Real-Time Load Profiling and Operational Pattern Recognition in a Smelting Industry

Authors:

Abstract
This study presents a novel three-stage analytical framework, combining data preprocessing, load profiling, and operational pattern recognition, specifically designed for the high-power, nonlinear environment of silicon smelting substations. Unlike existing approaches that focus on equipment-level analytics or generic industrial loads, the proposed framework provides a unified, substation-level diagnosis capable of quantifying operational stability, variability, and functional roles within the smelting process. Application of the framework to four substations revealed distinct numerical performance signatures: Substation_1 exhibited highly stable behavior with Load Factor 1.0, and minimal variability; Substation_2 showed extreme volatility (Load Factor (LF) 0.22–0.35, Variability Index (VI) > 3000) and clear dual daily peaks at 10:00 and 14:00; Substation_3 demonstrated more efficient operation (Load Factor 0.55–0.72, Peak-to-Base Ratio (PBR) 2.5–4); and the Smelter Outlet operated as a continuous base-load unit with Load Factor 0.62–0.70 and near-zero Peak-to-Base Ratio while sustaining 20–22 MW during production. Variability-band and clustering analyses further distinguished the stable, furnace-driven load of the smelter from the flexible, schedulable auxiliary loads. The key novelty of this work lies in linking quantitative temporal load signatures to functional process roles, offering a structural view of energy behavior across smelting infrastructure. The findings provide actionable insights: flexible substations can support real-time demand response, substations with Power Factors fluctuations (0.3–1.0) require targeted reactive compensation, and grid operators gain improved visibility for forecasting and stability management, thereby strengthening grid–industry integration.
Funding Information
  • Science and Technology Projects from State Grid Corporation (5400-202316228A-1-1-ZN)

This publication has 26 references indexed in Scilit: