Journal of Energy Systems Engineering

Abstract Lithium-ion batteries are commonly applied in many industries, particularly in electric vehicles and large-scale energy storage systems. Because of their crucial impact on the device's overall performance and operational safety, accurate estimation of battery capacity and State of Health is required. In recent years, artificial intelligence and machine learning have emerged as powerful tools for predicting battery health by capturing complex nonlinear patterns in operational data, enabling optimized charge cycles, improved maintenance, and enhanced overall system efficiency. This study focuses on predicting the State of Health of Oxford lithium-ion battery cells using machine learning, with voltage, capacity, and temperature identified as key predictive features. A novel hybrid model combining Random Forest and the Sparrow Search Algorithm was developed to improve prediction accuracy. The model utilized voltage, temperature, and capacity features, along with their derivatives, as predictors of SOH. The SSA-RF model outperformed all alternatives across four different cells, achieving R² scores of 0.992, 0.990, 0.994, and 0.988 for cells 1 through 4, respectively. Corresponding Root Mean Square Error (RMSE) percentages were as low as 0.690%, 0.863%, 0.555%, and 0.769%. The Mean Absolute Error (MAE) and Relative Absolute Error (RAE) also reached minimal values, with the best case recorded in cell 3 (MAE: 0.490%, RAE: 0.644%). Model interpretability, enhanced by SHAP and permutation analyses, highlighted voltage features as key state-of-health predictors, confirming the SSA-RF model’s robustness and suitability for predictive maintenance in electrical vehicle and energy storage systems.

Abstract The unpredictability, adverse environmental impact, and rapid rate of fossil fuel depletion necessitate a worldwide shift to alternative energy sources. In regions where there is a high potential for renewable energy, such as Al-Karak in Jordan, sustainable electricity deficit reduction is essential. This study performs a techno-economic feasibility assessment of an off-grid hybrid microgrid power system made up of solar photovoltaics, wind turbines, and hydrogen storage with a project duration of 20 years. The components of the system—Photovoltaic Panels, Wind Turbines, Converter, Electrolyzer, Hydrogen Tank, Proton Exchange Membrane (PEM) Fuel Cell, and Battery Energy Storage System—were optimized using HOMER Pro software. The system offered a Net Present Cost (NPC) of approximately 3,049,711  and a Levelized Cost of Electricity (LCOE) of 0.3642 , which made the system economical compared to conventional methods of energy. The research acknowledges the potential of hybrid renewable energy systems in reducing carbon emissions, enhancing energy security, and reducing dependence on imported fossil fuels. The research also provides a replicable model for the deployment of renewable energy in off-grid areas that aligns with Jordan's national renewable energy policy and the international overall sustainability agenda.

Abstract The global energy landscape is facing increasing challenges due to the depletion of conventional resources and the growing electricity demand. Microgrids offer a promising solution that enhances the sustainability and reliability of grid support and power generation by enabling the integration of renewable energy sources. However, operational uncertainty and competing stakeholder interests continue to make managing multiple microgrids difficult. Artificial intelligence (AI) techniques can assist in the integration, coordination, and control of distributed energy resources by addressing operational constraints and modeling complexity. This paper proposes the optimal microgrid management strategy for regional French energy networks, focusing on the representative winter (January) and summer (July) months. Variables like load demand, operating costs, and the availability of non-dispatchable resources are all taken into consideration by the optimization model. This research looks at two situations: Whereas Scenario 1 only optimizes for economic performance, Scenario 2 considers both environmental and economic constraints. The proposed method schedules unit operations, distributes energy contributions from each source, and regulates grid interactions to achieve economical and environmentally sustainable performance. The results demonstrate how combining emission limits with financial objectives significantly enhances the environmental sustainability of microgrid operations without compromising operational dependability.

Abstract Commonly used fossil fuels like coal, oil, and gas are incredibly costly, harmful to the environment, and negatively impact the ecosystem worldwide. However, some renewable energy sources—such as solar, wind, and hydro—are ecologically sustainable and secure. Among them, solar power is considered the most environmentally friendly renewable energy and has gained widespread acceptance worldwide. Accurate operational solar irradiance forecasts are crucial for solar energy system operators to make better decisions due to resource variability and fluctuating energy demand. However, accurately predicting Direct Normal Irradiance (DNI) is difficult due to several complex and dynamic influencing factors. To forecast DNI in Wuhai City, China, this study introduces a novel hybrid model that involves the integration of a Bidirectional Long Short-Term Memory (Bi-LSTM) network and the Coati Optimization Algorithm (COA). The COA adjusts the hyperparameters of the Bi-LSTM and improves prediction accuracy and robustness over the traditional approach. This synergy leverages the sequence-learning ability of Bi-LSTM and the global optima-seeking ability of COA, which is particularly suited to the nonlinear multivariate nature of solar irradiance forecasting. The input variables for the proposed model include wind speed, temperature, relative humidity, pressure, and cloud cover, while the output variable is DNI. These features were collected in Wuhai City from January 2023 to the end of the year. With the best R² of 0.9851 on the test set, the proposed model outperformed the other baseline models (Support Vector Regression, Long Short-Term Memory, Bi-LSTM, Battle Royale Optimization–Bidirectional Long Short-Term Memory). Therefore, the method presented in this study can be considered a reliable and effective tool for generating accurate forecasts of DNI. This work exhibits significant increases in accuracy and reliability, highlighting the advantages of the hybrid COA-Bi-LSTM approach for solar energy forecasting applications.

Abstract Lithium-ion‍‌ batteries (LIBs) are the main energy sources of today's energy storage systems and are the lifeblood and safety of the diverse applications, which range from portable consumer electronics to large-scale electric vehicles. To guarantee this, the need for an accurate forecasting of the state of health (SOH) is evident. SOH is a parameter that basically shows the battery’s aging condition and remaining usable life. Unfortunately, an accurate SOH is still as difficult as the nonlinear degradation behavior of LIBs in large-scale implementations. The present work introduces a novel hybrid method that pools the Water Cycle Algorithm (WCA) as an optimization technique with Categorical Boosting Regression (CBR) as a predictive model. In order to obtain features that can both explain degradation and predict SOH; health indicators derived from the charging-discharging process were used. The authors propose their method on the actual Oxford battery dataset. The comparative study revealed that the WCA-CBR model, the one the authors propose, was always superior to the other benchmark models in terms of predictive accuracy. In essence, the model reached extremely high determination coefficients (R²) for cells 1, 3, 5, and 7, where the values were 0.992, 0.991, 0.987, and 0.990, respectively. The findings demonstrate the flexibility and dependability of the hybrid approach in battery health prediction. By a further daring prediction of SOH, the WCA-CBR model will not only improve BMS, but also be a safe and sound means of LIBs pervading long-term practical usage in different environments, thus contributing to their lifespan and ‍‌safety.