Research and implementation of fault data recovery method for dry-type transformer temperature control sensor based on ISSA-LSTM algorithm

Abstract

This study proposes a temperature control sensor fault data recovery model based on the improved Sparrow Search Algorithm (ISSA) optimizing Long Short-Term Memory (LSTM) to address erroneous data output caused by temperature monitoring sensor failures in dry-type transformers. The model aims to recover faulty sensor data in dynamic processes. It enhances the algorithm by initializing the population using Tent chaotic mapping, employing t-distribution and differential variation perturbations, and introducing a dynamic step size factor. Simulation experiments evaluated the performance of ISSA-LSTM, SSA-LSTM, PSO-LSTM, GA-LSTM, and ACSA-LSTM algorithms. The ISSA-LSTM model outperforms the worst PSO-LSTM model by 82.6%, 82.1%, and 1.8% in terms of MAE, RMSE and $R^2$ values, respectively. Field experiments confirmed the algorithm's effectiveness in recovering data from faulty sensors and different fault types, improving the accuracy and stability of temperature control sensor fault data recovery in dry-type transformers.