Smart energy management system with AI suggestions

다국어 초록 (Multilingual Abstract)

The global imperative to reduce industrial carbon emissions and improve energy efficiency has intensified demand for intelligent energy management solutions. Factory environments, characterized by complex and heterogeneous energy consumption patterns across HVAC systems, industrial machinery, lighting, and production lines, represent one of the largest opportunities for energy optimization. This thesis presents the design, implementation, and evaluation of a Smart Energy Management System （SEMS） - a simulation-based platform that models factory energy profiles and employs machine learning techniques, specifically Random Forest （RF） Regression, Isolation Forest, and Random Forest, to generate actionable AI-powered ranked optimization suggestions. The system integrates a physics-grounded simulation engine that models eight device types - solar panels, wind turbines, batteries, HVAC units, lighting, EV chargers, appliances, and smart meters - driven by latitude-based weather and profile-aware occupancy models. The AI pipeline combines Random Forest （RF） Regression for multi-horizon load forecasting, a hybrid anomaly detector pairing four explicit rules with an Isolation Forest, a dual-model efficiency classifier （Random Forest for categorical labels and Random Forest （RF） for continuous scores）, and a tariff-driven load-shift optimizer. Evaluation results demonstrate that the regression model achieves a mean absolute percentage error （MAPE） of 4.7% on a 30-day holdout dataset, the clustering module identifies three distinct energy waste patterns leading to average simulated savings of 18.3% per factory zone, The suggestion engine produces ranked, deduplicated recommendations with estimated savings in both kWh and cost. A usability evaluation using the System Usability Scale （SUS） places the dashboard in the above-average to excellent range. The thesis contributes a replicable simulation framework, a validated ML pipeline for energy management, and an open-source dashboard implementation to the sustainable computing research community.

목차 (Table of Contents)

• Contents
• I. Introduction 1
• 1. Research Background 3
• 2. Research Importance 4
• 3. Research Motivations and Significance 6
• 4. Research Objectives 7
• 5. Research Questions 8
• 6. Scope of the Study 9
• II. Literature Review 11
• 1. Machine Learning for Energy Management Systems 16
• 2. AI for Industrial Energy Optimization 17
• 3. Digital Twins and Co-simulation for Energy Systems 18
• 4. Energy Management Dashboards and Visualisation 19
• 5. Limitations of Existing Approaches 20
• 6. Research Gap 21
• Ⅲ. Methodology 23
• 1. System Design and Architecture 23
• 2. Backend Service Design 28
• 3. Simulation Engine Design 29
• 4. AI Suggestion Pipeline 32
• 5. Data Strategy 35
• 6. Evaluation Metrics 37
• 7. Experimental Assumptions and Constraints 38
• IV. Results and Analysis 41
• 1. Forecasting Accuracy Results 41
• 2. Anomaly Detection Effectiveness Results 42
• 3. Efficiency Classification Quality Results 44
• 4. AI Suggestion Pipeline Results 45
• 5. System-Level Performance Results 47
• 6. Comparative Study Results 49
• 7. Discussion and Interpretation 55
• V. Conclusion 60
• 1. Summary of Findings 60
• 2. Contributions of Research 63
• 3. Limitations of the Study 64
• VI. References 66
• VII. Appendix 69
• 국문 초록 75