Maximum Entropy-Based Traffic Generation

doi:10.1109/tnsm.2026.3712637

Maximum Entropy-Based Traffic Generation

2026-07-13

0

PRE

AI

DOI：10.1109/tnsm.2026.3712637

Original

Original request for help

Save

Abstract

En 中文

The development of machine learning models and algorithms for many communication network optimization problems has generated a huge need for realistic traffic data generators, as real-world traffic datasets remain very few, especially compared to their size. We therefore propose a novel traffic generation framework based on the Maximum Entropy Principle (MEP). It explicitly incorporates empirical statistical constraints, ensuring generated traffic closely mirrors the complex patterns found in real-world data. Using vehicle traffic datasets of the City of Calgary, we explore multiple distributional assumptions, namely Gaussian, exponential, and mixture models. Our results demonstrate that the Gaussian and the Gaussian mixture models consistently achieve superior performance, capturing diverse temporal fluctuations and intricate statistical behaviors inherent in urban vehicle traffic. This study not only highlights the effectiveness and flexibility of MEP-based models but also establishes them as robust, interpretable, and data-efficient alternatives to existing generative methods in traffic synthesis.

Keywords:

Time series dataset

maximum entropy principle

traffic modeling

synthetic traffic generation

Maximum Entropy-Based Traffic Generation

Abstract

Journal

Organization