电网环境¶

`powerzoo.envs.base.BaseEnv(delta_t_minutes=1.0)` ¶

Bases: Env, ABC

Base environment for PowerZoo.

Provides: - np_random: Gymnasium's seeded RNG, initialised through super().reset(seed=seed, options=options). - observation_space / action_space placeholders (None until subclass initialises them in __init__). - Abstract step / obs methods that subclasses must implement. - Default reward / cost hooks for reward-shaping and CMDP cost.

Seed contract¶

Subclasses should call super().reset(seed=seed, options=options) as the first line of their reset() implementation. This delegates RNG management to Gymnasium and preserves its seed=None semantics: reuse the current generator when one already exists.

`reset(*, seed=None, options=None)` ¶

Apply common Gymnasium reset bookkeeping.

The base implementation seeds np_random (via super().reset), resets time_step to 0, and returns the placeholder pair (None, {}).

Subclass contract¶

Call super().reset(seed=seed, options=options) as the first line — this seeds self.np_random and resets the clock.
After super(), build and return a valid (observation, info) pair that satisfies self.observation_space.
Never return the base placeholder (None, {}) from a concrete environment — downstream wrappers and RL libraries will fail on None observations.

`step(action)` `abstractmethod` ¶

Apply action and return (obs, reward, terminated, truncated, info).

Subclasses are responsible for advancing self.time_step inside step() and for computing truncated from their own episode-limit policy (for example max_episode_steps).

Recommended implementation order¶

Action decode — map the RL agent's action (e.g. normalised values in [-1, 1]) to physical setpoints (MW, MVar, °C …).
Physics — run power flow / OPF / dynamic simulation for the current time step.
State extract — read voltages, flows, costs, and resource state-of-charge from the solver result.
Observation — call self.obs(state) to produce the agent- facing observation array.
Reward / cost — compute scalar reward and, if applicable, CMDP cost terms; populate info with cost_* keys.
Clock advance — increment self.time_step; set truncated when the episode step limit is reached.

`obs(state)` `abstractmethod` ¶

Convert internal state dict to a flat observation array.

Subclasses should implement this to transform the internal full physical state into the agent-facing observation, which may be local, partial, noisy, or otherwise filtered relative to the simulator state.

`reward(r, state, info)` ¶

Return the scalar reward after optional reward-shaping adjustments.

This is a lightweight base-class hook for reward shaping. Concrete environments or wrappers can override it to transform the scalar reward while leaving the main transition logic elsewhere.

For non-trivial reward shaping, prefer using a gym.Wrapper instead; this hook is intended as a lightweight escape hatch for environment-internal adjustments only.

`powerzoo.envs.grid.base.GridEnv(case=None, solver=None, delta_t_minutes=30.0, data_loader=None, start_date='2024-01-01', end_date='2024-01-31', load_columns=None, max_load_ratio=0.9, min_load_ratio=None, time_series=None, max_episode_steps=None, randomize_start_time=False, time_alignment=None)` ¶

Bases: BaseEnv

Base power grid environment (inner physical simulation engine).

GridEnv acts as the physical state machine layer of the PowerZoo stack. Its reset() and step() return a raw state dict (the full power-flow solution) rather than a flat numpy observation array, because the state is consumed by upper-layer components that each need different subsets of it:

PowerEnv — the benchmark-facing RL facade. Calls grid.step() to advance the simulation, then passes the returned state dict to grid.obs(state) and to each resource's obs() to assemble the agent observation.
ObsWrapper / gym_wrappers.GymWrapper — thin Gym-compatibility wrappers that call env.obs(state) directly on a GridEnv subclass for single-grid use-cases that don't need PowerEnv.

Because of this design, passing a bare GridEnv subclass to standard Gym validation tools (e.g. gymnasium.utils.env_checker.check_env) will produce an observation-type warning — the raw state dict does not conform to observation_space. Always wrap the env in PowerEnv or a Gym wrapper before running a standard RL training loop or check.

Subclasses must implement¶

_run_power_flow(action) — run the solver; return True on success.
_get_state() — build and return the full state dict after a solve.
obs(state) — project the state dict to a flat float32 array matching observation_space.
build_info(state) — build the step info dict.

Initialize grid environment

Parameters:

Name	Type	Description	Default
`case`	`Any`	Power system case	`None`
`solver`	`Any`	Power flow solver	`None`
`delta_t_minutes`	`float`	Time step in minutes (default: 30)	`30.0`
`data_loader`	`Optional[DataLoader]`	DataLoader instance. If None, creates default DataLoader. Ignored when `time_series` is provided.	`None`
`start_date`	`str`	Start date for data loading (format: 'YYYY-MM-DD'). Ignored when `time_series` is provided.	`'2024-01-01'`
`end_date`	`str`	End date for data loading (format: 'YYYY-MM-DD'). Ignored when `time_series` is provided.	`'2024-01-31'`
`load_columns`	`Optional[List[str]]`	Semantic signal names to load. Default: `[signals.LOAD_ACTUAL_MW, signals.SOLAR_AVAILABLE_MW, signals.WIND_AVAILABLE_MW]`. Distribution envs may also request `signals.LOAD_REACTIVE_MVAR` to provide an explicit feeder reactive-demand time series. Legacy raw column names (`'ActualDemand'`, `'Wind'`, …) are auto-mapped with a deprecation warning. Ignored when `time_series` is provided.	`None`
`max_load_ratio`	`float`	Maximum load as ratio of total generation capacity (default: 0.9)	`0.9`
`min_load_ratio`	`Optional[float]`	Minimum load as ratio of total generation capacity (default: None)	`None`
`time_series`	`Any`	Custom time-series data supplied directly by the user. Accepted formats: - `pandas.DataFrame`: must have a DatetimeIndex (or 'datetime' column) and at least a `signals.LOAD_ACTUAL_MW` (or legacy `'ActualDemand'`) column. An optional `signals.LOAD_REACTIVE_MVAR` (or legacy `'ReactiveDemand'`) column is used by distribution envs as an explicit Q-demand series. - `numpy.ndarray` of shape `(T,)` or `(T, 1)`: treated as a single-column `load.actual_mw` time series. A synthetic DatetimeIndex starting at `start_date` is created automatically. When provided, data_loader / start_date / end_date / load_columns are NOT used for loading data (max_load_ratio and min_load_ratio still apply for scaling).	`None`
`max_episode_steps`	`Optional[int]`	Maximum steps per episode before `truncated=True`. Defaults to one full day (`steps_per_day`).	`None`
`randomize_start_time`	`bool`	When True, choose a random intra-day starting offset at each reset instead of always beginning at `time_step=0`. Increases initial-state diversity from O(n_days) to O(n_timesteps). Default False to preserve backward compatibility (F5 fix).	`False`
`time_alignment`	`Optional[Dict[str, str]]`	Per-signal time-alignment overrides for cross-period data. Example: `{"solar.available_mw": "2024-01-01"}` maps the solar source's 2024 data onto the simulation's `start_date`. Profile-mode datasets (e.g. data-center traces) are tiled automatically and do not need an override.	`None`

`reset(*, seed=None, options=None, day_id=None)` ¶

Reset grid environment and all sub-resources.

Follows the Gymnasium v26+ convention: seed is the primary reproducibility parameter; day_id (legacy kwarg) or options={'day_id': N} both work for selecting a specific day.

Parameters:

Name	Type	Description	Default
`seed`	`Optional[int]`	Random seed. Passed through `super().reset(seed=seed, options=options)` so Gymnasium manages `self.np_random`.	`None`
`options`	`Optional[Dict]`	Optional dict. Recognised keys: `day_id` (int) – override which day to simulate.	`None`
`day_id`	`Optional[int]`	Legacy kwarg; takes precedence over `options['day_id']`.	`None`

`step(action=None)` ¶

Execute one time step

Parameters:

Name	Type	Description	Default
`action`	`Any`	Dict containing: - resource actions (key=resource_id, value=resource action) - grid control parameters (subclass-specific, e.g., unit_power_mw, node_load_mw)	`None`

Returns:

Type	Description
	state, reward, done, truncated, info

`obs(state=None)` ¶

Convert a raw state dict to a flat float32 observation array.

This method is part of GridEnv's public API and is called by:

PowerEnv._build_agent_observation(state) — assembles the combined grid + resource + time observation for the RL agent.
GymWrapper (powerzoo.wrappers.gym_wrappers) — provides a thin Gym-compliant wrapper for direct single-grid use.

Subclasses must implement this and define self.observation_space to match the returned array's shape and dtype. The default raises NotImplementedError.

`register_resource(resource, bus_id, name=None)` ¶

Register a resource and assign unique ID.

Calls _on_resource_changed() after updating internal state so that subclasses can rebuild observation_space and action_space.

Parameters:

Name	Type	Description	Default
`resource`	`Any`	Resource instance to register	required
`bus_id`	`int`	Bus ID where resource is connected	required
`name`	`Optional[str]`	Optional custom name. If None, auto-generated (e.g., 'solar_0')	`None`

Returns:

Name	Type	Description
`resource_id`	`str`	The assigned resource ID

`unregister_resource(resource_id)` ¶

Unregister a resource.

Calls _on_resource_changed() after updating internal state.

`cal_pf(*args, **kwargs)` ¶

Run power flow and return result dict (must be implemented).

If the solver does not converge, implementations should return a result dict with converged=False (or equivalent flag) rather than raising an exception, so that downstream safety_check and reward/cost logic can handle divergence gracefully.

`safety_check(*args, **kwargs)` ¶

Check physical constraints (must be implemented).

Implementations must be robust to non-converged or NaN-laden states (e.g. when cal_pf did not converge). In such cases the method should report maximum violation / unsafe status instead of crashing.

`powerzoo.envs.grid.trans.TransGridEnv(case=None, solver=None, delta_t_minutes=30.0, data_loader=None, start_date='2024-01-01', end_date='2024-01-31', load_columns=None, max_load_ratio=0.9, min_load_ratio=None, time_series=None, max_episode_steps=None, randomize_start_time=False, physics='dc', solver_mode='opf', solver_type='auto', normalize_actions=True, ac_config=None, difficulty=None, reward_scale=0.01, control_der=False)` ¶

Bases: GridEnv

Transmission grid environment.

Two orthogonal parameters control the solver behaviour:

physics — 'dc' (linearised, P only) or 'ac' (full AC with voltage and reactive power).
solver_mode — 'opf' (environment runs OPF internally; RL agent provides bids / commitments) or 'pf' (environment evaluates power flow only; RL agent provides unit dispatch directly).

The four resulting modes are DCOPF, ACOPF, DCPF, and ACPF.

Default case is Case5.

Initialize transmission grid environment.

Two orthogonal parameters control the solver behaviour:

physics — physical model: 'ac' (full AC equations) or 'dc' (linearised DC approximation).
solver_mode — solver role: 'opf' (environment runs optimal power flow internally; RL agent provides bids / commitments) or 'pf' (environment only evaluates power flow; RL agent provides unit dispatch directly via action['unit_power_mw']).

The four combinations are:

======= =========== ========== ======================================= physics solver_mode Solver RL use-case ======= =========== ========== ======================================= dc opf DCOPF Agent learns bidding / UC strategy ac opf ACOPF Agent learns bidding (with voltage) dc pf DCPF Agent learns dispatch (P only) ac pf ACPF (NR) Agent learns dispatch (P + V + Q) ======= =========== ========== =======================================

Parameters:

Name	Type	Description	Default
`case`	`ClearCase`	Power system case (default: Case5)	`None`
`solver`	`Any`	Power flow solver	`None`
`delta_t_minutes`	`float`	Time step in minutes (default: 30)	`30.0`
`data_loader`	`Optional[DataLoader]`	DataLoader instance	`None`
`start_date`	`str`	Start date for data loading	`'2024-01-01'`
`end_date`	`str`	End date for data loading	`'2024-01-31'`
`load_columns`	`Optional[List[str]]`	List of columns to load	`None`
`max_load_ratio`	`float`	Maximum load as ratio of total capacity	`0.9`
`min_load_ratio`	`Optional[float]`	Minimum load as ratio of total capacity (optional)	`None`
`time_series`	`Any`	Custom time-series data	`None`
`max_episode_steps`	`Optional[int]`	Max steps before truncation (default: one full day)	`None`
`physics`	`str`	Physical model — `'dc'` (default) or `'ac'`.	`'dc'`
`solver_mode`	`str`	Solver role — `'opf'` (default, environment optimises) or `'pf'` (agent provides dispatch, environment evaluates).	`'opf'`
`solver_type`	`str`	OPF LP solver - 'auto', 'gurobi', 'scipy', 'cvxpy'.	`'auto'`
`normalize_actions`	`bool`	Whether to normalise actions to [-1, 1].	`True`
`ac_config`	`ACConfig`	AC solver parameters as an :class:`ACConfig` object. Defaults to `ACConfig()` (standard voltage limits, built-in solver). See :class:`ACConfig` for all configurable fields.	`None`
`difficulty`	`Optional[str]`	Preset difficulty level - 'easy', 'medium', or 'hard'. When set, overrides `delta_t_minutes` and `max_load_ratio`.	`None`
`reward_scale`	`float`	Multiplicative scaling factor applied to generation cost in the reward signal (`economic_cost = -reward_scale * gen_cost`). Default 0.01 is calibrated for Case5 (typical cost ~100-1000 $/step). For larger systems (e.g. Case118) where cost can reach 10⁵-10⁶, pass a smaller value such as 0.0001 to keep rewards in a well-conditioned range for policy gradient algorithms.	`0.01`
`control_der`	`bool`	When `True`, flatten DER control dimensions into the environment's `action_space`. In PF modes the flat action becomes `[unit_power_mw (n_units), der_actions (n_resources)]`. In OPF modes with registered DER, the flat action contains only `der_actions` so ndarray agents cannot silently bypass the OPF by injecting `unit_power_mw`. Agents can then control DER through a single flat Box. DER states are always visible in `observation_space` regardless of this flag (when resources are registered). Default `False` (DER run as non-dispatchable injections).	`False`

`reset(*, seed=None, options=None, day_id=None)` ¶

Reset transmission grid and run initial power flow.

`obs(state=None)` ¶

Convert current (or provided) state to a flat float32 observation array.

Parameters:

Name	Type	Description	Default
`state`	`Any`	Internal state dict (as returned by `_get_state`). If None, the most recently cached state is used.	`None`

Returns:

Type	Description
`ndarray`	numpy.ndarray of shape `observation_space.shape` and dtype float32.

`cal_pf(unit_power_mw, node_load_mw, df=False)` ¶

Thin wrapper — see :func:trans_solve.cal_pf.

Side effects: updates self._power_imbalance_mw and self._slack_gen_violation_mw.

`safety_check(line_flow_mw, with_info=False)` ¶

Thin wrapper — see :func:trans_solve.safety_check.

`render(mode='human')` ¶

Render the transmission grid state.

Produces a two-panel figure (network topology + unit dispatch chart). See :func:powerzoo.envs.grid._render.render_trans_grid for details.

Parameters:

Name	Type	Description	Default
`mode`	`str`	`'human'` (interactive) or `'rgb_array'` (returns a `(H, W, 3)` uint8 ndarray without opening a window).	`'human'`

`powerzoo.envs.grid.dist.DistGridEnv(case=None, solver=None, delta_t_minutes=30.0, data_loader=None, start_date='2024-01-01', end_date='2024-01-31', load_columns=None, max_load_ratio=0.9, min_load_ratio=None, time_series=None, max_iter=100, tol=1e-06, max_episode_steps=None, randomize_start_time=False, v_slack=1.0, v_min=0.9, v_max=1.1, allow_mesh_pruning=True, difficulty=None, violation_penalty_weight=0.0, v_dev_penalty_weight=0.0, loss_penalty_weight=0.1)` ¶

Bases: _DistPFMixin, _DistLoadsMixin, GridEnv

Distribution grid environment using BFS (Backward/Forward Sweep) power flow.

Default case is Case33bw; handles both active (P) and reactive (Q) power. The physical core is a single-phase balanced radial DistFlow solver: explicit voltage-angle states, phase coupling, and unbalance are outside this benchmark surface.

Naming conventions (consistent with MATPOWER): baseMVA, baseKV — system base values slack_bus_id — slack/reference bus index (0-based); alias: ref_bus v_slack — slack bus voltage setpoint (p.u.); alias: v_ref_mag p_flow_MW — sending-end (from-bus) active power flow on each branch p_loss_MW — active power loss (I²R) on each branch p_slack_MW — total slack-bus active-power injection into the feeder q_slack_MVAr — total slack-bus reactive-power injection into the feeder is_diverged — BFS failed to satisfy the iteration tolerance before hitting max_iter (distinct from voltage collapse)

Resource control mode

All non-slack buses are treated as PQ buses by the BFS solver. When a resource is registered via register_resource(), it operates in PQ control mode: its P/Q setpoints are subtracted from the nodal load before the BFS solve.

Action space: only gymnasium.spaces.Box (continuous) is supported.

Initialize distribution grid environment

Parameters:

Name	Type	Description	Default
`case`	`ClearCase`	ClearCase instance (default: Case33bw)	`None`
`solver`	`Any`	Optional external solver	`None`
`delta_t_minutes`	`float`	Time step length in minutes (default: 30).	`30.0`
`data_loader`	`Optional[DataLoader]`	DataLoader for external time-series data.	`None`
`start_date`	`str`	Start date for data loading.	`'2024-01-01'`
`end_date`	`str`	End date for data loading.	`'2024-01-31'`
`load_columns`	`Optional[List[str]]`	Columns to load from DataLoader. Distribution envs also honour an optional `load.reactive_mvar` signal when provided.	`None`
`max_load_ratio`	`float`	Peak load as fraction of case capacity (default: 0.9).	`0.9`
`min_load_ratio`	`Optional[float]`	Minimum load ratio (optional).	`None`
`time_series`	`Any`	Custom time-series (numpy array or DataFrame).	`None`
`max_iter`	`int`	Maximum iterations for power flow (default: 100)	`100`
`tol`	`float`	Convergence tolerance (default: 1e-6)	`1e-06`
`max_episode_steps`	`Optional[int]`	Max steps per episode before truncation.	`None`
`randomize_start_time`	`bool`	Randomize intra-day start offset on reset.	`False`
`v_slack`	`float`	Slack bus voltage setpoint in p.u. (default: 1.0).	`1.0`
`v_min`	`float`	Minimum voltage limit (p.u., default: 0.90).	`0.9`
`v_max`	`float`	Maximum voltage limit (p.u., default: 1.10).	`1.1`
`allow_mesh_pruning`	`bool`	If True (default), extra lines in a non-radial input are pruned to the BFS first-visit spanning tree with a warning. If False, env initialization fails fast on mesh input.	`True`
`difficulty`	`Optional[str]`	Preset - 'easy', 'medium', or 'hard'. Overrides v_min/v_max.	`None`
`violation_penalty_weight`	`float`	Weight for soft-penalty mode. When > 0, `cost_voltage_violation` and `cost_thermal_overload` are added as negative reward terms (standard RL mode). When 0 (default), violations are exposed only via info cost fields (CMDP mode).	`0.0`
`v_dev_penalty_weight`	`float`	Weight for voltage-deviation penalty. When > 0, adds `-v_dev_penalty_weight * mean((v - 1.0)²)` to reward. MSE penalises large deviations quadratically with a natural soft deadband. Useful for Volt-VAR / voltage regulation tasks. When 0 (default), no voltage-deviation term is added.	`0.0`
`loss_penalty_weight`	`float`	Weight on active-loss reward shaping. Default reward uses `-0.1 * p_loss_MW`.	`0.1`

`reset(*, seed=None, options=None, day_id=None)` ¶

Reset distribution grid and run initial power flow.

`obs(state=None)` ¶

Return flat float32 observation array.

When state is provided, its nodes/lines/time_step are used instead of the live cache (useful for replaying a past step).

On PF failure the agent receives a penalty observation (below-normal voltages, zero flows) rather than the pre-divergence state, so that the catastrophic reward is correctly paired with an out-of-band observation. See _obs_should_use_failure_fallback() for trigger conditions.

Observation layout and normalisation: see _build_spaces().

`render(mode='human')` ¶

Render the distribution grid state.

Produces a two-panel figure (radial network + voltage profile chart). See :func:powerzoo.envs.grid._render.render_dist_grid for details.

Parameters:

Name	Type	Description	Default
`mode`	`str`	`'human'` (interactive) or `'rgb_array'` (returns a `(H, W, 3)` uint8 ndarray without opening a window).	`'human'`

`powerzoo.envs.grid.dist_3phase.DistGrid3PhaseEnv(case=None, solver=None, delta_t_minutes=30.0, data_loader=None, start_date='2024-01-01', end_date='2024-01-31', load_columns=None, max_load_ratio=0.9, min_load_ratio=None, time_series=None, max_iter=100, tol=1e-06, max_episode_steps=None, randomize_start_time=False, v_min=0.9, v_max=1.1, vuf_max=2.0, difficulty=None, violation_penalty_weight=0.0, loss_penalty_weight=0.1, vuf_dense_penalty_weight=0.0)` ¶

Bases: _Dist3PhasePhysicsMixin, _Dist3PhaseLoadsMixin, DistGridEnv

Three-phase distribution grid environment using BIBC/BCBV power flow.

Default case is Case123; handles three-phase unbalanced power flow. The solver stores a full three-state A/B/C vector at every non-slack bus, so branch phase availability must already be reflected in the case's 3x3 impedance data rather than inferred from a separate missing-phase topology. Solver vectors use node-major ABC order [node1_A, node1_B, node1_C, node2_A, ...]; the explicit node mapping is available via self.topo3ph.non_ref_node_ids and self.topo3ph.node_id_to_matrix_index.

Resource phase injection

Resources carry a phase attribute ('A', 'B', 'C', or 'ABC'). Power is injected only into the connected phase(s), enabling the RL agent to learn phase-balancing strategies.

Non-convergence contract

cal_pf() still returns the last BFS iterate for debugging, but those voltages/flows/losses are diagnostic only. RL-facing callers must check self._converged / info['pf_converged'] and treat a False value as a power-flow failure rather than a valid operating point.

Initialize three-phase distribution grid environment.

Parameters:

Name	Type	Description	Default
`case`	`ClearCase`	ClearCase instance (default: Case123).	`None`
`solver`	`Any`	Optional external solver.	`None`
`delta_t_minutes`	`float`	Time step length in minutes (default: 30).	`30.0`
`data_loader`	`Optional[DataLoader]`	DataLoader for external time-series data.	`None`
`start_date`	`str`	Start date for data loading.	`'2024-01-01'`
`end_date`	`str`	End date for data loading.	`'2024-01-31'`
`load_columns`	`Optional[List[str]]`	Columns to load from DataLoader.	`None`
`max_load_ratio`	`float`	Peak load as fraction of case capacity (default: 0.9).	`0.9`
`min_load_ratio`	`Optional[float]`	Minimum load ratio (optional).	`None`
`time_series`	`Any`	Custom time-series (numpy array or DataFrame).	`None`
`max_iter`	`int`	Maximum iterations for power flow (default: 100).	`100`
`tol`	`float`	Convergence tolerance (default: 1e-6).	`1e-06`
`max_episode_steps`	`Optional[int]`	Max steps per episode before truncation.	`None`
`randomize_start_time`	`bool`	Randomize intra-day start offset on reset.	`False`
`v_min`	`float`	Minimum voltage limit (p.u., default: 0.90).	`0.9`
`v_max`	`float`	Maximum voltage limit (p.u., default: 1.10).	`1.1`
`vuf_max`	`float`	Maximum voltage unbalance factor (%, default: 2.0). IEEE Std 1159 / EN 50160 typical limit is 2%.	`2.0`
`difficulty`	`Optional[str]`	Preset - 'easy', 'medium', or 'hard'. Overrides v_min/v_max.	`None`
`violation_penalty_weight`	`float`	Weight for soft-penalty mode (default: 0.0).	`0.0`
`loss_penalty_weight`	`float`	Weight on active-loss reward shaping `-loss_penalty_weight * p_loss_MW` (default: 0.1).	`0.1`
`vuf_dense_penalty_weight`	`float`	Weight for dense VUF penalty (default: 0.0). When > 0, adds `-vuf_dense_penalty_weight * max(max_vuf_percent - 0.75 * vuf_max, 0) / 100` to the reward at every step. This keeps a deadband over the benign low-VUF region (default: 1.5 % when `vuf_max=2.0`), so the dense shaping only activates near the safety boundary. Operates independently of `violation_penalty_weight`.	`0.0`

电网环境¶

powerzoo.envs.base.BaseEnv(delta_t_minutes=1.0) ¶

Seed contract¶

reset(*, seed=None, options=None) ¶

Subclass contract¶

step(action) abstractmethod ¶

Recommended implementation order¶

obs(state) abstractmethod ¶

reward(r, state, info) ¶

powerzoo.envs.grid.base.GridEnv(case=None, solver=None, delta_t_minutes=30.0, data_loader=None, start_date='2024-01-01', end_date='2024-01-31', load_columns=None, max_load_ratio=0.9, min_load_ratio=None, time_series=None, max_episode_steps=None, randomize_start_time=False, time_alignment=None) ¶

Subclasses must implement¶

reset(*, seed=None, options=None, day_id=None) ¶

step(action=None) ¶

obs(state=None) ¶

register_resource(resource, bus_id, name=None) ¶

unregister_resource(resource_id) ¶

cal_pf(*args, **kwargs) ¶

safety_check(*args, **kwargs) ¶

reset(*, seed=None, options=None, day_id=None) ¶

obs(state=None) ¶

cal_pf(unit_power_mw, node_load_mw, df=False) ¶

safety_check(line_flow_mw, with_info=False) ¶

render(mode='human') ¶

reset(*, seed=None, options=None, day_id=None) ¶

obs(state=None) ¶

render(mode='human') ¶

`powerzoo.envs.base.BaseEnv(delta_t_minutes=1.0)` ¶

`reset(*, seed=None, options=None)` ¶

`step(action)` `abstractmethod` ¶

`obs(state)` `abstractmethod` ¶

`reward(r, state, info)` ¶

`powerzoo.envs.grid.base.GridEnv(case=None, solver=None, delta_t_minutes=30.0, data_loader=None, start_date='2024-01-01', end_date='2024-01-31', load_columns=None, max_load_ratio=0.9, min_load_ratio=None, time_series=None, max_episode_steps=None, randomize_start_time=False, time_alignment=None)` ¶

`reset(*, seed=None, options=None, day_id=None)` ¶

`step(action=None)` ¶

`obs(state=None)` ¶

`register_resource(resource, bus_id, name=None)` ¶

`unregister_resource(resource_id)` ¶

`cal_pf(*args, **kwargs)` ¶

`safety_check(*args, **kwargs)` ¶

`reset(*, seed=None, options=None, day_id=None)` ¶

`obs(state=None)` ¶

`cal_pf(unit_power_mw, node_load_mw, df=False)` ¶

`safety_check(line_flow_mw, with_info=False)` ¶

`render(mode='human')` ¶

`reset(*, seed=None, options=None, day_id=None)` ¶

`obs(state=None)` ¶

`render(mode='human')` ¶