Artificial intelligence-driven smart grid optimization: A comprehensive review of machine learning, renewable integration, and cybersecurity applications

• Comprehensive synthesis of 205 AI studies across smart grid optimization domains. • Machine learning achieves < 2% MAPE in load forecasting under optimal conditions. • Cybersecurity frameworks demonstrate > 98% intrusion detection accuracy. • Computational footprint and scalability barriers challenge operational deployment. • Research gaps persist in safety verification and adversarial robustness. Smart grid modernization necessitates advanced computational approaches addressing renewable variability, distributed control, and cybersecurity challenges that exceed traditional methodologies. This review synthesizes 208 peer-reviewed publications (2020–2026) examining artificial intelligence applications across smart grid optimization and energy management. Machine learning architectures—including convolutional neural networks, recurrent networks, transformers, and hybrid ensembles—demonstrated load forecasting accuracy improvements with reported mean absolute percentage errors below 2% in optimal conditions. Deep reinforcement learning enabled autonomous control policies achieving cost reductions and emission minimizations across simulated environments, though transferability to operational systems requires validation. Renewable integration benefited from AI-enhanced forecasting, energy storage optimization, and microgrid coordination, with case studies reporting operational cost reductions ranging from 3.77% to 55%. Cybersecurity applications achieved intrusion detection accuracies exceeding 98%, though adversarial robustness against sophisticated attacks remains partially characterized. Emerging technologies—digital twins, large language models, federated learning—present opportunities alongside uncertainties regarding deployment timelines. Challenges span computational complexity, data quality limitations, scalability barriers, and environmental sustainability concerns as AI energy consumption potentially offsets efficiency gains. Research gaps persist across safety verification, explainability guarantees, multi-objective optimization frameworks, and lifecycle impact assessments. Successful deployment requires balancing technological capabilities with regulatory frameworks, workforce development, and stakeholder acceptance across diverse institutional contexts.