Energy and AI最新文献

To solve the optimal power flow (OPF) problem, reinforcement learning (RL) emerges as a promising new approach. However, the RL-OPF literature is strongly divided regarding the exact formulation of the OPF problem as an RL environment. In this work, we collect and implement diverse environment design decisions from the literature regarding training data, observation space, episode definition, and reward function choice. In an experimental analysis, we show the significant impact of these environment design options on RL-OPF training performance. Further, we derive some first recommendations regarding the choice of these design decisions. The created environment framework is fully open-source and can serve as a benchmark for future research in the RL-OPF field.

Machine learning (ML) is a rapidly growing tool even in the lithium-ion battery (LIB) research field. To utilize this tool, more and more datasets have been published. However, applicability of a ML model to different information sources or various LIB cell types has not been well studied. In this paper, an unsupervised learning model called variational autoencoder (VAE) is evaluated with three datasets of charge-discharge cycles with different conditions. The model was first trained with a publicly available dataset of commercial cylindrical cells, and then evaluated with our private datasets of commercial pouch and hand-made coin cells. These cells used different chemistry and were tested with different cycle testers under different purposes, which induces various characteristics to each dataset. We report that researchers can recognise these characteristics with VAE to plan a proper data preprocessing. We also discuss about interpretability of a ML model.

This study addresses the crucial challenge of monitoring the State of Health (SOH) of Lithium-Ion Batteries (LIBs) in response to the escalating demand for renewable energy systems and the imperative to reduce CO2 emissions. The research introduces deep learning (DL) models, namely Encoder-Long Short-Term Memory (E-LSTM) and Convolutional Neural Network-LSTM (CNN–LSTM), each designed to forecast battery SOH. E-LSTM integrates an encoder for dimensionality reduction and an LSTM model to capture data dependencies. CNN–LSTM, on the other hand, employs CNN layers for encoding followed by LSTM layers for precise SOH estimation. Significantly, we prioritize model explainability by employing a game-theoretic approach known as SHapley Additive exPlanations (SHAP) to elucidate the output of our models. Furthermore, a method based on pattern mining was developed, synergizing with the model, to identify patterns contributing to abnormal SOH decrease. These insights are presented through informative plots. The proposed approach relies on the battery dataset from the Massachusetts Institute of Technology (MIT) and showcases promising results in accurately estimating SOH values, in which the E-LSTM model outperformed the CNN–LSTM model with a Mean Absolute Error (MAE) of less than 1%.

Existing research on fault diagnosis for polymer electrolyte membrane fuel cells (PEMFC) has advanced significantly, yet performance is hindered by variations in data distributions and the requirement for extensive fault data. In this study, a cross-domain adaptive health diagnosis method for PEMFC is proposed, integrating the digital twin model and transfer convolutional diagnosis model. A physical-based high-fidelity digital twin model is developed to obtain diverse and high-quality datasets for training diagnosis method. To extract long-term time series features from the data, a temporal convolutional network (TCN) is proposed as a pre-trained diagnosis model for the source domain, with feature extraction layers that can be reused to the transfer learning network. It is demonstrated that the proposed pre-trained model can hold the ability to accurately diagnose the various fuel cell faults, including pressure, drying, flow, and flooding faults, with 99.92 % accuracy, through the effective capture of the long-term dependencies in time series data. Finally, a domain adaptive transfer convolutional network (DATCN) is established to improve the diagnosis accuracy across diverse fuel cells by learning domain-invariant features. The results show that the DATCN model, tested on three different target domain devices with adversarial training using only 10 % normal data, can achieve an average accuracy of 98.5 % (30 % improved over traditional diagnosis models). This proposed method provides an effective solution for accurate cross-domain diagnosis of PEMFC devices, significantly reducing the reliance on extensive fault data.

This paper explores the application of non-intrusive load monitoring techniques in the industrial sector for disaggregating the energy consumption of machinery in manufacturing processes. With an increasing focus on energy efficiency and decarbonization measures, achieving energy transparency in production becomes crucial. Utilizing non-intrusive load monitoring, energy data analysis and processing can provide valuable insights for informed decision-making on energy efficiency improvements and emission reductions. While non-intrusive load monitoring has been extensively researched in the building and residential sectors, the application in the industrial manufacturing domain needs to be further explored. This paper addresses this research gap by adapting established non-intrusive load monitoring techniques to an industrial dataset. By employing artificial neural networks for energy disaggregation, the determination of energy consumption of industrial machinery is made possible. Therefore, a generally applicable cross-energy carrier method to disaggregate the energy consumption of machinery in manufacturing processes is developed using a design science research approach and validated through a practical case study utilizing a compressed air demonstrator. The results show that the utilization of artificial neural networks is well-suited for energy disaggregation of industrial data, effectively identifying on and off states, multi-level states and continuously variable states. Non-intrusive load monitoring should be further considered in the research of emerging artificial intelligence technologies in energy consumption evaluation. It can be a viable alternative for intrusive load monitoring and is a prerequisite to installing energy meters for every machine.

The state of health (SOH) and remaining useful life (RUL) of lithium-ion batteries are crucial for health management and diagnosis. However, most data-driven estimation methods heavily rely on scarce labeled data, while traditional transfer learning faces challenges in handling domain shifts across various battery types. This paper proposes an enhanced vision-transformer integrating with semi-supervised transfer learning for SOH and RUL estimation of lithium-ion batteries. A depth-wise separable convolutional vision-transformer is developed to extract local aging details with depth-wise convolutions and establishes global dependencies between aging information using multi-head attention. Maximum mean discrepancy is employed to initially reduce the distribution difference between the source and target domains, providing a superior starting point for fine-tuning the target domain model. Subsequently, the abundant aging data of the same type as the target battery are labeled through semi-supervised learning, compensating for the source model's limitations in capturing target battery aging characteristics. Consistency regularization incorporates the cross-entropy between predictions with and without adversarial perturbations into the gradient backpropagation of the overall model. In particular, across the experimental groups 13–15 for different types of batteries, the root mean square error of SOH estimation was less than 0.66 %, and the mean relative error of RUL estimation was 3.86 %. Leveraging extensive unlabeled aging data, the proposed method could achieve accurate estimation of SOH and RUL.

Accurate energy consumption forecasting is crucial for reducing operational costs, achieving net-zero carbon emissions, and ensuring sustainable buildings and cities of the future. Despite the frequent use of Artificial Intelligence (AI) algorithms for learning energy consumption patterns and predictions in Building Science, relying solely on these techniques for energy demand prediction addresses only a fraction of the challenge. A drift in energy usage can lead to inaccuracies in these AI models and subsequently to poor decision-making and interventions. While drift detection techniques have been reported, a reliable and robust approach capable of explaining identified discrepancies with actionable insights has not been discussed in extant literature. Hence, this paper presents an Artificial Intelligence framework for energy consumption forecasting with explainable drift detection, aimed at addressing these challenges. The proposed framework is composed of energy embeddings, an optimized dimensional model integrated within a data warehouse, and scalable cloud implementation for effective drift detection with explainability capability. The framework is empirically evaluated in the real-world setting of a multi-campus, mixed-use tertiary education setting in Victoria, Australia. The results of these experiments highlight its capabilities in detecting concept drift, adapting forecast predictions, and providing an interpretation of the changes using energy embeddings.

The energy sector faces rapid decarbonisation and decision-makers demand reliable assessments of the security of electricity supply. For this, detailed simulation models with a high temporal and technological resolution are required. When confronted with increasing weather-dependent renewable energy generation, probabilistic simulation models have proven. The significant computational costs of calculating a scenario, however, limit the complexity of further analysis. Advances in code optimization as well as the use of computing clusters still lead to runtimes of up to eight hours per scenario. However ongoing research highlights that tailor-made approximations are potentially the key factor in further reducing computing time. Consequently, current research aims to provide a method for the rapid prediction of widely varying scenarios. In this work artificial neural networks (ANN) are trained and compared to approximate the system behavior of the probabilistic simulation model. To do so, information needs to be sampled from the probabilistic simulation in an efficient way. Because only a limited space in the whole design space of the 16 independent variables is of interest, a classification is developed. Finally it required only around 35 min to create the regression models, including sampling the design space, simulating the training data and training the ANNs. The resulting ANNs are able to predict all scenarios within the validity range of the regression model with a coefficient of determination of over 0.9998 for independent test data (1.051.200 data points). They need only a few milliseconds to predict one scenario, enabling in-depth analysis in a brief period of time.

Building geometry data is crucial for detailed, spatially-explicit analyses of the building stock in energy systems analysis and beyond. Despite the existence of diverse datasets and methods, a standardized and validated approach for creating a nation-wide unified and complete dataset of German building heights is not yet available. This study develops and validates such a methodology, combining different data sources for building footprints and heights and filling gaps in height data using an XGBoost machine learning algorithm. The XGBoost model achieves a mean absolute error of 1.78 m at the national level and between 1.52 m and 3.47 m at the federal state level. The goal is proving the applicability of the methodology at a large scale and creating a useful dataset. The resulting dataset is thoroughly evaluated on a building-by-building level and spatially resolved statistics on the quality of the dataset are reported. This detailed validation found that the building number and footprint area of German building stock is 90.31 % and 94.84 % correct, respectively, and the building height accuracy is 0.59 m at the national level. However, errors are not homogeneous across Germany and further research is needed into the impact of including additional datasets, especially for regions and building types with lower accuracies. This study proves that the chosen methodology is useful for generating a building height dataset and the workflow, with some modifications for regional data availability, can be transferred to other countries. The generated building dataset for Germany constitutes a valuable data basis for the research community in fields such as energy research, urban planning and building decarbonization policy development.

Proton exchange membrane (PEM) based electrochemical systems have the capability to operate in fuel cell (PEMFC) and water electrolyser (PEMWE) modes, enabling efficient hydrogen energy utilisation and green hydrogen production. In addition to the essential cell stacks, the system of PEMFC or PEMWE consists of four sub-systems for managing gas supply, power, thermal, and water, respectively. Due to the system's complexity, even a small fluctuation in a certain sub-system can result in an unexpected response, leading to a reduced performance and stability. To improve the system's robustness and responsiveness, considerable efforts have been dedicated to developing advanced control strategies. This paper comprehensively reviews various control strategies proposed in literature, revealing that traditional control methods are widely employed in PEMFC and PEMWE due to their simplicity, yet they suffer from limitations in accuracy. Conversely, advanced control methods offer high accuracy but are hindered by poor dynamic performance. This paper highlights the recent advancements in control strategies incorporating machine learning algorithms. Additionally, the paper provides a perspective on the future development of control strategies, suggesting that hybrid control methods should be used for future research to leverage the strength of both sides. Notably, it emphasises the role of artificial intelligence (AI) in advancing control strategies, demonstrating its significant potential in facilitating the transition from automation to autonomy.