| With the development of urbanization and the improvement of people’s living standards,the number of motor vehicles is increasing,but the increase in road mileage and urban infrastructure such as parking lots cannot support the increase in motor vehicles.This creates two problems,traffic congestion and parking difficulties.Advanced Traveler Information System（ATIS）,as an important component of intelligent transportation system,can release real-time traffic information and provide reasonable route planning to help travelers choose travel mode,travel time and travel route.This can alleviate traffic congestion and parking difficulties to some extent.Traffic condition prediction is a key step in ATIS.Accurate prediction results can help managers better analyze traffic trends and provide a good data basis for traffic control decisions.The existing traffic condition prediction methods only use the historical data of the prediction object when predicting the traffic flow or parking condition,or add the oneway influence of external factors such as weather and holidays.This single-task prediction method results in an insufficient grasp of the characteristics of regional traffic situation by the model,which has also become one of the bottlenecks in improving the prediction accuracy.Traffic phenomenon is an organic whole,and various factors are intertwined,such as pedestrians,non-motor vehicles,public transportation and parking cruises,etc.There is a certain interaction between these traffic activities.If we can combine these interrelated traffic data,we can fully understand the traffic characteristics and improve the prediction accuracy.Through phenomenon observation and data analysis,we find that there is an interaction between traffic flow and parking,especially during peak travel periods.On the one hand,traffic flow will bring pressure to the parking lot,and on the other hand,vehicles that park and cruise will run at low speed and affect normal traffic.However,the interaction between the two has been neglected in the existing research.Based on this objective phenomenon in urban traffic,this paper proposes a multitask integrated prediction model,which makes full use of the correlation between traffic flow and parking to complete the prediction task of traffic flow and parking occupancy,that is,the prediction of regional traffic conditions.At first,we put forward a basic model,in which we characterize the traffic flow and parking data in the form of topologies,and then use the graph convolutional neural network to extract the spatial features of the two topological networks respectively.Then,the fusion of the two types of spatial features is completed by concatenating.Finally,the gated recurrent unit extracts temporal features and completes the final prediction.Experiments show that the basic model has a good prediction effect,and we also confirm the reason for the improved prediction effect of the model,which is to capture the correlation between traffic flow and parking.In the basic model,the spatial feature fusion step of traffic flow and parking adopts the method of concatenating,which is also a commonly used method in neural networks,but it has certain shortcomings.Therefore,we then optimize the spatial feature fusion step,that is,the feature fusion of the two networks is completed according to the correlation between traffic flow and parking.To measure the correlation between traffic flow and parking in an area,we propose a correlation measure that considers spatial topology,also known as spatial-temporal mutual information.Then we obtained the interaction patterns of traffic flow and parking in different time periods according to the spatial-temporal mutual information and geographic location relationship.Finally,the feature fusion of the two types of nodes was completed by the heterogeneous graph neural network.This fusion method enables the model to further learn the interaction mechanism between traffic flow and parking.The experiment further proves that the spatiotemporal correlation between traffic flow and parking can promote the prediction accuracy of the model.Finally,we compare the basic model and optimization model with the existing classic neural network traffic prediction model.The experiments show that the multi-task integrated prediction model can capture the correlation between traffic flow and parking,and the prediction effect is better than the existing classic model. |
