Forecast And Analysis Of Short-term Traffic Flow Based On Machine Learning

By means of technical methods and the figures effectively used,traffic flow prediction in the city can forecast the short-flow of urban district traffic.It is essential to urban way project and traffic administration.Additionally,It is an exceedingly complicated question to make a projection of urban traffic,especially short-term issues at critical times.It is affected by date,climate,and wind velocity in room and other elements,and also by time cycle at the right moment,but within these elements has varying degrees of associativity among them.Set up a dimensional coupling pattern and a impermanent association mould of traffic stream is meaningful doubtlessly,and establish a short-term traffic flow forecast method for key dates and periods.The research work is as follows:1)Propose a method for analyzing the temporal and spatial correlation of multiple factors in urban traffic flow.Using R/S analysis,DFA analysis and DCCA technology,the auto-correlation and cross-correlation of the power law in the traffic sequence are studied.The redundant scale extreme difference approach is harnessed as the general situation long-distance relativity analysis,and appraise the long-distance relativity of the traffic flow time queue by using Hurst index,and the multiple fractal of the array is computed by MF-DFA and the spatial cross-correlation of the DCCA calculation sequence is analyzed.The divinable property of short,middling and long period traffic flow sequences.2)Propose an integrated neural network traffic flow prediction method based on IOWA operator.Aiming at the time characteristics of the historical traffic flow data of a single road,firstly,the initial parameters of the BP neural network of multiple functional modes are searched and optimized by the mind evolution algorithm,and then the Adaboost algorithm is used to integrate many networks,and finally based on the IOWA operator criteria The weight of each predictor in the integrated network is optimized and adjusted.3)A traffic flow prediction method combining multi-source data fusion and convolutional LSTM network is proposed.In view of the different temporal and spatial characteristics of multiple road traffic flows and the influence of external events on regional traffic,first combine the decomposition and recombination mechanism of CEEMDAN and sample entropy to screen out IMF sub-sequences with more concise fluctuation characteristics,and then combine CNN and LSTM to mine data Spatially related features.Finally,considering the impact of external events on traffic,The external features are implemented based on One-hot coding,which combines internal road features and external features,and establishes a traffic flow prediction model based on a convolutional LSTM network.