A Multivariate Combined Traffic Flow Forecasting System Integrating Feature Selection and Multi-objective Optimization

Transportation plays a crucial role in guiding urban development and providing mobility strategies, with short-term traffic prediction serving as the foundation for traffic allocation. However, the increase in car ownership has intensified the randomness and uneven distribution of trips, presenting traffic problems as a significant challenge. To address that, a short time traffic flow forecasting system (STFFS) was developed in this study, utilizing a dual-stage data processing approach to remove redundancy and select the optimal input threshold. The optimizer was used in point forecasting (PF) and the comprehensive interval forecasting (IF) level. Results indicated that the algorithm outperforms traditional methods in prediction accuracy, stability, and sensitivity. Combination of Pareto mechanism, archiving, and roulette was theoretically proven to achieve global optimality. Traffic flow in three regions of Yunnan Province, China, confirmed that STFFS significantly improved accuracy, reduced sensitivity, and realized the prediction of determinacy and uncertainty, enhancing the practical application of prediction data. Lastly, the study indicated that long-term prediction can’t be accomplished by determination of historical data.