Energy and AI最新文献

Amid the backdrop of carbon neutrality, traditional energy production is transitioning towards integrated energy systems (IES), where model-based scheduling is key in scenarios with multiple uncertainties on both supply and demand sides. The development of artificial intelligence algorithms, has resolved issues related to model accuracy. However, under conditions of high proportion renewable energy integration, component load adjustments require increased flexibility, so the mathematical model of the component must adapt to constantly changing operating conditions. Therefore, the identification of operating condition changes and rapid model updating are pressing issues. This study proposes a modeling and updating method for IES components based on knowledge distillation. The core of this modeling method is the light weighting of the model, which is achieved through a knowledge distillation method, using a teacher-student mode to compress complex neural network models. The triggering of model updates is achieved through principal component analysis. The study also analyzes the impact of model errors caused by delayed model updates on the overall scheduling of IES. Case studies are conducted on critical components in IES, including coal-fired boilers and turbines. The results show that the time consumption for model updating is reduced by 76.67 % using the proposed method. Under changing conditions, compared with two traditional models, the average deviation of this method is reduced by 12.61 % and 3.49 %, respectively, thereby improving the model's adaptability. The necessity of updating the component model is further analyzed, as a 1.00 % mean squared error in the component model may lead to a power deviation of 0.075 MW. This method provides real-time, adaptable support for IES data modeling and updates.

Mass renovation goals aimed at energy savings on a national scale require a significant level of public financial commitment. To identify target buildings, decision-makers need a thorough understanding of energy performance. Energy Performance Certificates (EPC) provide information about areas of space, such as land plots or a building’s footprint, without specifying exact locations. They cover only a fraction of dwellings. This paper demonstrates that learning from observed EPCs to predict missing ones at the building level can be viewed as a spatial interpolation problem with uncertainty both on input and output variables. The Kriging methodology is applied to random fields observed at random locations to determine the Best Linear Unbiased Predictor (BLUP). Although the Gaussian setting is lost, conditional moments can still be derived. Covariates are admissible, even with missing observations. We present applications using both simulated and real data, with a specific case study of a city in France serving as an example.

Dust accumulation on the surface of solar photovoltaic panels diminishes their power generation efficiency, leading to reduced energy generation. Regular monitoring and cleaning of solar photovoltaic panels is essential. Thus, developing optimal procedures for their upkeep is crucial for improving component efficiency, reducing maintenance costs, and conserving resources. This study introduces an improved Adam optimization algorithm designed specifically for detecting dust on the surface of solar photovoltaic panels. Although the traditional Adam algorithm is the preferred choice for optimizing neural network models, it occasionally encounters problems such as local optima, overfitting, and not convergence due to inconsistent learning rates during the optimization process. To mitigate these issues, the improved algorithm incorporates Warmup technology and cosine annealing strategies with traditional Adam algorithm, that allows for a gradual increase in the learning rate, ensuring stability in the preliminary phases of training. Concurrently, the improved algorithm employs a cosine annealing strategy to dynamically tweak the learning rate. This not only counters the local optimization issues to some degree but also bolsters the generalization ability of the model. When applied on the dust detection on the surface of solar photovoltaic panels, this improved algorithm exhibited superior convergence and training accuracy on the surface dust detection dataset of solar photovoltaic panels in comparison to the standard Adam method. Remarkably, it displayed noteworthy improvements within three distinct neural network frameworks: ResNet-18, VGG-16, and MobileNetV2, thereby attesting to the effectiveness of the novel algorithm. These findings hold significant promise and potential applications in the field of surface dust detection of solar photovoltaic panels. These research results will create economic benefits for enterprises and individuals, and are an important strategic development direction for the country.

Home Energy Management Systems (HEMS) are increasingly relevant for demand-side management at the residential level by collecting data (energy, weather, electricity prices) and controlling home appliances or storage systems. This control can be performed with classical models that find optimal solutions, with high real-time computational cost, or data-driven approaches, like Reinforcement Learning, that find good and flexible solutions, but depend on the availability of load and generation data and demand high computational resources for training. In this work, a novel HEMS is proposed for the optimization of an electric battery operation in a real, online and data-driven environment that integrates state-of-the-art load forecasting combining CNN and LSTM neural networks to increase the robustness of decisions. Several Reinforcement Learning agents are trained with different algorithms (Double DQN, Dueling DQN, Rainbow and Proximal Policy Optimization) in order to minimize the cost of electricity purchase and to maximize photovoltaic self-consumption for a PV-Battery residential system. Results show that the best Reinforcement Learning agent achieves a 35% reduction in total cost when compared with an optimization-based agent.

In this paper, we propose a graph model embedded with compact physical equations for modeling the thermal dynamics of buildings. The principles of heat flow across various components in the building, such as walls and doors, fit the message-passing strategy used by Graph Neural networks (GNNs). The proposed method is to represent the multi-zone building as a graph, in which only zones are considered as nodes, and any heat flow between zones is modeled as an edge based on prior knowledge of the building structure. Furthermore, the thermal dynamics of these components are described by compact models in the graph. GNNs are further employed to train model parameters from collected data. During model training, our proposed method enforces physical constraints (e.g., zone sizes and connections) on model parameters and propagates the penalty in the loss function of GNN. Such constraints are essential to ensure model robustness and interpretability. We evaluate the effectiveness of the proposed modeling approach on a realistic dataset with multiple zones. The results demonstrate a satisfactory accuracy in the prediction of multi-zone temperature. Moreover, we illustrate that the new model can reliably learn hidden physical parameters with incomplete data.

Artificial Neural Networks (ANNs) have emerged as a powerful tool in combustion simulations to replace memory-intensive tabulation of integrated chemical kinetics. Complex reaction mechanisms, however, present a challenge for standard ANN approaches as modeling multiple species typically suffers from inaccurate predictions on minor species. This paper presents a novel ANN approach which can be applied on complex reaction mechanisms in tabular data form, and only involves training a single ANN for a complete reaction mechanism. The approach incorporates a network architecture that automatically conserves mass and employs a particular loss weighting based on species depletion. Both modifications are used to improve the overall ANN performance and individual prediction accuracies, especially for minor species mass fractions. To validate its effectiveness, the approach is compared to standard ANNs in terms of performance and ANN complexity. Four distinct reaction mechanisms (H${}_2$, C${}_7$H${}_{16}$, C${}_{12}$H${}_{26}$, OME${}_{34}$) are used as a test cases, and results demonstrate that considerable improvements can be achieved by applying both modifications.

Supersonic aircraft requires thermal endurance of aviation fuel in the process of cooling engine and aircraft. As the composition of petroleum-based jet fuel (RP-3) is confined by crude oil and refining process, sustainable alternative jet fuel with green house gas reduction become to undertake the composition optimization for improving thermal stability. For designing aviation fuel with robust thermal stability and the detail understanding of thermal stability mechanism, RP-3, Fischer–Tropsch fuel, and additives with cyclic structure for absorbing free radical, were investigated thermal stability by modifying different blend ratios under different conditions. Thermal endurance degree was assessed by chroma and deposition tendency. FT blend with cyclic hydrocarbon can improve thermal endurance degree. In compliance with individual optimized blend ratio, the contribution follows methyl cyclopentane > decalin > methyl cyclohexane > tetralin > n-propyl-benzene > 1,2,4 trimethyl-benzene. The appropriate blend ratio could undertake hydrogen donors for terminating the propagation of oxygen-carrying radical, but hydrocarbons with cyclic structure could enhance deposition tendency. Methyl cyclopentane and its oxidation derivatives take the roles of solvent by anti-polymerization and hydrogen donor by opening cyclic structure in the thermal endurance process, and thus lead to a wide range of blend ratio for improving significantly thermal stability. β-scission leading to C–C bond cleavage is the major reaction at the early decomposition stage, which resulted in most abundant derivatives plus C2. The effects of additives on thermal stability are complex and nonlinear on the tendency of thermal deposits and thermal endurance degree, and thus the appropriate ANN-thermal stability model has been trained based on the experiment data and can achieve above 0.995 correlation coefficient. ANN - thermal stability model can predict not only the content of derivatives including ester, olefin, alcohol, ketone, cyclic oxide, aromatics but also the degree of thermal endurance.

Accurate short-term forecasting of heating energy demand is needed for achieving optimal building energy management, cost savings, environmental sustainability, and responsible energy consumption. Furthermore, short-term heating energy prediction contributes to zero-energy building performance in cold climates. Given the critical importance of short-term forecasting in heating energy management, this study evaluated six prevalent deep-learning algorithms to predict energy load, including single and hybrid models. The overall best-performing predictors were hybrid models using Convolutional Neural Networks, regardless of whether they were multivariate or univariate. Nevertheless, while the multivariate models performed better in the first hour, the univariate models often were more accurate in the final 24 h. Thus, the best-performing predictor of the first timestep was a multivariate hybrid Convolutional Neural Network–Recurrent Neural Network model with a coefficient of determination (R²) of 0.98 and the lowest mean absolute error. Yet, the best-performing predictor of the final timestep was the univariate hybrid model Convolutional Neural Network–Long Short-Term Memory with an R² of 0.80. Also, the prediction accuracy of the best-performing multivariate hybrid models reduced faster per hour compared to the univariate models. These findings suggest that multivariate models may be better suited for early timestep predictions, while univariate models may be better suited for later time steps. Hence, combining the models can enhance accuracy at various timesteps for achieving high fidelity in forecasting and offering a comprehensive tool for energy management.

The reliability and safety of fuel cell vehicle are crucial for the daily operation. Insulation resistance serves as a crucial index of vehicle reliability, especially when fuel cells operate at high voltages. Low insulation resistance can lead to vehicle malfunctions, exposing the operator to the risk of electric shock. In this study, long-term insulation resistance data from thirteen vehicles equipped with three different types of fuel cell systems are analyzed to diagnose possible low insulation resistance issues. For this purpose, a robust locally weighted scatterplot smoothing method is utilized to filter the original data. In this research, an insulation variation model is developed using a data-driven long short-term memory neural network to identify insulation resistance value anomalies caused by deionizer failure. The results indicate that the coefficient of determination of the failure model is 99.78 %. Moreover, current model efficiently identifies insulation faults resulting from reliability issues, such as conductivity issues of cooling pipes and erosion of vehicle wiring harnesses.

This study focuses on the integrated energy production system in industrial parks, addressing the problem of stable load dispatch of equipment under demand fluctuations. A cross-level method for steam load smoothing and optimization is proposed, aiming to achieve stable production and optimal economic performance through three levels of integration: load forecasting, load dispatch, and load regulation. Unlike traditional methods that directly use load forecasting values, heat network elasticity is presented as a buffer between demand and supply. Constraints for minimal changes in equipment load and operational parameters are established for smooth regulation. Industrial cases demonstrate that the load forecasting model has mean absolute percentage errors of 2.44% and 1.68% for medium-pressure and low-pressure steam, respectively, meeting accuracy requirements. The modified supply-side load smoothness is effectively improved by considering heat network elasticity. The method increases boiler efficiency by 1.92%, reducing average coal consumption by 0.92 t/h. Compared to manual operation, the proposed model leads to an average increase of 5.69 MW in power generation and an average reduction of 10.81% in coal-to-electricity ratio. This study verifies the importance of smooth integration across different levels and analyzes the effective response of the proposed method to the uncertainty in load forecasting. The method demonstrates the enormous potential of data-driven methods in achieving safe, economical, and sustainable production in industrial parks.