Adaptive neuro-fuzzy approach for modeling equilibrium speed–density relationship

Abstract

ABSTRACTThis paper endeavors to model the equilibrium speed–density relationship using a new fuzzy logic-based approach. To capture the randomness in traffic flow dynamics, it develops a single-input Adaptive Neuro-Fuzzy Inference System (ANFIS) trained with a hybrid algorithm. Furthermore, it proposes a new Premise-Consequent Conjugate Effect (PCCE) relationship to estimate fundamental diagram (FD) parameters from the ANFIS model. Several options (i.e. optimal split of data into training and testing data, selection of suitable membership function) offered by ANFIS are then illustrated. An experiment is performed to determine the maximum achievable goodness of fit without the occurrence of overfitting phenomenon by changing the number of clusters. The calibrated ANFIS model is compared against traditional and advanced speed–density models for five different freeway locations. Results show that the proposed model outperforms other preceding models by attaining goodness-of-fit values within a range of 0.82–0.92. Finally, the proposed PCCE relationship shows the ANFIS model's robustness in accurate estimation of FD parameters for all freeway locations.