Deep learning driven blocking prediction in elastic optical networks using spatio-temporal hybrid neural network
(English) The explosive growth of bandwidth-intensive services such as ultra-high-definition video streaming, cloud gaming, and virtual reality is pushing optical transport networks toward greater flexibility and efficiency. Elastic Optical Networks (EONs) address these demands by allocating spectru...
| Autor: | |
|---|---|
| Tipo de documento: | tese |
| Data de publicação: | 2026 |
| País: | España |
| Recursos: | Universitat Politècnica de Catalunya (UPC) |
| Repositório: | UPCommons. Portal del coneixement obert de la UPC |
| Idioma: | inglês |
| OAI Identifier: | oai:dnet:upcommonspor::0d92478ad77a4782eebba8ef4e84c572 |
| Acesso em linha: | https://hdl.handle.net/2117/461924 https://dx.doi.org/10.5821/dissertation-2117-461924 |
| Access Level: | Acceso aberto |
| Palavra-chave: | elastic optical networks convolutional neural networks sequential data blocking prediction spatio-temporal modeling CNN–LSTM CNN–BiLSTM deep learning spectrum allocation 621.3 - Enginyeria elèctrica. Electrotècnia. Telecomunicacions Àrees temàtiques de la UPC::Enginyeria de la telecomunicació |
| Resumo: | (English) The explosive growth of bandwidth-intensive services such as ultra-high-definition video streaming, cloud gaming, and virtual reality is pushing optical transport networks toward greater flexibility and efficiency. Elastic Optical Networks (EONs) address these demands by allocating spectrum in variable-sized frequency slots rather than fixed-grid channels. However, the dynamic setup and teardown of heterogeneous lightpaths fragment the optical spectrum, often leading to connection blocking even when sufficient total capacity exists. Accurate early prediction of blocking events enables network controllers to trigger proactive spectrum defragmentation and resource reallocation, improving service continuity and network utilization. This dissertation investigates deep learning-based approaches for blocking prediction in EONs using network state information represented as spectrum occupancy matrices. First, we introduce one-dimensional and two-dimensional Convolutional Neural Networks (1D-CNN and 2D-CNN) to automatically extract spatial fragmentation patterns from simulated EON topologies. The proposed 2D-CNN achieves a prediction accuracy of 92.17\% and outperforms conventional machine learning baselines such as Support Vector Machines and k-Nearest Neighbors. Building on this, we propose hybrid spatio-temporal architectures that integrate convolutional feature extraction with Long Short-Term Memory (LSTM) and Bidirectional LSTM (BiLSTM) units to capture both spatial and temporal evolution of spectrum fragmentation. The CNN–BiLSTM model achieves 94.1\% prediction accuracy while maintaining reasonable computational complexity, and the CNN–LSTM model offers a favorable trade-off between accuracy and training speed. By unifying spatial and temporal deep learning techniques, this work provides a comprehensive framework for early and reliable blocking prediction in dynamic EON environments. The proposed models can be integrated into elastic optical network controllers to enable timely defragmentation actions, reducing blocking probability and improving overall network efficiency. The findings advance the application of artificial intelligence to optical networking and lay the foundation for future integration with intelligent routing, spectrum assignment, and network automation strategies. |
|---|