FGPA Acceleration of Neural Networks for Transmission Fault Detection

This paper presents a fault detection system for power transmission lines using a Neural Network model integrated with the FPGA (Field Programmable Gate Array) platform in Xilinx. Transmission lines, essential for power system reliability, are susceptible to faults caused by adverse weather, equipment failure, or environmental interferences. Traditional fault detection methods lack the speed and accuracy required for processing real-time data, particularly in modern power grids. This work leverages the parallel processing capabilities of FPGAs, ensuring fast and efficient classification of faults based on electrical parameters such as the three individual phase currents (Ia, Ib, Ic) and voltages. This system implements MLP (Multi-Layer Perceptron) based ANN (Artificial Neural Network) model employing backpropagation for error minimization, along with advanced techniques like weight initialization, data regularization to prevent overfitting of data, and Adam optimizer to handle extensive datasets effectively. The architecture evaluates three distinct activation functions to optimize results of area occupation on the ZedBoard Zynq-7000 FPGA board, which reduces power consumption and facilitates efficient resource allocation. Hardware acceleration through Lookup Tables (LUTs) ensures optimal neuron performance, and the FPGA design flow, using Verilog for circuit synthesis and resource mapping, enables energy-efficient operation. Validated through regression factors and cross-examination of activation functions, this system achieves robust, real-time fault classification, offering a scalable and reliable solution.