Peak Hour Traffic Volume Prediction using a Hybrid Genetic Neural Method

Accurate prediction of short-term traffic flow plays a fundamental role in theIntelligent Transportation Systems (ITS) and Advanced Traffic ManagementSystems (ATMS). In this paper, a combination of multi-layer Back-PropagationNeural Networks (BPNN) and Genetic Algorithm (GA) is used to forecast thevolume of the traffic during peak hours. The real data used for modeling isobtained from Karaj-Qazvin freeway in the spring of 2013. Given the proposedmethod (BPNN-GA), Neural Network designed and be taught using trainingdata. In part train of neural network, which is typically are used internal functionsof the Neural Network's toolbox, Genetic Algorithms have been used in thisresearch. The purpose of train the network is determine the weight of its internalstructure. The Genetic Algorithm optimize network weights and improve thenetwork in learning the patterns which exist in traffic data. Thus the NeuralNetwork get the better answers in forecasting. Then the trained network isvalidated and is used to forecast the volume of traffic during peak hours in thefollowing week. To assess these forecasts, the conventional Back-PropagationNeural Network (BPNN) has also been developed and its results are comparedwith the proposed method. The results show that the proposed method (BPNNGA)forecasts the volume of peak hour traffic with greater stability and moreaccurately than conventional Neural Networks.

Keywords:

Neural network , prediction , volume of peak hour traffic , hybrid model , genetic algorithm