Enhancing Fault Detection and Classification in Wind Farm Power Generation Using Convolutional Neural Networks (CNN) by Leveraging LVRT Embedded in Numerical Relays

Real-time fault detection and classification are important for power system stability and resilience of the power grid to minimize downtime and prevent cascading failures. Numerical relays (NRs) are used to detect faults in power system networks and isolate the power to prevent grid instability in the event of a fault. Faults may occur due to many factors, such as lightning strikes, weather events (heavy rain, snow, and wind), accidents, or power surges. However, integrating renewable energy sources, particularly wind farms, introduces more complex and unpredictable fault conditions due to weather patterns that traditional numerical relays struggle to handle effectively. In addition, microgrids impose more complex types of faults, which may pose challenges for NRs and necessitate a “smart grid” where communication can happen between the power company and devices on the grid for fast responses. In this paper, a non-overlapping sliding window-based continuous online monitoring Convolutional Neural Network (CNN) technique is proposed by leveraging the Low Voltage Ride Through (LVRT) code to detect and classify the type of faults in wind farms and other renewable sources. Extensive simulations of different fault conditions were conducted using MATLAB/Simulink to evaluate the performance of the proposed model. Electrical fault data were converted to images to simulate the output of a numerical relay. These images were fed to a neural network. To validate the model, a detailed analysis was performed, comparing different combinations of classifiers and optimizers. The proposed model performs well in fault detection and classification, introducing a new feature of numerical relays.