Paper Title
Early Fire Detection in Wind Turbines

Abstract
Wind turbine fires are one of the most disastrous dangers in contemporary wind turbine power systems, leadingto irreparable damage to the turbine, shutdown, and massive financial losses. In traditional fire detection systems, manual interpretation, and threshold-based systems, which give notifica- tions once a fire is created, and fire suppression systems, which operate once a fire has occurred, are used. These systems do not possessthecapabilityforpredictivepremonitions.Inthispresent study, a novel work named FireHybridNet V2, a predictive fire danger early warning system possessing a predictive capability forfireincidents60minutesbeforetheiroccurrencethrough the Supervisory Control and Data Acquisition sensor, is offered. In this work, a huge real-time database from a wind turbine operation, comprising over 40 operational attributes, was pre- processedandtransformedintoaseriesof96-stepsequencesthat yield predictions for various attributes. These are to be used in ‘classimbalanceweightingforoptimization’inordertoovercome imbalances in data distribution and enable learning from both normal and fire data attributes. This experiment has produced a test accuracy rating within 81% for predictive capabilities even with highly noisy and imbalanced datasets, an effective proof of the system’s predictive capabilities for fire incidents. Keywords - Wind Turbine Fire Prediction, SCADA Data, Deep Learning, CNN-Bilstm-Attention Model, Fire Hybrid Net, Predictive Maintenance, Imbalanced Data Learning, Early Warning System, Industrial Iot Safety.