Applications of performance prediction, design optimization, and operational control using artificial intelligence for organic Rankine cycle: A review

The integration of Artificial Intelligence (AI) with Organic Rankine Cycle (ORC) technology marks a transformative step toward intelligent, efficient, and adaptive renewable energy conversion and waste-heat recovery. This review provides a current state-of-the-art advancement in AI-assisted ORC modeling addressing performance prediction, design optimization, and intelligent operational control strategies. This review systematically evaluates diverse AI methodologies including machine learning algorithms, neural network architectures, genetic algorithms, multi-objective optimization, intelligent control strategies, and hybrid metaheuristic frameworks highlights their role in enhancing ORC reliability and efficiency. The findings indicate that AI-based models exhibit outstanding accuracy and reliability in ORC performance prediction with ANN models consistently achieving the highest accuracy (96-99.8%) in predicting the power output, expander efficiency and thermophysical fluid properties while LSTM/RNN models demonstrate higher capability for transient and dynamic analyses and LSSVM architectures offer a high precision in heat transfer prediction. In optimization, GA and PSO exhibit moderate performance. However, hybrid frameworks and ANN-MILP regression models acknowledge better optimization and computation efficiency outperforming NSGA-II and NSGA-III hybrids. Advanced control strategies particularly model predictive control (MPC), deep reinforcement learning (DRL) and neuro-fuzzy inference systems (ANFIS) are rapidly improving the capabilities of ORC controllers integrated with machine learning surrogates significantly improve system performance and operational stability under transient conditions. The review highlights emerging directions such as digital twin frameworks, physics-informed neural networks (PINNs), Internet of Things (IoT) and reinforcement learning to enable real-time adaptability and autonomous control. The findings indicate that research on digital twin, hybrid AI architecture and distributed learning remains limited with few studies addressing the application of PINNs in ORC systems. Future research should prioritize hybrid AI architecture, distributed learning frameworks, digital twin integration, PINNs and sustainable AI implementation in distributed ORC technologies.