利用无监督预训练的轨迹深度关联

针对时空相似度算法关联轨迹的局限性，采用深度学习方法进行轨迹关联，并提出了一种基于无监督预训练的匹配神经网络训练方式。利用Geohash向量嵌入对轨迹信号做特征工程处理，构建自注意力机制神经网络结构，使用无标注轨迹数据基于遮蔽预测任务进行模型预训练；然后构建孪生匹配网络结构，加载预训练模型参数；最后使用标注轨迹对数据基于均方差损失函数微调预训练模型参数得到轨迹对匹配模型。采用Geolife GPS轨迹数据集作为评估数据集进行模型训练与测试，实验结果显示，利用无监督预训练的轨迹关联方法较现有最优算法匹配准确率提高了5个百分点，达到了96.3%,充分证明了该方法的有效性。目前轨迹关联领域基于深度学习预训练模型的研究较少，该方法具有重要的参考意义。

Deep Trajectory Matching Using Unsupervised Pre-training

Due to the limitations of spatiotemporal similarity algorithm for trajectory matching,a deep learning method is adopted and a matching neural network training method based on unsupervised pre-training is proposed.Geohash vector embedding is used to perform feature engineering processing on trajectory signals,a self-attention mechanism neural network structure is built,and unlabeled trajectory data is used to conduct model pre-training based on mask prediction task.Then a siamese matching network structure is constructed and the pre-training model parameters are loaded.Finally,labeled trajectory pair data is used to fine-tune the parameters of the pre-training model based on the mean square error loss function to obtain the trajectory matching model.The Geolife GPS trajectory data is used as the evaluation dataset for model training and testing.The experimental results show that the proposed method improves the matching accuracy by 5 percentage points(reaching 96.3%) compared with the existing optimal algorithm,thus fully demonstrating the effectiveness of the method.At present,there is limited research in the field of trajectory matching based on deep learning pre-training models,and the proposed method has important reference significance.