The forecast of the current urban road travel time is mostly limited to single-source data and the prediction accuracy is not high. Based on the floating car GPS data and microwave detector traffic data, a model of travel time was built in the fusion method. Wavelet neural network was optimized by using genetic algorithms, which can solve the blindness and the randomness of selecting wavelet neural network initial parameter, thus greatly improving Web search efficiency and the speed of training. The predicted travel time is in good agreement with video observed data. The results show that the model is effective and reliable.