Elsevier

Computers & Electrical Engineering

Volume 71, October 2018, Pages 452-464
Computers & Electrical Engineering

A cable fault recognition method based on a deep belief network

https://doi.org/10.1016/j.compeleceng.2018.07.043Get rights and content

Abstract

To meet the requirement of online diagnosis of a cable fault, certain problems should be addressed. Therefore, in this paper, we propose an online cable fault diagnosis method. First, we establish a simulation model of an underground cable distribution system for collecting fault signals. Second, a deep belief network (DBN) is created by the deep learning theory for identifying a cable fault. Finally, we extract the characteristics of the fault signal and classify them into a large number of fault data automatically. A comparison of the results of the cable fault recognition with the proposed method and conventional shallow neural network shows that the DBN is of 97.8%, the conventional back propagation (BP) network is of 86.6%, ACCLN is of 94.1%, which demonstrate that the DBN-based cable fault recognition method has distinct advantages compared with a shallow neural network.

Introduction

Power cables are mainly used to transmit and distribute electrical energy in many fields, such as in urban underground power grids, power plants, industrial and mining enterprises, and underwater transmission. They are the main lines that are used to transmit and distribute high power in a power system. Compared with the overhead lines for power transmission, power cables are plenty of advantageous: they occupy less space, creates less external environment disturbance, and serves to revamp an urban environment, which could be widely applied to many fields. According to the government research report of the Wire and Cable Industry in China of 2015, cable production and growth have gradually increased in 10 years.

Cable fault recognition is mainly divided into offline and online diagnostic modes; currently, the main method of cable fault recognition is offline. The basic principle of offline diagnostics is to shut off the power when a cable fault occurs, and some dedicated devices are used to detect the cable fault. The offline diagnostic mode has several shortcomings, such as wastage of manpower and time, low real-time performance, and damage to the cable. The online diagnostic mode represents a future trend of cable fault recognition. Over the past two decades, to ensure safe operation of power cables, various online monitoring technologies for cable insulation have been developed. Online monitoring technologies for power cables mainly include the DC component method [1], DC superposition method, partial discharge method, AC superposition method, low-frequency superposition method, and so on [2]. Various technologies, including wavelet analysis, neural networks, fuzzy theory, expert systems, and other methods, have been widely applied to online monitoring of cable faults. After reviewing the related literature and research, normal detection methods can be combined, for example, wavelet analysis is used to extract useful information from a fault signal, and the artificial neural network (ANN) [3], [4] or fuzzy logic system (FLS) method is used to identify the fault type [5], [6].

Although the existing research on cable fault recognition has achieved a certain effect, the research object is usually a single cable; hence, the amount of data collected is small. Considering that an actual cable system is composed of multiple sets of cables with a large amount of information, existing methods provide limited capabilities when a large amount of data is present. Deep learning (DL) refers to the use of a multi-layer network architecture model [7], [8], where data feature calculation, signal transformation, and pattern classification are performed [9], [10]. Unlike the conventional learning model, the learning concept simulates and imitates the human brain for establishing comprehensive and deep neural networks, such as text classification [11], image recognition [12], voice recognition [13], wind speed forecasting [14], and feature extraction [15], [16], [17]. An unsupervised cross-modal retrieval method is proposed, which provides an additional regularization by introducing adversarial learning [18], a deep adversarial metric learning approach nonlinearly maps the labeled data pairs of different modalities into a shared latent feature subspace, under which the intra- and inter-class variation are minimized and maximized, respectively, and the difference of each data pair captured from two modalities of the same class is minimized [19], and so on. With the intelligent development of power cable fault recognition, a large amount of information needs to be collected, which requires effective tools for analyzing the information. DL, as a powerful tool, is highly appropriate for analyzing cable faults.

We designed an underground distribution system with 16 sets of cables, which was used for stimulating the occurrence of various cable faults and collecting the corresponding fault information. Then, we proposed a deep belief network model and designed a training model by using a large number of fault signals. The simulation results showed that the proposed method is effective in identifying cable faults, with theoretical significance and usage value.

Section snippets

Deep belief networks

In a deep belief network (DBN), a restricted Boltzmann machine (RBM) is trained by the input data with features obtained from the random neurons in the hidden layer. The active training data are processed at the next layer as the input data. The learning process in layer by layer can be considered a set of features of learning features. When a new layer is added to the DBN, the lower bounds on the log probability of the raw training data are promoted.

Power cable fault recognition

Fig. 5 shows the recognition and classification process of cable faults. First, the fault types are identified. They can be a ground fault, short-circuit fault, or open-circuit fault. Then, the fault phase of the cable is identified, which is known as the ABC three-phase detection. Given that the distribution system considered in this study has 16 sets of cables, all possible categories were (7 + 4 + 7) * 16 * 3 = 864, when the multi-phase fault was considered. In this situation, the

Network training

For better identifying a cable failure, training the network first is essential, whose purpose is to obtain the parameters of the cable failure model. Thus, huge fault data are required to be learned for recognizing the characteristics of fault data. After the data is collected according to Table 4, the samples are randomly distributed into five sample sets as shown in Table 5, and the samples are inputted to the network for training. The tested data set is verified after the training has

Conclusions

In this paper, A distributed system consisting of 16 groups of cables is built to collect a large amount of fault data, which is used for stimulating the fault diagnosis of cable. The model has rich and adjustable performance to simulate the general situation of fault occurrence. It has a certain reference meaning to the actual power cable fault detection. A deep belief network is proposed to identify cable faults, in a way that enables more functions than conventional single cable fault

Acknowledgments

This paper is supported by the National Science Foundation for Young Scientists of ChinaNo. 51704229 and No. 51405381, the support from the Industrial Science and technology research foundation of Shaanxi province, No. 2015GY020, the Industrial Science and technology research foundation of Shaanxi province, No. 2016GY040.

Conflicts of interest

The authors declare that there is no conflict of interests regarding the publication of this paper.

Xuebin qin received his Ph.D. degree from Tottori University, Japan in 2013. After receiving his Ph.D., he worked at Xi'an University of Science and Technology as a lecturer in 2013. His research interests include computer/robot vision, brain-computer interface technology and mixed reality.

Cited by (44)

  • Transmission line fault detection and classification based on SA-MobileNetV3

    2023, Energy Reports
    Citation Excerpt :

    This end-to-end machine learning greatly improves the generalization ability of fault classifiers without manually designing features and choosing appropriate classifier, which is robust to different cases without repeated modification and adjustment. Considering that transmission line fault signals are typical time-series data, the autoencoder with unsupervised feature learning (Chen et al., 2018), adaptive deep belief network (ADBN) (Liang et al., 2019; Zhang et al., 2018), long short-term memory network (LSTM) (Omar et al., 2020) and one-dimensional convolutional neural network (1D-CNN) (Ye et al., 2020) are used for fault identification and classification in transmission line. Although models based on deep networks have been proved to be effective in fault detection and classification, this still imposes challenges such as recognition accuracy and computational complexity.

  • Fast and Accurate Fault Detection and Classification in Transmission Lines using Extreme Learning Machine

    2023, e-Prime - Advances in Electrical Engineering, Electronics and Energy
  • Power System Fault Diagnosis: A Wide Area Measurement Based Intelligent Approach

    2022, Power System Fault Diagnosis: A Wide Area Measurement Based Intelligent Approach
View all citing articles on Scopus

Xuebin qin received his Ph.D. degree from Tottori University, Japan in 2013. After receiving his Ph.D., he worked at Xi'an University of Science and Technology as a lecturer in 2013. His research interests include computer/robot vision, brain-computer interface technology and mixed reality.

Yizhe Zhang received his BS degree from Xi'an University of Science and Technology. He is a graduate student. His research interests include cable fault diagnosis.

Wang Mei received her M.S. and Ph.D. degrees in Electrical Automation, and Safety Technology and Engineering from Xi'an University of Science and Technology, China, in 1990 and 2006, respectively. She current research interests include the signal processing and analysis, electrical device, control system of cloud and brain, fault diagnosis, IoT.

Gang Dong received his Ph.D. degree from Xi'an University of Science and Technology in 2018. After receiving his Ph.D., he worked at Zhongping Information Technology Co., Ltd. in 2018. His research interests are the research of electrical equipment fault diagnosis in coal mine.

Jun Gao received his master degree from Xi'an University of Science and Technology in 2009. After receiving his Ph.D., he worked at CCTEG XI'AN Research Institute. His research interests is the research of coal mine safety.

Pai Wang received his M.S. and Ph.D. degree in Measurement Control and Instruments, with the Institute of Intelligent Control and Image Engineering (ICIE), Xidian University. His research interests include Nondestructive testing technology Electrical Resistance Tomography, The inverse problem in image processing, and Brain Computer Interface.

Jun Deng received his Ph.D. degree from Xian Jiaotong University in 2004. He is doctoral supervisor. He is mainly engaged in the research and application of coal fire disaster prevention theory and technology.

Hongguang Pan received his Ph.D. degree from Xi'an Jiaotong University in 2015, From September 2013 to March 2006, he was a visiting Ph.D. student in Lehigh University, America. Now, he is a lecture at Xi'an University of Science and Technology. His research interests include predictive control, brain-machine interface.

Reviews processed and recommended for publication to the Editor-in-Chief by Associate Editor Dr. Huimin Lu.

View full text