Adaptive Q-Learning for Safety Message Priority in Vehicular Fog Computing Based on Indonesia Transportation and Weather Data

Abstract

The development of Intelligent Transportation Systems demands fast, adaptive, and reliable communication mechanisms between vehicles, especially when the system has to process multiple emergency events simultaneously. This article proposes an adaptation of the Adaptive Q-Learning model for the priority dissemination of safety messages in the Vehicular Fog Computing environment by utilizing the context of Indonesian data. Adaptation was made to the Q-learning design based on incident priority, but modified in terms of state, action, and reward variables to match the characteristics of Indonesia's urban transportation, especially DKI Jakarta and national data. The data sources mapped in the model come from the Central Statistics Agency's Land Transportation Statistics, Jakarta Statistics Profile, and BMKG Open Weather Forecast Data. State agents are built from a combination of bucket delay, traffic density level, weather conditions, trust node scores, and types of safety events such as ambulances, accidents, road hazards, and inundation. Action space is represented as a choice of different Quality of Service weights to balance delay, packet delivery ratio, trust, and energy efficiency. The reward function is designed to give higher priority to stacked emergencies while penalizing delays and energy consumption. The results in the tables and graphs in this article are presented as an illustrative simulation based on a methodology design, not the results of direct field tests. With this approach, this article offers a relevant, original, and contextual research framework for the development of intelligent transportation systems in Indonesia.