一种基于联邦学习的时空图卷积神经网络协同训练算法

With the rapid development of industries such as big data and intelligent transportation, ride-hailing and taxis have become common modes of transportation for people. As a result, companies have generated a large amount of taxi order data. The prediction of taxi order data has important practical significance. In response to this problem, many researchers have proposed relevant deep learning models and algorithms. However, there are still some shortcomings in existing models and algorithms. First, taxi order data has extremely high privacy and commercial value. Companies cannot train better models through data sharing between them, resulting in the problem of \'data islands\'. Secondly, in existing spatio-temporal graph convolutional neural network models, the construction process of the graph structure does not protect the privacy of the data itself. This paper models the problem of predicting taxi order demand and proposes a collaborative training algorithm based on federated learning that protects data privacy and securityfor spatio-temporal graph convolutional neural networks. Experimental results show that the algorithm proposed in this paper can enable participating parties to collaboratively train models with lower errors without revealing local data.