"Implementing Deep Learning Algorithms for Predictive Maintenance in Electrical Power Systems to Reduce Downtime and Costs" CASE STUDY: Al-Rways Power Plant Network (220KV)

Abstract

A deep learning-based predictive maintenance model for electrical power systems is developed and evaluated to minimize downtime and operational costs. The method was validated using Python (VSCodeUserSetup-x64-1.93.1). A feedforward neural network trained on a multi-dimensional dataset derived from power grid components including generators, transformers and circuit breakers for Al-Rways power plant network in Libyan utility. The methodology encompassed data preprocessing, feature engineering, and model optimization to attain optimal predictive accuracy. The model was trained through 100 epochs and reached its peak performance during a specific epoch, demonstrating enhanced accuracy on both training and validation sets, alongside a reduced Mean Squared Error (MSE) for both. In comparison to traditional machine learning models such as Random Forest, SVM, and GBM, the trained model was shown to surpass these algorithms in decreasing downtime, lowering maintenance costs and enhancing Mean Time Between Failures (MTBF) for the equipments mentioned above. From the results obtained, the paper illustrates that deep learning significantly improves predictive maintenance in electrical power systems by providing superior accuracy, reducing down time, increased system availability, and considerable cost savings. Furthermore, it recommends extending failure prediction lead-time and enhancing data collection granularity to further boost the model’s performance.