Energy Informatics最新文献

This study utilizes the PM100 dataset to quantitatively estimate job-level carbon emissions and analyze efficiency across different resource configurations. A multilayer perceptron (MLP) regression model was applied to predict emissions using execution time along with CPU, memory, and node-level power consumption data. To evaluate efficiency, we proposed the Carbon Efficiency Score (CES), which enables the classification of jobs into efficiency tiers. The analysis revealed that long-running jobs with excessive memory usage tend to exhibit low efficiency, whereas jobs with balanced resource configurations demonstrate relatively higher efficiency. CES-based classification further showed a difference of more than 200-fold between the most and least efficient jobs. Overall, this study provides a foundational framework for developing carbon-aware scheduling strategies in HPC environments and offers practical insights for the design of sustainable supercomputing operational policies.

Buildings consume about 36% of global energy and contribute nearly 40% of CO? emissions, making them central to the challenges of energy and climate. Artificial intelligence (AI) offers transformative pathways to improve forecast accuracy, optimize consumption, and support low carbon transitions. Oriented by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) framework, this review systematically screened the literature from 2020 to July 2025, with selective inclusion of previous foundational studies. In total, 268 publications were reviewed and 70 analyzed in depth. The synthesis covers three domains: (i) energy forecasting, with machine learning (ML) and deep learning (DL) improving demand and renewable generation prediction; (ii) optimization, with AI improving Heating, Ventilation, and Air Conditioning (HVAC) control, renewable scheduling, storage management, and smart grid operations; and (iii) energy efficiency, with AI–Internet of Things (IoT) frameworks enabling predictive control, fault detection, and Net Zero Energy Building (NZEB) strategies. Reported impacts include energy savings for HVAC of up to 37%, solar scheduling that reduces costs by 35%, and AI - IoT integration that reduces emissions by 21%. Publication trends show rapid growth since 2020, reflecting accelerated technological progress. The remaining challenges include data fragmentation, interoperability, high computational demand, and cybersecurity risks. In general, the findings highlight AI as a key enabler of resilient, efficient and climate-adaptive building energy systems.

Under the “Dual Carbon” strategy, efficient recovery of waste heat from diesel equipment in open-pit mines is required. Existing cooling systems cannot handle the dynamic load fluctuations in 5G-enabled energy supply systems, leading to delayed response and low energy efficiency. This paper builds a multi-objective optimization model based on an improved Honey Badger Algorithm. The model uses a Lithium Bromide Absorption cooling system and integrates differential evolution and balancing pool adjustment strategies to enhance the global search ability of the Honey Badger Algorithm. It is also embedded into a 5G scheduling platform to achieve real-time response and intelligent optimization of cooling loads. Experimental results show that the model achieves an average response time of only 1.13 s. Comprehensive system performance indicators such as cooling output, unit cooling cost, and heat recovery rate all outperform traditional optimization methods. The average coefficient of performance reaches 1.78, and the unit cooling cost is as low as 0.40 yuan/kWh. These results demonstrate that the proposed multi-objective optimization model offers excellent performance and practicality in waste heat cooling systems in mining areas. It effectively addresses the problems of slow response and low energy efficiency found in traditional methods and provides a feasible technical path and theoretical support for building green and intelligent energy supply systems in open-pit mines.

This systematic review and meta-analysis critically evaluates artificial intelligence (AI) applications for energy optimization in smart buildings through comprehensive analysis of 126 peer-reviewed studies (2010–2024) from four major databases. We present a novel taxonomic framework categorizing AI implementations into five distinct approaches: predictive systems, adaptive control, pattern recognition, hybrid ensemble methods, and edge AI implementations. Our meta-analysis reveals significant performance variations: reinforcement learning achieves highest energy savings (22.3% ± 8.4%, 95% CI: 20.2–24.4%, I² = 73%), followed by hybrid methods (28.1% ± 12.3%, 95% CI: 23.4–32.8%, I² = 81%) and supervised learning (14.7% ± 5.2%, 95% CI: 12.9–16.5%, I² = 45%). However, substantial heterogeneity exists across building types and climate zones. Critical findings include limited real-world deployment (18% academic literature, 26% including industry reports), predominant focus on office buildings (78%) and temperate climates (67%), and insufficient multi-system integration (76% single-system studies). Economic analysis indicates ROI periods of 2.1–5.8 years (median 3.4 years) with implementation costs varying from $8,000-$47,000 per facility. We identify five persistent research gaps and propose a prioritized research agenda addressing implementation barriers, standardization needs, and occupant-centric optimization. This study provides the first comprehensive benchmarking framework for AI building energy systems and establishes evidence-based guidelines for practical deployment.

Graphical Abstract

Employing fossil fuels in electricity generation increases carbon emissions and worsens global warming. Relying solely on fossil fuels is not a sustainable solution, which is making renewable energy sources (RESs) an essential part of a sustainable power system. However, high RES integration into power systems poses stability challenges, primarily due to issues with balancing supply and demand. One technical solution is to transfer RES unpredictability to the natural gas network, thereby enhancing power system stability. Other complementary solutions include energy storage systems (ESS), such as power-to-gas (P2G) conversion and battery energy storage systems (BESS), which provide additional balancing capabilities. Peer-to-peer (P2P) energy trading facilitates the integration of distributed energy resources (DERs), while microgrid architectures enable their implementation in medium-voltage (MV) distribution systems. However, neglecting grid constraints—particularly active power losses and voltage stability limits—may compromise the sustainability and cost-effectiveness of RES integration. This paper proposes a multi-objective decentralized P2P energy transactive market framework for the integrated multi-energy microgrid (IEM). The proposed framework employs a double-objective optimization (DOO) approach, minimizing both operating costs and active power losses while incorporating AC power flow constraints and gas pipeline dynamics. A Continuous Double Auction (CDA) mechanism facilitates dynamic energy trading among various energy resources (ERs), containing gas-fired generators (GFGs), prosumers (PRs), P2G systems, and RES-based energy producers. The DOO P2P market outperforms single-objective approaches, achieving 40% lower energy losses and 47% lower peak power demand. Additionally, since the DOO framework reduced energy losses and peak demand significantly, the loss-aware energy dispatch improves the average nodal voltage by 1.5%.

This paper proposes an innovative congestion management method for Active Distribution Networks (ADNs) by integrating a Pre-trained Large Language Model (PLLM) with the Goose Optimization (GO) algorithm to tackle challenges posed by the uncertainty of flexible resources. Initially, a short-term power forecasting method is developed using PLLM enhanced with Low-Rank Adaptation (LoRA). This approach precisely captures the dynamic fluctuations of power sources and loads, achieving R2 of 0.96 and 0.95 in wind and load forecasting, respectively. Compared to traditional algorithms (BP, LSTM, BiLSTM), it reduces MAE and RMSE by at least 39.08% and 19.68% for wind forecasting, and 15.82% and 21.80% for load forecasting, respectively, ensuring high-accuracy inputs for the subsequent optimization process. Subsequently, to address the bottlenecks of slow convergence and inefficiency inherent in conventional algorithms for large-scale scheduling, the novel GO algorithm is utilized as the core engine to solve the complex optimal power flow problem. Multi-scenario simulations on the IEEE 33-bus system validate that the proposed framework effectively alleviates line overloads and voltage violations, with the GO algorithm improving scheduling efficiency by at least 65.50% compared to the improved PSO. These findings highlight the exceptional performance and considerable potential of the combined PLLM and GO approach for improving the operational security, economic viability, and scheduling efficiency of ADNs.

Under the large-scale integration of wind turbine and photovoltaic into the grid, the power system faces the challenge of insufficient flexibility for regulation. Coordinated planning of hydro-wind-solar-storage systems can effectively mitigate the output volatility of renewable energy sources. This paper proposes a distributionally robust planning method for hydro-wind-solar-storage systems based on the Wasserstein distance. First, taking into account the spatiotemporal correlations of factors such as wind speed and solar irradiance, an auxiliary classifier generative adversarial network (AC-GAN) is employed to generate a set of wind turbine and photovoltaic output scenarios. Then, a bilevel capacity planning model is constructed for the integrated system. The upper level aims to minimize investment costs by determining the optimal energy storage capacity, while the lower level focuses on minimizing operational costs through optimizing storage operation states and the output of various devices. Subsequently, an improved proximal policy optimization (PPO) algorithm, grounded in the Markov decision process framework, is used to solve the model. Finally, an actual case study based on a hydro-wind-solar system in Qinghai China is conducted to validate the effectiveness of the proposed method.

Accurate prediction of domestic electricity consumption, particularly in intelligent buildings, is crucial for optimal resource utilization and allowing data-driven energy management. The comparison of Long Short-Term Memory (LSTM), Seq2Seq, and Prophet time-series models for electricity consumption prediction using the Low Carbon London dataset is carried out in this study. Preprocessing tasks involving the removal of outliers and missing value replacement were conducted, and Optuna was utilized to optimize model performance by tuning hyperparameters. Evaluation reveals that the Prophet model did the best in accuracy, then the LSTM model, which was remarkably improved via hyperparameter tuning, followed by Seq2Seq, which improved but performed slightly less effectively than LSTM. This article demonstrates the capability of deep learning models to recognize complex temporal patterns and provides a foundation for scalable, data-driven solutions in sustainable energy management. The study also sets the stage for future research on household-specific predictions and cluster-based optimization.

The data management method for transmission line defects and hidden dangers enables timely identification and resolution of safety risks in transmission lines, thereby reducing the probability of failures. However, existing data on defects and hidden dangers are often affected by redundant interference, resulting in low mining accuracy. To address this issue, this paper proposes a data management approach for transmission line defects based on an improved isolation forest algorithm. The types of transmission line hidden dangers are analyzed, and a data governance framework for such hidden dangers is established. This framework collects basic data of transmission lines through multiple channels, performs denoising and normalization processing, and constructs a sample dataset for transmission lines. The isolation forest algorithm is selected as the method for detecting hidden trouble data in transmission lines. The algorithm is enhanced using binary particle swarm optimization to improve the detection of hidden trouble data. The detected defect data are applied to the early warning of transmission lines, thereby completing the defect data management process. Experimental results demonstrate that the proposed method can quickly and accurately detect defect data in transmission lines, and the detection results can effectively facilitate risk warning for transmission lines.

In response to the increasingly concealed and sophisticated methods of electricity theft, which are difficult to comprehensively cover and detect in a timely manner, a method for identifying high-risk electricity theft behaviors in low-voltage distribution station areas based on density-based clustering of IoT sensing data is investigated. An intelligent IoT power distribution terminal is deployed at the distribution transformer side within the station area to collect IoT sensor data reflecting electricity consumption behavior. The density-based clustering algorithm is employed to achieve comprehensive clustering of the IoT sensing data by determining the initial cluster centers and iteratively searching and updating these centers. The clustering results of the IoT sensing data are used as input to an LM-BP neural network, which classifies the electricity consumption behavior data in the station area into normal and abnormal categories. Based on optimal matching values, a feature matching approach is applied to determine whether abnormal electricity consumption samples correspond to high-risk theft behaviors, thereby enabling the detection of such behaviors in low-voltage distribution station areas. Experimental results demonstrate that the proposed method can accurately identify high-risk electricity theft behaviors, such as meter bypassing, by leveraging the density-based clustering results of IoT sensing data.