Advances in Applied Energy最新文献

Machine learning is widely recognized as a promising data-driven modeling technique for the model-based control and optimization of building energy systems. However, the generalizability of data-driven models often faces significant challenges, as the available training data from building operations usually only covers a limited range of working conditions. Active learning can proactively test unseen and informative working conditions to enrich the training set by adding new data samples, leading to improved generalization performance of data-driven models. A novel distance and information density-based sample strategy is developed that accounts for the real-time status of building operation and outdoor environment. Based on Mahalanobis distance, this strategy determines the sampling value of an unlabeled sample (unseen working condition) by assessing its similarity to both the training samples and other unlabeled samples. As collecting sufficiently representative samples can be difficult, costly, and time-consuming, a distance-based sampling cost metric is proposed to compare the efficiency of different sampling methods, considering the detrimental effects of the actively sampling process on the normal operation of building energy systems. This paper presents a comprehensive and in-depth comparison of five active learning methods, including one incorporating the distance-based sampling strategy, by conducting data experiments on the data collected from the cooling towers of a real high-rise building. The results show that active learning can effectively identify informative data samples and improve the generalization performance of data-driven models. The research outcomes are valuable for enhancing AI-enabled data-driven modeling of building energy systems with substantial decreases in costs on data sampling.

The integrated community energy system (ICES) has emerged as a promising solution for enhancing the efficiency of the distribution system by effectively coordinating multiple energy sources. However, the concept and modeling of ICES still remain unclear, and operational optimization of ICES is hindered by the physical constraints of heterogeneous integrated energy networks. This paper, therefore, provides an overview of the state-of-the-art concepts for techno–economic modeling of ICES by establishing a Multi-Network Constrained ICES (MNC-ICES) model. The proposed model underscores the diverse energy devices at community and consumer levels and multiple networks for power, gas, and heat in a privacy-protection manner, providing a basis for practical network-constrained community operation tools. The corresponding operational optimization in the proposed model is formulated into a constrained Markov decision process (C-MDP) and solved by a Safe Reinforcement Learning (RL) approach. A novel Safe RL algorithm, Primal-Dual Twin Delayed Deep Deterministic Policy Gradient (PD-TD3), is developed to solve the C-MDP. By optimizing operations and maintaining network safety simultaneously, the proposed PD-TD3 method provides a solid backup for the ICESO and has great potential in real-world implementation. The non-convex modeling of MNC-ICES and the optimization performance of PD-TD3 is demonstrated in various scenarios. Compared with benchmark approaches, the proposed algorithm merits training speed, higher operational profits, and lower violations of multi-network constraints. Potential beneficiaries of this work include ICES operators and residents who could be benefited from improved ICES operation efficiency, as well as reinforcement learning researchers and practitioners who could be inspired for safe RL applications in real-world industry.

Linear programs are frequently employed to optimize national energy system models, which are used to find a minimum-cost energy system. For the operation, they assume perfect forecasting of the weather and demands over the whole optimization horizon and can therefore perfectly fit the energy systems’ design and operation. Therefore, they will yield lower costs than any real energy system that only has partial forecasting available. We compare linear programming with a priority list, a heuristic operation strategy which uses no forecasting at all, in a model of a climate-neutral German energy system. We find a 28% more expensive energy system under the priority list. Optimizing the same energy system model with both strategies envelopes the cost and design of any energy system that has partial forecasting. We demonstrate this by incorporating some rudimentary forecasting into a modified priority list, which actually reduces the gap to 22%. This is thus an approach to find an upper bound for how much a linear program possibly underestimates the costs of a real energy system in Germany in regard to imperfect forecasting. We also find that the two approaches differ mainly in the dimensioning and operation of energy storage. The priority list yields 63% less batteries, 73% less thermal storage and 54% more hydrogen storage. The use of renewables and other components in the system is very similar.

This paper proposed a user-friendly and adaptive home intelligent agent with self-learning capability for optimal scheduling of smart home energy systems. The intelligent agent autonomously identifies model parameters based on system operation data, eliminating the need for manual input and making it more user-friendly and practical to implement. It can also self-learn the latest energy consumption information from an updated dataset and adaptively adjust model parameters to accommodate changing conditions. Utilizing these determined models as input, the intelligent agent performs day-ahead optimal scheduling using the proposed many-objective integer nonlinear optimization model and automatically controls system operation. Experimental studies were conducted on a laboratory-based smart home energy system to verify the effectiveness of the developed intelligent agent in different scenarios. The results consistently demonstrate Mean Absolute Percentage Errors below -12.7 % across all three scenarios, indicating the accuracy of the intelligent agent. Furthermore, the optimal scheduling significantly enhances system performances. After optimization, daily operational costs, peak-valley differences, and CO₂ emissions were reduced by 34.1 % to 81.6 %, 29.2 % to 36.7 %, and 19.6 % to 43.2 %, respectively. Moreover, the PV generation self-consumption rate and self-sufficiency rate improved by 29.6 % to 38.0 % and 40.5 % to 49.4 %, respectively. The proposed intelligent agent provides invaluable guidance for optimal dispatch of smart home energy systems in real-world settings.

Soaring solar cell temperature hindered photovoltaic (PV) efficiency, but a novel radiative cooling (RC) cover developed in this study offered a cost-effective solution. Using a randomly particle-doping structure, the radiative cooling cover achieved a high “sky window” emissivity of 95.3% while maintaining a high solar transmittance of 94.8%. The RC-PV system reached a peak power output of 147.6 W/m${}^2$. A field study to explore its potential in various provinces in China revealed significant efficiency improvements, with yearly electricity outputs surpassing those of ordinary PV systems by a relative improvement of 2.78%–3.72%. The largest increases were observed under clear skies and in dry, cool climates, highlighting the potential of RC-PV systems under real weather and environmental conditions. This work provided the theoretical foundation for designing scalable radiative cooling films for PV systems, unlocking the full potential of solar energy.