EnergyX: An Agentic Framework for Explainable Energy Forecasting and Anomaly Detection

Sanyat Fale, Naman Srivastava, Pandarasamy Arjunan

Presentation Attendance: No, we cannot present in-person

Keywords: Time Series Forecasting, Anomaly Detection, Agentic, Explainable AI

Abstract: Energy time series forecasting and anomaly detection are critical for grid stability, demand response, and operational reliability. However, deploying effective solutions remains challenging due to complex model selection, hyperparameter tuning, and limited explainability, which restrict accessibility for domain users. We present EnergyX, an agentic framework that translates natural language user intent into executable forecasting and anomaly detection workflows. EnergyX performs automated data profiling, model selection with cross-validated performance-based aggregation, and generates explanations grounded in statistical diagnostics, feature attribution methods (e.g., SHAP, feature importance), and counterfactual analysis The system integrates decomposition, lag analysis, residual diagnostics, and anomaly consensus scoring to provide transparent and actionable insights. Through forecasting and anomaly detection scenarios, we demonstrate that EnergyX enables accurate, explainable, and user-steerable time series analysis, highlighting the potential of LLM-driven agentic systems for accessible energy analytics.

Track: Research Track (max 4 pages)

Submission Number: 97