Adaptive quantum inspired deep reinforcement learning for multi objective low carbon CCHP optimization

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:Novel integration of quantum-inspired evolutionary algorithms with DRL for multi-objective optimization; the hierarchical digital twin and C-OPF model are methodological contributions.

🏢実務担当者:Potential to reduce operational costs and emissions for CCHP plant operators; the 24.44% cost reduction and 40% emission cuts are compelling, but field validation is needed.

🏛政策担当者:Demonstrates the value of AI-driven optimization for demand response and carbon trading schemes; could inform incentives for smart grid and distributed energy resources.

A novel deep reinforcement learning (DRL)-based multi-objective optimization framework for low-carbon combined cooling, heating, and power (CCHP) systems is proposed. The framework integrates adaptive Quantum-Inspired Evolutionary Algorithms (QIEA) with digital twin technology. The proposed Adaptive Quantum-DRL Multi-Objective Optimization (AQ-DRLMO) model addresses the complex scheduling challenges inherent in the CCHP systems by incorporating flow monitoring of carbon emission, ladder-type carbon trading, and demand response scheduling. A hierarchical digital twin structure enables predictive optimization through Physics-Informed Neural Networks (PINNs), ensuring accurate modeling of thermodynamic interactions. Three enhanced control strategies Electric-Thermal Equivalent Following (ETEF), Electric Equivalent Following (EEF), and Thermal Equivalent Following (TEF) are introduced and employed attention-based transformer networks for temporal pattern recognition. A novel Carbon-Conscious Optimal Power Flow (C-OPF) model tracks carbon flows across multi-stage energy conversion pathways. Extensive simulations conducted under summer and winter operating conditions demonstrate that the proposed AQ-DRLMO framework achieves greenhouse gas emission reduction of 40.08%, primary energy saving of 34.04%, and cost reduction of 24.44% compared to conventional individual generation systems across different control strategies and seasonal scenarios. The quantum-inspired optimization achieves 67.3% faster convergence compared to conventional genetic algorithms while maintaining solution diversity in the Pareto front, converging in 45 iterations versus 137 iterations for standard genetic algorithms under identical test conditions. This study relies on synthetic load profiles and simulation-generated validation data; therefore, the findings are best interpreted as a simulation-based demonstration of day-ahead scheduling potential rather than a validated real-time control solution. Subject to field validation, the framework shows promise as an efficient solution for smart grid energy management in low-carbon distributed energy systems.

• openalex https://doi.org/10.1038/s41598-026-51604-wfirst seen 2026-06-12 05:12:08