MUMBAI, India, Sept. 28 -- Intellectual Property India has published a patent application (202641111011 A) filed by Pragati Engineering College on September 16, 2026, for Spatiotemporal Graph Neural Network For Short-Term Urban Traffic Congestion Prediction.
Inventors include Dr. V Anantha Lakshmi; Mr. G. Srinivasa Siva Kumar; Mr. Marukurthi Manishankar; and Mrs. P. Sandhya.
The application for the patent was published on September 25, 2026, under issue no. 39/2026.
Abstract: An IIoT sensor stream consists of data generated by various machine-related sensors, production-related sensors, electric-system sensors, environmental sensors, and industrial-control sensors. The streams of sensor data may include valuable information about the state of industrial equipment; however, the information contained in sensor streams can be abnormal in case of degraded equipment, malfunctioning sensors, communication problems, cyber attacks, unexpected conditions of operation, or process disturbances. Early detection of abnormalities can reduce the amount of time lost during equipment downtime and help to improve industrial safety, predictive maintenance, and maintain production quality. Traditional approaches to the problem of anomaly detection usually rely on threshold-based methods, assumptions about statistical properties of data or machine learning techniques. It is difficult for the traditional approaches to detect complicated and time-varying abnormalities in sensor data collected in multivariate industrial sensor streams. The goal of this paper is to propose an attention-based anomaly detection framework for industrial IoT sensor streams which integrates temporal modeling of sequences of events with the attention mechanism capable of assigning more weights to those sensor observations and time intervals which are involved in abnormalities. The proposed approach preprocesses sensor data collected continuously through data cleaning, normalization, filling-in missing values, and time windowing. Anomaly score is obtained for each time window, and the decision-making process adaptively determines whether the time window is normal or anomalous. The scheme can detect not only abrupt anomalies but also any anomalies that deviate from the normal operation pattern gradually. Experimental assessment of the model can be carried out by accuracy, precision, recall, F1-score, ROC-AUC, false alarm rate, and anomaly detection latency. The presented attention-based approach is intended to increase the detection performance of the system, but at the same time, to make it more understandable by highlighting what sensor properties and time intervals led to the anomaly detection.
Disclaimer: Curated by HT Syndication.
