Deep Learning loss model for large-scale low voltage smart grids. (October 2020)

• Record Type: Journal Article
• Authors: Velasco, Jose Angel; Amaris, Hortensia; Alonso, Monica
• Is Part Of: International journal of electrical power & energy systems. Volume 121(2020)
• Publication Date: 2020-10
• Subjects: Deep learning -- Smart grids -- Uncertainty -- Probabilistic modelling -- Clustering -- Phase imbalance; Electrical engineering -- Periodicals; Electric power systems -- Periodicals; Électrotechnique -- Périodiques; Réseaux électriques (Énergie) -- Périodiques; Electric power systems; Electrical engineering; Periodicals; 621.3
• DOI: 10.1016/j.ijepes.2020.106054
• Languages: English
• ISSNs: 0142-0615

Abstract

Highlights: A generic deep learning model for losses estimation in large scale LV smart grids. A reduced dimension clustering process for obtaining representative feeders. A proof-of-concept in a large-scale LV smart grids with distributed generation and unbalanced operation. Abstract: Distribution systems operators (DSOs) encounter the challenge of managing network losses in large geographical areas with hundreds of secondary substations and thousands of customers and with an ever-increasing presence of renewable energy sources. This situation complicates the estimation process of power loss, which is paramount to improve the network energy efficiency level in the context of the European Union energy policies. Thus, this article presents a methodology to estimate power losses in large-scale low voltage (LV) smart grids. The methodology is based on a deep-learning loss model to infer the network technical losses considering a large rollout of smart meters, a high penetration of distributed generation (DG) and unbalanced operation, among other network characteristics. The methodology has been validated in a large-scale LV distribution area in Madrid (Spain). The proposed methodology has proven to be a potential network loss estimation tool to improve the energy efficiency level in large-scale smart grids with a high penetration of distributed resources. The accuracy of the proposed methodology outperforms that of the state-of-the-art loss estimation methods, exhibiting a rapid convergence which allows for its use in real-time operations.