MUMBAI, India, July 7 -- Intellectual Property India has published a patent application (202641079050 A) filed by Sr University on June 25, 2026, for A Hybrid Deep Learning System For Improved Time Series Forecasting Using Integrated Cnn–lstm–attention Architecture.
Inventors include S T V S S Yadunandan; Dr. Sudersan Behera; and Dr. Sarat Chandra Nayak.
The application for the patent was published on July 03, 2026, under issue no. 27/2026.
Abstract: A HYBRID DEEP LEARNING SYSTEM FOR IMPROVED TIME SERIES FORECASTING USING INTEGRATED CNN– LSTM–ATTENTION ARCHITECTURE The present invention relates to a hybrid deep learning system for improved time-series forecasting using an integrated CNN–BiLSTM–Attention architecture. The system comprises a data preprocessing module, a convolutional neural network feature extraction module, a bidirectional long short-term memory temporal learning module, an attention-based weighting module, and a forecasting output module. The preprocessing module handles missing values, irregular sampling intervals, noise reduction, and normalization of input data. The CNN module extracts localized spatial correlations, while the BiLSTM module captures forward and backward temporal dependencies. The attention mechanism dynamically prioritizes influential time steps, seasonal effects, anomalies, and contextual patterns to improve prediction accuracy and interpretability. The system generates accurate point and interval forecasts for complex multivariate time-series datasets and can be deployed across cloud, fog, and edge computing environments. The invention provides robust, scalable, interpretable, and real-time forecasting suitable for finance, healthcare, environmental monitoring, industrial IoT, and smart-grid applications
Disclaimer: Curated by HT Syndication.
