基于深度学习的联锁测试序列推荐方法

A Deep Learning-Based Interlocking Test Sequence Recommendation Method

测试序列的组织与选取是保证铁路信号联锁系统测试质量和效率的关键环节,现阶段主要采用人工方式进行,工作量大且容易遗留漏洞。通过结合专家经验及历史测试信息,提出一种基于深度学习的联锁功能测试序列推荐方法,用以辅助测试人员开展测试工作,提高测试效率。针对联锁拓扑结构特征难以通过深度学习提取和训练的问题,构建了基于站场拓扑结构的树形递归神经网络,提高模型对联锁测试数据的特征提取能力;针对联锁测试序列特点,设计加入注意力机制的编码器-解码器框架,解决模型训练时产生的信息压缩问题,提高模型对联锁逻辑及测试案例间逻辑关联的学习能力。基于实际站场数据进行了一系列对比实验,结果表明,将递归神经网络与加入注意力机制的编码器-解码器框架相结合的测试序列推荐模型在训练集及测试集精确度方面均高于其他模型,达到90%以上并通过测试序列的执行时间和缺陷检测速率证明了该推荐模型的可行性。

The organization and selection of test sequences are critical steps to ensure the quality and efficiency of testing in railway signal interlocking systems.At the present stage,this process is mainly carried out manually,resulting in heavy workload and potential omissions.By combining expert experience and historical test data,this paper proposes a deep learning-based interlocking test sequence recommendation method to assist testers in improving testing efficiency.To address the challenge of extracting and training the topological features of interlocking structure through deep learning,a tree-structured recursive neural network based on station yard topology was constructed to enhance the model's ability to extract features from interlocking test data.Considering the characteristics of interlocking test sequences,an encoder-decoder framework with attention mechanism was designed to solve the issue of information compression during model training,improving the model's ability to learn the logical connections between the interlocking system and test cases.Based on real station yard data,a series of comparative experiments were conducted.The results showed that the test sequence recommendation model combining a recursive neural network with the attention-enhanced encoder-decoder framework achieved over 90%accuracy in both the training and test sets,which is much higher than that of other models.The model's feasibility was further demonstrated through the execution time of test sequences and defect detection rates.