Dynamic Path Planning Based On Urban Road Traffic Flow Prediction

With the deepening of urbanization,the impact of urban traffic congestion becomes more and more obvious.Traffic flow prediction has become the key to the development of urban intelligent transportation system.The prediction of traffic flow of the urban road network by using big data and artificial intelligence algorithm can recommend the optimal travel route for travelers and assist traffic management departments to carry out traffic control in advance,which is of great significance to the construction of smart cities.This thesis mainly studies urban traffic flow prediction and optimal path planning.Based on spatio-temporal data,the deep learning algorithm is used to realize the traffic flow prediction with the superposition of temporal and spatial features,and the prediction results are used in urban dynamic path planning to obtain the optimal driving path.The main work of this thesis is as follows:1)An urban traffic flow prediction model based on spatiotemporal fusion convolutional network is proposed.According to the urban road network,the model constructs the spatial neighbor subgraph containing local road relationship and the similar neighbor subgraph containing global road relationship.The sub-graph matrix is dynamically weighted by the graph attention network,and then the spatial features of the two dynamic sub-graph matrices are extracted by the graph convolutional network.In order to extract the temporal characteristics of traffic flow data,the model uses the temporal convolutional network with attention mechanism to extract multiple potential temporal information from the data,and obtains the local and remote time dependence relations of traffic flow data.Then high-dimensional spatial and temporal features are obtained by series aggregation of temporal and spatial features using the spatial-temporal fusion layer,and multi-step prediction is realized through the linear layer.The results of experiment prove that the prediction model has more accurate performance than the current mainstream prediction models.2)In order to improve the performance of the spatiotemporal fusion convolutional network,a parameter optimization algorithm based on particle swarm optimization was proposed.The algorithm optimizes the performance of the spatiotemporal fusion convolutional network by changing the value of network parameters,with the aim of finding the parameter values that make the network performance reach the optimal level and improving its prediction accuracy.The results of experiment prove that the optimized spatiotemporal fusion convolutional network has smaller prediction error and faster convergence speed than before.3)A dynamic path planning algorithm based on road network state information is proposed.In this algorithm,the shortest travel time is taken as the goal,and the comprehensive travel time of the road is taken as the spatio-temporal congestion coefficient.An optimization A*algorithm limited by driving angle is used to realize dynamic path planning in the process of vehicle driving,so as to obtain the global optimal driving path.This algorithm predicts the state information of the urban road network before entering the next intersection according to the position of the current vehicle.The network model of the dynamic road network state information obtained through the analysis of historical data is used to judge whether the spatio-temporal congestion coefficient matrix of the road needs to be updated.If so,the route should be re-planned.The results of experiment prove that that the model proposed in this thesis can obtain the path with shorter travel time than the current mainstream path planning algorithm,and has shorter running time.