Adaptive Differentiated Parrot Optimization: A Multi-Strategy Enhanced Algorithm for Global Optimization with Wind Power Forecasting Applications.

The Parrot Optimization Algorithm (PO) represents a contemporary nature-inspired metaheuristic technique formulated through observations of Pyrrhura Molinae parrot behavioral patterns. PO exhibits effective optimization capabilities by achieving equilibrium between exploration and exploitation phases through mimicking foraging behaviors and social interactions. Nevertheless, during iterative progression, the algorithm encounters significant obstacles in preserving population diversity and experiences declining search effectiveness, resulting in early convergence and diminished capacity to identify optimal solutions within intricate optimization landscapes. To overcome these constraints, this work presents the Adaptive Differentiated Parrot Optimization Algorithm (ADPO), which constitutes a substantial enhancement over baseline PO through the implementation of three innovative mechanisms: Mean Differential Variation (MDV), Dimension Learning-Based Hunting (DLH), and Enhanced Adaptive Mutualism (EAM). The MDV mechanism strengthens the exploration capabilities by implementing dual-phase mutation strategies that facilitate extensive search during initial iterations while promoting intensive exploitation near promising solutions during later phases. Additionally, the DLH mechanism prevents premature convergence by enabling dimension-wise adaptive learning from spatial neighbors, expanding search diversity while maintaining coordinated optimization behavior. Finally, the EAM mechanism replaces rigid cooperation with fitness-guided interactions using flexible reference solutions, ensuring optimal balance between intensification and diversification throughout the optimization process. Collectively, these mechanisms significantly improve the algorithm's exploration, exploitation, and convergence capabilities. Furthermore, ADPO's effectiveness was comprehensively assessed using benchmark functions from the CEC2017 and CEC2022 suites, comparing performance against 12 advanced algorithms. The results demonstrate ADPO's exceptional convergence speed, search efficiency, and solution precision. Additionally, ADPO was applied to wind power forecasting through integration with Long Short-Term Memory (LSTM) networks, achieving remarkable improvements over conventional approaches in real-world renewable energy prediction scenarios. Specifically, ADPO outperformed competing algorithms across multiple evaluation metrics, achieving average R(2) values of 0.9726 in testing phases with exceptional prediction stability. Moreover, ADPO obtained superior Friedman rankings across all comparative evaluations, with values ranging from 1.42 to 2.78, demonstrating clear superiority over classical, contemporary, and recent algorithms. These outcomes validate the proposed enhancements and establish ADPO's robustness and effectiveness in addressing complex optimization challenges.