Fault location in AC transmission lines with back-to-back MMC-HVDC using ConvNets

Abstract

The back-to-back modular multi-level converter–high-voltage DC (MMC-HVDC) system is gaining popularity in the field of enhancing power transmission capacity but poses challenges to fault diagnosis. A novel method for the fault location for such a system based on convolutional networks (ConvNets) is presented. The proposed method uses voltages and currents of only one terminal of transmission lines. Compared with existing methods, the proposed method automatically learns features from a dataset of voltage and currents signals. The fault location is then achieved by linear regression with an L(1) penalty using the features extracted by ConvNets. Additionally, the fault location network is trained jointly with fault-type information to improve performance. The feasibility of the proposed method has been verified on a 220 kV back-to-back MMC-HVDC transmission system for various fault locations and under different fault conditions using power systems computer-aided design (PSCAD)/electro magnetic transient design and control (EMTDC). Results show that the proposed method can locate faults using one cycle data with high accuracy.

Keywords

• fault diagnosis
• feature extraction
• learning (artificial intelligence)
• HVDC power convertors
• fault location
• HVDC power transmission
• power transmission faults
• power transmission lines
• convolution
• feedforward neural nets
• power engineering computing
• regression analysis
• AC transmission lines
• ConvNets
• power transmission capacity
• fault diagnosis
• fault location network
• fault-type information
• MMC-HVDC transmission system
• fault conditions
• convolutional networks
• modular multilevel converter–high-voltage DC system
• linear regression
• automatic learning
• feature extraction