Smart Campus Energy Management System using Federated Reinforcement Learning

Project Abstract

This study presents a novel Smart Campus Energy Management System (SCEMS) that leverages Federated Reinforcement Learning (FRL) to optimize energy consumption across diverse campus facilities while preserving data privacy and reducing operational costs. The proposed framework integrates edge devices, building management systems, and cloud-based coordination to create a distributed learning environment where agents at individual buildings learn consumption control policies locally and share model updates rather than raw data. This approach mitigates privacy concerns inherent in centralized data aggregation while enabling cross-building knowledge transfer to accelerate convergence and improve generalization under varying occupancy patterns, weather conditions, and utility tariffs. The FRL architecture employs hierarchical coordination with building-level agents responsible for electrical loads (HVac, lighting, climate control, and plug loads) and a campus-wide aggregator that synthesizes gradient information to refine a global policy without exposing sensitive data. A combination of model-free and model-based reinforcement learning techniques is explored to balance sample efficiency with policy robustness in dynamic campus scenarios. We formulate the management problem as a constrained Markov decision process, where the objective is to minimize total energy cost and peak demand charges while maintaining occupant comfort and building envelope integrity. The system incorporates predictive control using time-series forecasting for weather and occupancy, and online adaptation to fluctuating energy prices and renewable generation. To evaluate performance, we develop a comprehensive simulation environment calibrated with real campus data, complemented by a small-scale testbed on one or two buildings to validate feasibility. Baseline comparisons include conventional rule-based EMS, centralized RL, and non-FRL distributed approaches. Key performance metrics include total energy cost reduction, peak demand shaving, CO2 emissions, occupant comfort indices, training efficiency, model communication overhead, and privacy leakage resistance. Experimental results demonstrate that FRL achieves superior cost savings and peak shaving compared to centralized and non-federated baselines, while maintaining comparable occupant comfort and significantly reducing data exposure through local training and encrypted model exchanges. Sensitivity analyses reveal the resilience of the FRL framework to non-stationary occupancy, inaccurate forecasts, and partial device failures. The study also investigates communication strategies, such as selective parameter sharing and quantization, to further reduce bandwidth requirements without compromising policy quality. Practical deployment guidelines are provided, including integration steps with existing campus infrastructure, data governance policies, and privacy-preserving techniques. The research contributes to the fields of smart buildings and federated learning by demonstrating a scalable, privacy-aware, and energy-efficient solution for campus-wide energy management, offering a blueprint for broader adoption in university and corporate campuses seeking to optimize energy performance while safeguarding sensitive operational data.

Project Overview

What This Project Is About

---

A plain-language overview of using smart technologies to manage a university campus’s energy use. The project explores how rooms, buildings, and utilities can learn to use electricity more efficiently by sharing information without exposing private data, and how this can reduce waste and costs.

---

---

The Problem It Addresses

---

Many campuses waste energy because building systems operate with limited coordination and run on fixed schedules. Centralized controls raise privacy and data-sharing concerns. The project investigates an approach where each building learns locally and only shares learning updates, not raw data, to improve overall energy use while protecting privacy.

---

---

Objectives of the Project

---

1. Understand how campus energy systems work (lighting, HVAC, sensors).
2. Explain federated learning and why it helps with privacy.
3. Develop a simple model that optimizes energy use across buildings.
4. Test how the model adapts to changing campus occupancy and weather.
5. Evaluate energy savings, comfort, and system reliability.

---

---

What You Will Do Step by Step

---

1. Study background materials on energy management and federated learning.
2. Collect or simulate campus energy data (without sharing private details).
3. Build a basic local controller for each building and set up the federated framework.
4. Train the learning model with local updates and aggregate them securely.
5. Run experiments to compare with non-federated methods.
6. Analyze energy savings, peak demand reduction, and user comfort metrics.
7. Document results and discuss practical deployment considerations.
8. Prepare a final report and presentation.

---

---

Expected Outcome

A working, privacy-preserving energy management approach that can reduce campus electricity use while keeping occupants comfortable, along with practical guidance for real-world deployment.