摘要
中点钳位(Neutral point clamped,NPC)逆变器因具备电压应力抑制、谐波优化等优势,在新能源领域应用广泛,但其多开关拓扑易引发绝缘栅双极型晶体管开路故障。针对传统诊断方法的局限性,提出了一种融合S变换(S-transformation,ST)与残差并行多维注意力模型的高效诊断方法。首先通过MATLAB/Simulink仿真提取d轴电压–电流信号,通过d轴电压–电流信号的参数化处理,建立以虚拟电阻为表征量的故障特征提取机制,有效实现原始信号降维与特征解耦;随后,采用具有时频局部化特性的ST算法,将一维电信号映射为二维时频矩阵,通过时–频域联合分析揭示隐含故障模式;最后,设计具有残差跳跃连接与并行多维注意力机制的深度网络模型,利用通道–空间双域注意力机制强化故障敏感特征提取,并通过残差结构抑制梯度弥散问题。实验结果表明,该方法在无噪环境下故障分类准确率达100%,在30 dB噪声干扰下仍保持94.93%的识别率。该结果充分验证了其卓越的特征提取能力和抗噪性能。
Neutral point clamped(NPC)inverters are widely used in the new energy systems owing to their advantages in voltage stress suppression and harmonic optimization.However,their multi-switch topology is prone to open-circuit faults in insulated gate bipolar transistors(IGBTs).Aiming at the limitations of traditional diagnostic methods,this paper proposes an efficient diagnostic method that integrates the S-transform(ST)and residual parallel multidimensional attention(ResNet-PMDA)model.First,the d-axis voltage-current signals are extracted by MATLAB/Simulink simulation.By parameterizing these signals,a fault feature extraction mechanism based on virtual resistance is established,which effectively reduces the dimensionality of the original signals and the decoupling the features.Subsequently,the S-transform(ST)algorithm known for its time-frequency localization capability is employed to map the one-dimensional signal into a two-dimensional time-frequency matrix.This facilitates the revelation of hidden fault patterns through joint time-frequency analysis.Finally,a deep network model incorporating residual skip connection and a parallel multidimensional attention mechanism is designed.The model enhance fault-sensitive feature extraction through a channel-space dual-domain attention mechanism and mitigate gradient vanishing through the residual structure.Experimental results demonstrate that the proposed method achieves 100%fault classification accuracy in a noise-free environment and maintains a recognition rate of 94.93%under 30 dB noise interference,confirming its strong feature extraction capability and noise robustness.
作者
王小玲
毕贵红
杨楠
陈世语
李玉洪
陈仕龙
WANG Xiaoling;BI Guihong;YANG Nan;CHEN Shiyu;LI Yuhong;CHEN Shilong(Faculty of Electric Power Engineering,Kunming University of Science and Technology,Kunming 650500,China)
出处
《电力科学与工程》
2025年第12期23-36,共14页
Electric Power Science and Engineering
基金
国家自然科学基金资助项目(51767012)。
关键词
故障诊断
中点钳位逆变器
S变换
开路故障
残差并行多维注意力
fault diagnosis
neutral point clamped inverter
S-transform
open-circuit fault
residual network with parallel multidimensional attention
