Decentralised Regression Model for Intelligent Forecasting in Multi-agent Traffic Networks

The distributed nature of complex stochastic systems, such as traffic networks, can be suitably represented by multi-agent architecture. Centralised data processing and mining methods experience difficulties when data sources are geographically distributed and transmission is expensive or not feasible. It is also known from practice that most drivers rely primarily on their own experience (historical data). We consider the problem of decentralised travel time estimation. The vehicles in the system are modelled as autonomous agents consisting of an intellectual module for data processing and mining. Each agent uses a local linear regression model for prediction. Agents can adjust their model parameters with others on demand, using the proposed resampling-based consensus algorithm. We illustrate our approach with case studies, considering decentralised travel time prediction in the southern part of the city of Hanover (Germany).