TRAMESINO: Traffic Memory System for Intelligent Optimization of Road Traffic Control ? Cristian Axenie 1 , Rongye Shi 2 ( ), Daniele Foroni 1 , Alexander Wieder 1 , Mohamad Al Hajj Hassan 1 , Paolo Sottovia 1 , Margherita Grossi 1 , Stefano Bortoli 1 , and G¨otz Brasche 1 1 Intelligent Cloud Technologies Lab, Huawei Munich Research Center Riesstrasse 25, 80992 Munich, Germany cristian.axenie@huawei.com 2 EI Intelligence Twins Program, Huawei Cloud BU, Shenzhen, China shirongye@huawei.com Abstract. Whether efficient road traffic control needs accurate mod- elling is still an open question. Additionally, whether complex models can dynamically adapt to traffic uncertainty is still a design challenge when optimizing traffic plans. What is certain is that the highly non- linear and unpredictable real-world road traffic situations need timely actions. This study introduces TRAMESINO (TRAffic Memory System INtelligent Optimization). This novel approach to traffic control mod- els only relevant causal action-consequence pairs within traffic data (e.g. green time - car count) in order to store traffic patterns and retrieve plausible decisions. Multiple such patterns are then combined to fully describe the traffic context over a road network and recalled whenever a new, but similar, traffic context is encountered. The system acts as a memory, encoding and manipulating traffic data using high-dimensional vectors using a spiking neural network learning substrate. This allows the system to learn temporal regularities in traffic data and adapt to abrupt changes, while keeping computation efficient and fast. We evaluated the performance of TRAMESINO on real-world data against relevant state- of-the-art approaches in terms of traffic metrics, robustness, and run- time. Our results emphasize TRAMESINO’s advantages in modelling traffic, adapting to disruptions, and timely optimizing traffic plans. 1 INTRODUCTION Solving traffic congestion in urban agglomerations is still a problem resistant to straightforward solutions despite the large amount of research and systems developed to analyze [20], model [23], and control road traffic [26]. Systems de- ployed in real-world [10,14,6] use a traffic model [24,7] that heavily influences the run-time performance of the overall system. Basically, the role of the traffic model is to describe the dynamics of the traffic flow and to cope, eventually, with ? C. Axenie, R. Shi, D. Foroni, A. Wieder, M. A. H. Hassan, P. Sottovia, M. Grossi and S. Bortoli—Authors contributed equally to this research.