Dalarna University's logo and link to the university's website

Data-driven fault detection and diagnosis (FDD) for chiller systems is often constrained by the severe scarcity of labeled fault data under practical operating conditions, which significantly degrades model robustness and generalization performance. To address this challenge, this study presents a GAN-enhanced transfer learning framework for cross-condition chiller fault diagnosis, integrating synthetic sample generation, high-quality sample filtering, target-domain data augmentation, and cross-condition diagnosis into a unified workflow. Specifically, an auxiliary classifier Wasserstein GAN with gradient penalty (ACWGAN-GP) is employed to generate labeled fault samples in the target domain, while an adaptive weight dual-classifier fusion mechanism is designed to filter high-quality synthetic data. The filtered samples are then incorporated into both fine-tuning (FT) and domain-adversarial neural network (DANN) pipelines to improve diagnostic performance under severe data scarcity. Experimental results show that the proposed filtering mechanism improves fault diagnosis performance of the baseline model by approximately 10% when only 1/7 of real fault data are available. In cross-condition scenarios with a target-domain data ratio as low as 1/20, the proposed method achieves additional performance gains of approximately 2%-8% over conventional transfer learning methods. The results indicate that the proposed framework offers an effective solution for enhancing data-driven fault diagnosis performance in building energy systems under severely limited data conditions. © 2026 Elsevier Ltd

Amidst the growing emphasis on building energy system optimization and carbon neutrality goals, efficient fault diagnosis for chillers is of paramount importance. However, the scarcity of labeled data and the variability of operating conditions significantly constrain the generalization capability of existing diagnostic methods. Conventional strategies, such as transfer learning, semi-supervised learning, and data augmentation, have been explored to alleviate these challenges, but further improvement is still needed under complex and varying operating conditions. To address this issue, this paper proposes a novel semi-supervised Dynamic Weight Adaptive Graph Convolutional Network (DWAGCN), in which both labeled and unlabeled samples are jointly embedded into a graph structure, and neighbor aggregation weights are dynamically generated based on node features. By deeply integrating DWAGCN with a Long Short-Term Memory (LSTM) network, an end-to-end spatiotemporal feature learning framework is constructed, enabling the synergistic capture of temporal dynamics and spatial structures in system operational data. Experimental results demonstrate the strong robustness of the proposed method, with a diagnostic accuracy consistently exceeding 93% across multiple fault types and severity levels, outperforming the compared baseline models. Furthermore, under cross-conditions, the proposed semi-supervised transfer learning framework achieves an accuracy ranging from 75.39% to 85.00% with only 10% of the target-domain data, demonstrating strong generalization capability and providing an effective solution for intelligent diagnostics in data scarcity energy systems. © 2026 Elsevier Ltd.

This paper focuses on near-zero energy residential buildings is in hot summer and cold winter regions. The operation model of the solar-air source heat pump for underfloor radiant heating system is established on the TRNSYS simulation platform, and the related evaluation method for energy flexibility potential of the system is constructed. The characteristics of flexible energy use of the system under different adjustment strategies are analyzed through two evaluation indices of“energy consumption reduction during peak hours” and“increase in solar heat collection”. The results show that the system has significant potential of flexible energy use, which can effectively reduce the energy consumption during peak hours and increase the solar energy utilization. Among them, when the adjustment strategy of“the heat pump does not run during the peak period from 08:00 to 22:00”is implemented, the flexible energy use potential of the system is the highest, and the energy consumption reduction of the system during the peak period of the heating season is 125 kW·h, with a reduction ratio of 63%. At the same time, the solar heat collection improves by 98 kW·h, with an increase ratio of 20%. © 2026 Science Press. All rights reserved.

The airfoil fin printed circuit heat exchanger (PCHE) has great potential for deployment in advanced Brayton cycles with supercritical CO<inf>2</inf>-based mixtures as working fluid. However, accurately predicting the complex flow and heat transfer behavior in the PCHEs remains challenging. This study employs two prominent machine learning (ML) methods – Artificial Neural Networks (ANN) and Extreme Gradient Boosting (XGBoost) – to predict the thermal-hydraulic performance (i.e., Nusselt number Nu and Fanning friction factor f) of supercritical CO<inf>2</inf>-based mixtures in airfoil fin channels. Beyond conventional predictive modeling, an innovative application of SHAP (SHapley Additive exPlanations) analysis is introduced to provide novel physical understanding of model outputs. The results demonstrate that both ANN and XGBoost exhibit excellent prediction performance, significantly outperforming conventional correlations, with R2 exceeding 0.99 for Nu and approaching 0.95 for f as well as small root mean square error (RMSE) and weighted mean absolute percentage error (wMAPE). The models also effectively capture the structurally-induced fluctuations of Nu and f along the flow direction. The feature selection based on the Spearman correlation coefficient method yields a more compact feature space without compromising predictive capability. SHAP analysis reveals a consistent and dominant influence of heat flux (q) and Reynolds number (Re) on the predictions for two targets, with q primarily affecting Nu, while Re and the densities ratio (ρ<inf>b</inf>/ρ<inf>w</inf>) are crucial for f. Notably, a previously overlooked positive impact of the buoyancy effect on improving hydraulic performance is identified in this study. These findings demonstrate significant potential of explainable ML models in predicting the complex supercritical thermal-hydraulic performance, promoting reliable design and optimization of novel heat exchangers in supercritical Brayton cycles. © 2025 Elsevier B.V., All rights reserved.

Solar photovoltaic (PV) is a key technology for decarbonization. However, these systems cause an environmental impact along their life cycle. Greenhouse gas (GHG) emissions and mitigation from PV have been studied on a world level, with a yearly resolution, concluding that there are significant differences in decarbonization depending on where PV systems are deployed. This study explores the life-cycle GHG mitigation potential in the European Union countries with hourly resolution. The lifecycleGHGmitigation potential from PV can vary from 0.6 tCO2ekWp−1 in Sweden to 18.6CO2ekWp−1 in Cyprus. Furthermore, a difference between calculating with an hourly and yearly resolution up to 49% is found. Despite this estimation being dependant on future decarbonization scenarios and other limitations, the variation between countries suggest that these results could be used as a decisionmaking tool to prioritize PV deployment in regions with higher mitigation potential. 

Methane, a highly essential industrial raw material, plays a pivotal role in safeguarding national energy security and advancing sustainable development. Due to the expansion of industrial scale and increased integration in modern methane production, the production data exhibits complex multiscale variability over time, which poses great challenges for accurate methane production prediction. Therefore, a novel production prediction model is proposed by employing an improved Long Short-Term Memory Network (LSTM) combining with the multiscale feature fusion method (MSFF) (MSFF-LSTM). The MSFF decomposes the raw industrial process data into multiple two-dimensional tensors based on periods, which can ravel out the complex temporal fluctuations into multiple intraperiod-and interperiod-variations. Then, the methane prediction model is constructed utilizing multiple LSTM models to extract interactive features at various scales. Finally, using a feature fusion module to fuse the prediction results at different scales can fully aggregate local and global features for complementary prediction. Experimental results demonstrate that, compared with other prediction models, the MSFF-LSTM achieves the state-of-the-art results with the mean absolute error (MAE), the mean square error (MSE), coefficient of determination (R2) and the root mean square error (RMSE) of 0.1056, 0.0300, 0.9199 and 0.1733, respectively, which offers the optimization direction for the anaerobic digestion process of straw for methane production.

Park-level load exhibiting nonlinearity, multi-coupling and stochastic fluctuations, which pose challenges for accurate load forecasting. To address these issues, this study proposes a hybrid short-term load forecasting model based on CNN-LSTM-SAO-MHA. In this model, Convolutional Neural Networks (CNN) extract local temporal features, while Long Short-Term Memory (LSTM) captures long-term dependencies. The Multi-Head Attention (MHA) mechanism strengthens the model’s ability to assign adaptive weights to different time steps, thereby enhancing feature representation and improving the capture of temporal dependencies. Additionally, the Snowmelt Optimisation Algorithm (SAO) is employed for hyperparameter optimisation, enabling automatic adjustment of key parameters to enhance prediction accuracy and computational efficiency. To validate the effectiveness of the proposed model, experiments were conducted using real-world cooling load data from a typical park. The results demonstrate that the proposed CNN-LSTM-SAO-MHA hybrid model significantly outperforms benchmark models, achieving reductions of 28.74% in RMSE and CV-RMSE compared to CNN-LSTM, highlighting its superior performance in short-term park-level load forecasting. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2026.

Polypropylene (PP) has excellent physical and chemical properties, and its products are widely used in various fields. In actual PP production processes, process parameters are usually determined by a single-objective optimization algorithm to have a unique optimal strategy set, but factories typically choose different switching strategies at different times. In order to meet the requirements of the actual grade switching process, the paper proposes a dynamic multi-objective grade switching model based on the polymerization reaction mechanism of the PP. The polymerization reaction mechanism of the PP is used to correlate the cumulative melt index (MI) and operating parameter. Then, the improved multi-objective slime mold algorithm incorporating the Lorentz distribution (IMOSMA-L) is proposed to efficiently optimize the grade switching process. Moreover, the IMOSMA-L introduces the Lorentz distribution function to improve the problem of the arctanh function converging too quickly. Meanwhile, the random search guided by the individual historical optimum of slime molds is introduced to prevent the optimization process from falling into local optimum. Finally, the IMOSMA-L algorithm is applied to optimize the transition processes from L5D98 to ZMK1870 and from L5D98 to ZM-FC801. Comparative experiments with several classical algorithms confirm that the IMOSMA-L achieves superior performance. Furthermore, when benchmarked against actual production data, the optimized switching process reduces PP waste by 48.77 tons and 78.3 tons under temperature fluctuation ranges of 5℃ and 2℃, respectively. © 2026 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.