Research On Urban Area Flow Short-term Forecasting

With the acceleration of China's urbanization process and the arrival of the internet of things era,more and more multi-source and heterogeneous data can be obtained.These data provide the possibility to forecast urban regional traffic effectively.Urban regional traffic forecast means to use the acquired data to predict the future short-term urban regional traffic through technical,and it plays an important role in the city's road planning,traffic management,and security early warning.Regional traffic forecasting is a multi-factor-influenced spatial-temporal sequence forecasting problem with high dimensionality,strong nonlinearity,etc.The time dimension and space dimension mining increase the complexity of regional traffic forecasting tasks,making the task research more challenging.Because of this,it is particularly important to study a reasonable and effective urban regional flow forecasting method.This paper conducts the research from the perspective of various factors affecting regional traffic.The main research methods are machine learning and deep learning.The specific research work is as follows:1)Aiming at the feature that urban traffic is affected by many factors,a method of short-term urban traffic prediction based on gradient-boosting tree(GBDT)is proposed.First,the regional traffic data set is described and statistically analyzed.Then,feature extraction is used to construct a new data set using existing spatiotemporal data sets.The experimental results show that the GBDT method can effectively predict the regional traffic.At the same time,the importance of the feature is analyzed,and the related factor sequence is given,which provides a basis for subsequent research.2)The time-dependent and spatially related problems of urban regional traffic are studied separately,and a method for urban regional traffic prediction based on time-dependent LSTM network is proposed.Given different input forms,the long short-term memory(LSTM)neural network is used.Get the time dependency of the data.The experimental results show that the regional flow has the characteristics of long-term time dependence,and it is more reasonable and effective to use the time statistical characteristics as input.Then,a method of realizing urban regional traffic prediction based on spatial correlation convolutional LSTM network is proposed.This method uses the time statistics feature as input and adds convolution operation on the basis of LSTM network.The experimental results show that compared to the LSTM method,the convolution effectively obtains the spatial correlation,avoids the redundancy of the spatial features,considers the space-time attributes more comprehensively,and both methods realize the end-to-end prediction.3)Aiming at the impact of external events on regional traffic and the betterspatio-temporal characteristics of data mining,this paper proposes an urban regional traffic prediction method based on multi-source data fusion combined with CRNN.This method uses the combination of CNN and RNN to mine data.Spatial correlation,taking into account the impact of external events on the regional traffic,to achieve the external feature mining in a fully connected network.The experimental results show that the convolutional recurrent neural network(CRNN)can extract the spatial features of the data,compared with the Conv LSTM method,the extraction ability is stronger,and the integration of external features effectively improves the prediction accuracy of the method.