04 / SELECTED WORKAI / Multi-agent systems
Smart home energy.
Smart Home Energy Management System
A multi-agent smart home energy system in JADE coordinating five specialized agents with an LSTM load forecaster reaching 92% accuracy.
MULTI-AGENT ENERGY MANAGEMENT
Load forecastingDemand
responseJADE5 AGENTSP2P
tradingBehavioral clustering
responseJADE5 AGENTSP2P
tradingBehavioral clustering
92%
LSTM forecasting accuracy
75% traditional baseline
01 Overview
A multi-agent smart home energy management system built in JADE, coordinating five specialized agents for load forecasting, demand response, P2P trading, and behavioral clustering. Integrated an LSTM hourly load forecaster benchmarked against Random Forest and XGBoost.
02 What I built
- Multi-agent system in JADE with five specialized agents: load forecasting, demand response, P2P trading, behavioral clustering
- LSTM hourly load forecaster reaching 92% accuracy vs. 75% traditional baseline
- Benchmarking pipeline comparing LSTM against Random Forest and XGBoost
- MySQL-backed data layer and TensorFlow model integration
03 Engineering decisions
- LSTM forecaster achieved 92% accuracy vs. 75% traditional baseline — a 17pp improvement
- Used JADE multi-agent architecture to model independent agent behaviors and coordination
- Benchmarked against Random Forest and XGBoost to validate LSTM performance gains