Development of Uboard: A Real-Time IoT Analytics Platform for Smart Energy Management

This capstone project addresses the critical gap in industrial manufacturing where traditional energy monitoring remains reactive and disconnected from operational planning. To transition facilities from reactive observation to proactive optimization, this project develops Uboard, a "schedule-aware" smart energy management platform designed for the manufacturing sector. While the project was initially framed around continuous time-series architectures (such as SARIMA), initial data analysis revealed that the manufacturing datasets are fundamentally job-based rather than purely time-based. Consequently, the modeling strategy has pivoted to prioritize regression-based ensemble methods (Linear Regression, Random Forest, XGBoost, and LightGBM), with preliminary experiments yielding a peak R^2 score of 0.869 using the Random Forest Regressor. Upon progression of the project, the team has finalized the choice of model as a random forest model, as XGBoost and Linear regression proved to be lacking in terms of accuracy and had a tendency to show overfitting, due to the quantitative lack of provided resources. Tested against real-world manufacturing datasets, this dual-module pipeline delivers both technical reliability and actionable business metrics, establishing a scalable, reproducible framework aligned with Vietnam's national priority of AI-driven energy optimization and Net Zero sustainability goals.