高压断路器是整个电网中运行的最核心电力设备之一，其正常工作是电力系统稳定运行的重要保障。根据国际大电网会议（CIGRE）的统计结果，高压断路器机械故障占其故障总数的63%，严重威胁到电力系统的运行可靠性。随着智能电网的发展以及先进传感技术的逐渐普及，对高压断路器进行在线状态监测与智能识别成为智能电网新的发展方向，基于在线监测数据对高压断路器进行故障分类预测，能尽可能提前发现其潜在故障，避免更大系统安全威胁与经济损失，意义重大。首先，本文针对多种高压断路器机械特性监测数据，提出了基于多源状态参量的相关性分析算法，将不同类别的机械特性监测数据融合进行统一建模，进行机械特性状态参量与故障类型之间的相关性分析，并进一步提出相应的故障分类预测算法。其次，分合闸线圈电流作为高压断路器机械特性检测中最重要的一环，本文提出了基于无标签分合闸线圈电流数据的故障分类预测算法。通过线圈电流变化与高压断路器动作过程的一一对应关系，对电流信号进行特征值提取，并通过异常点检测算法剔除异常数据，通过聚类算法打标签，最终训练神经网络进行故障分类预测，并对算法的鲁棒性进行探究。最后，针对有标签的分合闸线圈电流数据，以一维卷积神经网络（CNN）为基础，提出基于电流原始信号序列的故障分类预测算法，通过模型实现特征自动提取与高效处理，避免了人工特征提取的过程，实现高准确率的故障分类预测。同时对此算法的鲁棒性进行探究。

High-voltage circuit breakers are one of the most core electrical equipment operating in the power grid. Its normal operation is an important guarantee for the stable operation of the power system. According to the statistical results of CIGRE, the mechanical faults of high-voltage circuit breakers account for 63% of its total faults, which seriously threatens the reliability of power system. With the development of smart grid and the gradual popularization of advanced sensing technology, online status monitoring and intelligent identification of high-voltage circuit breakers have become a new development direction. It is of great significance to discover its potential faults in advance as much as possible and avoid greater system security threats and economic losses.First, aiming at a variety of high-voltage circuit breaker mechanical characteristic monitoring data, a correlation analysis algorithm based on multi-source state parameters is proposed , which integrate different types of mechanical characteristic monitoring data for unified modeling. The correlation between mechanical characteristic state parameters and fault types is analyzed, and the corresponding fault classification and prediction algorithm is further proposed. Next, the tripping and closing coil current is the most important part in the detection of the mechanical characteristics of the high-voltage circuit breaker. A fault classification and prediction algorithm based on unlabeled tripping and closing coil current data is proposed in this paper. Through the one-to-one correspondence between the current waveform and the action process of the high-voltage circuit breaker, the characteristic value of the current signal is extracted. The abnormal data is eliminated through the abnormal point detection algorithm, and the clustering algorithm is used to label the data. A neural network is trained for fault classification and prediction, and the robustness of the algorithm is explored.Finally, aiming at the labeled tripping and closing coil current data, this paper proposes a fault classification and prediction algorithm based on one-dimensional convolution neural network, which realizes automatic feature extraction and efficient processing, avoiding the process of manual feature extraction. This algorithm realizes high accuracy of fault classification and prediction. The robustness of the algorithm is explored at the same time.