Research on online monitoring of grid current transformer based on transformers and BiGRU

Abstract

Transformers play a crucial role in ensuring the safety of power grids. It is of great value to diagnose faults using the large amount of power data generated by the grid. It is possible to detect internal latent faults in transformers in advance of their occurrence if the normal operating condition of the transformer is detected in a timely manner. To perform online fault diagnosis of grid current transformers, we combine the Transformer and BiGRU methods. There is a temporal component to the fault input sample sequences. By using Transformer’s multi-headed attention mechanism to extract deep features from fault input sample sequences, the temporal association between latent variables can be fully exploited. As a result of the extraction of features, BiGRU is used to generate fault category coding as an output. The experimental results indicate that using the proposed algorithm achieves better results than using a single model, which is useful for the study and application of fault diagnosis in power grids for current transformers.

Keywords

Current transformer transformer BiGRU latent variable fault sample sequence fault category coding

1. Introduction

In power systems, current transformers are used to measure and provide critical information to protection devices. This is essentially a special type of transformer, which unifies and standardizes various instruments and protection devices by transforming large currents on the primary side into small currents on the secondary side. The current transformer can greatly avoid the disadvantage that the shunt has no electrical isolation. the output accuracy and voltage can achieve the effect of the shunt at the same time. Its non-contact detection function can greatly enhance the safety of workers in construction or power grid maintenance, and it is very convenient to use in practical projects.It is important to note that the safe operation of current transformers is not only important for the equipment itself but also directly affects the safe and stable operation of the power grid, which holds an essential and important position in the electricity distribution syste [1]. Voltage transformer fault diagnosis is currently mainly based on routine outage inspections on the power grid. As a result of the outage inspection cycle not being able to detect latent defects in the equipment in time, equipment failures and power outages often result. It is for this reason that online monitoring technology has been developed for current transformers. Using fuzzy clustering state ranking and uniformity measures, Huang et al. [2] mined and analyzed real-time monitoring state vector data. The algorithm complexity is not high, but the operational efficiency is high. However, the number of state clusters must be selected manually, the algorithm results are highly dependent on the initial state value selected, and the applicability of the algorithm is limited. An improved artificial fish swarm algorithm was used by Jia et al. [3] to optimize the wavelet neural network for diagnosing current transformer faults. Based on the excellent window scale performance of wavelet function bases in both the time and frequency domain, Xing et al. [4] conducted fault diagnosis research using data obtained from current transformers, which has certain advantages in convergence accuracy. However, it is difficult to select the wavelet parameters. Lin et al. used thermal imaging depth convolution technology to perform fault classification semantic detection on current transformer infrared images, and achieved good results. However, the problems of high fault noise and low super pixel detection efficiency of infrared images have not been solved well. Xing et al. combined infrared image depth convolution, CCD optical image and other multi-source sensor data to detect the defects of power equipment, and used ZigBee wireless sensor network networking for information transmission and multi-source Kalman filtering data analysis methods to achieve the effect of reducing errors, but it has the problem of real-time asynchronous transmission delay. Liu et al. used multi-scale wavelet deep neural network and multi data fusion methods to analyze the fault of power equipment signals, and made full use of the high time resolution of wavelet basis change, which has greatly improved the accuracy of fault classification, but it is also difficult to select wavelet basis function parameters. Our study explores the online monitoring and fault diagnosis of current transformers in a grid environment using a transformer combined with the BiGRU real-time prediction method based on the time scale, secondary output voltage amplitude, phase angle, frequency, temperature, humidity, and other types of data collected from current transformers in the equivalent test environment. As a result of the experimental results, the proposed method shows some improvement in diagnostic accuracy and reliability and has practical value for online monitoring and metering diagnosis of transformers in power grids [5].

2. System overview

2.1 Composition of the overall system

The system consists of three components: a data acquisition unit, a data transmission unit, and a data analysis and monitoring unit. Distributed storage and big data calculation technologies have been introduced to meet the needs of storing and optimizing massive power data. By using the online monitoring sensor head, the data acquisition unit obtains status communication information, which includes the insulation gas pressure, insulation temperature, and humidity. Using the secondary voltage and secondary current digital signals, the data package is then formed and encoded into the data transmission unit, It mainly considers the selection standards and requirements of the industry to choose actual characteristics of the grid sensor data. and is then transmitted. There are two main parts to the data transmission unit, namely the data routing gateway and the big data distributed storage. By combining the data collected by each sensor with noise reduction, the former determines a fixed packet length limit taking into account the transmission period and the actual transmission rate. The result is that not all the information in the message is transmitted, but rather in a rotating cycle in the form of “state quantity $+$ state quantity sequence number” [6, 7, 8]. A big data distributed storage unit integrates structured and unstructured data in the grid by combining the distributed database Hbase with a relational database MySQL to handle online analysis and processing of the upper layer applications. Big data complex analysis and data application layers make up the data analysis and monitoring unit. Spark technology is primarily used for big data analysis. Figure 1 illustrates the architecture of the data application layer, which consists primarily of three components: decision support, statistical query, and data mining.

Figure 1.

Diagram of the system architecture.

2.2 Current transformer

Figure 2.

Diagram of the simple current transformer.

3. Principles of transformers and BiGRUs

In order to realize the online diagnosis of current transformer faults of power equipment and the acquisition of fault data, the equivalent test environment platform established will store and filter the collected data in all dimensions, and deeply combine the strong sequence data processing capabilities of Transformer and BiGRU to divide training and test set data for training and reasoning of fusion algorithm.

Figure 3.

Technical framework.

3.1 Principles of transformers

Transformer is not only widely used in natural language processing. As a result of its sequence feature concept, Transformer has gradually penetrated other application fields, such as deep feature extraction. A breakthrough has been achieved through the application of the parallel Attention mechanism. In the case of sequence-to-sequence extraction feature networks, the Encoder-Decoder architecture is generally used before the Transformer is generated. RNN is fundamental to the entire approach and is used as the codec’s basic structure. The input sequence is converted into a Context Vector by the Encoder, and the Context Vector is used as the input to the Decoder to predict the output. As a fundamental part of the method, RNN is used as the basic structure of the encoder and decoder. The encoder encodes the input data into a low-dimensional vector, which is then used as input for the decoder. RNN has achieved good results, but it has several obvious shortcomings.

The classical network structure Transformer was born out of a paper published by Google researchers in 2017 called Attention is All You Need [18]. To address the problems related to sequence models, an attention-only based structure is proposed. Transformers consist of an encoder and a decoder. Encoders consist of a stack of six encoders, and decoders consist of a stack of six decoders. The classical Transformer structure is illustrated in Fig. 4.

Figure 4.

Transformer structure diagram.

The encoder consists of $N=$ 6 identical layers. In the figure above, the layer is represented by the cell on the left side. There are six of them in this example. There are two sub-layers in each layer, which are the multi-headed attention mechanism and the fully connected layer. The output of each sub-layer is connected with residuals and regularized, $x$ represents the input characteristic code so that it can be expressed as follows:

$\displaystyle\textit{sub\_layer}=\textit{LN}(x+\textit{sublayer}(x))$ (1)

There are n attentional computation units in multi-headed attention. Parallel computations are used to determine the long term and short term dependencies of the input sequence. To click on the attention mechanism, the n weight matrices are randomly initialized and scaled by matrix linear operations, then output to the next layer of the Transformer by concatenation operations. The multi-headed attention mechanism is illustrated in Fig. 5.

Figure 5.

Multi-headed attention mechanism.

Following is a description of the computational procedure for the multi-headed attention mechanism:

$\displaystyle\textit{attention\_value}=\textit{Attention(Q,K,V)}$ (2) $\displaystyle\textit{Attention}(Q,K,V)=\textit{soft}\max\left(\frac{QK^{T}}{% \sqrt{d_{k}}}\right)V$ (3) $\displaystyle\textit{head}_{i}=\textit{Attention}(\textit{QW}_{i}^{Q},\textit{% KW}_{i}^{K},\textit{VW}_{i}^{V})$ (4) $\displaystyle\textit{Mulhead}(Q.KV)=\textit{Concat}(\textit{head}_{1},\textit{% head}_{2},\ldots,\textit{head}_{n})W^{O}$ (5)

Where LN(.) represents the layer regularization operator. $Q$ , $K$ , and $V$ represent the query matrix, key matrix, and value matrix, respectively. $d_{k}$ represents the balance penalty parameter. $W^{Q}_{i}$ , $W^{K}_{i}$ , $W^{V}_{i}$ represent linear transformation matrices. Concat(.) represents the matrix concatenation operator, and $W^{O}$ represents the parameter matrix.

3.2 BiGRU principle

A variation of GRU is BiGRU. Gated Recurrent Neural Networks (GRU) are proposed to solve the problems of long-term memory and gradient in back propagation. Compared with the long and short-term memory model, it reduces the original three control gates to two, and its structure is relatively simple, with few parameters and high calculation efficiency. Under the allowable cost of calculation error, it is a good choice to select GRU model to improve the convergence speed of the modelBiGRU’s specific structure can be seen in Fig. 6.

Figure 6.

BiGRU neural network structure.

The current hidden layer state $h(t)$ of BiGRU is determined $x_{t}$ by the current input, the output of the forward hidden $\overrightarrow{h}\to(t-1)$ layer state at moment $t-1$ , and the output $\overleftarrow{h}(t-1)$ of the reverse hidden layer state. $\omega$ , $\sigma$ , $b(t)$ represents the weight and offset parameters of the output of the forward unit and backward unit respectively, and $\overrightarrow{h}(t)$ , $\overleftarrow{h}(t)$ represens the state values of the forward and backward hidden layers respectively. Since the BiGRU can be considered as two one-way GRUs [19, 20, 21, 22]. The summation process is shown in Eqs (6)–(8):

$\displaystyle\overrightarrow{h}(t)=\textit{GRU}(x_{t},\overrightarrow{h}(t-1))$ (6) $\displaystyle\overleftarrow{h}(t)=\textit{GRU}(x_{t},\overleftarrow{h}(t-1))$ (7) $\displaystyle h(t)=\omega\overrightarrow{h}(t)+\sigma\overleftarrow{h}(t)+b(t)$ (8)

A one-way gated recurrent neural network is referred to as a GRU. A diagram of the GRU structure is shown in Fig. 7.

Figure 7.

GRU neural network structure.

Where the computational procedure is shown below [23, 24, 25, 26]:

$\displaystyle r=\sigma\left(W^{r}\frac{x(t)}{h(t-1)}\right)$ (9) $\displaystyle z=\sigma\left(W^{z}\frac{x(t)}{h(t-1)}\right)$ (10) $\displaystyle h^{\ast}=\tanh\left(W\frac{x(t)}{h(t-1)^{\prime}}\right)$ (11) $\displaystyle h(t-1)^{\prime}=h(t-1)\odot r$ (12) $\displaystyle h(t)=(1-z)\odot h(t-1)+z\odot h(t-1)^{\prime}$ (13)

Where () represents Hadamard Product, $h(t-1)$ represents the state value of the hidden layer at time $t-1$ . $r, z$ represent the reset and update gates [27, 28, 29, 30], respectively, which are responsible for the “trade-off” between the implicit layer and the input state values. $x(t)$ represents the input value at the moment. $y(t)$ represents the output value at time t.

4. Result of experiments and experiments

4.1 Data acquisition

It has been established to monitor the current transformer of the power grid online by developing a test environment platform based on the four types of faults mentioned above, namely, short circuits between turns or layers of the primary and secondary coils, short circuits in secondary circuits, overheating faults, and others. Furthermore, five important parameters were selected and collected as samples, including insulation gas pressure, insulation temperature and humidity, secondary voltage, and secondary current. There were 3200 sets of data selected. Two thousand data sets are used as training samples, and eight hundred data sets are used as test samples. To achieve standardization of the data scale, the voltage, current, pressure, temperature, and humidity parameters were mapped to the interval of [0,1]. In general, the formula is as follows:

$\displaystyle y=\frac{x-x_{\min}}{x_{\max}-x_{\min}}$ (14)

$\displaystyle\textit{MSE}=\frac{\sum\limits_{i=1}^{N}{(y_{i}-\bar{y}_{i})}}{N}$ (15)

Where $y$ represents the output after linear transformation. $x,x_{\min},x_{\max}$ represent the original indicator input, the minimum indicator input, and the maximum indicator input, respectively. MSE represents the mean square error of the model results. $y_{i}$ , $\bar{y}_{i}$ represents the actual and predicted values, respectively, and $N$ represents the sample size.

4.2 Experimental analysis

In order to collect sensor data required for fault detection experiment in multiple dimensions and solve the problem of few samples of traditional fault data, an equivalent test environment platform has been established. This platform is fast and pluggable, and can obtain a large number of enough data for many times. The information about the equivalent test environment platform’s actual environment and important experimental sensor parameter models are as follows:

Table 1
Important information of sensor

Sensor type	Sensor type	Important parameters
Test platform	HMQB-C	(1%–149%)5A, (5%–149%)100V
Temperature and humidity sensor	GXHT35-DIS	( $-$ 40 ${{}^{\circ}}$ C–90 ${{}^{\circ}}$ C) (0–100%)
Current sensor	HZIB-C11	0 $\pm$ 4 V, 0 $\pm$ 5 V
Pressure sensor	24PC	0–250 psi

Figure 8.

Experiment platform diagram.

While obtaining the test data, we also debugged the overall architecture parameters. After many experimental comparisons and grid parameter analysis, the overall structure of Transformer and BiGRU is shown as follows:

Table 2

Schema parameter

Structure name	Transformer	BiGRU
Specific structure	5 layers of Transformer, each layer (5-8-5)	2 layers of GRU, each layer (5-5)

The five states of the test circuit are binary coded as 0000, 0001, 0010, 0011, 0100 for normal operation, short circuit between turns or layers of the primary or secondary coil, short circuit fault in the secondary circuit, overheating fault, and other faults respectively. The deep learning framework of Pytorch is used, the GPU is Nvidia 2060, and the version of Cuda is Cuda11.3. Python version is Python 3.6. The training inference is performed on Windows 10 operating system. The training learning rate is set to polynomial decay mode, initial learning rate 0.1, decay steps 50, minimum learning rate 0.005, decay factor 0.5. The Transformer module is experimentally tested with 3 layers, the BiGRU module is 6 layers in both directions, the output neurons are 5, the corresponding binary code is output, batchsize $=$ 5, epoch $=$ 200 The following. Figure 8 shows the convergence curve of Transformer combined with BiGRU for training iterations.

Figure 9.

Transformer $+$ GRU neural network structure.

From Fig. 9, we can see that the Transformer combined with BiGRU network converges after about 100 Epoch iterations. The convergence accuracy is high, about 0.01. Transformer extracts the fault feature sequences in depth, including local near features and long-range correlation features, and then extracts the predictions by BiGRU’s bi-directional information. From the above figure, we can see that the proposed method is efficient in terms of convergence speed and convergence accuracy.

In order to further investigate the performance difference between the proposed method and the traditional method, a comparison experiment on GRU, Transformer algorithm is conducted. The 800 test samples are divided into 5 equal parts. Each of them is 160 test samples. The following Figs 9–11 and Table 1 show the actual grid current transformer diagram and experimental results respectively.

Table 3

Comparison of accuracy of algorithms

Fault type	GRU			Transformer			GRU+Transformer
	Total	Predicted correctly	Recall rate	Total	Correct prediction	Recall rate	Total	Correctly predicted	Recall rate
0000	160	140	87.50%	160	145	90.60%	160	150	93.75%
0001	160	142	88.75%	160	143	89.37%	160	151	94.37%
0010	160	143	89.30%	160	144	90%	160	151	94.37%
0011	160	142	88.75%	160	144	90%	160	152	95%
0100	160	136	85%	160	140	87.50%	160	146	91.25%
Average recall rate	800	703	87.86%	800	716	89.50%	800	750	93.75%

Table 4

Comparison of accuracy of algorithms

Fault type	SVM			6-layer convolution neural network
	Total	Predicted correctly	Recall rate	Total	Predicted correctly	Recall rate
0000	160	134	83.75%	160	143	89.37%
0001	160	139	86.87%	160	141	88.12%
0010	160	140	87.50%	160	140	87.5%
0011	160	136	85%	160	139	86.87%
0100	160	131	81.87%	160	139	86.87%
Average recall rate	800	680	85%	800	702	87%

Figure 10.

Grid current transformer operating diagram.

Figure 11.

Comparison chart of discount effect of three methods.

Figure 12.

Comparison of discounted effects of five methods.

It can be seen from Tables 3, 4 and Fig. 11, the GRU+Transformer combined model achieves higher accuracy in four fault states and normal states than only using GRU or Transformer algorithm. Especially, for 0001 fault and 0011 fault, The accuracy of GRU+Transformer combined model is about 5% higher than that of Transformer. The average recall of Transformer is 1.64% higher than that of GRU, and the average recall of GRU+Transformer is 4.25% higher than that of Transformer. It shows that GRU+Transformer combined model has a good effect on the detection of the above five states. Compared with SVM algorithm and 6-layer convolutional neural network, the GRU+Transformer combined model shows a better effect. The average recall rate of the combined model is increased by 8.75% and 6.75% respectively, which shows that the combined model is superior to both traditional machine learning and deep neural network.

As can be seen from Fig. 12, the green dotted line represents the GRU+Transformer composite model, which reaches the convergence state after iteration of 100 Epochs, with the convergence accuracy of 0.01. The black solid line and blue dotted line represent the GRU algorithm and the Transformer algorithm respectively. The GRU algorithm converges after iteration of 80 Epochs, with the convergence speed rising, but the convergence accuracy is lower than that of the Transformer algorithm. The Transformer algorithm reaches the convergence after iteration of 120 Epochs, The convergence accuracy is 0.015 higher than that of GRU algorithm. The red dotted line and the yellow implementation represent the iterative effects of SVR algorithm and 6-layer convolution neural network algorithm respectively. The final results of the two algorithms are in the upper left corner of the decision plane, and their iteration times and algorithm accuracy are poor. It can be seen from the broken line graph of the effects of the five algorithms that the GRU+Transformer combined model has greater generalization and effectiveness.

5. Concluding remarks

In this paper, four fault states and normal states of current transformers in power grid are studied by combining the combined model of BiGRU [31, 32, 33] and Transformer [34, 35], and the experimental data of current transformers are obtained off-line. Through the analysis of fault experiments, it can be seen that the combined model of BiGRU and Transformer has improved convergence accuracy and convergence speed compared with the single algorithm, but the source of experimental data used in this paper is narrow, and the comparative experiment is not complete and sufficient , which is the next stage.

References

Guojun

Wenzhou

Qun

Lei

. Single-phase ground fault selection method for small current grounding system based on compensation parameters. Power System Protection and Control. 2021; 2(49): 1-9.

Wenxia

Hui

Ming

Yuan

Daixiao

Xiang

. Broadband nonlinear model of capacitive voltage transformer considering saturation characteristics of intermediate transformer. Proceedings of the CSEE. 2021; 14(41): 5034-5043.

Wenxia

Hui

Ming

Yuan

Daixiao

Xiang

. Broadband nonlinear model of capacitive voltage transformer considering saturation characteristics of intermediate transformer. Proceedings of the CSEE. 2021; 14(41): 5034-5043.

Chao

Xinsheng

Guoxiong

Zhirong

Xiong

. Research on application of transformer winding short-circuit reactance online measurement engineering. High Voltage Engineering. 2020; 11(46): 3943-3950.

Jie

Ming

Xuan

Zhirong

Xiong

. Research on nonlinear correction method of current transformer based on support vector machine. Electric Machines and Control. 2020; 10(24): 130-138.

Hongyu

Xiaoning

Yanlei

. Research on adaptive three-phase reclosing of distribution network using injection signal. Power System Technology. 2021; 7(45): 2623-2630.

Yuwen

Cuihua

Fan

Yaojun

Baichao

. High Voltage Engineering. Power System Technology. 2020; 9(46): 3154-3163.

Hanli

Danqing

Zhenxing

Jingguang

Yongbo

Hao

CHEN

. Line longitudinal protection applying two-dimensional spatial mapping distribution characteristics of current sampling values at both ends. Electric Power Automation Equipment. 2020; 8(40): 133-141.

Xuan

Jiajing

Xichao

Shipu

Hua

Sujie

. Analysis and compensation of error source of power supply type voltage transformer with automatic voltage adjustment function. High Voltage Engineering. 2020; 7(46): 2554.

10.

Fubin

Peng

Shipu

Jieqing

. Fault mechanism analysis of all-fiber current transformer with sine wave modulation. Automation of Electric Power Systems. 2020; 17(44): 153-160.

11.

Nan

Qifeng

Qiao

. Phase delay detection method of optical voltage transformer using Newtonian ring grating. Automation of Electric Power Systems. 2020; 8(44): 185-191.

12.

Kun

Hao

Xiangjun

Yuling

Feng

Xiaoping

. A new method for measuring the double-ended resonance of ground insulation parameters in resonant grounding distribution network. Automation of Electric Power Systems. 2020; 12(44): 154-161.

13.

Chao

Xiangjun

Kun

Yanxia

Guojie

. New technology for real-time measurement of ground parameters in neutral point ungrounded distribution network based on dual voltage transformer. Power System Technology. 2020; 7(44): 2657-2664.

14.

Xiang

Haiyang

Xian

Mingming

Jianpeng

Jianwei

. Effect of lightning impulse voltage on measurement accuracy of primary and secondary fusion equipment for power distribution. Electric Power Automation Equipment. 2020; 2(40): 214-221+224.

15.

Peng

Wei

Bingyin

Mingna

Dong

Shengpeng

. Design method of current transformer energy extraction power supply based on adaptive power output control. Automation of Electric Power Systems. 2020; 3(44): 194-200.

16.

Peng

Mingmin

Guanchen

Zhichao

Henglin

. Experimental study on electromagnetic immunity performance of electronic current transformer acquisition card under extreme temperature conditions. Power System Technology. 2020; 2(44): 718-724.

17.

Xuhong

Jierui

Shaosheng

Lin

. Online power extraction method of current transformer using dual magnetic circuit. Automation of Electric Power Systems. 2020; 4(44): 187-194.

18.

Hai’an

Zhu

Huinan

Hongbin

Xue

Shao

. Online evaluation of metering performance of capacitive voltage transformer based on principal element analysis. Electric Power Automation Equipment. 2019; 5(39): 201-206.

19.

Feng

Xiaodong

Chunyang

Min

Fuchang

. Measurement method of error voltage coefficient of high accuracy standard voltage transformer. Proceedings of the CSEE. 2019; 24(39): 7421-7428+7515.

20.

Zhenhua

Yuan

Siqiu

Zhenxing

. Combined electronic instrument transformer based on high-precision digital integration method. Electric Power Automation Equipment. 2019; 39(3): 137-142.

21.

Yongyan

Jian

Qionglin

Xiaopeng

Kaipei

. Research on harmonic measurement error of capacitive voltage transformer. Electric Power Automation Equipment. 2017; 3(37): 167-174.

22.

Zhihong

Song

Guoqing

Zhizhong

Peili

Wenbin

. Electric Power Automation Equipment. 2017; 1(37): 212-216.

23.

Kaipei

Yue

Bin

. Transient overcurrent analysis of neutral point ungrounded grid voltage transformer based on instantaneous symmetric component method. Electric Power Automation Equipment. 2016; 11(36): 157-164.

24.

Yang

Jiandong

Xiaoqing

Yunge

Yiyang

Wenchao

Zhuanting

. J-A model parameter identification and MAXUS flow mode test of current transformer. Proceedings of the CSEE. 2016; S1(36): 240-245.

25.

Qingxin

Hongbin

Tongtian

Boming

Wenchuan

Qinglai

. Analysis of false data injection attack of transformer in substation state estimationAutomation. of Electric Power Systems. 2016; 17(40): 79-86.

26.

Lesniewska

Lisowiec

. Estimation of Influence of External Magnetic Field on New Technology Current-to-Voltage Transducers Operating in Protection System of AC Power Networks. IEEE Transactions on Power Delivery. 2016; 2(31).

27.

Hang

Yan

Xia

Feng

Guoying

. On-site calibration method and key issues of HVDC electronic current transformer. Proceedings of the CSEE. 2016; 19(36): 5227-5235+5404.

28.

Qiang

Wei

Songnong

Xingzhe

Dengyun

. High Voltage Engineering. 2016; 6(42): 1781-1789.

29.

Ripka

Draxler

Styblikova

. Measurement of DC Currents in the Power Grid by Current Transformer. IEEE Transactions on Magnetics, 2013; 1(49).

30.

Xuedong

. Research on electronic transformer fault diagnosis method based on wavelet-fractal theory. Chongqing University, 2012.

31.

Huanzhang

Xingguo

Zexin

Yarong

Lilac

. A saturation identification method for current transformer using sampling value of current mutation. Power System Technology. 2016; 11(40): 3574-3579.

32.

Santos

Slomovitz

. Error Compensation Of Current Transformers Used In High Voltage Networks. IEEE Latin America Transactions. 2006; 3(4).

33.

Yuxiang

Qixin

. Analysis of the influence of secondary side crimping on the automatic verification of current transformer. Chinese Journal of Scientific Instrument. 2016; 2(37): 316-322.

34.

Yin

Lin

Cui

. Wireless Traffic Prediction With Scalable Gaussian Process: Framework, Algorithms, and Verification. IEEE Journal on Selected Areas in Communications. 2019; 37(6): 1291-1306.

35.

Yin

Lin

Kong

Theodoridis

Cui

. FedLoc: Federated Learning Framework for Data-Driven Cooperative Localization and Location Data Processing. IEEE Open Journal of Signal Processing. 2020; 1(37): 187-215.

Research on online monitoring of grid current transformer based on transformers and BiGRU

Abstract

Keywords

1. Introduction

2. System overview

2.1 Composition of the overall system

4.1 Data acquisition

Table 1 Important information of sensor

References

Table 1
Important information of sensor