Track destination prediction method based on knowledge graph and self-attention mechanism

Abstract

The invention discloses a trajectory destination prediction method based on a knowledge graph and a self-attention mechanism, and belongs to the field of intersection of deep learning and an urban traffic planning technology. According to the method, a multi-layer knowledge graph is constructed from bottom to top, the first layer represents a basic road network structure and basic attributes, the second layer represents upstream and downstream relations and selection preferences of road sections in trajectory data, and the third layer uses a Mean Shift algorithm to cluster a trajectory starting point and a trajectory ending point to divide functional areas of a road network. Secondly, a Graph-Bert algorithm is improved, a TrafficGraph-Bert algorithm is proposed to carry out graph representation on the multi-layer traffic knowledge graph, and a road network is divided according to the density of destinations, so that the problem of sparse trajectory data is solved. And finally, the trajectory sequence is learned by using a self-attention mechanism, contribution degrees of different road sections in the trajectory to result prediction is learned through the attention mechanism, and a location where the final trajectory arrives is predicted.

Description

technical field [0001] The invention belongs to the intersecting field of deep learning and urban traffic planning technology, and relates to a method for predicting a trajectory destination based on a traffic knowledge map and a self-attention mechanism based on partially known trajectories. Background technique [0002] Recently, transportation, as an eye-catching and unavoidable problem in contemporary society, frequently appears in our field of vision. As one of the key points of urban development, transportation has a remarkable influence on both ecological environment and economic development. The connectivity between cities has been realized through transportation, which has promoted the rapid development of industries such as transportation and commerce. In addition to the benefits of traffic itself, advertising on vehicles can also bring economic benefits to advertisers. Of course, while transportation has brought us a lot of convenience and good economic effects,...

AI Technical Summary

Problems solved by technology

But this kind of destination prediction algorithm based on low-order Markov chain usually has two main problems. The first problem is that in the low-order Markov algorithm, only the latest timestamp is merged, that is, the future state is only Depends on the current state regardless of the past state, so cannot capture long-term dependencies in the sequence

The second problem is that due to the sparsity of trajectory data in the road network, the accuracy of prediction is likely to be affected for trajectories that have never appeared in the data set

However, similarly, data-driven algorithms also face the problem of sparse trajectories

For those trajectories that have never appeared, when using data-driven algorithms for prediction, the efficiency of the algorithm will drop significantly

Images