Streamlined Traffic Flow Uncertainty Quantification Based on Interval Type-II Fuzzy Inference System with Gradient Descent Optimization

Uncertainty quantification is essential for building reliable traffic management and control applications, with current methods broadly classified into distribution-based methods that rely on probability distribution function and nondistribution-based methods that depend on other theoretical instruments. Among nondistribution-based methods, interval Type-II fuzzy inference system–based methods are promising in modeling uncertainty in traffic condition data. However, these methods generally lack a mechanism for parameter adaptation to accommodate evolving traffic patterns. Therefore, in this paper, an interval Type-II fuzzy inference system–based traffic flow uncertainty quantification method is introduced that can dynamically adjust model parameters via the gradient descent algorithm. First, an interval Type-II fuzzy inference system is proposed to model traffic flow uncertainty structure. Subsequently, built upon this structure, an uncertainty quantification method is developed that can produce prediction intervals using streaming traffic flow data. In the empirical study, real-world highway traffic flow data are used to validate the proposed method. Experimental results reveal that the proposed method maintains workable uncertainty prediction performance with respect to different time of day and traffic flow levels. In addition, the proposed method demonstrates a better performance compared with its Type-I counterpart. Discussions are provided on uncertainty quantification, its characteristics, and applications in transportation systems.