goal-level energy accounting

This study addresses the limitations of existing AI energy consumption assessments, which typically focus on single inference or training runs and fail to capture the real-world energy dynamics of goal-oriented agent systems involving multi-step execution, retries, and recovery. To bridge this gap, the authors propose the A-LEMS framework, introducing two novel metrics: Energy per successful Goal (EpG) and Orchestration Overhead Index (OOI). A-LEMS integrates a cross-layer observation pipeline with a time-bounded attribution model to enable end-to-end, reproducible energy evaluation. Experimental results demonstrate that agent workflows incur an average EpG of 888.1 joules—4.33 times higher than that of linear baselines—while achieving OOI values below 1.0 in tool-augmented tasks, confirming EpG’s sensitivity and effectiveness in reflecting the energy impact of orchestration structures.

Existing web sustainability reports widely rely on simplified energy models (e.g., DIGEST, DIMPACT), yet their accuracy under real-user interaction scenarios lacks empirical validation. Method: We conduct end-device energy measurements across four representative website categories—e-commerce, booking, navigation, and news—using realistic user workflows on four mainstream laptop models, and systematically compare measured energy consumption against estimates from simplified models. Contribution/Results: We identify that the constant-power assumption inherent in such models introduces substantial systematic bias, with estimation errors varying significantly across website categories and device hardware characteristics. To address this, we propose— for the first time—a model calibration framework that jointly incorporates website functional features (e.g., interaction complexity, media load) and device-specific energy-efficiency parameters. This empirically grounded approach enhances the accuracy, reproducibility, and cross-platform comparability of web sustainability assessments.

This study addresses the disconnect between bottom-up energy consumption metrics and top-down reporting in cloud carbon accounting, which impedes unified accountability. To bridge this gap, we propose the Reconciled Software Carbon Intensity (rSCI) framework. By designing residual decomposition and allocation algorithms grounded in physical drivers and integrating cloud provider audit data, rSCI re-anchors bottom-level energy estimates to top-level emissions reports. The research exposes limitations in existing accounting methodologies and establishes standardized procedures for rSCI implementation. This approach effectively reconciles dual-source signals while preserving optimization incentives, enabling precise attribution of idle capacity and embodied carbon emissions. Ultimately, it offers a new paradigm for carbon accountability in cloud computing.

Current environmental impact assessments of the AI lifecycle suffer from tool heterogeneity, insufficient coverage of water usage and value-chain stages, and poor cross-study comparability. To address these gaps, this paper proposes a unified operational definition of Green AI and introduces a five-stage lifecycle framework—encompassing hardware manufacturing, model development, training, deployment, and reuse. It explicitly distinguishes Green AI (energy-efficiency–focused) from Sustainable AI (holistically sustainable across environmental, social, and economic dimensions). Methodologically, the framework integrates Life Cycle Assessment (LCA), PDCA-based governance, edge–cloud co-optimized hardware strategies, and a calibrated multi-level metrics system combining estimation and empirical measurement. The resulting methodology enables vendor-agnostic, reproducible quantification of energy consumption, carbon emissions, water use, and embodied impacts. This significantly enhances transparency and comparability, providing researchers, engineers, and policymakers with an evidence-driven, actionable guide for sustainable AI development and deployment.

Existing Android energy measurement approaches lack open-source, reliable hardware-level tools, resulting in time-consuming experiments and poor reproducibility. This paper introduces the first automated evaluation framework that deeply integrates physical power measurement hardware—based on high-precision current/voltage sampling—with mainstream IDEs (Android Studio/IntelliJ). Leveraging a custom-built IDE plugin, the framework unifies test triggering, real-time data acquisition, automatic data cleaning and aggregation, multi-run statistical analysis, and visualization reporting. Its key innovation lies in seamlessly embedding hardware-level power measurement directly into the development environment, enabling reproducible, high-accuracy energy-efficiency comparisons under realistic usage scenarios (e.g., cross-version or cross-app evaluations). The solution substantially lowers the barrier to rigorous energy measurement while improving reliability and efficiency. As an open-source infrastructure, it advances both mobile energy-efficiency research and industrial engineering optimization.

This study addresses the long-overlooked indirect carbon emissions generated by the information and communications technology (ICT) sector’s support for the digitalization of high-carbon oil and gas (O&G) industries, thereby challenging prevailing underestimations of the environmental paradox inherent in green technologies. Integrating input-output and macroeconomic data from 2000 to 2022, this work systematically uncovers the previously neglected economic and technological coupling between ICT and O&G sectors and introduces the concept of “self-defeating decarbonization” to describe how digital technologies inadvertently reinforce fossil fuel activities. By developing a classification framework for O&G digitalization activities and an associated embodied emissions assessment methodology, the study finds that approximately 2% of annual ICT expenditures flow into O&G—a figure that in 2022 exceeded combined investments in renewable and nuclear energy by more than fourfold—thus establishing a methodological foundation for quantifying the hidden carbon costs of digital technologies.

Existing software energy measurement tools struggle to balance accuracy and overhead while often being constrained to specific hardware or programming languages, limiting their cross-platform portability. This work proposes CodeGreen, a modular energy measurement platform that innovatively integrates Tree-sitter–based AST queries to enable automatic, multi-language instrumentation. By decoupling instrumentation from measurement through an asynchronous producer-consumer architecture, CodeGreen supports fine-grained energy analysis for languages including Python, C/C++, and Java. Its Native Energy Measurement Backend (NEMB) unifies polling of hardware sensors such as Intel RAPL, NVIDIA NVML, and AMD ROCm. Evaluated on the Computer Language Benchmarks Game, CodeGreen achieves an energy estimation accuracy with a coefficient of determination of R² = 0.9934 and demonstrates near-perfect workload linearity (R² = 0.9997), offering both high precision and low overhead.

研究通过在线调查134名开发者，探讨软件开发中对能耗的关注度及优化方法，揭示了间接优化、认知盲点和采用障碍，并提出改进措施。

AI infrastructure confronts multidimensional physical and economic constraints—including power, thermal management, water usage, interconnect bandwidth, memory capacity, and data throughput—while existing metrics (e.g., PUE, TCO) are siloed and fail to capture the coupled trade-offs among energy efficiency, performance, and cost, hindering cross-layer co-optimization. To address this, we propose a unified measurement architecture grounded in a 6×3 cross-layer taxonomy—spanning facility, network, compute, storage, software, and application layers, each annotated with physical, computational, and economic semantics—and introduce the Measurement Propagation Graph (MPG) to enable, for the first time, system-level, three-dimensional relational modeling. Leveraging systematic literature review, meta-analysis, and graph-based modeling, our framework integrates heterogeneous, multi-source metrics. It supports benchmarking, capacity planning, and total cost of ownership analysis, substantially enhancing interpretability of AI cluster efficiency frontiers and enabling rigorous multi-objective optimization.

This work addresses the challenge of balancing emission constraints, operational cost, and service quality in dynamic power grids where carbon intensity varies over time. Traditional fixed emission rate strategies prove inadequate under such conditions. To overcome this limitation, the authors propose a time-window-based emission budgeting mechanism that replaces static rates, enabling applications to accrue emission allowances during low-carbon periods and flexibly consume them during high-carbon intervals. Integrated within a MAPE-K adaptive control architecture, the approach leverages real-time monitoring of grid carbon intensity and system power consumption to dynamically schedule resources while adhering to long-term emission caps. Simulations using six weeks of real-world data from Germany, France, and Poland demonstrate that the method improves task completion rates by up to 36% in volatile grids while matching the performance of existing approaches in stable grids, achieving effective co-optimization of emissions, cost, and performance.