⚡ Eco-Agentic Micro-Grid Negotiator
📌 Project Overview
The Eco-Agentic Micro-Grid Negotiator is a high-fidelity Reinforcement Learning environment designed to simulate a Decentralized Neighborhood Energy Market. The environment challenges an AI agent to act as a "Smart Home Controller" that must balance energy production, consumption, and economic trade-offs in real-time.
🎯 Problem Statement
In a decentralized grid, energy availability is volatile (solar-dependent) and demand is fluctuating. The agent must learn an optimal policy to:
Maximize Financial Gain: Sell excess energy when market prices are high.
Ensure Grid Stability: Use battery storage to prevent "blackouts" during low-production periods.
Promote Sustainability: Minimize reliance on the primary carbon-heavy grid by optimizing local solar usage.
⚙️ Technical Implementation
Framework: Built using the OpenEnv specification and Gymnasium API.
Observation Space: A 5-dimensional continuous vector consisting of:
Current Hour (Temporal state)
Solar Production (Environmental variable)
House Demand (Stochastic load)
Battery State of Charge (Internal state)
Market Price (Economic variable)
Action Space: A discrete space of 4 agentic decisions: Charge, Discharge, Sell, and Buy.
Reward Shaping: A multi-objective reward function that balances profit, energy security, and green energy utilization.
🚀 Complexity Metrics
Temporal Dynamics: The environment uses a sinusoidal function to simulate solar cycles, forcing the AI to learn "diurnal patterns."
Economic Volatility: Market prices fluctuate based on a time-of-day curve, requiring the agent to perform "Arbitrage" (buying low, selling high).
Constraint Satisfaction: The agent must manage a hard constraint (Battery Capacity) while avoiding a critical failure state (Blackout).
