【目的】传统隔离开关故障特征提取方法不够可靠，故障识别准确率不高。针对此问题，提出了一种基于多特征图谱和卷积神经网络(CNN)-黑翅鸢算法(BKA)-最小二乘支持向量机(LSSVM)的隔离开关故障诊断方法。【方法】首先，通过马尔科夫转移场(MTF)、格拉姆角场(GAF)和短时傅里叶变换(STFT)分别构建频域、时域和时频域三种特征图谱；然后，分别建立三种CNN模型，输入三种特征图谱，通过卷积、池化等步骤提取有效的故障特征，并在CNN模型的全连接层使用t分布随机邻域嵌入算法对特征数据进行降维；最后，将提取出的特征向量进行融合拼接，采用BKA优化的LSSVM代替Softmax层，并将融合后的特征向量输入到BKA-LSSVM中进行故障识别。【结果】通过现场故障模拟试验，采集了四种状态下的隔离开关振动信号数据，并进行了对比分析。结果表明，本文所提模型相较其他故障诊断模型具有更高的精度、更强的可靠性和泛化能力，其运行8次的平均诊断准确率达到97.08%。试验结果验证了所提模型的可行性。【结论】本文所提方法从多维度提取特征图谱，克服了单一维度方法的局限性；通过引入CNN-BKA-LSSVM模型，能够更好地提取关键特征，提高故障识别的精度和准确性。所提方法为隔离开关的故障诊断提供了可靠的理论依据和技术参考，也为隔离开关设备维护提供了新的思路，具有重要的应用价值。

[Objective] The traditional methods for extracting fault characteristics of disconnector are not reliable enough, and the fault identification accuracy is not high. To address this issue, a fault diagnosis method for disconnector based on multi-characteristic map and convolutional neural network (CNN)-black-winged kite algorithm (BKA)-least squares support vector machine (LSSVM) is proposed. [Methods] Firstly, three characteristic maps were constructed in the frequency domain, time domain, and time-frequency domain using the Markov transition field (MTF), Gramian angular field (GAF), and short-time Fourier transform (STFT) respectively. Then, three CNN models were established separately, three characteristic maps were input, effective fault characteristics were extracted through convolution, pooling, and other steps. And the t-distributed stochastic neighbor embedding algorithm was used to reduce the dimension of the characteristic data in the fully connected layer of the CNN model. Finally, the extracted characteristic vectors were fused and spliced, the BKA-optimized LSSVM was used instead of the Softmax layer, and the fused characteristic vectors were input into the BKA-LSSVM for fault identification. [Results] Through on-site fault simulation tests, vibration signal data of the disconnector in four different states were collected, and comparative analysis was carried out. The results showed that the model proposed in this paper has higher accuracy, stronger reliability and generalization ability compared with other fault diagnosis models, and the average diagnostic accuracy of the proposed model for 8 runs reached 97.08%. The experimental results verified the feasibility of the proposed model. [Conclusion] The proposed method in this paper extracts the characteristic maps from multiple dimensions, which overcomes the limitations of the single dimension method. By introducing the CNN-BKA-LSSVM model, it can better extract the key characteristics and improve the precision and accuracy of fault identification. The proposed method provides a reliable theoretical basis and technical reference for the fault diagnosis of disconnectors, and also provides new ideas for the maintenance of disconnector equipment, which has important application value.

国网江苏省电力有限公司重点科技项目资助(J2024047)