In this paper, a performance optimization and fault prevention model based on big data analysis is proposed to solve the problems of accelerated insulation aging and increased fault risk faced by 220kV spare transformer under complex working conditions. Firstly, a multi-source data acquisition system is constructed, and the data are aligned in time and space by dynamic time warping (DTW) algorithm, and the joint feature space is constructed by using Variational Autoencode (VAE) to optimize feature fusion. Secondly, an improved random forest algorithm is used to establish a dynamic performance evaluation model, and a time-varying weight mechanism is introduced to enhance the perception of the dynamic characteristics of equipment aging. Combined with the load change trend predicted by Long Short-Term Memory (LSTM) model, an operation optimization model is constructed to realize the dynamic adjustment of load and cooling system and reduce energy consumption and aging rate. Thirdly, the LSTM-Attention time series model is constructed to realize early fault warning, attention mechanism is introduced to enhance the contribution of key time points to fault prediction, and the physical prior knowledge of dissolved gas analysis (DGA) is introduced as a constraint condition through the auxiliary channel to realize the integration of data-driven and mechanism model. Finally, the transfer learning enhancement module is designed, and the maximum mean difference (MMD) is used as the domain adaptation metric. By minimizing the loss of MMD, the generalization ability of the model on the target device is improved, and the over-fitting problem under the condition of small samples is alleviated. In the pilot application of 12 220kV spare transformers in a provincial power grid, the model has achieved the expected goals in key indicators such as fault early warning accuracy, average early warning lead time, no-load loss reduction rate, standard deviation of hot spot temperature, etc., effectively reducing the unplanned outage rate and annual maintenance cost, verifying the remarkable effect of the model in improving energy efficiency, prolonging equipment life and adapting to small sample scenarios, and providing a new solution for intelligent operation and maintenance of transformers in power system.
Luo et al. (Sun,) studied this question.