Year: 2026
Type: article
Abstract: Electricity distribution utilities must examine large volumes of smart-meter readings to find consumption patterns that may indicate non-technical losses, yet unusual readings can equally arise from changes in occupancy, seasonality, solar generation, meter faults or data errors. Published work concentrates mainly on detection models, whereas the workflow around data quality, explanation and human review is less often documented as one reproducible system. This paper presents GridIntel, a web-based platform that integrates comma-separated-value (CSV) data ingestion, data-quality auditing, sixteen customer-level statistical features, feeder-month energy-balance analysis, Isolation Forest anomaly scoring and, when sufficient labels exist, Random Forest, XGBoost and ensemble scoring. Scores are expressed as cohort-relative percentile rankings with feature-based explanations and feed an investigation workflow with case snapshots, an append-only audit trail and CSV and portable-document-format (PDF) reports. The platform also provides a separate past-bill analysis using browser-side optical character recognition (OCR) and a dataset-grounded assistant. It is implemented with React, TypeScript, FastAPI and Python analytics libraries. In a recorded verification run, 263 of 264 backend tests and all 17 frontend unit tests passed, one opt-in live-Ollama test was skipped, the production build succeeded, and a manual 12-route browser smoke test showed no console errors or warnings. No independent labelled utility dataset was used, so no claim is made about real-world detection accuracy. The contribution is an integrated software architecture and investigation workflow, not a new learning algorithm. Keywords— Smart Grid Analytics; Non-Technical Loss Detection; Electricity Theft Investigation; Anomaly Detection; Explainable Machine Learning; Data Quality Assessment.
Institution: Nagpur Institute of Technology
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7237346239
