Urban road network travel time estimation method based on adaptive multi-task deep learning

Abstract

The invention discloses an urban road network travel time estimation method based on adaptive multi-task deep learning and belongs to the technical field of intelligent transportation. Firstly, features affecting the travel time of a road segment are extracted, then a deep network formed by multiple feature learning layers is constructed with a noise reduction sparse self-encoder as a component, and the feature representation of input is learned layer by layer. Finally, a probability-based method is used to carry out modeling on the uncertainty of the travel time, a multi-task regression layeris constructed, an error from an observation value is outputted by a minimized model, and thus the model automatically adjusts the weight of each task. According to the method, an existing method hasthe defects that a shallow network is difficult to describe a complex non-linear travel time in an urban road network, the manual adjustment of the weight of a joint learning task and only data witha complete task tag can be used are overcome, the efficiency of travel time estimation is improved, and the method has an important practical significance in the estimation of the travel time of the urban road network.

Description

technical field [0001] The invention relates to a method for estimating travel time of an urban road network using a multi-task deep learning method with adaptive weights, and belongs to the technical field of intelligent transportation. Background technique [0002] In real life, we usually divide a large problem into several highly related sub-problems. Considering the learning of each sub-problem as a task, the common method is to learn one task at a time and then combine these learning tasks, but this ignores the high correlation between these sub-tasks, and it is these correlations This allows us to achieve good learning results even with less data available. Compared with single-task learning, multi-task learning adopts the following mode: first, the complex learning problem is decomposed into theoretically independent sub-problems, and then each sub-problem is learned separately, and finally the complex problem is established by combining the learning results of the ...

AI Technical Summary

Problems solved by technology

At the same time, in actual situations, due to the limitation of the number of vehicles and the routes traveled by vehicles, it is difficult to obtain the observed values ​​of the travel time of all road sections. This kind of incompletely labeled data will cause many problems in the process of training the model, making The training effect of the model is not ideal

Images