A Global Cyber-Resilient Model for Dynamic Line Rating Forecasting Based on Deep Federated Learning

Uncertainty in renewable energy generation and energy demand reduces generation flexibility, which should be compensated by increasing flexibility on the supply grid and demand sides. Dynamic line rating (DLR) forecasting, in which the maximum current carrying capacity of overhead transmission lines is accurately forecasted, is important to enable secure grid-side flexibility. This article introduces a novel model for DLR forecasting using a federated learning (FL) approach. By generating a global model, the FL can accurately forecast the DLR values without the need for data from regions that are not available. In addition, data security protection and data protection from various types of cyberattacks are prominent features of the FL technique. In this article, the proposed federated network is trained based on data from nine different regions of Iran that have different geographical and meteorological features. The process of training and recognizing patterns in each local client from FL is based on convolutional neural networks using deep learning techniques. Based on the features extracted from each client, the federated network generates a supermodel that is able to forecast the DLR values for new and unknown regions. Generating the supermodel is performed using geographical and meteorological data from three new regions, namely Boroujen, Nahavand, and Rafsanjan, which had no effect on the construction of the supermodel. The results of the forecasts are analyzed using various performance evaluation metrics. The evaluations show that the generated global supermodel is able to forecast DLR values for the regions of Boroujen, Nahavand, and Rafsanjan, with correlation coefficients of 96%, 94%, and 97%, respectively.