TrustFedCL: Adaptive Federated Continual Learning with Trust-Aware Aggregation for Blockchain Intrusion Detection

Abstract

The growing need for power as a result of artificial intelligence infrastructures, cloud computing, and renewable energy deployment poses several difficulties in sustaining smart grids' stability and sustainability. Predictive analysis and intelligent energy management have thus become crucial in ensuring efficient power management and avoiding risks during operation. In this paper, a model based on deep learning is designed for predicting energy constraints in intelligent smart grids with respect to key energy indicators such as AI data centers energy consumption, renewable energy percentage, and AI grid stress index. Several energy attributes and smart grid characteristics form the dataset. Missing value treatment, feature selection, label encoding, and feature scaling were done during pre-processing stages. An Artificial Neural Network was used for multi-class classification of energy constraints levels with dense hidden layers and rectified linear unit activation functions. The results showed highly accurate predictions with 94.74% overall classification accuracy rate. From the classification report, high precision, recall, and F1-score were observed while from the confusion matrix, successful classification with few mispredictions was obtained. Overall, deep learning models can be useful in analyzing energy sustainability and smart grid monitoring.