An electrical arc is a high-energy discharge that produces electromagnetic radiation, acoustic emission, and heat generation. Arc discharges in transformers can lead to a range of serious malfunctions. Therefore, it is necessary to ascertain the category of arc discharge. The conventional diagnostic methods for partial discharges are not applicable to arc discharges the distinctive characteristics of the latter. In order to achieve an efficient diagnosis of arc discharges, this paper presents a lightweight identification method based on an improved convolutional neural network model. In this model, the modularized depthwise separable convolution layers are employed as a replacement for the conventional convolution layers, thereby reducing the operands, parameters and elapsed time. Additionally, the activation function is refined with the objective of further reducing the inferring time. Experiments are conducted on defect models to simulate the arc discharges and capture the multiple signals. The adaptive general Stockwell transform is employed to convert the time domain signals into feature maps, which serve as the sample database. The results indicate that the identification model exhibits a 99.3% accuracy rate with multi-channel inputs. The proposed diagnostic method demonstrates an improvement in accuracy of 20% relative to the deep linear neural network, and exhibits an 88% reduction in the number of parameters and a 47% reduction in the elapsed time compared to the conventional convolutional neural network.

The frequency of extreme weather events has become higher, and electricity consumption has also become more complex. These changes increase the risk of overload in distribution transformers (DTs), and this risk threatens the stability and reliability of the power grid. Existing methods have significant limitations. Traditional static threshold methods (based on DGA gas ratios and electrical signal thresholds) fail to consider temporal changes and complex links between factors, while modern machine learning models lack cause–effect relationships over time and clear ways to describe uncertainty. With such motivations, this paper proposes a causal-enhanced hybrid framework, which combines Long Short-Term Memory (LSTM) networks and Random Forest (RF) algorithms. The framework uses causal Seasonal Trend decomposition using Loess (STL) to reveal load patterns at different time scales. The mutual information index and spatiotemporal graph convolutional network (ST-GCN) are used to explore nonlinear relations and reveal how temperature affects load changes. The LSTM model captures time dependence in load series, and the Bayesian optimized Random Forest is used to solve the problem of data imbalance and quantify uncertainty. In addition, the framework constructs an early warning system that combines data from many sources in real time. Test results show that the proposed algorithm exhibits excellent performance in multi-source data environments.