[BUG] DHW Storage is never used

[LorenzoBonanni]

Issue Description

There is an incorrect variable reference in citylearn/building.py that prevents the Domestic Hot Water (DHW) storage system from updating its state of charge or energy balance. DHW actions are multiplied by self.heating_storage.capacity instead of self.dhw_storage.capacity.

When heating_storage is disabled (as in the 2023 challenge dataset), heating_storage.capacity == 0, causing DHW storage to remain unused even though dhw_storage appears as an active action.

Expected Behavior

When the agent outputs non-zero dhw_storage actions, the DHW tank should charge or discharge according to its configured capacity. Its SOC and energy balance should vary over time.

Actual Behavior

• dhw_storage actions have no effect.
• DHW SOC stays strictly zero for the entire episode.
• DHW energy balance remains zero.
• All DHW demand is supplied directly by the electric heater.
• Changing one line of code to reference self.dhw_storage.capacity produces
  the expected non-zero behavior.

Output with original code (incorrect)

DHW Storage SOC: [0. 0. 0. 0. ... 0.]

Output after fixing the line

DHW Storage SOC: [0.9780134 1. 1. 1. ... 0.62974083 1. 0.]

This clearly shows that DHW storage behavior is completely suppressed in the current version.

Steps to Reproduce

1. Use the citylearn_challenge_2023_phase_1 dataset where:

   • dhw_storage is active,
   • heating_storage is inactive → capacity = 0.

2. Create any agent that outputs non-zero dhw_storage actions.

3. Run a full episode and log:

   • env.buildings[0].dhw_storage.soc

4. Observe that SOC stays zero throughout the entire simulation.

Minimal Reproducible Example

from typing import List
from citylearn.agents.base import Agent
from citylearn.citylearn import CityLearnEnv

class CustomRBC(Agent):
    def __init__(self, env: CityLearnEnv, **kwargs):
        super().__init__(env, **kwargs)

    def predict(self, observations: List[List[float]], deterministic=True):
        actions = []
        for i, o in enumerate(observations):
            available_obs = self.observation_names[i]
            available_act = self.action_names[i]
            action = [0.0 for _ in range(len(available_act))]
            hour = o[available_obs.index('hour')]

            if 'dhw_storage' in available_act:
                if 6 <= hour <= 9 or 18 <= hour <= 22:
                    action[available_act.index('dhw_storage')] = -0.2
                else:
                    action[available_act.index('dhw_storage')] = 1.0

            actions.append(action)
        return actions

env = CityLearnEnv(
    schema='citylearn_challenge_2023_phase_1',
    central_agent=True,
)

agent = CustomRBC(env)

obs, _ = env.reset(seed=0)
while not env.terminated:
    action = agent.predict(obs)
    obs, reward, _, _, _ = env.step(action)

print("DHW Storage SOC:", env.buildings[0].dhw_storage.soc)

Environment

• CityLearn version: 2.4.2
• Dataset: citylearn_challenge_2023_phase_1
• Operating System: Ubuntu 24.04
• Python version: 3.10.19
• Issue present also in latest main branch at time of reporting.

Root Cause

In building.py around line 1765:

energy = action * self.heating_storage.capacity * self.algorithm_action_based_time_step_hours_ratio

This uses the heating storage capacity even when processing DHW actions.

Possible Solution

Replace the incorrect line with:

energy = action * self.dhw_storage.capacity * self.algorithm_action_based_time_step_hours_ratio

This immediately restores correct DHW storage behavior.

Additional Notes

This bug makes DHW storage unusable in all configurations where heating storage is inactive (capacity = 0), including the official challenge dataset. Fixing it is necessary for agents to learn or implement any DHW shifting strategies.

[LorenzoBonanni]

In addition to the DHW storage bug described above, I found two closely related issues in the same file (citylearn/building.py):

1. Incorrect attribute used in update_heating_storage (around line 1720):

   energy = action * self.cooling_storage.capacity * self.algorithm_action_based_time_step_hours_ratio

   This appears to reference cooling_storage instead of heating_storage.
   It should likely be:

   energy = action * self.heating_storage.capacity * self.algorithm_action_based_time_step_hours_ratio

2. Missing hourly ratio in update_cooling_storage:
   The function currently uses:

   energy = action * self.cooling_storage.capacity

   but, for consistency with both update_heating_storage and update_dhw_storage, and to correctly scale the energy based on the action-time-step ratio, it should probably be:

   energy = action * self.cooling_storage.capacity * self.algorithm_action_based_time_step_hours_ratio