Intellectual Property India Publishes Patent Application for "ai-Driven Multimodal Traffic Prediction And Adaptive Route Optimization System Using Transformer Networks, Hypergraph Convolutions, Spiking Neural Networks, And Meta-Learning Reinforcement Learning''

MUMBAI, India, Sept. 22 -- Intellectual Property India has published a patent application (202521023408 A) filed by Ms. Sweta Kahurke on March 17, 2025, for "ai-Driven Multimodal Traffic Prediction And Adaptive Route Optimization System Using Transformer Networks, Hypergraph Convolutions, Spiking Neural Networks, And Meta-Learning Reinforcement Learning''.

Inventors include Ms. Sweta Kahurke; and Prof. Dr. Ktv Reddy.

The application for the patent was published on September 18, 2026, under issue no. 38/2026.

Abstract: ABSTRACT OF INVENTION: The present invention relates to an advanced multimodal traffic prediction and optimization system that integrates multiple machine learning and deep learning architectures to improve reaMime traffic forecasting, adaptive route optimization, and uncertainty estimation. The system utilizes a Multimodal Transformer Joint Embedding Network (MT-JEN) to fuse heterogeneous transportation data, including traffic sensors, GPS signals, pedestrian movement, and real-time weather conditions. By leveraging self-attention mechanisms, MT-JEN effectively aligns multimodal features, ensuring a cohesive representation of spatial-temporal traffic conditions. To model complex transportation interactions, the fused embeddings are processed by a Hypergraph Spatial-Temporal Graph Convolutional Network (HG-STGCN), which represents the traffic system as a hypergraph to capture higher-order dependencies between road segments, intersections, and various transportation modes. This enables a more granular understanding of traffic flow patterns, making it robust to missing or noisy data samples. For real-time congestion forecasting, the system employs a Spiking Temporal Convolutional Network (S-TCN), which integrates event-driven spiking neurons to efficiently capture traffic fluctuations. This approach reduces computational overhead while improving the responsiveness of congestion predictions. Based on real-time forecasts, an adaptive reinforcement learning model using Meta-Learning Deep Deterministic Policy Gradient (MAML-DDPG) is employed to dynamically optimize route planning. By leveraging meta-learning principles, the model quickly adapts to new cities and road networks, significantly reducing training time while improving traffic management efficiency. To enhance prediction reliability, the system incorporates a Neural Ordinary Differential Equation Uncertainty Estimator (NODE-UE), which models traffic dynamics as a continuous-time process to generate probabilistic congestion forecasts with high-precision uncertainty bounds. This integrated Al-driven system offers improved prediction accuracy (92-95%), optimized travel times (20-25% reduction), and increased computational efficiency (35% lower cost), making it a scalable solution for smart cities and next-generation transportation networks.

Disclaimer: Curated by HT Syndication.