I am trying to run a whole multifamily building simulation in ResStock by setting whole_sfa_or_mf_building_sim = true in measure.rb.
Initially, the simulation failed with this error:
pandas.errors.MergeError: Passing 'suffixes' which cause duplicate columns {'schedules_No Space Heating_x', 'schedules_hot_water_fixtures_x', 'schedules_Vacancy_x', 'schedules_hot_water_clothes_washer_x', 'schedules_plug_loads_other_x', 'schedules_lighting_interior_x', 'schedules_clothes_washer_x', 'schedules_ceiling_fan_x', 'schedules_occupants_x', 'schedules_plug_loads_tv_x', 'schedules_lighting_garage_x', 'schedules_No Space Cooling_x', 'schedules_Power Outage_x', 'schedules_cooking_range_x', 'schedules_clothes_dryer_x', 'schedules_hot_water_dishwasher_x', 'schedules_dishwasher_x'} is not allowed.
I traced this back to how schedules are merged in buildstockbatch/base.py. The original merging code was:
for schedules_filepath in schedules_filepaths:
schedules = read_csv(schedules_filepath, dtype=np.float64)
schedules.rename(columns=lambda x: f"schedules_{x}", inplace=True)
schedules["TimeDST"] = tsdf["Time"]
tsdf = tsdf.merge(schedules, how="left", on="TimeDST")
To fix the issue, I modified it as follows:
for schedules_filepath in schedules_filepaths:
schedules = read_csv(schedules_filepath, dtype=np.float64)
time_cols = {"TimeDST", "Time", "time", "TimeUTC"}
schedules.rename(
columns=lambda c: c if c in time_cols else f"schedules_{c}",
inplace=True,
)
schedules["TimeDST"] = tsdf["TimeDST"]
dup = [c for c in schedules.columns if c in tsdf.columns and c != "TimeDST"]
if dup:
logging.debug(f"Dropping duplicate schedule columns: {dup}")
schedules.drop(columns=dup, inplace=True)
tsdf = tsdf.merge(schedules, how="left", on="TimeDST")
With this change, the simulation runs successfully when max_num_units_modeled in measure.rb is left at its default value of 5.
However, as soon as I increase max_num_units_modeled to any value greater than 5 (even 6), the simulation fails due to running out of RAM.
My questions:
- Is there any straightforward way to run whole multifamily building simulations without modifying the base code?
- If not, does my fix make sense? Or is there a better long-term fix for the issue?
- Is there a known limitation with higher max_num_units_modeled values in ResStock (e.g., performance, memory constraints)?
- Has anyone successfully run whole multifamily building simulations without hitting these memory problems?
Thank you!
