Multi-model hybrid traffic flow forecast algorithm based on multivariate data

Abstract

Traffic flow forecast is a fine-grained task in urban intelligent transportation systems. Accurate traffic flow forecast can effectively support the development of intelligent transportation systems, reduce congestion, and improve the quality of residents’ travel. The forecast of traffic flow is affected by many random factors such as weather, holidays and seasons. It has a certain degree of randomness and uncertainty, which makes the traditional single model prediction result extremely unstable, and the consideration of random factors is incomplete. As a result, the final forecast results are quite different from the actual situation. To address this problem, this paper proposes a multi-model hybrid traffic flow forecast algorithm based on multivariate data, which considers various random factors from multiple aspects, and captures different features through multiple models to improve the accuracy. The experiments on the dataset of KDD CUP 2017 demonstrate the effectiveness of our approach.