Energy Informatics最新文献

The power grid infrastructure serves as the backbone of modern society, providing essential electricity supply to meet the demands of various sectors. Ensuring a reliable and efficient power grid amidst increasing demand remains paramount. This paper provides a literature assessment of the United Kingdom’s (UK) power grid, with a focus on fault occurrences, maintenance techniques, and the use of new technology for monitoring and maintenance. According to the research, insulation degradation is the most common source of power grid problems. The power grid’s maintenance cycle is then investigated, including preventive, predictive, and corrective maintenance techniques. The study emphasises the significance of regular inspections, condition-based monitoring, and asset management strategies in improving grid dependability and longevity. The paper then addresses the concept of Integrated Vehicle Health Management (IVHM) and how it relates to power grid infrastructure. It studies the role of IVHM systems in real-time monitoring, diagnostics, and prognostics for grid assets, allowing for predictive maintenance and informed decision-making. Furthermore, the article studies the use of machine learning approaches to power grid health monitoring and maintenance. This article discusses machine learning methodologies such as supervised and unsupervised learning, as well as reinforcement learning, and how they are used in defect detection, classification, and predictive maintenance. Overall, this paper provides an overview of the UK power grid, its fault management strategies, maintenance cycles, and the integration of machine learning techniques for health monitoring and maintenance, offering insights into enhancing grid reliability and performance in the face of evolving challenges.

The transition to battery electric buses (BEBs) offers a significant opportunity to reduce greenhouse gas (GHG) emissions in public transit. However, the limited driving range of BEBs presents operational challenges, making accurate energy demand prediction essential for effective deployment. Despite advances in machine learning and data-driven modeling, an integrated framework for real-world BEB energy demand prediction remains underdeveloped. Most existing research in this domain relies heavily on simulated or controlled datasets, limiting practical applicability. This study addresses this gap by presenting a comprehensive approach to predicting the energy demands of a BEB fleet under actual service conditions, grounded in real-world operational data collected from the Toronto Transit Commission’s (TTC) BEB trial, one of the largest of its kind in North America. At the core of this approach is a novel data processing framework specifically designed for streaming high-resolution vehicle telematics data, which integrates diverse contextual sources such as weather conditions, route topology, passenger loads, and bus schedules. This integrated framework enables the construction of a large-scale BEB dataset derived from in-service operational data of the TTC’s BEB fleet, encompassing 149,813 hours of driving and 2.56 million kilometers traveled. The dataset is leveraged to train and evaluate several machine learning models to predict energy demands along TTC routes. Results demonstrate that the best-performing model achieves a 38% reduction in mean absolute error compared to a baseline method and explains 87% of the variance in net energy demand. Additionally, an analysis of seasonal effects reveals heightened prediction challenges during colder months, driven by increased variability in energy consumption across different BEB makes and models. Finally, a physics-informed hybrid modeling approach is proposed, which integrates energy estimates from vehicle longitudinal dynamics into the data-driven pipeline, yielding further improvements in prediction accuracy and underscoring the value of domain knowledge in machine learning applications for transit.

In contemporary power grid infrastructure, the stability and health of DC systems are critical for uninterrupted energy delivery. As these systems become more complex, traditional monitoring methods are inadequate for detecting early warning signs and critical failures. Integration of cognitive service technologies provides promising capabilities for intelligent monitoring and fault detection in such systems. Despite the availability of raw sensor data, power grid operators struggle to accurately identify and predict faults in DC systems in real-time. The absence of intelligent classification and predictive mechanisms frequently results in a delayed response to system abnormalities, jeopardizing operational reliability. This research aims to develop a machine learning-based monitoring and identification framework for evaluating the operational status of DC systems using sensor-driven datasets. The primary goal is to predict the system’s health status—Healthy, Fault Detected, or Critical Fault—using electrical and environmental parameters. A new algorithm, SmartDC-FaultMonitor, is proposed for analyzing the SmartDC-Monitoring Dataset, which includes voltage, current, temperature, battery condition, communication signal strength, fault alarms, and load status. The methodology includes data preprocessing (missing value handling, encoding, and normalization), hybrid feature selection using Mutual Information and Recursive Feature Elimination (RFE), and classification with an ensemble voting classifier that combines a Light Gradient Boosting Machine (LightGBM), Categorical Boosting (CatBoost), and TabNet. Model tuning is done using grid search, and performance is measured on a hold-out test set. The proposed ensemble model achieved high-performance metrics on the test dataset, with an accuracy of 94.00%, precision of 93.75%, recall of 94.50%, F1-score of 94.12%, and a Matthews Correlation Coefficient (MCC) of 0.91. These results demonstrate the model’s ability to accurately classify system health statuses, including the early detection of critical faults. The study confirms the effectiveness of cognitive service technology in improving the monitoring and identification of DC power grid systems. The SmartDC-FaultMonitor algorithm provides a dependable and scalable approach for real-time fault detection, giving grid operators timely insights and enabling proactive maintenance in smart energy infrastructures.

The integration of renewable energy into power systems has introduced significant complexity and dynamism, particularly in the interaction between distribution network and VPP. Existing methods struggle to capture the complex and dynamic characteristics, while machine learning techniques like LSTM remain underutilized in this context. This study proposes a methodology for evaluating distribution network-VPP interaction in uncertain environments. The methodology integrates a multi-dimensional evaluation index system with a dynamic weighting approach that combines the entropy method for initial weight generation and LSTM for optimization. The evaluation index system covers economic, safety, and flexibility dimensions, with specific indicators designed to capture the complex interdependencies and dynamic characteristics. The LSTM, leveraging its ability to process sequential data and capture temporal dependencies, dynamically adjusts the weights of evaluation indicators based on historical operational patterns, thereby enhancing the accuracy and adaptability of the assessment. Implementation results demonstrate that the proposed method achieves high accuracy and reliability, with MSE of 0.0012, MAE of 0.0056, and WRC of 96.2%. Testing using real-world operational data from a regional distribution network confirms a 95.0% match with expert argumentation, highlighting the practical applicability and robustness of the methodology. This study contributes to the advancement of data-driven decision-making frameworks for power system planning and operation, particularly in the context of integrating distributed energy resources and achieving carbon neutrality goals.

System restoration is aimed at ensuring continuity of the electric supply to all loads in a distribution system under abnormal conditions without violating electrical-constraints. This adds the feature of “self-healing” to the distribution system to make it as smart system. This paper presents a literature survey of published research techniques on electric supply restoration over the period 1981–2024. Four categories of distribution systems with different attributes are proposed by the present authors to compare fairly among these techniques through implementation and running the necessary codes for each restoration technique. Comparisons are concerned with contribution, adopted technique, test model, advantages and disadvantages as well as utilization of renewables. To meet the electrical-constraints on electric supply restoration, fifteen challenges are selected, reviewed and discussed within the comparisons. The algorithms based on graph theory showed better performance regarding the challenges related to minimizing the energy-not-supplied, achieving self-healing dream, preventing feeder overloading and maintaining the voltage profile within limits when compared with other algorithms. The algorithms based on linear and nonlinear programming showed better performance concerning the challenges related to minimizing restoration time and preventing in-supply load shedding when compared with other algorithms. The algorithms based on heuristics and metaheuristics showed better performance concerning the challenges related to system configuration, generating optimal sequence of switches, minimizing the number of ordered switches and reducing the restoration cost when compared with other algorithms. The future trends of the supply restoration in smart distribution systems are also discussed. The present survey is concluded with a summary of the findings from the literature survey and outlines potential directions for future research. It highlights the key opportunities to support researchers in advancing more intelligent restoration strategies for electric supply in smart distribution systems.

Accurate and robust wind power prediction for wind farms could significantly decrease the substantial effect on grid operating safety caused by integrating high-permeability intermittent power supplies into the power grid. The article introduces a new wind power multistep prediction model combining Variational Mode De-composition (VMD) with the Long Short-Term Enhanced Forget Gate (LSTM_EFG) network. The VMD is occupied to break down the initial wind power and speed data into various sub-layers. The LSTM_EFG network predicts the low-frequency sub-layers extracted from the VMD. In contrast, the Artificial Bee Colony optimization algorithm fine-tunes the network for the high-frequency sub-layers acquired from the VMD-LSTM-EFG model. The high performance of projected methods in multistep prediction was evaluated by comparing them with eight different models. Results from four experiments show that: (a) the projected model exhibits the most superior multistep prediction performance out of all models tested; (b) in comparison to other models, the proposed model proves to be more efficient and resilient in capturing trend information. The implementation of accurate wind power prediction models continues to pose challenges due to the unpredictable, sudden, and seasonal changes in wind patterns.

In the contemporary landscape of complex industrial processes, the efficient utilization of renewable energy has emerged as a crucial concern, captivating the attention of researchers, industries, and policymakers alike. However, integrating these renewable energy sources into traditional AC distribution networks has proven to be a formidable challenge. Against this backdrop, this paper presents an innovative optimal control method tailored for energy routers (ERs) in flexible AC/DC distribution networks. To effectively harness the capabilities of ERs, a Long-Short-Term Memory (LSTM) network augmented with an attention mechanism is employed. The attention mechanism allows the LSTM network to focus on the most relevant information in the time-series data, thereby improving the prediction accuracy. Subsequently, an optimization model is constructed to maximize the utilization of renewable energy by ERs. To validate the effectiveness of the proposed method, a two-week field test was conducted as part of an energy retrofit project in China. When compared with conventional methods, the proposed approach has been shown to enhance the local absorption of PV generation by over 24.7%.

Accurate electricity load forecasting is essential for the stability, efficiency, and sustainability of modern power systems. However, individual forecasting models often lack generalization across temporal and regional variations and offer limited interpretability. This study proposes a comprehensive meta-learning-based forecast combination framework to enhance both prediction accuracy and model transparency. Using hourly load data from 20 European countries spanning 2018 to 2024, the framework incorporates time-aware features such as hour of the day, day of the week, month, and public holidays. Ten diverse base models—including XGBoost, LightGBM, Random Forest, and LSTM—are trained globally, from which the top five performers are selected (based on R², MAE, and MAPE) and fed into five meta-learners: Ridge Regression, Lasso, Random Forest, Gradient Boosting, and MLP. These meta-models are trained using both model predictions and engineered time features. Experimental results demonstrate superior performance, with the best-performing meta-learner (Random Forest Regressor) achieving a coefficient of determination (R²) of 0.9998 and a Mean Absolute Percentage Error (MAPE) of 0.79%, significantly outperforming traditional ensemble methods. Furthermore, the inclusion of lag features and 5-fold cross-validation led to substantial improvements across all models, including dramatic reductions in MAE (up to 87%), MAPE (up to 88%), and MSE (up to 97%), along with near-perfect R² scores (~ 1.000). Additionally, SHAP-based explainability reveals the contribution of individual time-based features and the influence of each base model within the ensemble, thereby enhancing transparency and supporting practical decision-making.

The proliferation of electric vehicles (EVs) requires accurate and context-aware forecasting of charging times to maximize user satisfaction and optimize energy resource planning. Existing predictive models, however, sometimes ignore dynamic elements including battery health degradation, ambient temperature variations, and charger variability by depending just on static statistics and simple heuristics. This work presents a robust artificial intelligence-based system integrating data-driven modelling with computer vision for automatic recognition of EV models and adaptive charging time estimate. Multi-angle visual data helps to optimize a refined ResNet50 architecture for strong EV classification. The model guarantees consistent performance under real-world conditions including occlusion, lighting variation, and non-standard viewing angles by using transfer learning, residual feature propagation, and extensive data augmentation. With a top-1 classification accuracy of 96%, an F1-score of 96%, and a recall of 95%, experimental data show that the proposed ResNet50 model beats conventional models including VGG16, VGG19, and YOLOv8. Following recognition, a module driven by metadata retrieves important battery properties. These are then fed into a dynamic power-flow-based charging time calculator that modulates predictions depending on real-time criteria including state-of- charge (SoC), charger rating, and ambient conditions. Through reduction of idle charging times and improvement of user-level decision-making, this combined approach offers a scalable and intelligent answer to EV infrastructure planning. The integration of deep learning-based image recognition with real-time parameterized analytics demonstrates strong potential for advancing smart transportation systems and enabling more adaptive, personalized electric mobility experiences.

Energy communities consist of decentralized energy production, storage, consumption, and distribution and are gaining traction in modern power systems. However, these communities may increase the vulnerability of the grid to cyber threats. We propose an anomaly-based intrusion detection system to enhance the security of energy communities. The system leverages LSTM autoencoders to detect deviations from normal operational patterns in order to identify anomalies induced by attacks or faults. Operational data for training and evaluation are derived from a Simulink-based model of an energy community. The results show that the autoencoder-based intrusion detection system achieves good detection performance across multiple attack scenarios, up to 0.9270 and 0.9735 in precision and recall respectively. We also demonstrate potential for real-world application of the system by training a federated model that enables distributed intrusion detection while preserving data privacy.