MUMBAI, India, Nov. 2 -- Intellectual Property India has published a patent application (202641123916 A) filed by Vellore Institute Of Technology on October 16, 2026, for A Regime-Aware Multi-Station Pm2.5 Forecasting And Severe-Smog Early Warning System.
Inventors include Krishnamoorthy A; and Lakshmi S.
The application for the patent was published on October 30, 2026, under issue no. 44/2026.
Abstract: The present invention provides an artificial-intelligence-based, sensor-quality-aware and regime-aware PM2.5 forecasting system for multi-station urban air-quality monitoring. Multi-station sensor streams (101) comprising PM2.5, CO, RH and time information are converted into hourly station-level records through quality control (102), while missingness and temporal gaps are detected (103). Gap-aware sliding windows (104) are generated from contiguous segments using a 72-hour input window. Train-only normalization and target transformation (105) produce a feature tensor (106), which is processed using station embedding and a shared temporal encoder (107). A routine-pollution Expert A (108) and peak-pollution Expert B (110) generate respective forecasts, which are dynamically combined using a learned gate (109) according to (y=(1-g)y_A+gy_B) (111). Peak-aware learning (115) assigns higher loss weights to severe PM2.5 target positions. The system generates a next-24-hour PM2.5 forecast (113) and a severe-smog alert (114) when a predicted concentration exceeds a threshold. No forecasting sample is generated across a temporal discontinuity, while missingness indicators distinguish observed and reconstructed covariates (112).
Disclaimer: Curated by HT Syndication.
