1 Application of ANFIS in Transportation Problems Ali Gholami, PhD Student Center for Advanced Transportation Education and Research University of Nevada, Reno Weekly Seminar Nov 13, 2014 2 Fuzzy Sets Center for Advanced Transportation Education and Research University of Nevada, Reno Project Panel Meeting Weekly Seminar Nov 13, 2014 March 27, 1012 3 How to present a fuzzy set in a computer Center for Advanced Transportation Education and Research University of Nevada, Reno Project Panel Meeting Weekly Seminar Nov 13, 2014 March 27, 1012 4 Fuzzy rules If x is A Then y is B For example: Rule 1: Rule 2: IF Green time is > 30 sec IF Green time is < 10 sec AND It is during peak-hour THEN “traffic_volume” is low THEN “traffic_volume” is high Center for Advanced Transportation Education and Research University of Nevada, Reno Project Panel Meeting Weekly Seminar Nov 13, 2014 March 27, 1012 5 Fuzzy inference A process of mapping from a given input to an output, using the theory of fuzzy sets Mamdani-style inference Sugeno-style inference Fuzzification Center for Advanced Transportation Education and Research University of Nevada, Reno Rule evaluation Aggregation of the rule outputs Defuzzification Project Panel Meeting Weekly Seminar Nov 13, 2014 March 27, 1012 6 Fuzzy inference: Mamdanistyle inference Center for Advanced Transportation Education and Research University of Nevada, Reno Project Panel Meeting Weekly Seminar Nov 13, 2014 March 27, 1012 7 Fuzzy inference: Sugenostyle inference Center for Advanced Transportation Education and Research University of Nevada, Reno Project Panel Meeting Weekly Seminar Nov 13, 2014 March 27, 1012 8 Neural network Center for Advanced Transportation Education and Research University of Nevada, Reno Project Panel Meeting Weekly Seminar Nov 13, 2014 March 27, 1012 Summation Defuzzification ANFIS Normalisation Rules Fuzzification Inputs 9 Adaptive Neuro-Fuzzy Inference System A fuzzy model with a systematic approach to generate fuzzy rules using neural network Center for Advanced Transportation Education and Research University of Nevada, Reno Project Panel Meeting Weekly Seminar Nov 13, 2014 March 27, 1012 10 Case Study: Incentive Mean Absolute Percentage Error 80 60 40 20 0 Thru Left Thru Left Thru Left Thru Left NB SB EB WB Kietzke and Moana, Reno, NV Center for Advanced Transportation Education and Research University of Nevada, Reno SB WB NB EB Sparks and Prater, Sparks, NV Project Panel Meeting Weekly Seminar Nov 13, 2014 March 27, 1012 11 Research Question Can turning volumes be estimated based on signal information without using loop detector data? Center for Advanced Transportation Education and Research University of Nevada, Reno Project Panel Meeting Weekly Seminar Nov 13, 2014 March 27, 1012 12 Methodology Modeling the intersection in VISSIM Center for Advanced Transportation Education and Research University of Nevada, Reno Developing a code in COM interface (VISSIM tool) with different parameters Making a prediction model for the data Project Panel Meeting Weekly Seminar Nov 13, 2014 March 27, 1012 13 Data Analysis Center for Advanced Transportation Education and Research University of Nevada, Reno Project Panel Meeting Weekly Seminar Nov 13, 2014 March 27, 1012 14 Modeling Center for Advanced Transportation Education and Research University of Nevada, Reno Project Panel Meeting Weekly Seminar Nov 13, 2014 March 27, 1012 15 R Center for Advanced Transportation Education and Research University of Nevada, Reno Project Panel Meeting Weekly Seminar Nov 13, 2014 March 27, 1012 16 Accuracy 𝑀𝐴𝑃𝐸 % = 𝑛 𝐷𝑖 −𝐵𝑖 𝑖=1 𝐵 𝑖 𝑛 MAPE: Mean Absolute Percentage Error Di: the detector data value Bi: the reference (base) data value n: the total number of intervals Center for Advanced Transportation Education and Research University of Nevada, Reno Project Panel Meeting Weekly Seminar Nov 13, 2014 March 27, 1012 17 Case Study Intersections 9th St. and Sierra St. 8th Street and Center Street N McCarran Blvd and Clear Acre Ln Center for Advanced Transportation Education and Research University of Nevada, Reno Project Panel Meeting Weekly Seminar Nov 13, 2014 March 27, 1012 18 Results Center for Advanced Transportation Education and Research University of Nevada, Reno Project Panel Meeting Weekly Seminar Nov 13, 2014 March 27, 1012 19 Conclusion - • Needs additional effort for learning the ANFIS concept - • Needs more data records to have better training and data preparation + • More flexibility for modeling complex data sets + • More accuracy especially when data pattern is not clear + • Ready toolbox in MATLAB Center for Advanced Transportation Education and Research University of Nevada, Reno Project Panel Meeting Weekly Seminar Nov 13, 2014 March 27, 1012 20 Thank you for not being like these! Any comment or question? Center for Advanced Transportation Education and Research University of Nevada, Reno Project Panel Meeting Weekly Seminar Nov 13, 2014 March 27, 1012