A hybrid prediction method for short-duration traffic flow on urban roads based on integrated neural networks

Urban road traffic flow is affected by many factors, such as road topology, road state and traffic control, among which there are complex spatial correlations and spatiotemporal dynamics, making it difficult to predict the results and accuracy. Therefore, a hybrid forecasting method for short-term traffic flow of urban roads based on integrated neural network is proposed. Collect and preprocess urban road traffic data to build a historical traffic flow dataset. Elman neural network is used to construct periodic sequence prediction sub network, long-term and short-term memory neural network is used to construct continuity prediction sub network, and ELM neural network is used to construct correlation prediction sub network. According to the traffic flow characteristics of the historical traffic flow data set, the related sequence set is constructed and allocated to each sub network. Through the global hybrid forecasting process, using the integration method based on parameter matrix, the outputs of three neural networks are mixed to obtain the final short-term traffic flow hybrid forecasting results of urban roads. Experiments have shown that the proposed algorithm can achieve relatively accurate traffic flow forecasting results, with an absolute percentage deviation of less than 1.6 % and high forecasting accuracy.