Process neural network modeling for real time short-term traffic flow prediction

In order to fully utilize the spatio-temporal process characteristic of traffic flow and predict traffic flow in real time, both process neural network and the online learning technology of data stream were imported into short-term traffic prediction. Considering the inherent traffic features of daily-periodicity and weekly-periodicity, process neural network and wavelet transform were combined to deal with the multi-scale process characteristic of historical data. A road network prediction model was constructed, and was optimized by adopting principal component analysis and utilizing the influence of traffic flow space similarity. An online learning algorithm was proposed based on Hart wavelet technology, which has the characteristics of selfadaptability and real-time prediction. Experimental result shows that the forecasting accuracy of the model is better than ordinary neural networks, its relative error of mean percentage reduces by 6%-8%, and its prediction time reduces by 67% at least, so the model has good performance and can meet the demand of real-time prediction of short-term traffic flow.