Toward unified intelligence in renewable-dominant power systems: a survey of hybrid forecasting, explainable AI, and multi-agent optimization

The transition toward renewable-dominant power systems has introduced significant operational challenges due to increased uncertainty, reduced system inertia, and the proliferation of distributed energy resources. While artificial intelligence (AI) techniques have been widely adopted for renewable energy forecasting and grid optimization, existing studies primarily address these components in isolation, leading to fragmented system-level intelligence. This paper presents a comprehensive survey that systematically integrates three critical dimensions of modern intelligent energy systems: hybrid forecasting, explainable artificial intelligence (XAI), and multi-agent optimization. First, we review advanced hybrid learning architectures, including decomposition-based models, ensemble frameworks, and deep learning-metaheuristic hybrids, highlighting their effectiveness in capturing nonlinear and stochastic renewable dynamics. Second, we examine XAI techniques that enhance transparency, interpretability, and trust in AI-driven energy systems. Third, we analyze the evolution from centralized optimization to distributed collective intelligence using multi-agent reinforcement learning (MARL), swarm intelligence, and federated learning paradigms. Unlike conventional surveys, this work provides a unified system-level perspective by explicitly linking predictive uncertainty with decentralized decision-making and explainability mechanisms. A conceptual integration framework is presented to demonstrate how probabilistic forecasting, risk-aware MARL policies, and interpretable models can jointly enable robust and scalable grid operation. Furthermore, we identify key research challenges related to scalability, computational efficiency, uncertainty propagation, and real-time deployment. This survey offers a structured reference for researchers and practitioners and outlines future directions toward trustworthy, adaptive, and cooperative AI-driven energy systems.