2026-05-04T19:54:15.132Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:1:11779 2026-01-06T16:54:22Z 1

Solar Power Forecasting Using Modified Particle Swarm Optimization and Long Short-Term Memory 2025 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/1/11779 https://ieeexplore.ieee.org/document/11314424 L. Jovanovic J. Arsic I. Kosta V. Thomas N. Bacanin M. Zivkovic Accurate forecasting of photovoltaic power production is essential for reliable grid integration and efficient energy management. This work explores long short-term memory (LSTM) networks optimized through an adaptive β-hill climbing particle swarm optimization (AHCPSO) strategy for hyperparameter tuning. The proposed approach addresses two common PSO limitations-premature convergence and weak early exploration-by adaptively regulating search diversification and intensification. Comparative experiments against well established and recently proposed algorithms as well as the conventional PSO were conducted on real PV output data from publically available sources. Results show that LSTM-AHCPSO achieves the most stable optimization performance with the lowest variance, while delivering superior accuracy across error metrics (R2=0.8757, MAE =0.0444, RMSE =0.0817). Forecasts produced by the best model closely track actual PV generation, demonstrating its robustness and consistency. conferenceObject 1 4 10.1109/TELFOR67910.2025.11314424 2025 33rd Telecommunications Forum (TELFOR), IEEE, Belgrade, Serbia