A multi-stream multi-scale lightweight SwinMLP network with an adaptive channel-spatial soft threshold for online fault diagnosis of power transformers

Abstract

The fault diagnosis of power equipment is extremely crucial to the stability of the power grid system. However, complex operating environments, high costs and limitations of single-modal signals are the biggest bottlenecks. To this end, a Multi-Stream Multi-Scale Lightweight SwinMLP Network (MLSNet) with adaptive channel-spatial soft threshold is proposed in this paper. First, a Res2net-based feature-enhanced method is used to learn the correlated features of vibration and voltage multi-modal signals. Second, a novel MLSNet isdesigned to combine the benefits of SwinMLP with lightweight CNN, and employs a staged model to extractvarious scale features. Third,module is intended an adaptive deep fusion approachto integrate employing a channel-spatial soft thresholdand recalibrate staged information at different scales. The overall accuracy ofthe proposed method can reach 99.20% by various experiments, aiming to make it an effective method for online fault diagnosis of power transformers.

Hierarchical dynamic aggregation graph modeling of vibration for semi-supervised fault diagnosis of power transformer

“…The swin transformer block is shown in figure 3(a) using the shifted windows as a foundation [21]. It consists of a two-layer MLP, a layer normalization (LN) layer, a window-based MSA module, and a residual connection that makes up each swin transformer block.…”

“…The combination of graphic modeling techniques and deep learning (DL) techniques have proven to be successful and beneficial in fault identification. Various DL-based techniques, such as CNNs [17][18][19], Res2Net [20], RNNs, transformer [16], and swin transformer [21], have been extensively employed for this purpose. For instance, the deep residual shrinkage network [22] is an innovative DL architecture that utilizes a soft threshold approach, serving as a CNN variant specifically designed to handle noise.…”

“…Additionally, the CLFormer [24] networks leverage a linear attention mechanism and convolutional embedding to maximize the extraction of global features. The swin transformer is utilized as the network's backbone in [21,25,26] to address domain shifts and improve fault identification performance by capturing global dependencies effectively. However, DL methods with high anti-interference and multi-domain fusion ability are still difficult to study.…”

A novel vision transformer network for rolling bearing remaining useful life prediction

“…Wang et al [22] reported a method combining 1D CNN and ViT to recognize engine faults, and the experimental results showed its effectiveness. Liu and He [23] presented a multi-stream multi-scale lightweight SwinMLP network for online fault diagnosis of power transformers. However, the ViT also exists the shortcomings, for example the complex model structures and the too large number of parameters.…”

Hierarchical spiking neural network auditory feature based dry-type transformer fault diagnosis using convolutional neural network

“…Power systems are critical infrastructures in modern society. As the core device of a power system, regular operation of transformers is crucial for stable operation [1]. However, the occurrence of abnormal faults in transformers is inevitable, timely repairs to prevent economic losses.…”