Refined Travel Demand Forecast Based On Deep Learning

The transportation network is like the "vessel" of a city,which plays a very important role in the progress and development of the city.With the rapid development of cities,the transportation network has become denser and denser.On the one hand,it has promoted the further development of the city;on the other hand,the problem of "blood vessel blockage" has gradually emerged,affecting the health of the city and has become a key issue that needs to be solved urgently.The study of traffic data can reveal the inherent laws of traffic flow,and its development and utilization are of great significance to the public safety of the entire city,urban planning and development,and people’s travel planning.It can provide very effective data support and better Promote the construction of smart cities.Traffic flow data is generally spatio-temporal data,and taxi flow data that can reflect people’s daily travel needs is one of them.This article focuses on the time dependence and spatial correlation existing in this type of data,and studies and establishes corresponding neural network predictions.The method aims to learn the evolutionary laws behind the time-space sequence,and to reflect the future flow trends and travel needs of user groups by predicting the future inflow and outflow of taxis in various areas of the city.The research content of this article mainly includes the following aspects:1.In order to better understand user travel behavior and traffic distribution trends,this paper uses the original taxi navigation and positioning data to mine the attributes of different areas and the corresponding user travel characteristics.First,perform preprocessing operations such as cleaning and sorting the original data,and then finely divide the entire urban area,and define the outflow and inflow of the region in terms of people’s travel demand forecasting,and define each area.Statistics on the historical taxi traffic data of,analyzes the differences in travel distribution of users in different regions in time and space.2.Since the refined grid area traffic has the characteristics of grid data similar to the picture,the refined travel demand forecast,that is,the problem of fine-grained regional inbound and outbound traffic forecast,can also be regarded as a mapping of low entropy data to high The information entropy data mapping problem,that is,the flow super-resolution problem,has the same characteristics as the image superresolution problem.Therefore,the reference image super-resolution technology is considered to solve the flow super-resolution problem,and three different flow superresolution problems are used.The resolution prediction model has been simulated and predicted and analyzed;3.In consideration of the time series trend of the researched data,this article first constructed a traffic prediction model based on long-and short-term memory networks.In addition,considering the balance of short-term and long-term spatio-temporal characteristics in the spatiotemporal sequence prediction task,starting from the randomness and periodicity of the time series,the input data is divided into different time steps.One is the three-dimensional two-step misalignment prediction method.,The second is a multi-step forecasting method based on short-term,medium-term,and long-term.The latter is suitable for seasonal spatiotemporal data with inherent dynamic patterns but relatively complex trend information,such as traffic flow series,etc.;longshort-term memory network Although the temporal characteristics of regional flow data can be captured well,as spatio-temporal data,its spatial information and spatial correlation are equally important and cannot be ignored.On the basis of the basic neural network structure proposed above,in order to consider the characteristics of hierarchical deep network in the unified modeling of spatio-temporal correlation,this paper studies the cross-layer transfer method of memory state based on recurrent neural network,and establishes The transfer type cyclic convolutional neural network realizes the deep fusion of the time memory state and the multi-layer spatial feature inside the cyclic network node unit,and further improves the accuracy of the regional traffic prediction;it also takes into account the input and output traffic between various regions.It will affect each other,but when it is sent directly to the neural network as two-channel matrix data,although the network can learn the correlation between the two spontaneously,the specific learning mechanism and learning rate cannot be controlled,so the attention mechanism is introduced,Split the inflow and outflow into two parts of input,so that the neural network can learn the emphasis relationship between the inflow and outflow between regions.This paper establishes a deep neural network traffic prediction model to predict the inbound and outbound traffic of taxis(that is,people’s travel needs)in various areas of the city in the future.The study of urban regional flow is helpful to grasp the distribution trend and flow law of human traffic flow.If we can perceive and predict the future human travel flow in any area of the city in advance,it will be important for strengthening urban public safety and improving urban traffic efficiency.Has an important role.