Please use this identifier to cite or link to this item:
|Title:||Development of neuro fuzzy techniques for traffic conditions||Authors:||Chandrasekaran, Dikshita||Keywords:||DRNTU::Engineering::Electrical and electronic engineering::Control and instrumentation::Control engineering||Issue Date:||2014||Abstract:||The transport system in Singapore is well developed. Singapore is facing a first world public transport and is critically suffering due to increase in congestion, overcrowding and crowd bottle neck. In many developed countries, the major challenge faced is the monitoring efficiently and control of the traffic of the city. The conventional method of using a pre-timed controller at signalized intersection, cycle timing preset along with closed loop sensors are no longer efficient in effectively managing the traffic conditions. This facilitates the need to employ intelligent techniques to model the traffic conditions to achieve efficient and effective control and management of the traffic resources. To overcome this problem, this project suggests an alternative fuzzy neural networks learning memory techniques to design intelligent controllers. The learning algorithm to generate fuzzy rules using the Generalised Dynamic Fuzzy Neural Network(GDFNN) is studied and a Matlab code is developed for the neural network. An online open source signalised traffic intersection simulator is observed and an isolated traffic signal light intersection is designed. The extension of signal timing at the intersection is predicted using fuzzy logic. The traffic light intersection data thus generated is used as inputs to evaluate the performance of training and testing the architecture. Another fuzzy neural network architecture, DENFIS, Dynamic Evolving Neural Fuzzy Inference system is studied and used to benchmark the performance of GDFNN. The prediction of signal timing extension is obtained from the fuzzy logic controller designed for an isolated signalized intersection. The performance of both the architectures is found be promising for the traffic data collected from simulation and prediction timing. Performance is based on the mean number of rules, mean training time and the prediction error of the outputs. GDFNN produced good results with generation of a compact fuzzy rule base and low prediction error of outputs. DENFIS architecture generated fuzzy rules at lower computational time.||URI:||http://hdl.handle.net/10356/61515||Rights:||Nanyang Technological University||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||EEE Student Reports (FYP/IA/PA/PI)|
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.