2026-05-04T19:54:14.767Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:3:11835 2026-02-03T17:51:20Z 3

Metaheuristics Applied to Recurrent Neural Network Optimization in Photovoltaic Production Forecasting, Chapter in LNNS Lecture Notes in Networks and Systems: CML 2025: Computing and Machine Learning, Springer, volume 1613 2026 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/3/11835 https://link.springer.com/chapter/10.1007/978-981-95-2875-2_32 M. Antonijevic L. Jovanovic M. Zivkovic S. Andjelic T. Zivkovic S. Malisic D. Kavitha N. Bacanin In order to fight climate change, reduce greenhouse gas emissions, and promote sustainable development, there must be a global transition to renewable energy. Communities may improve energy independence, use local resources, and become less dependent on centralized power grids via small-scale distributed solar systems. However, there are significant obstacles to connecting these systems with conventional grids. Grid management is made more difficult by solar energy’s unpredictability and fluctuation as well as its poor forecasting skills. Distributed systems frequently don’t synchronize with centralized networks, which leads to inefficiency, energy restriction, and grid instability. These optimization problems are successfully resolved using metaheuristic methods. In contrast to traditional methods, they are able to effectively explore large and intricate search areas for near-optimal answers. In this study, a unique framework especially tailored for solar energy time-series data is proposed, which combines a modified red fox optimization (RFO) method with recurrent neural networks (RNNs). Simulations on real-world datasets show promising results with a RMSE as low as 0.090128. bookPart Springer, Singapore 403 415 10.1007/978-981-95-2875-2_32 LNNS Lecture Notes in Networks and Systems: CML 2025: Proceedings of International Conference on Computing and Machine Learning, volume 1613