Research On Short Time Traffic State Prediction Model Of Urban Road Based On LSTM Deep Network

With the ever-accelerating process of urbanization,traffic congestion has become serious increasingly.Governance and mitigation of traffic jams in urban areas are thus crucial and have been regarded as one of the most important social problem worldwide.Actually.there are still a large number of roads in non-congested state during the urban traffic jam period.As long as we can grasp the traffic conditions of the traffic road in time,we can make use of the prediction of the long and short time to alleviate the traffic pressure of the main congested sections in the urban areas and improve the utilization rate of existing urban road resources,since it can lead to scientific decisions on the guidance of effective traffic control.In view of this?the Intelligent Transportation System(ITS).which incorporates a variety of advanced technologies,has become the preferred method to solve the problem of urban road congestion.Short-term traffic condition forecasting.as a branch of ITS.plays an important role in the intelligent management and dynamic control of traffic and is the key to traffic control and guidance in intelligent traffic systems.However.how to achieve effective short-term traffic status prediction of urban roads restricts the rapid development of intelligent transportation.Reasons are many,such as low prediction accuracy of traditional prediction methods.high acquisition cost,poor quality.and small coverage of the traffic data utilized and complex road network structure of urban area,to name but a few.The massive traffic data obtained by the use of floating car technology effectively overcomes the above problems.making the research and prediction of the traffic conditions into the era of big data.which makes it possible to use traffic big data to predict short-term traffic conditions.In this paper,a traffic flow parameter extraction method is designed for traffic sparse characteristics of massive floating cars.This method can effectively solve the difficult problem of traffic flow parameters extraction due to the sparse characteristics of floating car data.and can accurately extract the speed of the road section reflecting the actual situation.Moreover,this paper proposes a comprehensive prediction model based on the speed of road segment for short-term prediction of urban traffic conditions.including the prediction of traffic flow parameters of urban roads and the identification of traffic conditions.This model utilizes deep learning for the prediction of traffic flow parameter sequences.and builds a prediction model of urban road traffic flow based on an optimized Long Short-Term Memory(LSTM)deep network.The optimized algorithm improves the overall prediction accuracy by about 2.1%compared with the conventional LSTM network.Meanwhile,the stability of the prediction model has also been greatly enhanced.In addition.this paper applies the fuzz)y set theory to the identification of traffic conditions and adopts the fuzzy C-means clustering traffic state identification model combined with paste theory and clustering ideas.The model can not only use the traffic flow state parameters to classify the traffic state categories but also calculate the attribution of samples and categories.In order to verify the proposed prediction model.this paper relies on the traffic simulation platform of the Nanning Public Security and Communications Corps to carry out in-depth pre-processing of the collected mass floating car data used for the verification of the road speed parameter extraction and prediction model.The verification results effectively demonstrate the validity and accuracy of the model prediction.The results of this paper can provide basic data for the dynamic control and induction of urban traffic.which has certain practical reference and application value.