Apppublic

channu07/microgrid-env

sourceHugging Faceupdated 6mo agoView on Hugging Face

0likes

App README

AI Custodire Resilience Engine for Next-Gen Microgrids

Why This Matters

Power grids are failing. Climate change brings unpredictable solar generation. EV adoption spikes demand overnight. A single fault can cascade into blackouts affecting millions. Today, human operators make split-second decisions about battery dispatch, load shedding, and fault isolation under pressure, with incomplete information.

AI Custodire trains RL agents to do this automatically. An agent that masters microgrid management could prevent blackouts, reduce energy waste, and accelerate the clean energy transition. This is not a toy problem - utilities like ERCOT, NTPC, and Adani Green face these exact challenges daily.

Environment Description

A simulated microgrid with solar generation, battery storage, variable load demand, and circuit switching. The agent acts as an autonomous grid operator, balancing power supply and demand while responding to faults and optimizing operational cost.

Action Space

Field	Type	Range	Description
battery_dispatch	float	-10 to +10 MW	Positive = discharge (supply power), Negative = charge (store power)
load_shed	float	0.0 to 1.0	Fraction of load to curtail (0 = none, 1 = cut all)
switch_cmd	int	0-3	0=no change, 1=open seg1, 2=open seg2, 3=restore all

Observation Space

Field	Type	Normal Range	Description
voltage_pu	float	0.95-1.05	Grid voltage in per-unit
frequency_hz	float	49.5-50.5	Grid frequency in Hz
load_mw	float	2-15 MW	Total demand
solar_mw	float	0.5-10 MW	Solar generation
battery_soc	float	0-1	Battery state of charge
netbalancemw	float	-	Generation minus demand
fault_active	bool	-	Active fault present
fault_segment	int	0-2	Faulted segment (0=none)
seg1_energized	bool	-	Segment 1 powered
seg2_energized	bool	-	Segment 2 powered

Tasks

Task	Difficulty	Steps	Objective
load_balance	Easy	20	Keep netbalancemw near zero across variable solar and load
fault_recovery	Medium	30	Detect fault at step 6, isolate faulted segment, restore grid
optimal_dispatch	Hard	40	Minimize battery cycling cost while maintaining voltage and frequency stability

Reward Design

Rewards are dense every step - agents get continuous feedback, not just end-of-episode:

—load_balance: 0-1 per step based on balance quality. Penalizes load shedding and active faults
—fault_recovery: -0.4 per step with unhandled fault. +0.4 bonus for correct isolation. 0-0.6 for post-recovery stability
—optimal_dispatch: Stability score minus operational cost. Clamped to [0.0, 1.0]

Setup

pip install -e . uvicorn server.app:app --host 0.0.0.0 --port 7860 --reload

Docker

docker build -t microgrid-env . docker run -d -p 7860:7860 microgrid-env

API Endpoints

Endpoint	Method	Description
/health	GET	Health check
/reset	POST	Reset with task name
/step	POST	Execute action
/state	GET	Current state
/tasks	GET	List tasks with graders
/grader	POST	Score a task episode
/metadata	GET	Environment metadata
/schema	GET	Action and observation schemas

Baseline Scores (Qwen2.5-72B-Instruct)

Task	Score	Success
load_balance	0.920	true
fault_recovery	0.351	true
optimal_dispatch	0.776	true