Document Type
Student Coursework
Publication Date
Winter 11-1-2026
Abstract
Commercial and institutional buildings account for roughly one-third of global energy demand and nearly 37% of energy-related greenhouse gas emissions. While modern smart facilities deploy dense Internet of Things (IoT) sensor meshes to monitor power draw and environmental conditions, transforming continuous, heterogeneous streams into reliable analytical intelligence requires rigorous data engineering and precise semantic modeling. In this paper, we formulate, implement, and critically audit an end-to-end cloud pipeline deployed on Google Cloud Platform (GCP). The architecture adapts the Lambda design pattern, separating bulk historical ingestion from an asynchronous real-time streaming lane (Pub/Sub, Eventarc, 2nd-generation Cloud Functions) with an explicit dead-letter error quarantine, converging at a unified BigQuery analytical serving layer that strictly prevents duplicate message replay. Using a six-month empirical IoT dataset comprising 7,761 records (172 days) from the M5 Building at the University of Sharjah, we execute in-database time-series forecasting and unsupervised anomaly detection via BigQuery ML ARIMA_PLUS, coupled with automated natural-language narrative synthesis powered by Gemini 2.5 Flash-Lite through Vertex AI. A post-hoc forensic audit reveals a profound methodological constraint: the ingested energy series represents an unkeyed average of cumulative readings across seven sub-meters rather than true differential daily interval consumption. Consequently, the 14-day forecast extrapolates an accumulating running average, all nine detected statistical anomalies (p >= 0.95) coincide with changes in which sub-meters reported, with that attribution inferred from meter totals rather than directly recorded, and observed environmental correlations (temperature r = -0.371, humidity r = +0.349, motion-inferred occupancy r = -0.242) are strongly attenuated after removing the linear time trend, consistent with confounding by elapsed calendar time (r = 0.994 with energy). This investigation provides both a validated architectural blueprint for serverless IoT telemetry and a rigorous cautionary case study on how loss of sensor identity during ingestion can invalidate otherwise executable downstream analytics.
Program or Discipline Name
Computer and Information Sciences
Recommended Citation
Gudalwar, L., Kore, N., Wang, Z., & Cha, S. (2026). Smart Building IoT Analytics: An Integrated Cloud Pipeline and the Consequences of Cumulative-Meter Data Semantics for Forecasting and Anomaly Detection. Smart Building IoT Analytics: An Integrated Cloud Pipeline and the Consequences of Cumulative-Meter Data Semantics for Forecasting and Anomaly Detection, 1-12. Retrieved from https://digitalcommons.harrisburgu.edu/other-works/36
Publication Title
Smart Building IoT Analytics: An Integrated Cloud Pipeline and the Consequences of Cumulative-Meter Data Semantics for Forecasting and Anomaly Detection
Start Page No.
1
End Page No.
12
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Submission aggreement