2026-10-11T05:21:58.225Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:3:12178 2026-09-18T18:16:05Z 3

Enhancing Solar Power Generation Prediction: A Modified Variable Neighborhood Search Optimized Long Short-Term Memory Approach, Chapter in SIST Smart Innovation, Systems and Technologies: BIDA 2025: Business Intelligence and Data Analytics, Springer, volume 472 2026 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/3/12178 https://link.springer.com/chapter/10.1007/978-3-032-15996-0_26 M. Zivkovic A. Petrovic L. Jovanovic V. Marevic M. Stankovic M. Tomic V. Zeljkovic N. Bacanin The problem of predicting solar power generation is hindering the progress of renewable energy resources (RES) integration into the existing power grid. To solve this issue, this research proposes a modified variable neighborhood search (VNS) metaheuristic algorithm as an optimizer for the long short-term memory (LSTM), a subtype of recurrent neural networks (RNNs). This approach has been chosen due to its high performance as a predictor for time-series-related problems, while metaheuristics have proven as strong optimizers to determine suboptimal hyperparameter settings for deep learning models such as LSTM. Proposed methodology is validated on a real-world photovoltaic dataset, and rigid comparative analysis with other contemporary algorithms in terms of standard regression metrics has been conducted. Research findings reveal potential of tuned LSTM models for solar power generation prediction. bookPart Springer, Cham 337 351 10.1007/978-3-032-15996-0_26 SIST Smart Innovation, Systems and Technologies: BIDA 2025: Proceedings of International Conference on Business Intelligence and Data Analytics, volume 472