Fault Area Diagnosis Of Ultra-high Voltage Multi-terminal Hybrid DC Transmission Lines Based On Deep Learnin

With the increasing efforts to transmit clean energy across regions in line with China’s "carbon peak and carbon neutrality" strategy,multi-terminal hybrid DC transmission is gradually becoming an important development direction for the future power grid.LCC-MMC type multi-terminal hybrid DC transmission,which combines the mature commutation technology of LCC-HVDC and MMC-HVDC,has the advantages of low investment cost,no commutation failure,and flexible reactive power control.Therefore,it has become an effective way to solve the power consumption and new energy integration in China.In practical engineering,multi-terminal hybrid DC transmission systems use large-scale DC transmission lines,with long-distance transmission across regions,harsh operating environments,and higher probability of faults.In addition,parallel multi-terminal hybrid DC transmission systems have T-junction busbars,but the boundary conditions of the T-junction are weak,which can make it difficult for the lines on both sides of the T-junction to be directly identified by transient traveling waves.The fault characteristics of such a system differ significantly from those of traditional DC power systems and flexible DC power systems.Therefore,in order to ensure the safe,stable,and reliable operation of multi-terminal hybrid DC power systems,it is necessary to conduct in-depth research on the identification methods of fault areas in multi-terminal hybrid DC power systems,and to provide theoretical support for the protection methods of such systems.The main research contents are as follows:In this paper,two different identification schemes for protection measurement points placed at the left and right of the T-zone and the head of the rectifier side are built,and three different algorithms are constructed to identify their fault areas to provide reference and supplement to the existing protection schemes.This paper proposed a method for diagnosing faults in high voltage three-terminal hybrid DC transmission lines based on PCNN-LSTM.For the first solution,First,the amplitude-frequency characteristics of the voltage transfer functions of the Kunbei side,Longmen side,and Liubei T-section lines are analyzed to identify the characteristic differences of faults in different areas of the high voltage multi-terminal hybrid DC transmission lines.Then,the PCNN-LSTM diagnostic model is constructed using transient fault voltage data collected from four measurement points on the left and right sides of Liubei T-section as training samples for deep learning.The model has the ability to extract spatial features with a dual-branch CNN and temporal features with LSTM,and its structural parameters are determined through parameter comparison experiments.The method achieves end-to-end area diagnosis through "offline training and online diagnosis." Simulation experiments show that the proposed fault region recognition method can accurately identify faults inside and outside different lines,and has a certain tolerance for transient resistance and anti-interference ability.This paper proposes a fault region diagnosis method for extra-high voltage three-terminal hybrid DC transmission lines based on Res-GRU network.For solution 1,firstly,the amplitude-frequency characteristics of current transfer functions of Kunbei side,Longmen side line and Liubei T area are analyzed to infer the characteristic differences of faults in different areas of the UHV multi-terminated hybrid DC transmission line,secondly,the fault transient currents are decomposed by wavelet decomposition to obtain the set of current subcomponents,which are used as the input samples of the deep learning model,and subsequently,the Res-GRU model is built,which model combines the residual network and GRU network,through the residual connection can effectively reduce the gradient disappearance and explosion problem during the training process,while the GRU network can effectively capture the dynamic changes in the transient current sequence,and the combination of the two makes the model more stable.After simulation tests,it is verified that the fault area diagnosis method based on multi-mode decomposition and Res-GRU network for EHV three-terminal hybrid DC transmission lines can correctly diagnose faults inside and outside the line area and T-zone converging bus faults.By building different machine learning and deep learning models for comparison and analysis,it is known that the proposed method is still applicable in high resistance grounding and noise background.This paper proposed a fault area diagnosis method for ultra-high-voltage three-terminal hybrid DC transmission lines based on multi-mode decomposition and a multi-branch parallel residual network.Firstly,the differences in fault current waveforms between different fault areas on the Kunbei side,Longmen side lines,and Liubei T-area are analyzed,and the preliminary features of the fault current are extracted using the multi-mode decomposition algorithm.The complementary rule of different decomposition algorithms is discovered through feature visualization.Then,the SSA-MRes-GRU model is built,with the decomposed current component feature matrix as the input.The multi-branch residual network with different sizes of receptive fields extracts multi-scale spatial coupling and interaction features of the fault current components,and the GRU module further excavates the time sequence features in the time dimension,significantly enhancing the features.The Sparrow algorithm is introduced to optimize the key parameters of the model,improving the model’s adaptability and scalability.Through experiments and comparisons with different models,multi-mode decomposition experiments,and anti-noise interference experiments,it is confirmed that the proposed method has high diagnostic accuracy,is basically unaffected by fault distance and transition resistance.