2026-07-18T08:50:28.926Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:3:12112 2026-07-16T00:46:54Z 3

Photovoltaic Generation Time Series Forecasting Optimization with a Modified Metaheuristic Algorithm, Chapter in LNNS Lecture Notes in Networks and Systems: WCAIAA 2025: World Conference on Artificial Intelligence: Advances and Applications, Springer, volume 1909 2026 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/3/12112 https://link.springer.com/chapter/10.1007/978-3-032-22289-3_16 V. Zeljkovic N. Bacanin L. Jovanovic M. Zivkovic M. Antonijevic V. Marevic M. Markovic Blagojevic M. Tomic Renewable energy plays an important role in sustainable development. Photovoltaic power generation in particular plays an important role in shifting energy dependence form fossil fuels to renewable sources. However, the sporadic production, as well as the high costs of energy storage limited their usability in practice. Through careful planing and resource allocations solar power can be integrated into existing grids, supplementing power production and reducing the burden on traditional power generation and distribution systems. An accurate and robust system for production forecasting could greatly benefit sustainable development. This paper looks at how to use artificial intelligence (AI) and long short-term memory (LSTM) neural networks to predict voltages at solar production substations. A modified optimization metaheuristic based on the VNS algorithm is suggested since the accuracy of forecasting models depends on choosing the right hyperparameters. We run simulations on real-world data, and the top models have mean absolute error (MAE) scores as low as 0.022953. bookPart Springer, Cham 191 202 10.1007/978-3-032-22289-3_16 LNNS Lecture Notes in Networks and Systems: WCAIAA 2025: Proceedings of World Conference on Artificial Intelligence: Advances and Applications, volume 1909