QSVM-RQNN：用于状态监测和故障分类的低量子比特循环量子相似度学习
QSVM-RQNN: Low-Qubit Recurrent Quantum Similarity Learning for Condition Monitoring and Fault Classification
浏览论文内容
中文总结 AI 辅助
该研究针对NISQ设备的量子资源限制问题，提出QSVM-RQNN框架，结合QSVM与RQNN实现低量子比特需求，在故障分类任务上达到有竞争力的性能，实现了性能与效率的良好权衡。
中文摘要 AI 辅助
在有噪声的中等规模量子（NISQ）时代，有限的量子比特可用性和硬件噪声制约着量子机器学习（QML）的实际部署。现有量子神经网络（QNN）和量子卷积神经网络（QCNN）架构通常需要随输入维度增长而增加量子资源，限制了其在近期设备上的可扩展性。我们提出QSVM-RQNN，这是一种集成了基于质心的量子支持向量机（QSVM）相似度学习与循环量子神经网络（RQNN）的低量子比特框架，用于故障分类。该框架使用主成分分析（PCA）缩减特征空间，将缩减后的表示划分为时序步长，并使用具有共享参数的紧凑三量子比特循环量子架构处理这些步长。我们开发了两种互补变体：QSVM-RQNN-V1对输入和质心片段执行类条件联合量子编码，而QSVM-RQNN-V2对逐时序步长的量子相似度表示执行循环学习。在多个故障诊断数据集上的实验评估显示，与QSVM、QNN、QCNN、QSVM-QNN、QSVM-QCNN和RQNN模型相比，该框架具有竞争力，且在若干情况下达到了最先进的性能。所提出的架构实现了良好的性能-效率权衡、改进的召回率，以及在高度不平衡数据集上增强的故障检测能力。这些结果表明，将基于质心的量子相似度学习与低量子比特循环表示学习相结合，为资源受限的NISQ设备上的状态监测和故障分类提供了一种有效且可扩展的方法。
英文摘要
In the Noisy Intermediate-Scale Quantum (NISQ) era, limited qubit availability and hardware noise constrain the practical deployment of quantum machine learning (QML). Existing quantum neural network (QNN) and quantum convolutional neural network (QCNN) architectures often require increasing quantum resources as the input dimension grows, limiting scalability on near-term devices. We propose QSVM-RQNN, a low-qubit framework integrating centroid-based Quantum Support Vector Machine (QSVM) similarity learning with Recurrent Quantum Neural Networks (RQNNs) for fault classification. The framework reduces the feature space using principal component analysis (PCA), partitions the reduced representation into sequential timesteps, and processes them using a compact three-qubit recurrent quantum architecture with shared parameters. Two complementary variants are developed: QSVM-RQNN-V1 performs class-conditioned joint quantum encoding of input and centroid segments, whereas QSVM-RQNN-V2 performs recurrent learning over timestep-wise quantum similarity representations. Experimental evaluation on multiple fault diagnosis datasets shows competitive and, in several cases, state-of-the-art performance compared with QSVM, QNN, QCNN, QSVM-QNN, QSVM-QCNN, and RQNN models. The proposed architectures provide favorable performance-efficiency trade-offs, improved recall, and enhanced fault detection on highly imbalanced datasets. These results demonstrate that integrating centroid-based quantum similarity learning with low-qubit recurrent representation learning provides an effective and scalable approach to condition monitoring and fault classification on resource-constrained NISQ devices.
