Investigation on the fault monitoring of high-voltage circuit breaker using improved deep learning.

Mechanical faults are the main causes of abnormal opening, refusal operation, or malfunction of high-voltage circuit breakers. Accurately assessing the operational condition of high-voltage circuit breakers and delivering fault evaluations is essential for the power grid's safety and reliability. This article develops a circuit breaker fault monitoring device, which diagnoses the mechanical faults of the circuit breaker by monitoring the vibration information data. At the same time, the article adopts an improved deep learning method to train vibration information of high-voltage circuit breakers, and based on this, a systematic research method is employed to identify circuit breaker faults. Firstly, vibration information data of high-voltage circuit breakers is obtained through monitoring devices, this vibration data is then trained using deep learning methods to extract features corresponding to various fault types. Secondly, using the extracted features, circuit breaker faults are classified and recognized with a systematic analysis of the progression traits across various fault categories. Finally, the circuit breaker's fault type is ascertained by comparing the test set's characteristics with those of the training set, using the vibration data. The experimental results show that for the same type of circuit breaker, the accuracy of this method is over 95%, providing a more efficient, intuitive, and practical method for online diagnosis and fault warning of high-voltage circuit breakers.

Fault Diagnosis of Medium Voltage Circuit Breakers Based on Vibration Signal Envelope Analysis.

In modern power systems or new energy power stations, the medium voltage circuit breakers (MVCBs) are becoming more crucial and the operation reliability of the MVCBs could be greatly improved by online monitoring technology. The purpose of this research is to put forward a fault diagnosis approach based on vibration signal envelope analysis, including offline fault feature training and online fault diagnosis. During offline fault feature training, the envelope of the vibration signal is extracted from the historic operation data of the MVCB, and then the typical fault feature vector M is built by using the wavelet packet-energy spectrum. In the online fault diagnosis process, the fault feature vector T is built based on the extracted envelope of the real-time vibration signal, and the MVCB states are assessed by using the distance between the feature vectors T and M. The proposed method only needs to handle the envelope of the vibration signal, which dramatically reduces the signal bandwidth, and then the cost of the processing hardware and software could be cut down.

Identification of three metabolic subtypes in gastric cancer and the construction of a metabolic pathway-based risk model that predicts the overall survival of GC patients.

Gastric cancer (GC) is highly heterogeneous and GC patients have low overall survival rates. It is also challenging to predict the prognosis of GC patients. This is partly because little is known about the prognosis-related metabolic pathways in this disease. Hence, our objective was to identify GC subtypes and genes related to prognosis, based on changes in the activity of core metabolic pathways in GC tumor samples. Differences in the activity of metabolic pathways in GC patients were analyzed using Gene Set Variation Analysis (GSVA), leading to the identification of three clinical subtypes by non-negative matrix factorization (NMF). Based on our analysis, subtype 1 showed the best prognosis while subtype 3 exhibited the worst prognosis. Interestingly, we observed marked differences in gene expression between the three subtypes, through which we identified a new evolutionary driver gene, CNBD1. Furthermore, we used 11 metabolism-associated genes identified by LASSO and random forest algorithms to construct a prognostic model and verified our results using qRT-PCR (five matched clinical tissues of GC patients). This model was found to be both effective and robust in the GSE84437 and GSE26253 cohorts, and the results from multivariate Cox regression analyses confirmed that the 11-gene signature was an independent prognostic predictor (p < 0.0001, HR = 2.8, 95% CI 2.1-3.7). The signature was found to be relevant to the infiltration of tumor-associated immune cells. In conclusion, our work identified significant GC prognosis-related metabolic pathways in different GC subtypes and provided new insights into GC-subtype prognostic assessment.

Keyword spotting techniques to improve the recognition accuracy of user-defined keywords.

The existing keyword spotting (KWS) techniques can recognize pre-defined keywords well but have a poor recognition accuracy for user-defined keywords. In real use cases, there is a high demand for users to define their keywords for various reasons. To address the problem, in this work, three techniques have been proposed, including incremental training with revised loss function, data augmentation, and fine-grained training, to improve the accuracy for the user-defined keywords while maintaining high accuracy for pre-defined keywords. The proposed techniques are applied to a classical KWS model (cnn-trad-fpool3) and a state-of-the-art KWS model (res15) respectively. The experimental results show that the proposed techniques have better recognition accuracy than several existing methods for the recognition of use-defined keywords. With the proposed techniques, the recognition accuracy of user-defined keywords on cnn-trad-fpool3 and res15 are significantly improved by 21.78% and 24.42%, respectively.