._params.extend(['qw', 'qu'])
self._descr.update({
'tg': 'idx for turbine governors',
'vsc': 'idx for VSC dynamic models',
'qw': 'the coeff for minimizing frequency deviation',
'qu': 'the coeff for minimizing input deviation'
})
self._units.update({'tg': 'list', 'vsc': 'list'})
self._mandatory.extend(['tg', 'avr'])
self.calls.update({
'init1': True,
'gcall': True,
'jac0': True,
'fxcall': True
})
self._service.extend([
'xg10', 'pin0', 'delta0', 'omega0', 't', 'dpin0', 'x0', 'xlast'
'xidx', 'uidx', 'yxidx', 'sfx', 'sfu', 'sfy', 'sgx', 'sgu', 'sgy',
'A', 'B', 'Aa', 'Ba', 'obj', 'domega', 'du', 'dx', 'x', 'xpred'
'xa'
])
self._algebs.extend(['dpin'])
self._fnamey.extend(r'\Delta P_{in}')
self.solver = Solver(system.config.sparselib)
self.H = 6
self.uvar = None
self.op = None
self._linearized = False
self._interval = 0 # AGC apply interval in seconds. 0 - continuous
self._init()
def init1(self, dae):
if globals()['cp'] is None:
try:
globals()['cp'] = importlib.import_module('cvxpy')
except ImportError:
logger.error(
'CVXPY import error. Install optional package `cvxpy` to use AGCMPC'
)
sys.exit(1)
self.t = -1
self.tlast = -1
# state array x = [delta, omega, xg1]
# input array u = [dpin]
self.copy_data_ext('Governor', field='gen', dest='syn', idx=self.tg)
self.copy_data_ext('Synchronous',
field='delta',
dest='delta',
idx=self.syn)
self.copy_data_ext('Synchronous',
field='omega',
dest='omega',
idx=self.syn)
self.copy_data_ext('Synchronous',
field='e1d',
dest='e1d',
idx=self.syn)
self.copy_data_ext('Synchronous',
field='e1q',
dest='e1q',
idx=self.syn)
self.copy_data_ext('Synchronous',
field='e2d',
dest='e2d',
idx=self.syn)
self.copy_data_ext('Synchronous',
field='e2q',
dest='e2q',
idx=self.syn)
self.copy_data_ext('Governor', field='xg1', dest='xg1', idx=self.tg)
self.copy_data_ext('Governor', field='xg2', dest='xg2', idx=self.tg)
self.copy_data_ext('Governor', field='xg3', dest='xg3', idx=self.tg)
self.copy_data_ext('Governor', field='pin', dest='pin', idx=self.tg)
self.copy_data_ext('AVR', field='vm', dest='vm', idx=self.avr)
self.copy_data_ext('AVR', field='vr1', dest='vr1', idx=self.avr)
self.copy_data_ext('AVR', field='vr2', dest='vr2', idx=self.avr)
self.copy_data_ext('AVR', field='vfout', dest='vfout', idx=self.avr)
dae.y[self.dpin] = 0
self.dpin0 = zeros(self.n, 1)
# build state/ input /other algebraic idx array
self.xidx = matrix([
self.delta, self.omega, self.e1d, self.e1q, self.e2d, self.e2q,
self.xg1, self.xg2, self.xg3, self.vm, self.vr1, self.vr2,
self.vfout
])
self.x0 = dae.x[self.xidx]
self.x = zeros(len(self.xidx), 1)
self.dx = zeros(len(self.xidx), 1)
self.xlast = dae.x[self.xidx]
self.uidx = matrix([self.dpin])
self.ulast = zeros(self.n, 1)
self.dpin_calc = zeros(self.n, 1)
self.widx = self.system.PQ.a
self.w0 = self.system.PQ.p0
self.wlast = matrix(self.w0)
self.yidx = self.omega
self.yidx_in_x = [index(self.xidx, y)[0] for y in self.yidx]
yidx = np.delete(np.arange(dae.m), np.array(self.uidx))
self.yxidx = matrix(yidx)
# optimization problem
self.uvar = cp.Variable((len(self.uidx), self.H + 1), 'u')
self.uzero = cp.Parameter((len(self.uidx), ), 'u0')
self.xazero = cp.Parameter((2 * len(self.xidx), 1), 'xa')
self.prob = None
self.t_store = []
self.xpred_store = []
def gcall(self, dae):
if self.t == -1:
self.t = dae.t
return
if not self._linearized:
# update the linearization points
self._linearized = True
self.t = dae.t
self.tlast = dae.t
self.sfx = dae.Fx[self.xidx, self.xidx]
self.sfu = dae.Fy[self.xidx, self.uidx]
self.sfy = dae.Fy[self.xidx, self.yxidx]
self.sgx = dae.Gx[self.yxidx, self.xidx]
self.sgu = dae.Gy[self.yxidx, self.uidx]
self.sgw = spmatrix(1, self.widx, list(range(len(self.widx))),
(len(self.yxidx), len(self.widx)))
self.sgy = dae.Gy[self.yxidx, self.yxidx]
# create state matrices
self.gyigx = matrix(self.sgx)
self.gyigu = matrix(self.sgu)
self.gyigw = matrix(self.sgw)
self.solver.linsolve(self.sgy, self.gyigx)
self.solver.linsolve(self.sgy, self.gyigu)
self.solver.linsolve(self.sgy, self.gyigw)
self.A = (self.sfx - self.sfy * self.gyigx)
self.B = (self.sfu - self.sfy * self.gyigu)
self.C = -(self.sfy * self.gyigw)
self.A = self.system.tds.h * self.A
self.Aa = sparse([[self.A, self.A],
[
spmatrix([], [], [],
(self.A.size[0], self.A.size[1])),
spdiag([1] * len(self.xidx))
]])
self.Ba = sparse([self.B, self.B])
self.Ca = sparse([self.C, self.C])
# formulate optimization problem
nx = len(self.xidx)
nu = len(self.uidx)
obj_x = 0
xa_0 = self.xazero
for i in range(self.H):
# calculate Xa for each step in horizon H
# du = cp.reshape(self.uvar[:, i+1], (nu, 1)) - self.uvar[:,i]
du = cp.reshape(self.uvar[:, i + 1] - self.uvar[:, i], (nu, 1))
xa_i = matrix(self.Aa) * xa_0 + matrix(self.Ba) * du
obj_x += cp.multiply(
self.qw,
cp.square(xa_i[nx:][self.yidx_in_x] -
self.x0[self.yidx_in_x]))
xa_0 = xa_i
# construct the optimization problem
self.obj_x = cp.sum(obj_x)
self.obj_u = 0
self.obj_u += cp.sum(
cp.multiply(
np.array(self.qu).reshape((nu, )),
cp.sum(cp.square(self.uvar[:, 1:] - self.uvar[:, :-1]),
axis=1)))
constraints = [
self.uvar[:, 0] == self.uzero,
self.uvar[:, 1:] - self.uvar[:, :-1] <= 0.5,
self.uvar[:, 1:] - self.uvar[:, :-1] >= -0.5
]
self.prob = cp.Problem(cp.Minimize(self.obj_x + self.obj_u),
constraints)
if dae.t != self.t:
self.t = dae.t
nx = len(self.xidx)
nu = len(self.uidx)
# # update Delta x and x for current step
self.x = dae.x[self.xidx]
self.dx = self.x - self.xlast
self.xa = matrix([self.dx, self.x])
# assign values to self.uzero and self.xazero
self.uzero.value = np.array(self.ulast).reshape((-1, ))
self.xazero.value = np.array(self.xa).reshape((-1, 1))
# use warm_start when possible
if dae.t == 0:
self.prob.solve()
else:
self.prob.solve(warm_start=1)
self.dpin_calc = matrix(self.uvar.value[:, 1])
# update every interval
if (self.t - self.tlast) >= self._interval:
self.tlast = self.t
self.dpin0 = self.dpin_calc
opt_val = self.prob.solution.opt_val
logger.debug("t={:.4f}, obj={:.6f}, u[0]={:.6f}".format(
dae.t, opt_val, self.uvar.value[0, 0]))
self.t_store.append(self.t)
xa_post = matrix(self.Aa) * self.xa + matrix(
self.Ba) * (matrix(self.uvar.value[:, 0]) - self.ulast)
self.xpred_store.append(xa_post[nx:][self.yidx_in_x][0])
# # post-optimization evaluator
# # u_val = matrix([[0, 0], [0, 0], [0, 0]])
# u_val = matrix(self.uvar.value)
# u_val = zeros(2, self.H)
# obj_x = 0
# xa_0 = self.xa
# u_0 = self.ulast
# for i in range(self.H):
# # calculate Xa for each step in horizon H
# du = np.reshape(u_val[:, i], (-1, 1)) - u_0
# xa_i = matrix(self.Aa) * xa_0 + matrix(self.Ba) * matrix(du) #+ matrix(self.Ca) * self.dw
# obj_x += mul(self.qw, (xa_i[nx:][self.yidx_in_x] - self.x0[self.yidx_in_x]) ** 2)
# xa_0 = xa_i
# u_0 = np.reshape(u_val[:, i], (-1, 1))
# self.obj_x = sum(obj_x)
# u2 = np.array(mul(u_val, u_val))
# self.obj_u = sum(mul(self.qu, matrix(np.sum(u2, 1))))
#
# eval_obj = self.obj_x + self.obj_u
# print("Post eval, t={:.4f} obj = {:.6f}, u = {:.6f}, {:.6f}".format(self.t, eval_obj, u_val[0, 0],
# u_val[1, 0]))
# print(" obj_x = {}, obj_u = {}".format(self.obj_x, self.obj_u))
# record data for the current step
self.ulast = self.dpin_calc
self.xlast = dae.x[self.xidx]
dae.g[self.dpin] = dae.y[self.dpin] - self.dpin0
dae.g[self.pin] += dae.y[
self.dpin] # positive `dpin` increases the `pin` reference
def jac0(self, dae):
dae.add_jac(Gy0, 1, self.dpin, self.dpin)
dae.add_jac(Gy0, 1, self.pin, self.dpin)
class EIG(RoutineBase):
"""
Eigenvalue analysis routine
"""
def __init__(self, system, rc=None):
self.system = system
self.solver = Solver(system.config.sparselib)
self.config = Eig(rc=rc)
# internal flags and storages
self.As = None
self.eigs = None
self.mu = None
self.part_fact = None
def calc_state_matrix(self):
"""
Return state matrix and store to ``self.As``
Returns
-------
matrix
state matrix
"""
system = self.system
Gyx = matrix(system.dae.Gx)
self.solver.linsolve(system.dae.Gy, Gyx)
self.As = matrix(system.dae.Fx - system.dae.Fy * Gyx)
# ------------------------------------------------------
# TODO: use scipy eigs
# self.As = sparse(self.As)
# I = np.array(self.As.I).reshape((-1,))
# J = np.array(self.As.J).reshape((-1,))
# V = np.array(self.As.V).reshape((-1,))
# self.As = csr_matrix((V, (I, J)), shape=self.As.size)
# ------------------------------------------------------
return self.As
def calc_eigvals(self):
"""
Solve eigenvalues of the state matrix ``self.As``
Returns
-------
None
"""
self.eigs = numpy.linalg.eigvals(self.As)
# TODO: use scipy.sparse.linalg.eigs(self.As)
return self.eigs
def calc_part_factor(self):
"""
Compute participation factor of states in eigenvalues
Returns
-------
"""
mu, N = numpy.linalg.eig(self.As)
# TODO: use scipy.sparse.linalg.eigs(self.As)
N = matrix(N)
n = len(mu)
idx = range(n)
W = matrix(spmatrix(1.0, idx, idx, (n, n), N.typecode))
gesv(N, W)
partfact = mul(abs(W.T), abs(N))
b = matrix(1.0, (1, n))
WN = b * partfact
partfact = partfact.T
for item in idx:
mu_real = mu[item].real
mu_imag = mu[item].imag
mu[item] = complex(round(mu_real, 4), round(mu_imag, 4))
partfact[item, :] /= WN[item]
# participation factor
self.mu = matrix(mu)
self.part_fact = matrix(partfact)
return self.mu, self.part_fact
def run(self):
ret = False
system = self.system
if system.pflow.solved is False:
logger.warning(
'Power flow not solved. Eig analysis will not continue.')
return ret
elif system.dae.n == 0:
logger.warning('No dynamic model. Eig analysis will not continue.')
return ret
t1, s = elapsed()
logger.info('-> Eigenvalue Analysis:')
system.dae.factorize = True
exec(system.call.int)
self.calc_state_matrix()
self.calc_part_factor()
self.dump_results()
self.plot_results()
ret = True
t2, s = elapsed(t1)
logger.info('Eigenvalue analysis finished in {:s}.'.format(s))
return ret
def plot_results(self):
try:
plt = importlib.import_module('matplotlib.pyplot')
except ImportError:
plt = None
if plt is None:
logger.warning('Install matplotlib to plot eigenvalue map.')
return
mu_real = self.mu.real()
mu_imag = self.mu.imag()
p_mu_real, p_mu_imag = list(), list()
z_mu_real, z_mu_imag = list(), list()
n_mu_real, n_mu_imag = list(), list()
for re, im in zip(mu_real, mu_imag):
if re == 0:
z_mu_real.append(re)
z_mu_imag.append(im)
elif re > 0:
p_mu_real.append(re)
p_mu_imag.append(im)
elif re < 0:
n_mu_real.append(re)
n_mu_imag.append(im)
if len(p_mu_real) > 0:
logger.warning(
'System is not stable due to {} positive eigenvalues.'.format(
len(p_mu_real)))
else:
logger.info(
'System is small-signal stable in the initial neighbourhood.')
if self.config.plot and len(p_mu_real) > 0:
fig, ax = plt.subplots()
ax.scatter(n_mu_real, n_mu_imag, marker='x', s=26, color='green')
ax.scatter(z_mu_real, z_mu_imag, marker='o', s=26, color='orange')
ax.scatter(p_mu_real, p_mu_imag, marker='x', s=26, color='red')
plt.show()
def dump_results(self):
"""
Save eigenvalue analysis reports
Returns
-------
None
"""
system = self.system
mu = self.mu
partfact = self.part_fact
if system.files.no_output:
return
text = []
header = []
rowname = []
data = []
neig = len(mu)
mu_real = mu.real()
mu_imag = mu.imag()
npositive = sum(1 for x in mu_real if x > 0)
nzero = sum(1 for x in mu_real if x == 0)
nnegative = sum(1 for x in mu_real if x < 0)
numeral = []
for idx, item in enumerate(range(neig)):
if mu_real[idx] == 0:
marker = '*'
elif mu_real[idx] > 0:
marker = '**'
else:
marker = ''
numeral.append('#' + str(idx + 1) + marker)
# compute frequency, undamped frequency and damping
freq = [0] * neig
ufreq = [0] * neig
damping = [0] * neig
for idx, item in enumerate(mu):
if item.imag == 0:
freq[idx] = 0
ufreq[idx] = 0
damping[idx] = 0
else:
freq[idx] = abs(item) / 2 / pi
ufreq[idx] = abs(item.imag / 2 / pi)
damping[idx] = -div(item.real, abs(item)) * 100
# obtain most associated variables
var_assoc = []
for prow in range(neig):
temp_row = partfact[prow, :]
name_idx = list(temp_row).index(max(temp_row))
var_assoc.append(system.varname.unamex[name_idx])
pf = []
for prow in range(neig):
temp_row = []
for pcol in range(neig):
temp_row.append(round(partfact[prow, pcol], 5))
pf.append(temp_row)
text.append(system.report.info)
header.append([''])
rowname.append(['EIGENVALUE ANALYSIS REPORT'])
data.append('')
text.append('STATISTICS\n')
header.append([''])
rowname.append(['Positives', 'Zeros', 'Negatives'])
data.append([npositive, nzero, nnegative])
text.append('EIGENVALUE DATA\n')
header.append([
'Most Associated', 'Real', 'Imag', 'Damped Freq.', 'Frequency',
'Damping [%]'
])
rowname.append(numeral)
data.append(
[var_assoc,
list(mu_real),
list(mu_imag), ufreq, freq, damping])
cpb = 7 # columns per block
nblock = int(ceil(neig / cpb))
if nblock <= 100:
for idx in range(nblock):
start = cpb * idx
end = cpb * (idx + 1)
text.append('PARTICIPATION FACTORS [{}/{}]\n'.format(
idx + 1, nblock))
header.append(numeral[start:end])
rowname.append(system.varname.unamex)
data.append(pf[start:end])
dump_data(text, header, rowname, data, system.files.eig)
logger.info('report saved.')
class PFLOW(RoutineBase):
"""
Power flow calculation routine
"""
def __init__(self, system, rc=None):
self.system = system
self.config = Pflow(rc=rc)
self.solver = Solver(system.config.sparselib)
# store status and internal flags
self.solved = False
self.niter = 0
self.iter_mis = []
self.F = None
def reset(self):
"""
Reset all internal storage to initial status
Returns
-------
None
"""
self.solved = False
self.niter = 0
self.iter_mis = []
self.F = None
self.system.dae.factorize = True
def pre(self):
"""
Initialize system for power flow study
Returns
-------
None
"""
if self.solved and self.system.tds.initialized:
logger.error(
'TDS has been initialized. Cannot solve power flow again.')
return False
logger.info('-> Power flow study: {} method, {} start'.format(
self.config.method.upper(),
'flat' if self.config.flatstart else 'non-flat'))
t, s = elapsed()
system = self.system
dae = self.system.dae
system.dae.init_xy()
for device, pflow, init0 in zip(system.devman.devices,
system.call.pflow, system.call.init0):
if pflow and init0:
system.__dict__[device].init0(dae)
# check for islands
system.check_islands(show_info=True)
# reset internal storage
self.reset()
t, s = elapsed(t)
logger.debug('Power flow initialized in {:s}.'.format(s))
return True
def run(self, **kwargs):
"""
call the power flow solution routine
Returns
-------
bool
True for success, False for fail
"""
ret = None
# initialization Y matrix and inital guess
if not self.pre():
return False
t, _ = elapsed()
# call solution methods
if self.config.method == 'NR':
ret = self.newton()
elif self.config.method == 'DCPF':
ret = self.dcpf()
elif self.config.method in ('FDPF', 'FDBX', 'FDXB'):
ret = self.fdpf()
self.post()
_, s = elapsed(t)
if self.solved:
logger.info(' Solution converged in {} in {} iterations'.format(
s, self.niter))
else:
logger.warning(' Solution failed in {} in {} iterations'.format(
s, self.niter))
return ret
def newton(self):
"""
Newton power flow routine
Returns
-------
bool
success flag
"""
dae = self.system.dae
while True:
inc = self.calc_inc()
dae.x += inc[:dae.n]
dae.y += inc[dae.n:dae.n + dae.m]
self.niter += 1
max_mis = max(abs(inc))
self.iter_mis.append(max_mis)
self._iter_info(self.niter)
if max_mis < self.config.tol:
self.solved = True
break
elif self.niter > 5 and max_mis > 1000 * self.iter_mis[0]:
logger.warning('Blown up in {0} iterations.'.format(
self.niter))
break
if self.niter > self.config.maxit:
logger.warning('Reached maximum number of iterations.')
break
return self.solved
def dcpf(self):
"""
Calculate linearized power flow
Returns
-------
bool
success flag, number of iterations
"""
dae = self.system.dae
self.system.Bus.init0(dae)
self.system.dae.init_g()
Va0 = self.system.Bus.angle
for model, pflow, gcall in zip(self.system.devman.devices,
self.system.call.pflow,
self.system.call.gcall):
if pflow and gcall:
self.system.__dict__[model].gcall(dae)
sw = self.system.SW.a
sw.sort(reverse=True)
no_sw = self.system.Bus.a[:]
no_swv = self.system.Bus.v[:]
for item in sw:
no_sw.pop(item)
no_swv.pop(item)
Bp = self.system.Line.Bp[no_sw, no_sw]
p = matrix(self.system.dae.g[no_sw], (no_sw.__len__(), 1))
p = p - self.system.Line.Bp[no_sw, sw] * Va0[sw]
Sp = self.solver.symbolic(Bp)
N = self.solver.numeric(Bp, Sp)
self.solver.solve(Bp, Sp, N, p)
self.system.dae.y[no_sw] = p
self.solved = True
self.niter = 1
return self.solved
def _iter_info(self, niter, level=logging.INFO):
"""
Log iteration number and mismatch
Parameters
----------
level
logging level
Returns
-------
None
"""
max_mis = self.iter_mis[niter - 1]
msg = ' Iter {:<d}. max mismatch = {:8.7f}'.format(niter, max_mis)
logger.info(msg)
def calc_inc(self):
"""
Calculate the Newton incrementals for each step
Returns
-------
matrix
The solution to ``x = -A\\b``
"""
system = self.system
self.newton_call()
A = sparse([[system.dae.Fx, system.dae.Gx],
[system.dae.Fy, system.dae.Gy]])
inc = matrix([system.dae.f, system.dae.g])
if system.dae.factorize:
try:
self.F = self.solver.symbolic(A)
system.dae.factorize = False
except NotImplementedError:
pass
try:
N = self.solver.numeric(A, self.F)
self.solver.solve(A, self.F, N, inc)
except ValueError:
logger.warning('Unexpected symbolic factorization.')
system.dae.factorize = True
except ArithmeticError:
logger.warning('Jacobian matrix is singular.')
system.dae.check_diag(system.dae.Gy, 'unamey')
except NotImplementedError:
inc = self.solver.linsolve(A, inc)
return -inc
def newton_call(self):
"""
Function calls for Newton power flow
Returns
-------
None
"""
# system = self.system
# exec(system.call.newton)
system = self.system
dae = self.system.dae
system.dae.init_fg()
system.dae.reset_small_g()
# evaluate algebraic equation mismatches
for model, pflow, gcall in zip(system.devman.devices,
system.call.pflow, system.call.gcall):
if pflow and gcall:
system.__dict__[model].gcall(dae)
# eval differential equations
for model, pflow, fcall in zip(system.devman.devices,
system.call.pflow, system.call.fcall):
if pflow and fcall:
system.__dict__[model].fcall(dae)
# reset islanded buses mismatches
system.Bus.gisland(dae)
if system.dae.factorize:
system.dae.init_jac0()
# evaluate constant Jacobian elements
for model, pflow, jac0 in zip(system.devman.devices,
system.call.pflow, system.call.jac0):
if pflow and jac0:
system.__dict__[model].jac0(dae)
dae.temp_to_spmatrix('jac0')
dae.setup_FxGy()
# evaluate Gy
for model, pflow, gycall in zip(system.devman.devices,
system.call.pflow, system.call.gycall):
if pflow and gycall:
system.__dict__[model].gycall(dae)
# evaluate Fx
for model, pflow, fxcall in zip(system.devman.devices,
system.call.pflow, system.call.fxcall):
if pflow and fxcall:
system.__dict__[model].fxcall(dae)
# reset islanded buses Jacobians
system.Bus.gyisland(dae)
dae.temp_to_spmatrix('jac')
def post(self):
"""
Post processing for solved systems.
Store load, generation data on buses.
Store reactive power generation on PVs and slack generators.
Calculate series flows and area flows.
Returns
-------
None
"""
if not self.solved:
return
system = self.system
exec(system.call.pfload)
system.Bus.Pl = system.dae.g[system.Bus.a]
system.Bus.Ql = system.dae.g[system.Bus.v]
exec(system.call.pfgen)
system.Bus.Pg = system.dae.g[system.Bus.a]
system.Bus.Qg = system.dae.g[system.Bus.v]
if system.PV.n:
system.PV.qg = system.dae.y[system.PV.q]
if system.SW.n:
system.SW.pg = system.dae.y[system.SW.p]
system.SW.qg = system.dae.y[system.SW.q]
exec(system.call.seriesflow)
system.Area.seriesflow(system.dae)
def fdpf(self):
"""
Fast Decoupled Power Flow
Returns
-------
bool
Success flag
"""
system = self.system
# general settings
self.niter = 1
iter_max = self.config.maxit
self.solved = True
tol = self.config.tol
error = tol + 1
self.iter_mis = []
if (not system.Line.Bp) or (not system.Line.Bpp):
system.Line.build_b()
# initialize indexing and Jacobian
# ngen = system.SW.n + system.PV.n
sw = system.SW.a
sw.sort(reverse=True)
no_sw = system.Bus.a[:]
no_swv = system.Bus.v[:]
for item in sw:
no_sw.pop(item)
no_swv.pop(item)
gen = system.SW.a + system.PV.a
gen.sort(reverse=True)
no_g = system.Bus.a[:]
no_gv = system.Bus.v[:]
for item in gen:
no_g.pop(item)
no_gv.pop(item)
Bp = system.Line.Bp[no_sw, no_sw]
Bpp = system.Line.Bpp[no_g, no_g]
# Fp = self.solver.symbolic(Bp)
# Fpp = self.solver.symbolic(Bpp)
# Np = self.solver.numeric(Bp, Fp)
# Npp = self.solver.numeric(Bpp, Fpp)
exec(system.call.fdpf)
# main loop
while error > tol:
# P-theta
da = matrix(div(system.dae.g[no_sw], system.dae.y[no_swv]))
# self.solver.solve(Bp, Fp, Np, da)
da = self.solver.linsolve(Bp, da)
system.dae.y[no_sw] += da
exec(system.call.fdpf)
normP = max(abs(system.dae.g[no_sw]))
# Q-V
dV = matrix(div(system.dae.g[no_gv], system.dae.y[no_gv]))
# self.solver.solve(Bpp, Fpp, Npp, dV)
dV = self.solver.linsolve(Bpp, dV)
system.dae.y[no_gv] += dV
exec(system.call.fdpf)
normQ = max(abs(system.dae.g[no_gv]))
err = max([normP, normQ])
self.iter_mis.append(err)
error = err
self._iter_info(self.niter)
self.niter += 1
if self.niter > 4 and self.iter_mis[-1] > 1000 * self.iter_mis[0]:
logger.warning('Blown up in {0} iterations.'.format(
self.niter))
self.solved = False
break
if self.niter > iter_max:
logger.warning('Reached maximum number of iterations.')
self.solved = False
break
return self.solved
