MUMBAI, India, Aug. 17 -- Intellectual Property India has published a patent application (202621074079 A) filed by Ashokrao Mane Group Of Institutions, An Autonomous Institute, Kolhapur on June 15, 2026, for “predictive Traffic Orchestration Using Temporal Fusion Transformers And Deep Q-Learning In Sdn”.
Inventors include Prof. Arati V. Patil; Prof. Shubhangi R. Patil; Dr. Chettan R. Dongarsane; Prof. Prachi G. Chavan; Prof. Mrunal M. Pawar; and Ms. Sanjana L. Mane.
The application for the patent was published on August 14, 2026, under issue no. 33/2026.
Abstract: A predictive-cognitive traffic orchestration system (100) and method (200) for software-defined networks is disclosed, in which traffic prediction, adaptive routing and centralized SDN control are unified within a single controller (110). A telemetry-ingestion subsystem (103) collects per-link statistics from SDN switches (101) over a southbound interface (102) and forms a normalised time-series. A forecasting engine (104) implementing a Temporal Fusion Transformer produces a calibrated, multi- horizon, per-link load forecast (302). Distinctively, this forecast is structurally injected into the state vector and reward function of a Deep Q-Learning cognitive routing engine (105), so that forwarding paths are selected to avoid links anticipated to congest before congestion occurs. A rule-compilation module (106) installs the resulting flow rules into the switches (101), and a feedback module (107) compares measured against predicted load to drive periodic re-training of both models, forming a self- correcting closed loop. The system is realised purely in software on commodity hardware and integrates with existing SDN frameworks, thereby achieving proactive congestion avoidance, improved bandwidth utilisation, and reduced latency and packet loss in dynamic networks such as data centres, ISP backbones, enterprise networks and smart-city IoT infrastructures. FIG. 1 is recommended for publication with the abstract.
Disclaimer: Curated by HT Syndication.
