Betta agonistic optimization algorithm: A novel metaheuristic approach for photovoltaic power forecasting

Accurate photovoltaic (PV) power forecasting is essential for efficient renewable energy integration because of the nonlinear, dynamic, and uncertain characteristics of solar power generation. This paper proposes a novel Evolutionary Memory-Augmented MambaFormer (EvoMem-MambaFormer) optimized using the proposed Betta Agonistic Optimization Algorithm (BAOA) for short-term PV power prediction. The proposed EvoMem-MambaFormer combines selective state-space modeling with memory-guided attention to effectively capture both short-term and long-term temporal dependencies. A Dual Predictive Space Generator is introduced to learn observation-oriented and future-oriented feature representations, while an Error-Adaptive Selective State Evolution mechanism enables recursive memory updates through error feedback. In addition, a Memory Echo Attention Transformer integrates evolved memory states into the attention mechanism to improve temporal feature extraction, and a Quantum-inspired Phase Memory Resonance module enhances the learning of periodic solar generation patterns. The proposed BAOA, inspired by the territorial competition behaviour of Betta fish, is used to optimize the model hyperparameters, improving convergence and forecasting results. Experimental results on a real-world PV dataset demonstrate that the proposed BAOA-EvoMem-MambaFormer attains an RMSE of 7.448, MAE of 4.602, and R² of 0.948, outperforming existing state-of-the-art forecasting models. These outcomes shows that the model provides an accurate, robust, and efficient solution for photovoltaic power forecasting in renewable energy and smart grid applications.