def test_update_thermal_demand_forecast(ctrl, obs, ua_mocks, monkeypatch):
ctrl_proxy = ua_mocks[0].ctrl_agent
# Note: Controller for testing is initialized with current system time.
# *start* has thus to be set in the future of utcnow()
start = arrow.utcnow().replace(days=1,
hour=0,
minute=0,
second=0,
microsecond=0)
res = 900
data = np.arange(96 * 2)
fc = util.TimeSeries(start, res, data)
target_schedule = util.TimeSeries(start, res, [0 for i in range(96)])
weights = util.TimeSeries(start, res, [0 for i in range(96)])
# Patch ControllerAgent
@coroutine
def run_negotiation(neg_start):
return start
monkeypatch.setattr(ctrl, '_run_negotiation', run_negotiation)
# Call method and assert that a new negotiation has been started
yield from ctrl_proxy.update_thermal_demand_forecast(
target_schedule, weights, fc)
ret = yield from ctrl._task_negotiation
assert ret == start
# Assert that all UnitAgents got the (same) data
futs = [ua.update_forecast_called for ua in ua_mocks]
ret = yield from gather(*futs)
for i in ret:
assert i == fc
def get_day_ahead_plan(self, dap_start, dap_end):
"""Return the day ahead plan (which is the result of a negotiation).
"""
fc = self._forecast_p
if dap_start < fc.start:
raise ValueError('dap_start must be >= the forecast start')
if dap_end > fc.end:
raise ValueError('dap_end must be <= the forecast end')
# TODO FIXME: is this still correct?
# FIXME cont: Maybe _forecast_p should be handled different?
#
# Update target_schedule and weights array.
# Because we communicate our planned generation to an external party,
# we have to set it as new target schedule for our VPP:
self._target_schedule = util.TimeSeries(fc.start, fc.res,
np.zeros(len(fc)))
self._target_schedule[:dap_end] = fc[:dap_end]
self._weights = util.TimeSeries(fc.start, fc.res, np.zeros(len(fc)))
self._weights[:dap_end] = 1
# Extract DAP
dap = fc[dap_start:dap_end]
return dap
def test_user_get_thermal_demand_forecast(user):
start = arrow.get(demand_meta['start_time'])
fc = user._get_thermal_demand_forecast(start, 2)
assert fc == util.TimeSeries(start, demand_meta['interval_minutes'] * 60,
[0, 25])
start = start.replace(hours=1)
fc = user._get_thermal_demand_forecast(start, 2)
assert fc == util.TimeSeries(start, demand_meta['interval_minutes'] * 60,
[25, 60])
def _get_target(self, date):
"""Sets and returns self._ts and self._weights, both TimeSeries,
depending on the given date. If no target has been given for date,
zero target with zero weights is given.
Raises a RunTimeError if more than one target is passed for one day,
as this has to be handled by the intraday routines."""
# get appropriate target
start_keys = []
for start_key in self._target_gens:
date_diff = start_key - date
if date_diff.days >= 0 and date_diff.days < 1:
start_keys.append(start_key)
if len(start_keys) > 1:
raise RuntimeError('More than 1 target given for day %s' % date)
# TODO FIXME: We have hardcoded stuff here for dap far too much
if len(start_keys) == 0:
logger.info('UserAgent:_get_target: no target for day %s,'
'taking zero ts' % date)
target_res = 15 * 60 # TODO FIXME: hardcoded default resolution
# day-ahead-planning relies on targets starting at midnight local
# time, so we have to use this for zero target as well
dap_target_start = util.get_tomorrow(date, 'local')
self._ts = util.TimeSeries(dap_target_start, target_res,
[0 for i in range(96)])
self._weights = util.TimeSeries(dap_target_start, target_res,
[0 for i in range(96)])
else:
# date_str = start_keys[0].format('YYYY-MM-DD HH:mm:ss ZZ')
logger.info('UserAgent:_get_target: starting with target %s' %
start_keys[0])
target_start = start_keys[0]
target_res, target_gen = self._target_gens[target_start]
# read all target and weights data entries from file
ts_data = []
weights_data = []
for line in target_gen:
data = line.strip().split(',')
val, weight = map(float, data)
ts_data.append(val)
weights_data.append(weight)
self._ts = util.TimeSeries(target_start, target_res, ts_data)
self._weights = util.TimeSeries(target_start, target_res,
weights_data)
# target has to be dap-compatible:
assert self._ts.start.to('local').hour == 0
return (self._ts, self._weights)
def test_update_thermal_demand_forecast_argchecks(ctrl, obs, err_kwargs):
"""Check various bad arguments."""
kwargs = {
'start': ctrl._da_scheduler._forecast_p.start,
'res': ctrl._da_scheduler._scheduling_res,
'data': np.zeros(ctrl._da_scheduler._scheduling_intervals),
}
kwargs.update(err_kwargs)
fc = util.TimeSeries(**kwargs)
target_schedule = util.TimeSeries(**kwargs)
weights = util.TimeSeries(**kwargs)
with pytest.raises(ValueError):
yield from ctrl.update_thermal_demand_forecast(target_schedule,
weights, fc)
def _get_thermal_demand_forecast(self, date, hours):
"""Return a :class:`~util.TimeSeries` with a thermal demand forecast
for the period from *date* for *hours* hours."""
# The forecast usually covers a period of 48h and we request a new
# forecast every 24 hours.
#
# The forecast data comes from a large file. We don't want to keep the
# full content of that file in memory.
#
# So in order to create a new forecast, we
# 1. reuse the old one,
# 2. strip all data from before the new start date, and
# 3. extend it with new data from the data file
#
demand_start, demand_res, demand_gen = self._demand_gen
if self._fc is None:
# Create an initial, empty forecast if there is none
self._fc = util.TimeSeries(date, demand_res, [])
# Strip old data from the forecast
self._fc.lstrip(date)
# We need data for [start_date, end_date) from the data file
# (in the first run, we load 2 days, but on second day, we only
# have to add the data for the second, as the first has been
# read already)
period = hours * 3600 - self._fc.period
start_date = self._fc.end
end_date = start_date.replace(seconds=period)
# Skip lines until we reach "start_date"
start_diff = util.get_intervals_between(start_date, demand_start,
demand_res)
[next(demand_gen) for i in range(start_diff)]
# Get the data from the file and extend the forecast
n_lines = util.get_intervals_between(end_date, start_date, demand_res)
fc_data = []
for i in range(n_lines):
data = next(demand_gen).strip().split(',')
p_el, p_th_heat, p_th_water = map(float, data)
sum_p_th = p_th_heat + p_th_water
fc_data.append(sum_p_th)
new_data = util.TimeSeries(start_date, demand_res, fc_data)
self._fc.extend(new_data)
self._demand_gen = (end_date, demand_res, demand_gen)
return self._fc
def test_get_day_ahead_plan(ctrl_obs):
ctrl, obs = ctrl_obs
ctrl._da_scheduler._forecast_p = util.TimeSeries(
arrow.get('2015-01-01T09:00:00'), ctrl._da_scheduler._scheduling_res,
np.arange(ctrl._da_scheduler._scheduling_intervals))
ctrl._task_negotiation = asyncio.Future()
ctrl._task_negotiation.set_result(None)
# 15h, because its 09:00
r_start = int(15 * 3600 / ctrl._da_scheduler._scheduling_res)
r_end = r_start + (24 * 3600 // ctrl._da_scheduler._scheduling_res)
expected_dap = list(range(r_start, r_end))
# Assert that the ts returned is the correct sub-range of the forecast:
dap_start = arrow.get('2015-01-02')
dap_end = dap_start.replace(days=1)
dap = yield from ctrl.get_day_ahead_plan(dap_start, dap_end)
assert list(dap) == expected_dap
# Assert that the correct target_schedule and weight vector have been set
exp_len = ctrl._da_scheduler._scheduling_intervals
pad_len = exp_len - r_end # Length of 0-padding
assert len(ctrl._da_scheduler._target_schedule) == exp_len
assert list(ctrl._da_scheduler._target_schedule.data) == \
list(range(r_end)) + [0] * pad_len
assert len(ctrl._da_scheduler._weights) == exp_len
assert list(
ctrl._da_scheduler._weights.data) == [1] * r_end + [0] * pad_len
def test_run_negotiation_timeout(ctrl, obs, ua_mocks):
"""UAs don't detect termination and we time out."""
ctrl._agents = collections.OrderedDict(
sorted(ctrl._agents.items(), key=lambda i: i[1]))
ctrl._neg_timeout = 0.1
start = arrow.get('2015-01-01')
res = ctrl._da_scheduler._scheduling_res = 15
ctrl._da_scheduler._forecast_p = util.TimeSeries(start, res, [0] * 4)
ctrl._da_scheduler._target_schedule = util.TimeSeries(
start, res, list(range(4)))
ctrl._da_scheduler._weights = util.TimeSeries(start, res, [1] * 4)
yield from aiomas. async (ctrl._run_negotiation(start))
# Check solution returned to the BGA
assert ctrl._da_scheduler._forecast_p == util.TimeSeries(
start, res, [18, 22, 26])
def test_user_get_dap_dates(user, now, exp_start, exp_end):
now, exp_start, exp_end = [arrow.get(d) for d in [now, exp_start, exp_end]]
ts = util.TimeSeries(exp_start, 900, [0 for i in range(96)])
user._ts = ts
s, e = user._get_dap_dates(user.container.clock)
assert s.tzinfo == dateutil.tz.tzutc()
assert s == exp_start
assert e.tzinfo == dateutil.tz.tzutc()
assert e == exp_end
def test_udpate_thermal_demand_forecast_neg_running(ctrl, obs):
"""Raise an error if the forecast is update while a negotiation is
currently running."""
ctrl._task_negotiation = asyncio.Future()
kwargs = {
'start': ctrl._da_scheduler._forecast_p.start,
'res': ctrl._da_scheduler._scheduling_res,
'data': np.zeros(ctrl._da_scheduler._scheduling_intervals),
}
fc = util.TimeSeries(**kwargs)
target_schedule = util.TimeSeries(**kwargs)
weights = util.TimeSeries(**kwargs)
with pytest.raises(RuntimeError):
yield from ctrl.update_thermal_demand_forecast(target_schedule,
weights, fc)
# Try again:
ctrl._task_negotiation.set_result(None)
yield from ctrl.update_thermal_demand_forecast(target_schedule, weights,
fc)
def test_dummy(uas, ctrl_obs, target):
# logging.basicConfig(level=logging.DEBUG)
ctrl, obs = ctrl_obs
results = []
msgs = []
for i in range(NUM_EXP):
start = arrow.get().replace(minute=0, second=0, microsecond=0)
ctrl._target_schedule = util.TimeSeries(start, 900, target)
ctrl._weights = util.TimeSeries(start, 900, [1] * 5)
ctrl._forecast_p = util.TimeSeries(start, 900, [0] * 5)
ctrl._scheduling_period = 900 * 5
target_schedule = util.TimeSeries(start, 900, [0] * 5)
weights = util.TimeSeries(start, 900, [0] * 5)
fc = util.TimeSeries(start, 900, [0] * 5)
yield from ctrl.update_thermal_demand_forecast(target_schedule,
weights, fc)
dap = yield from ctrl.get_day_ahead_plan(
start, start.replace(seconds=5 * 900))
diff_abs = np.abs(dap - target)
diff_mean = np.mean(diff_abs / target)
assert diff_mean <= 0.02
results.append(diff_mean)
assert np.mean(results) <= 0.05
def __init__(self, controller, scheduling_res, scheduling_period):
self._controller = controller
self._scheduling_res = scheduling_res
self._scheduling_period = scheduling_period
self._scheduling_intervals = int(scheduling_period / scheduling_res)
# We need a basic TimeSeries as long as we have no real data.
# We take utc 7a.m., as if the schedule were the result of a dap
utctime = self._controller.container.clock.utcnow()
utctime = utctime.replace(minute=0, second=0, microsecond=0, hour=7)
zero_ts = util.TimeSeries(utctime, self._scheduling_res,
np.zeros(self._scheduling_intervals))
# TODO wording: Maybe this should be called "planned_p", because its
# the result of a negotiation – the aggregate power output of the VPP:
self._forecast_p = zero_ts
self._target_schedule = zero_ts.copy()
# The *weights* list allows us to express that intervals closer to
# "now" must match the target schedule closer than intervals farther in
# the future:
self._weights = zero_ts.copy()
self._start = None
def test_run_negotiation(ctrl_obs, ua_mocks):
"""Test a simple DAP from the ctrl's point of view.
No negotitation at unit agents is performed (mocks used).
"""
ctrl, obs = ctrl_obs
obs._termination_detector._num_agents = len(ua_mocks)
# General config
ctrl._agents = collections.OrderedDict(
sorted(ctrl._agents.items(), key=lambda i: i[1]))
start = arrow.get('2015-01-01')
res = ctrl._scheduling_res = 15
ctrl._da_scheduler._forecast_p = util.TimeSeries(start, res, [0] * 4)
ctrl._da_scheduler._target_schedule = util.TimeSeries(
start, res, list(range(4)))
ctrl._da_scheduler._weights = util.TimeSeries(start, res, [1] * 4)
ts = list(ctrl._da_scheduler._target_schedule.data)
weights = list(ctrl._da_scheduler._weights.data)
# conn = {}
# t = [0, 1, 2, 3]
# w = [0, 0, 0, 0]
neg_task = aiomas. async (ctrl._run_negotiation(start))
# Test init_negotiation()
futs = [ua.init_negotiation_called for ua in ua_mocks]
ret = yield from gather(*futs)
assert ret == [
(['tcp://127.0.0.1:5556/1',
'tcp://127.0.0.1:5556/2'], start, res, ts, weights, True), # NOQA
(['tcp://127.0.0.1:5556/0',
'tcp://127.0.0.1:5556/2'], start, res, ts, weights, False), # NOQA
(['tcp://127.0.0.1:5556/0',
'tcp://127.0.0.1:5556/1'], start, res, ts, weights, False), # NOQA
]
# UA agents found a solution
for ua in ua_mocks:
yield from ua.obs_agent.update_stats(ua, 0, 0, 0, 1, 1, True)
# Retrieve final solution
yield from neg_task
assert ctrl._da_scheduler._forecast_p == util.TimeSeries(
start, res, [18, 22, 26])
# Check solutions sent to each UA
futs = [ua.stop_negotiation_called for ua in ua_mocks]
yield from gather(*futs)
# Sent final solution to observer
futs = [ua.obs_agent.update_final_cand(ua.solution) for ua in ua_mocks]
yield from gather(*futs)
futs = [ua.set_schedule_called for ua in ua_mocks]
ret = yield from gather(*futs)
solution = ua_mocks[0].solution
for i, schedule_id in enumerate(ret):
assert schedule_id == solution.sids[i]
def generate_schedules(self, start, res, intervals, state):
# TODO for electrical target schedule: generate more than one possible
# schedule / first in a simple manner to verify adaptation to
# electrical target schedule optimization
# TODO: unify resolution check?
# TODO: schedule resolution is 60 (thus a minute)
if res < self.model.res or res % self.model.res != 0:
raise ValueError('Schedule resolution must be >= %d and a '
'multiple of it' % self.model.res)
# The latest unit state should be from time "start", but we want
# the state from the interval before.
# Example: "start" is 09:00 o'clock, then the last state is for the
# interval [09:00, 09:01). If we want to simulate from 09:00, we need
# the previous state from [08:59, 09:00).
assert state[-1][0] == start
state = state[-2][1]
self._schedules = {}
self._schedule_id = itertools.count()
# Start index in the array with the forecast data
res_ratio = int(res // self.model.res)
n_sim_steps = res_ratio * intervals
# Generate initial schedule
setpoint = self._setpoint_for(state['chp_p_el'])
storage_e = state['storage_e_th']
self.model.reset(chp_setpoint=setpoint, storage_e_th=storage_e)
p_el = np.zeros(n_sim_steps)
p_th_gen = self._forecast_demand_p_th.iter(start, self.model.res,
n_sim_steps)
for i, p_th in enumerate(p_th_gen):
self.model.step(p_th)
p_el[i] = self.model.chp_p_el
schedule_id = next(self._schedule_id)
unit_schedule = util.TimeSeries(start, self.model.res, p_el)
utility = 1
# reshaping to 15 min resolution
negotiation_schedule = p_el.reshape(intervals, res_ratio).mean(axis=1)
self._schedules[schedule_id] = unit_schedule
possible_schedules = [(schedule_id, utility, negotiation_schedule)]
# TODO: quick hack for testing multiple schedules - remove
# np.random.seed(7) # at least .. deterministic hack
# for i in range(10):
# tmp_data = np.random.permutation(unit_schedule.data)
# tmp = util.TimeSeries(start, self.model.res, tmp_data)
# # print(tmp.data)
# sched_id = next(self._schedule_id)
# self._schedules[sched_id] = tmp
# # reshaping to 15 min resolution
# negotiation_schedule = tmp_data.reshape(
# intervals, res_ratio
# ).mean(axis=1)
# possible_schedules.append((sched_id, utility,
# negotiation_schedule))
# print("sched 6:")
# for val in self._schedules[6].data:
# print(val)
# end quick hack
# return [(schedule_id, utility, negotiation_schedule)]
return possible_schedules
def finalize_negotiation(self, solution):
# TODO wording: forecast_p = aggregierter Einsatzplan
forecast_p = self._aggregate_solution(solution)
assert self._start is not None
res = self._target_schedule.res
self._forecast_p = util.TimeSeries(self._start, res, forecast_p)