Abstract

This paper presents a comparison of different statistical and machine learning models using five groups of input features to forecast the future demand of a region, comprising 5 adjacent cities (Lahore, Sheikhupura, Kasur, Okara, Nankana), managed by a single government-operated utility. The proposed methodology was constructed using hourly data of electricity load in MW and hourly data of three weather variables with the highest correlation with load, for more than 10 years i.e. from July 2013 to Dec 2023. We have applied LM, SVM, and LSTM models, the most widely used techniques in literature for long term load forecasts. In three years of out-of-sample forecasts with 26041 timesteps, the SVM model performed best with a MAPE value of 2.98% applying ARIMA with Kalman smoothing to fill missing load values for 4 months. However, LSTM performed 2.5 times better than LM and SVM if input features are only weather variables and historical hourly load is not provided as input to training and test sets.

Abstract

The goal of concurrently addressing green transformation (GX), in this study, we aimed to reconcile exploring regional optimization in smart grids and renewable energy variability in Hitachi, Japan. The breakdown of CO2 emissions among Hitachi City’s sectors was estimated because of the lack of detailed data in the Ministry of Environment’s Municipal Emissions Carte. To foster GX, solar power generation using renewable energy are unstable because they are affected by the weather, and depending on the scale of installation, EVs are equipped with storage batteries, and depending on the time of day, they can be charged as a buffer for surplus electricity from renewable energy. If Hitachi City were to install about 1,000MW of solar panels, it could become a zero-carbon city by 2050.

Abstract

Reversible solid oxide cells (ReSOCs) stack consume electrical energy to produce chemical fuel in the electrolyser (SOEC) mode of operation and convert stored fuel into electrical energy during the fuel cell (SOFC) mode of operation. On the whole, the stack operation is endothermic in the SOEC mode and exothermic in the SOFC mode. A mutually compatible balance-of-the-plant (BoP) with thermal integration for the dual mode of operation is necessary for the ReSOC stack. A detailed process model of ReSOC of MWe scale capacity is developed with two layouts for the BoP: an oxygen-based system and an air-based system. Auxiliary power generation in both modes improves the efficiency of the system. Plant-level compatibility shows the plant can be operated at a 1:2.5 charging-to-discharging time ratio, thus making it highly suitable for islanded power operation of the daily residential energy requirement driven primarily by a solar PV-based renewable energy system. In terms of overall performance, the oxygen-based system shows a roundtrip efficiency of around 70% compared to around 55% for the air-based system. This difference may be attributed to air compression in both modes which requires significantly higher parasitic power consumption.

Abstract

Data centers are energy-intensive infrastructures that generate, manage, and store information for our interconnected society. Models based on Artificial Intelligence (AI) such as ChatGPT are increasingly accessible, leading to significant energy consumption and associated carbon emissions.
Assessing the carbon footprint of AI data centers is essential for evaluating their environmental impact and, consequently, promoting responsible AI development and encouraging sustainable practices. In this work, we evaluate an AI data center’s carbon footprint using a life cycle assessment approach. Unlike existing literature, we analyze the entire data center architecture rather than solely focusing on the servers’ footprint. Additionally, we assess the impact of varying the electricity mix and extending the lifetime of servers, providing potentials for emission reductions.

Abstract

This study evaluates the effectiveness of SHAP (SHapley Additive exPlanations) as a feature selection tool for national demand forecasting, addressing the need for a comprehensive understanding of its strengths and limitations. We benchmark SHAP against common selection methods, including variance-based selection, XGBoost, and NSGA-II. Using a 1D Convolutional Neural Network to forecast hourly national energy demand in Panama, we compare the performance of features selected by each method. Our results indicate that SHAP is generally outperformed by XGBoost and may not be the most effective initial approach for feature selection, nor does it consistently rank features by their predictive value. However, SHAP proves more effective in refining smaller sets of features, maintaining model accuracy by eliminating non-contributive features

Abstract

Shallow geothermal utilization systems (especially GSHP systems) are used for heating and cooling, hot water supply, snow melting, and other applications. They are an energy-saving technology with the potential to decarbonize and reduce utility costs. However, GSHP systems have been slower to spread in Japan than in the other countries due to the high cost of GSHP system installation, lack of awareness, and lack of hydrogeological information. Therefore, the Shallow Geothermal and Hydrogeology Team of AIST is engaged in research on social implementation to popularize shallow geothermal utilization system, and this report introduces an example of such research.

Abstract

With the widespread adoption of Carbon Capture and Storage (CCS) technologies and investigations into advanced power cycles such as the supercritical CO2 (sCO2) Brayton cycle, it is particularly crucial to focus on the risks and dynamics associated with accidental leaks under supercritical conditions. One key aspect is the accurate prediction of critical flow rates of supercritical fluids during leakage events. Owing to the scarcity of experimental data on critical flow in sCO2 microchannels, this study harnesses the abundant experimental data from critical flow studies of water, employing machine learning algorithms enhanced by transfer learning to predict the critical flow rates of sCO2. The approach involves pre-training a neural network on critical flow data from water, followed by fine-tuning with sCO2 data, thereby bridging the gap in data availability and enhancing the model’s generality across various parameters. The principal research findings indicate that transfer learning can improve prediction accuracy and adaptability, suggesting that transferring knowledge from extensively studied fluids like water can effectively enhance the predictive performance of less-documented supercritical fluids. This research provides a valuable tool for designing safer CCS and energy systems.

Abstract

Biogas, produced through the anaerobic digestion (AD) process of biomass, is one of the most promising renewable energy sources. Besides CH4, biogas contains other components that negatively impact its direct application, such as H2S, CO2, NH3, and siloxanes. Among these, CO2 has the most significant effect on reducing the calorific value of biogas, with its content reaching 30%-45%. Therefore, the separation of CO2 from biogas using solid adsorbents for biogas upgrading has been a widely studied topic in recent years. Previous theoretical studies have found strong interactions between Fe3O4 and CO2, indicating a high CO2-specific adsorption capacity. However, the low specific surface area of Fe3O4 limits its CO2 loading capacity. Coating the surface of Fe3O4 with a carbon layer to form carbon-coated magnetite (C@Fe3O4) can significantly increase its specific surface area, making Fe3O4 a promising new CO2 adsorption material. In this study, a novel C@Fe3O4 material was prepared using concentrated organic matter from biogas effluents and iron oxide through a hydrothermal-carbonization process. This material achieved a specific surface area of up to 187.50 m²/g. Under conditions of 1 bar and 37 °C, the material demonstrated a selectivity for CO2 over CH4 of up to 10.5, outperforming most adsorbent materials. Its CO2 adsorption capacity was measured at 0.56 mmol/g (0.99 mmol/cm³).This material supports the development of a new, efficient, low-energy, and low-cost pressure swing adsorption (PSA) process for CO2 and CH4 purification.

Abstract

With the increase in industrial carbon emissions, the efficient recovery of waste heat has become imperative. Thermal Energy Storage (TES) systems utilizing Phase Change Materials (PCMs) present a viable solution, with Packed Bed Latent Heat Storage Systems (PBLHS) being particularly noted for their effectiveness. However, current PBLHS designs face challenges in terms of accuracy and adaptability. This research introduces a MachineLearning (ML) approach to overcome these obstacles. By leveraging data from a validated Computational Fluid Dynamics (CFD) model, a deep ML model was developed and trained, achieving an R2value of 0.975 and a MAPE of less than 9.14%. The Harmony Search algorithm emerged as the most effective optimization technique, which, after refinement, enhanced design efficiency by over 40%. The optimized model improved existing experimental setups by up to 84%. This study underscores the potential of ML in advancing TES system designs for efficient waste heat recovery.

Abstract

Transportation electrification can reduce dependence on fossil fuels and promote energy transition. In the literature, numerous research efforts have been dedicated to investigating the influence of diverse energy and climate policies on EV penetration. However, limited attention has been given to the specific effects of ride-hailing services on EV penetration and the broader sustainable energy transition. In this study, we combine a Stated Preference (SP) survey with a bottom-up transportation model to examine the role of ride-hailing services on EV penetration and energy transition. Our method provides the benefit of analyzing how individuals opt for ride-hailing services over other transportation modes for different trips, while also endogenously determining the technology mix. The results reveal that ride-hailing services have the potential to increase EV penetration by 2.46%, or by 2.89% if the ride is shared, thereby contributing to the transition towards clean energy. Policy interventions such as the implementation of carbon pricing and the promotion of public transport can further drive transportation electrification, showcasing a synergistic effect with ride-hailing services.