Bidirectional Recurrent Neural Network based on Multi-Kernel Learning Support Vector Machine for Transformer Fault Diagnosis

Abstract

Traditional neural network has many weaknesses, such as a lack of mining transformer timing relation, poor generalization of classification, and low classification accuracy of heterogeneous data. Aiming at questions raised, this paper proposes a bidirectional recurrent neural network model based on a multi-kernel learning support vector machine. Through a bidirectional recurrent neural network for feature extraction, the features of the before and after time fusion and obvious data are outputted. The multi-kernel learning support vector machine method was carried out on the characteristics of data classification. The study of multi-kernel support vector machines in the weighted average of the way nuclear fusion improves the accuracy of characteristic data classification. Numerical simulation analysis of the temporal channel length for sequential network diagnostic performance, the effects of multi-kernel learning on the generalization ability of support vector machine, the influence on heterogeneous data processing capabilities, and transformer fault data classification experiment verifies the correctness and effectiveness of the bidirectional recurrent neural network based on multi-kernel learning support vector machine model. The experiment result shows that the diagnosis performance of bidirectional recurrent networks based on a multi-kernel learning support vector machine is better, and the prediction accuracy of the model is improved by more than 1.78% compared with several commonly used neural networks.

Interactive spatiotemporal LSTM approach for enhanced industrial fault diagnosis

“…The current issue and full text archive of this journal is available on Emerald Insight at: https://www.emerald.com/insight/0036-8792.htm feature representation, struggle to capture long-span temporal structures within a single temporal window (Ji et al, 2012;Tran et al, 2015;Zhu et al, 2022), rendering them less suitable for spatiotemporal sequence prediction problems where time-series relationships are crucial. Conversely, convolutional LSTMs (ConvLSTMs), despite their capacity to process spatial characteristics in three-dimensional tensors, often fail to retain detailed information in raw data, thereby compromising their predictive accuracy (Miao et al, 2022;Zhou et al, 2022;Zhang et al, 2022;Zhao et al, 2023). To address these challenges, we propose a novel spatiotemporal LSTM (ST-LSTM) that introduces a groundbreaking memory state transfer method, enabling interaction and updating of memory states across different network layers (Goodfellow et al, 2016).…”