Research On Short-term Traffic Flow Prediction Based On Deep Learning

With the acceleration of China’s urbanization process,the number of motor vehicles has increased year by year,which has gradually led to some traffic problems,such as traffic congestion and traffic pollution.In order to alleviate these problems,it is necessary to establish a smart traffic system and achieve real-time traffic planning and guidance.Accurate short-term traffic flow forecast information is one of the important references for real-time traffic planning.However,traffic flow data has complex spatial-temporal features and is also affected by external factors such as weather and holidays.Its prediction task has always been quite challenging.Existing prediction methods are difficult to fully extract the spatial-temporal features of traffic flow data.To solve this problem,this paper constructs a prediction model that integrates multiple temporal features and a prediction model based on graph convolution network for two different environments of single monitoring point and road network to predict short-term traffic flow.The main contents are as follows:(1)this paper introduces the parameters and features of traffic flow data,analyzes its temporal and spatial correlation,introduces the classic deep learning models of short-term traffic flow prediction,preprocesses the traffic flow dataset and weather dataset,analyzes the correlation between weather features and traffic flow data,and screens out weather features with strong correlation with traffic flow data,which lays the data and theoretical foundation for the construction of short-term traffic flow prediction model in the subsequent sections.(2)Aiming at the problem that it is difficult to determine the hyperparameters of the deep learning model,a method for optimizing the hyperparameters of the deep learning model based on the improved bald eagle search algorithm is proposed.This method makes the following improvements to the bald eagle search algorithm: by improving the Sine chaotic map,the initial distribution position of the bald eagle is more uniform;through the adaptive inertia weight,the global and local search capabilities of the algorithm are balanced;by introducing a Cauchy mutation operator,the global search ability of the algorithm is strengthened.The IBES-LSTM model is obtained by optimizing the hyperparameters of the long-term short-term memory network through the improved bald eagle search algorithm.The model has achieved good results on the data set and laid the foundation for the later construction of the prediction model.(3)In order to fully extract the temporal features of individual monitoring point,this paper proposes a forecasting model SACIL that integrates multiple temporal features.The model fully considers the influence of the adjacent time period,the same time period of the previous day and the previous week on the predicted time period.External factors such as weather and holidays are also incorporated.The convolutional neural network and IBES-LSTM network are used to fully extract the temporal features of each time period,and the self-attention mechanism is used to obtain effective traffic flow feature expression.Finally,the prediction results of the three periods are fused,and the superiority of the model in predicting short-term traffic flow data is verified on a real highway dataset.(4)Aiming at the level of the entire urban road network,considering the complex topological spatial structure of the road network,in order to fully extract the spatial features of traffic flow data,two short-term traffic flow prediction model SASTGCN and SASTGAN based on the graph convolutional network are constructed.The two models use graph convolutional neural network and graph attention network to extract the spatial features of the road network,and use the IBES-LSTM network and convolutional neural network to extract the temporal features of traffic flow and the self-attention mechanism to analyze the long-term dependence relationships of time series,using three submodules to analyze the relationship between different time periods and forecast periods.It is verified on the road network dataset that these two short-term traffic flow prediction models are superior to various comparison models,and SASTGAN has smaller prediction errors and higher accuracy than SASTGCN,effectively extracts the spatial-temporal features of traffic flow data.,and effectively extracts the spatial features of the road network.These models have reference significance for short-term traffic flow prediction of road network.