Travel Time Prediction Model Based On Traffic Data Fusion Technology

Travel information and path of efficient induction system will play a more and moreimportant role in traffic operation and management. Travel time as the key parameter, canprovide reference data for the release of traffic state estimation and city road networkcongestion. At the same time, travel time is widely used in intelligent transportation system，which is an important basis for road traffic efficiency and a visual indicator reflecting roadtraffic condition.At present, travel time prediction is limited to a single data source. Traffic data collectionwith single instrument is easily affected by instrument precision、sampling method、samplesize and human error. While the multi-source data fusion can make up the defect of singlesource data and improve the prediction accuracy. This paper established data fusion model oftravel time prediction with the GPS floating car data and the microwave detector traffic data.This paper established prediction model of data fusion with wavelet neural network.Wavelet neural network has stronger information extraction、nonlinear approximation andtolerance Compared with BP neural network, but it has a problem in randomly selected initialparameters, and is easy to form the local minima but not the overall optimal value. Thegenetic algorithm can be used to optimize wavelet neural network initial parameters.Optimized Wavelet neural network for data fusion can improve the convergence speed andgeneralization ability of the model.In this paper, travel time of Guangqing avenue of Qingyuan in Guangdong was predicted.The result shows that travel time of GPS taxi is lower compared to the video observationvalue, and the error exceeds the target threshold about15%. Taxi speed is higher than that ofthe road traffic flow because of frequent overtaking, and the travel time of microwavedetector is larger than the observed values, the error is more than15%of the range. Thereason is that the experimental road has many buses、large trucks, which partially block thesmall car. But the travel time of data fusion is good agree with the video observation value,the error is less than8%, in the target threshold requirement of15%. Travel time has beengreatly improved in accuracy and stability Compared to GPS taxi data or microwave detectordata.The training data fusion model was applied in a small network of Beijiang district ofQingyuan in Guangdong. Compared with the travel time of simulation output, the error ofsimulation results and the predicted results were in the range of15%, which shows that the simulation model and the prediction model are effective and reliable. But under the sameexperimental conditions, the travel time data fusion model gets better prediction precisioncompared with simulation software output.