Single-ended fault positioning method and system for high-voltage direct-current transmission line of hybrid network

Patent metadata

Field	Value
Publication number	US-12228601-B2
Application number	US-202117496774-A
Country	US
Kind code	B2
Filing date	Oct 8, 2021
Priority date	Dec 18, 2020
Publication date	Feb 18, 2025
Grant date	Feb 18, 2025

Abstract

The disclosure discloses a single-ended fault positioning method and system for a HVDC power transmission line based on a hybrid deep network. The method comprises the following: collecting rectification side bus output voltage and current signals of a HVDC power transmission system under different fault types, fault distances and transition resistances as an original data set; eliminating electromagnetic coupling of the bipolar direct-current line by using phase-mode transformation, extracting IMF components of fault voltage and current signals under different fault scenes by using variational mode decomposition, and calculating TEO of the IMF components to obtain a fault data set after feature engineering; normalizing the fault data set, and dividing the fault data set into a training set and a test set; and successively inputting the training set and the test set into a hybrid network of a convolutional neural network and a long short-term memory network for training and testing.

First claim

What is claimed is:

1. A single-ended fault positioning method for a reliable operation of high-voltage direct-current (HVDC) transmission line based on a hybrid deep network, comprising:
(1) establishing a simulation model of a HVDC bipolar transmission system based on a voltage source converter, and selecting an output voltage and current signals of a rectifier side bus under different fault types, fault distances and transition resistances as an original data set, and labeling classification of fault segments and labeling a location of a fault position according to the fault segments of a transmission line and its precise fault position where the fault occurs;
(2) performing variational modal decomposition (VMD) on the selected voltage and current on the rectifier side in various fault scenarios after phase-mode transformation, obtaining an effective intrinsic mode function (IMF) component of the signal, and calculating a Teager energy operator (TEO) of the IMF component to obtain a fault data set;
(3) performing normalized data preprocessing on the fault data set after performing VMD and TEO, and dividing the preprocessed fault data set into a training set and a test set;
(4) inputting the training set and the test set to a convolutional neural network (CNN)-long short-term memory (LSTM) network model in sequence for model training and test respectively, wherein the CNN is used as a classifier to identify the fault segments, and the LSTM network is used as a regressor to position faults of the HVDC transmission line with minimum effects cause by fault types, noise, sampling frequency and different HVDC topologies.

2. The method according to claim 1 , wherein step (1) comprises:
in the simulation model of the HVDC bipolar transmission system based on the voltage source converter, the different fault scenarios are set to construct the original data set, wherein a transmission line on a direct-current (DC) side is set with one type of the fault scenarios every several kilometers from the rectifier side bus, the fault types comprise positive grounding, negative grounding, and positive and negative short-circuit grounding; grounding resistances are set at equal intervals from a minimum value to a maximum value within a preset range, every combination of the various fault distances, the fault types and the grounding resistances is one type of the fault scenarios, and the output voltage and the current signals of the rectifier side bus under all of the fault scenarios are measured as the original data set.

3. The method according to claim 1 , wherein step (2) comprises:
noise is added to the voltage and the current signals respectively on the bipolar bus on the rectifier side under various fault scenarios to simulate noise interference scenarios of a measuring equipment, a noise-containing voltage signal undergoes phase-to-mode conversion to obtain a line-mode voltage component, and a noise-containing current signal undergoes phase-to-mode conversion to obtain a line-mode current component, each of the line-mode voltage components and each of the line-mode current components are subjected to VMD decomposition to obtain a first IMF component of high frequency, and the TEO of the first IMF component of each high frequency is calculated to obtain the fault data set after feature engineering.

4. The method according to claim 3 , wherein step (3) comprises:
the first IMF component of the voltage after the TEO is subjected to max-min normalization to obtain a preprocessed voltage component, and the calculated first IMF component of the current after the TEO is subjected to max-min normalization to obtain a preprocessed current component, the preprocessed voltage component and the preprocessed current component are constructed into a 2-dimensional tensor (2D-Tensor), and all of the 2D-Tensors are divided into the training set and the test set.

5. The method according to claim 4 , wherein step (4) comprises:
the training set is input into a hybrid deep model for training, the 2D-CNN in the hybrid deep model is used as a classifier to complete a task of identifying the fault segments, and the LSTM is used as a regressor to integrate a fault segment information in the classifier, wherein, division of the fault segments needs to be determined according to an accuracy rate of fault position for model training, a plurality of fault samples corresponding to the fault distance are selected for model training, a tolerance range of ±a % is set for a fault distance label, and the accuracy rate of the selected samples for fault position is calculated, an optimal number of the fault segments is determined through comparative experiments,
wherein a % comprises 5%.

6. The method according to claim 2 , wherein step (2) comprises:
noise is added to the voltage and the current signals respectively on the bipolar bus on the rectifier side under various fault scenarios to simulate noise interference scenarios of a measuring equipment, a noise-containing voltage signal undergoes phase-to-mode conversion to obtain a line-mode voltage component, and a noise-containing current signal undergoes phase-to-mode conversion to obtain a line-mode current component, each of the line-mode voltage components and each of the line-mode current components are subjected to VMD decomposition to obtain a first IMF component of high frequency, and the TEO of the first IMF component of each high frequency is calculated to obtain the fault data set after feature engineering.

7. The method according to claim 6 , wherein step (3) comprises:
the first IMF component of the voltage after the TEO is subjected to max-min normalization to obtain a preprocessed voltage component, and the calculated first IMF component of the current after the TEO is subjected to max-min normalization to obtain a preprocessed current component, the preprocessed voltage component and the preprocessed current component are constructed into a 2-dimensional tensor (2D-Tensor), and all of the 2D-Tensors are divided into the training set and the test set.

8. The method according to claim 7 , wherein step (4) comprises:
the training set is input into a hybrid deep model for training, the 2D-CNN in the hybrid deep model is used as a classifier to complete a task of identifying the fault segments, and the LSTM is used as a regressor to integrate a fault segment information in the classifier, wherein, division of the fault segments needs to be determined according to an accuracy rate of fault position for model training, a plurality of fault samples corresponding to the fault distance are selected for model training, a tolerance range of ±a % is set for a fault distance label, and the accuracy rate of the selected samples for fault position is calculated, an optimal number of the fault segments is determined through comparative experiments,
wherein a % comprises 5%.

9. A non-transitory computer-readable storage medium storing a computer program, when the computer program is executed by a processor, causing the processor to execute a single-ended fault positioning method for a reliable operation of high-voltage direct-current (HVDC) transmission lines based on a hybrid deep network, the method comprising:
establishing a simulation model of a HVDC bipolar transmission system based on a voltage source converter, and selecting an output voltage and current signals of a rectifier side bus under different fault types, fault distances and transition resistances as an original data set, and labeling classification of fault segments and labeling a location of a fault position according to the fault segments of a transmission line and its precise fault position where the fault occurs;
performing variational modal decomposition (VMD)

Assignees

• Univ Wuhan

Inventors

• He Yigang
• Wang Lei
• Li Lie
• Zhao Yingying
• Du Bolun
• He Liulu

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G01R31/088 Primary

  Aspects of digital computing · CPC title