DBLnet: Navigation algorithm based on dynamic Bayesian LSTM
Xiaoshuo Jia, Siyuan Li, Qingzhen Xu · Communications on Analysis and Computation · 2024
In long-distance navigation and path planning tasks, traditional algorithms and some SOTA (state of the art) algorithms are unable to provide accurate node information in real time due to problems such as the complex environment and a large number of nodes. To address this problem, this paper proposes a DBLnet model that combines a Static LSTM (long short term memory) module and a Dynamic Bayesian module. The Static LSTM module mainly predicts node feature information, and the Dynamic Bayesian module mainly makes decisions to determine whether the node is passable. In the experiment, we selected two data sets of TLC (Taxi and Limousine Commission) trip record data in May and July 2022 for training and testing. DBLnet was evaluated through two indicators: the traffic flow accuracy rate and congestion rate of the node. After experiments, it was demonstrated that DBLnet has certain generalization and robustness, and has the ability to process directed graph.