Energy and AI最新文献

The temperature field within porous media is considerably affected by different boundary conditions, and effective thermal conductivity varies with spatial structure morphologies. At present, traditional prediction methods for the temperature field are expensive and time consuming, particularly for large structures and dimensions, whereas deep learning surrogate models have limitations related to constant boundary conditions and two-dimensional input slices, lacking the three-dimensional topology and spatial correlations. Herein, a constraint-incorporated model using U-Net architecture as the backbone is proposed to predict the temperature field and effective thermal conductivity of sphere-packed porous media, considering diverse external heat fluxes. A total of 510 original samples of temperature fields are generated through lattice Boltzmann method (LBM) simulations, which are further augmented to 33,150 samples using the self-amplification method for the training. Physical prior knowledge is incorporated into the model to constrain the training direction by adding physical constraint terms as well as adaptive weights to the loss function. Input vectors with different heat fluxes and porosities are embedded into latent features for predicting different boundary conditions. Results indicate that the constraint-incorporated model has a mean relative error ranging between 1.1 % and 5.7 % compared with the LBM results in the testing set. It exhibits weak dependence on the database size and substantially reduces computational time, with a maximum speedup ratio of 7.14 × 10⁶. This study presents a deep learning model with physical constraints for predicting heat conduction in porous media, alleviating the burden of extensive experiments and simulations.

Three proposed reaction mechanisms can occur in an amine-CO₂ system: either zwitterionic or termolecular mechanisms for primary/secondary amines and base-catalyzed hydration for tertiary amines. The intricacy of this system hinders the construction of a general model for all types of amines. This study attempts to build an artificial neural network model that predicts the equilibrium solubility of any nonblended aqueous amine-CO₂ system under given operating conditions, regardless of the reaction mechanism. This is a novel approach that has not yet been reported. The amines were characterized using molecular descriptors derived from COSMO theory through density functional theory calculations to incorporate molecular structures as model features. Our model achieved performance metrics (R²) of 0.9645 and 0.9481 for the training and validation sets, respectively. For unfamiliar amines that were absent in both the training and validation sets, our model achieved an R² of 0.8601. Model benchmarking was performed using a previously established thermodynamic model. Interpretations of the model are also provided based on the chosen features. This study also offers exploratory insight into how the molecular structure and operating conditions affect the CO₂ equilibrium solubility in amines. The model developed in this study has the potential to reduce the solvent screening time in determining appropriate amines for larger-scale applications.

A novel approach for generative time series simulation of electricity price scenarios is presented. A “Time Series Simulation Conditional Generative Adversarial Network” (TSS-CGAN) generates short-term electricity price scenarios. In particular, the network is capable of generating a 24-dimensional output vector that corresponds to the expected behavior of electricity markets. The model can replace typical approaches from financial mathematics like statistical factor models to model the price distribution around a given forecast. The data cover a 3-year period from 2020 to 2023. Our empirical study is conducted on the EPEX SPOT market in Europe. An electricity price scenario includes the prices of the hourly contracts of a day-ahead auction at the EPEX SPOT power exchange. The model uses multivariate time series as input factors, consisting of point forecasts of electricity prices and fundamental data on generation and load profiles. The architecture of a TSS-CGAN is based on the idea of Conditional Generative Adversarial Networks combined with 1D Convolutional Neural Networks and Bidirectional Long Short-Term Memory. The model is evaluated using qualitative and quantitative criteria. For the evaluation, 10,000 simulations of a test period are carried out. Qualitative criteria are whether the model follows certain electricity market-specific regularities and depicts them adequately. The quantitative analysis includes common error metric, compared to benchmark models, like DeepAR, Prophet and Temporal Fusion Transformer, the examination of the quantile ranges, the error distribution and a sensitivity analysis. The results show that the TSS-CGAN outperforms benchmark models such as DeepAR by reducing the continuous ranked probability score by 50% and considers market-specific circumstances such as the production of fluctuating energies and reacts correctly to changes in the corresponding variables.

Sustainable aviation fuel (SAF) blend has been confirmed to benefit for greenhouse gases reduction, and thus the property of blend fuel should be understanded the detail to support the utilization in aircraft. Low flammability limit (LFL) is a key property of jet fuel which should be sufficiently flammable to burn in combustor of aeroengine and meanwhile should be non-flammable for safety storage in fuel tank of aircraft. LFL of fuel could be influenced by integrating effects including molecule structure, intramolecular chemical bond energy and binding energy of molecule to molecule. Three types of theoretical models, based on different individual view including LFL of every pure hydrocarbon, stoichiometric concentration, and combustion enthalpy, present unsatisfactory simulation results, which can be deduced without integrating all potential influence factors together. The artificial neural network (ANN) approaches have been involved to bridge the relationship of the complex compositions in jet fuels with LFL. For providing adequate and available composition input, the boundary of fuel compositions has been extracted based on constrains of boiling point, flash point and freeze point coupling with statistic petroleum-based jet fuels. By clustering analysis, 43 critical classes of compositions, extracted as surrogate hydrocarbons based on with similar LFL within 1 % deviation, have been deployed as input matrix. ANN-LFL model, trained by only drop-in fuel with feature of Sigmoid function as an activation function, can distinguish drop-in fuel with non-drop-in fuel. ANN LFL model can predict LFL of drop-in fuel with 0.988 accuracy. The predict output value of non-drop-in fuel could present obvious deviation with traditional jet fuel. The optimization methodologies of ANN-LFL model could be improved the understanding of LFL and extend ANN in SAF utilization.

Smart control techniques have been implemented to address fluctuating power levels within isolated microgrids, mitigating the risk of unstable frequencies and the potential degradation of power supply quality. However, a challenge lies in the fact that employing these computationally complex methods without stability preservation might not suffice to handle the rapid changes of this highly dynamic environment in real-world scenarios over communication delays. This study introduces a flexible real-time approach for the frequency control problem using an artificial neural network (ANN) constrained to stabilized regions. Our solution integrates stabilizing PID controllers, computed through small-signal analysis and tuned via an automated search for optimal ANN weights and reinforcement learning (RL)-based selected constraints. First, we design stabilizing PID controllers by applying the stability boundary locus method and the Mikhailov criterion, specifically addressing communication delays. Next, we refine the controller parameters online through an automated process that identifies optimal coefficient combinations, leveraging a constrained ANN to manage frequency deviations within a restricted parameter range. Our approach is further enhanced by employing the RL technique, which trains the tuning system using an interpolated stability boundary curve to ensure both stability and performance. This one-of-a-kind combination of ANN, RL, and advanced PID tuning methods is a big step forward in how we handle frequency control problems in isolated AC microgrids. The experiments show that our solution outperforms traditional methods due to its reduced parameter search space. In particular, the proposed method reduces transient and steady-state frequency deviations more than semi- and unconstrained methods. The improved metrics and stability analysis show that the method improves system performance and stability under changing conditions.

Addressing real-world challenges in battery diagnostics, particularly under incomplete or inconsistent boundary conditions, has proven difficult with traditional methodologies such as first-principles and atomistic calculations. Despite advances in data assimilation techniques, the overwhelming volume and diversity of data, coupled with the lack of universally accepted models, underscore the limitations of these traditional approaches. Recently, deep learning has emerged as a highly effective tool in overcoming persistent issues in battery diagnostics by adeptly managing expansive design spaces and discerning intricate, multidimensional correlations. This approach resolves challenges previously deemed insurmountable, especially with lost, irregular, or noisy data through the design of specialized network architectures that adhere to physical invariants. However, gaps remain between academic advancements and their practical applications, including challenges in explainability and the computational costs associated with AI-driven solutions. Emerging technologies such as explainable artificial intelligence (XAI), AI for IT operations (AIOps), lifelong machine learning to mitigate catastrophic forgetting, and cloud-based digital twins open new opportunities for intelligent battery life-cycle assessment. In this perspective, we outline these challenges and opportunities, emphasizing the potential of innovative technologies to transform battery diagnostics, as demonstrated by our recent practice and the progress made in the field. This includes promising achievements in both academic and industry field demonstrations in modeling and forecasting the dynamics of multiphysics and multiscale battery systems. These systems feature inhomogeneous cascades of scales, informed by our physical, electrochemical, observational, empirical, and/or mathematical understanding of the battery system. Through data assimilation efforts, meticulous craftsmanship, and elaborate implementations—and by considering the wealth and spatio-temporal heterogeneity of available data—such AI-based intelligent learning philosophies have great potential to achieve better accuracy, faster training, and improved generalization.

The performance degradation is a crucial factor affecting the commercialization of proton exchange membrane electrolyzer. However, it is difficult to establish a mechanism model incorporating all degradation categories due to their different time and spatial scales. In this paper, the data-driven method is employed to predict the electrolyzer voltage variation over time based on a convolutional neural network-long short term memory (CNN-LSTM) model. First, two datasets including constant operation for 1140 h and start-stop load for 660 h are collected from the durability tests. Second, the data-driven models are trained through the experimental data and the model hyper-parameters are optimized. Finally, the electrolyzer degradation in the next few hundred hours is predicted, and the prediction accuracy is compared with other time-series algorithms. The results show that the model can predict the degradation precisely on both datasets, with the R² higher than 0.98. Compared to conventional models, the algorithm shows better fitting characteristic to the experimental data, especially as the prediction time increases. For constant and start-stop operations, the electrolyzers degradate by 4.5 % and 2.5 % respectively after 1000 h. The proposed method shows great potential for real-time monitoring in the electrolyzer system.

Offshore wind energy offers several advantages relative to its onshore counterpart – not least stronger and steadier winds, the possibility of larger turbines, and no land occupation. The operational complexities, environmental challenges, and higher maintenance costs of offshore wind turbines necessitate innovative solutions. Traditional approaches are insufficient, and new ”big data” techniques, notably machine learning and deep learning, are poised to play a significant role in the design and optimisation of offshore wind turbines and farms. The objective of this paper is to conduct a scientometric analysis of machine learning and deep learning techniques applied to offshore wind energy. The research methodology employs a circular framework, integrating data acquisition and statistical analysis to provide a comprehensive scientometric insight into the state of the art. As regards the country of origin, most of the publications stem from just five countries, which signals a need of greater geographical diversity in this field of research. Most importantly, the rapid, steady increase in the annual number of publications since 2017 reveals the interest of the research community in this novel topic.

In this study, a Neural Network-Enhanced Gene Modification Optimization Technique was introduced for multi-objective energy resource management. Addressing the need for sustainable energy solutions, this technique integrated neural network models as fitness functions, representing an advancement in artificial intelligence-driven optimization. Data collected in the European Union covered greenhouse gas emissions, energy consumption by sources, energy imports, and Levelized Cost of Energy. Since different configurations of energy consumption by sources lead to varying greenhouse gas emissions, costs, and imports, neural network prediction models were used to project the effect of new energy combinations on these variables. The projections were then fed into the gene modification optimization process to identify optimal configurations. Over 28 generations, simulations demonstrated a 46 percent reduction in energy costs and a 9 percent decrease in emissions. Human bias and subjectivity were mitigated by automating parameter settings, enhancing the objectivity of results. Benchmarking against traditional methods, such as Euclidean Distance, validated the superior performance of this approach. Furthermore, the technique's ability to visualize chromosomes and gene values offered clarity in optimization processes. These results suggest significant advancements in the energy sector and potential applications in other industries, contributing to the global effort to combat climate change.

Electrified transportation requires batteries with high energy density and high-rate capability for both charging and discharging. Li-ion batteries (LiBs) face a dilemma: increasing areal capacity and reducing electrode porosity to boost energy density often reduces rate capability due to a longer and more tortuous ion transfer path. Tailoring cell design parameters to balance these metrics is essential but challenging. Here, we present a multi-objective optimization framework targeting energy density, fast charging, high-rate discharging, and lifespan simultaneously. Four cell parameters—cathode areal capacity, N-P ratio, cathode porosity, and anode porosity—along with operating temperature, are selected as design variables. A physics-based pseudo-2D model, validated against experimental data, generates data to train the surrogate model, which is combined with the NSGA-II algorithm for rapid optimization. Three different objective calculation methods are compared to identify the maximum sum of energy densities, lowest polarization, and most balanced performance, respectively. Cell design parameters are optimized at different temperatures using the most balanced optimization method. Results demonstrate that elevating cell operating temperature achieves high-rate capability while maintaining high energy density, mitigating the energy-power trade-off and broadening battery design parameter ranges.