Abstract
It is still very difficult to exploit possibilistic concepts to identify the strict parameters of vehicular ad-hoc networks (VANET) while minimizing the dispersion of its constraints. In this paper, as a first contribution, an empirical study is presented within the framework of the development of a predictive model based on a linear regression model to interpret the phenomenon of VANET network congestion in a precise way by minimizing the dispersion of predicted outputs. The objective of the model developed is twofold: (1) to model the interactions between the traffic variables involved in the model and to exploit possibilistic concepts to identify the constraints of the VANET network based on a linear optimization under constraints, (2) to anticipate with precision the probability of occurrence of an unforeseen incident on VANET traffic. In addition, the traffic data involved in our model are observed vehicle speed, predicted travel time, observed travel time, and delay time. In addition, the accuracy of the prediction is proven by checking the relevance of the model according to the goodness of fit and the statistical significance of each explained variable. Our second contribution focuses on the design of a new technique for collecting, aggregating, and predicting real-time traffic flow. The main objective is to optimize the prediction error rate under strict conditions whose traffic parameters have unstable values. For this, we propose a traffic flow prediction approach based on fuzzy logic and regression analysis. The approach incorporates approved traffic parameters with the appearance of fuzzy least squares. To increase the prediction accuracy of the traffic flow and then arrive at a quadratic resolution under constraints, we integrate the Fuzzy Support Vector Regression for VANET networks (FSVRNET) model and Unified Particle Swarm Optimization (UPSO) algorithm in our approach. The chosen approach aims to model the interactions between traffic data observed from multiple data sources (e.g., connected loop detectors) to adjust the stability of traffic parameters in the prediction process. Our experimental study shows a strong correlation between the predicted data and the actual state of the VANET traffic flow. In addition, the prediction error of regression analysis is significantly reduced. Prediction performance using UPSO-FSVRNET is better with the proposed test set, as assumed by most identification methods.