Artificial intelligence-driven assessment of salt caverns for underground hydrogen storage in Poland

Scientific Reports, Год журнала: 2024, Номер 14(1)

Опубликована: Июнь 20, 2024

Abstract This study explores the feasibility of utilizing bedded salt deposits as sites for underground hydrogen storage. We introduce an innovative artificial intelligence framework that applies multi-criteria decision-making and spatial data analysis to identify most suitable locations storing in caverns. Our approach integrates a unified platform with eight distinct machine-learning algorithms—KNN, SVM, LightGBM, XGBoost, MLP, CatBoost, GBR, MLR—creating rock deposit suitability maps The performance these algorithms was evaluated using various metrics, including Mean Squared Error (MSE), Absolute (MAE), Percentage (MAPE), Root Square (RMSE), Correlation Coefficient (R 2 ), compared against actual dataset. CatBoost model demonstrated exceptional performance, achieving R 0.88, MSE 0.0816, MAE 0.1994, RMSE 0.2833, MAPE 0.0163. novel methodology, leveraging advanced machine learning techniques, offers unique perspective assessing potential is valuable asset stakeholders, government bodies, geological services, renewable energy facilities, chemical/petrochemical industry, aiding them identifying optimal

Язык: Английский

---

Reinforcement learning for battery energy management: A new balancing approach for Li-ion battery packs

Results in Engineering, Год журнала: 2024, Номер 23, С. 102532 - 102532

Опубликована: Июль 11, 2024

This study investigates the challenge of cell balancing in battery management systems (BMS) for lithium-ion batteries. Effective is crucial maximizing usable capacity and lifespan packs, which essential widespread adoption electric vehicles reduction greenhouse gas emissions. A novel deep reinforcement learning (deep RL) approach proposed passive with switched shunt resistors. Notable RL algorithms capable handling discrete actions, such as Trust Region Policy Optimization (TRPO), Proximal (PPO), Deep Q-Network (DQN), Augmented Random Search (ARS), Asynchronous Advantage Actor Critic (A3C), are investigated. TRPO demonstrates superior performance compared to other rule-based methods both charging discharging scenarios without requiring fine-tuning, optimizing balance between switch changes. It achieves up 16.8% improvement pack capacity, 69.4% state-of-charge variance among cells, 40.4% decrease number switching operations simulation results five li-ion cells connected series. The introduces an innovative application balancing, a comprehensive modeling technique, tailored multi-objective reward function that balances costs. work represents significant advancement applying systems, providing framework further research practical implementation energy storage systems.

Язык: Английский

---

Integrating Autoencoder and Decision Tree Models for Enhanced Energy Consumption Forecasting in Microgrids: A Meteorological Data-Driven Approach in Djibouti

Results in Engineering, Год журнала: 2024, Номер unknown, С. 103033 - 103033

Опубликована: Окт. 1, 2024

Язык: Английский

---

Review of energy management systems and optimization methods for hydrogen‐based hybrid building microgrids

Energy Conversion and Economics, Год журнала: 2024, Номер 5(4), С. 259 - 279

Опубликована: Авг. 1, 2024

Abstract Renewable energy‐based microgrids (MGs) strongly depend on the implementation of energy storage technologies to optimize their functionality. Traditionally, electrochemical batteries have been predominant means storage. However, technological advancements led recognition hydrogen as a promising solution address long‐term requirements microgrid systems. This study conducted comprehensive literature review aimed at analysing and synthesizing principal optimization control methodologies employed in hydrogen‐based within context building infrastructures. A comparative assessment was evaluate merits disadvantages different approaches. The techniques for management are categorized based predictability, deployment feasibility, computational complexity. In addition, proposed ranking system facilitates an understanding its suitability diverse applications. encompasses deterministic, stochastic, cutting‐edge methodologies, such machine learning‐based approaches, compares discusses respective merits. key outcome this research is classification various strategy MG, along with mechanism identify which will be suitable under what conditions. Finally, detailed examination advantages strategies controlling optimizing hybrid systems emphasis utilization provided.

Язык: Английский

---

Strategic Participation of Electric Vehicles in Vehicle-to-Grid within a Microgrid System: A Decentralized Optimization Approach

Results in Engineering, Год журнала: 2024, Номер unknown, С. 103144 - 103144

Опубликована: Окт. 1, 2024

Язык: Английский

---

Software Defect Prediction Based on a Multiclassifier with Hyperparameters: Future Work

Results in Engineering, Год журнала: 2025, Номер unknown, С. 104123 - 104123

Опубликована: Янв. 1, 2025

Язык: Английский

---

Self-adjustable Nonlinear Galloping Energy Harvester Under Actual Wind Conditions

Results in Engineering, Год журнала: 2025, Номер unknown, С. 104452 - 104452

Опубликована: Фев. 1, 2025

Язык: Английский

---

Decision Making and Energy Storage System Management

Advances in computational intelligence and robotics book series, Год журнала: 2025, Номер unknown, С. 69 - 80

Опубликована: Фев. 28, 2025

This study evaluates the effectiveness of various machine learning strategies in managing energy Fuel Cell Electric Vehicles (FCEVs), focusing on fuel cell and battery inverter behaviour. The analysis compares four methods Gaussian Naive Bayes (NB), Random Forest, k-NN, AdaBoost using key metrics: Recall, f1-score, precision. NB Forest achieve identical performance for (Recall: 0.87, f1-score: 0.82, precision: 0.89) 0.66, 0.57, 0.5). In contrast, k-NN achieves a precision 0.74, while excels with 0.98 0.94. also outperforms other f1-score (0.98 cell, 0.90 battery) recall (0.95 0.84 battery), highlighting its superior behaviour control.

Язык: Английский

---

Enhanced Fault Detection in Photovoltaic Systems Through Hybrid SVM Evolutionary Optimization Techniques

Advances in computational intelligence and robotics book series, Год журнала: 2025, Номер unknown, С. 127 - 156

Опубликована: Фев. 28, 2025

Fault detection in photovoltaic (PV) systems is vital for maintaining optimal performance. Early of faults can prevent downtime and minimize energy loss. In this study, An approach fault PV proposed. The method integrates a hybrid support vector machine (SVM) with optimization algorithms such as particle swarm (PSO), genetic algorithm (GA), Bayesian (BO), Randomized Search CV (RS). Experimental results demonstrate the effectiveness approach, notably SVM-PSO variant achieving significant precision accuracy. Specifically, employing RBF kernel, model exhibits an accuracy 98.24%, 98.29%, recall 98.25%, F1 score 98.08%. contrast, utilizing linear kernel yields slightly lower performance, 89.47%, 89.82%, 89.51%. proposed system enhances performance reliability, ultimately leading to increased generation reduced maintenance costs.

Язык: Английский

---

From data to durability: Evaluating conventional and optimized machine learning techniques for battery health assessment

Results in Engineering, Год журнала: 2024, Номер 23, С. 102445 - 102445

Опубликована: Июнь 18, 2024

In the electronic era, demand for efficient storage systems has rapidly increased, making health and durability of batteries crucial. This research investigates performance distinct Machine Learning (ML) techniques—namely, Logistic Regression (LR), Convolutional Neural Network (CNN), CNN tuning using Particle Swarm Optimization (PSO)—for Battery Health Analysis (BHA). The dataset comprises various parameters related to battery health, with Remaining Useful Time (RUL) as target variable. proposed work is evaluated Root Mean Squared Error (RMSE), Absolute (MAE), R-squared (R2) scores. Initially, basic LR Model employed BHA, followed by capture complex data patterns. Subsequently, Model's optimized PSO algorithm, aiming improved performance. Experimental results demonstrate that significantly outperforms approach in terms accuracy, lower RMSE MAE, higher R2 conventional model outperformed approach, resulting a 20.11, MAE 15.26, score 0.996; whereas, PSO-Optimized-CNN further enhanced metrics 14.97, 8.03 0.998. Henceforth, PSO-optimized exhibits compared standalone Model. findings offer valuable insights into ML approaches BHA suggest methods optimizing management applications, including renewable energy systems, electric vehicles, portable electronics.

Язык: Английский

---