62970 research outputs found
Sort by
Advanced Machine Learning Frameworks for Renewable Energy Forecasting and Storage System Optimization
No full textThis dissertation presents a comprehensive investigation into the application of machine learning techniques for enhancing renewable energy systems through accurate forecasting and intelligent modeling, with a particular focus on microgrids, photovoltaic (PV) systems, and battery energy storage technologies. The core objective is to improve the reliability, predictability, and efficiency of renewable energy sources by leveraging both shallow and deep learning models in a variety of real-world scenarios.
The first part of this work focuses on temperature forecasting in microgrids, a critical factor affecting energy generation and operational reliability. Using Long Short-Term Memory (LSTM) neural networks, temperature predictions were generated based on historical weather datasets, including humidity, dew point, pressure, and wind speed. The study also introduces a rolling LSTM approach, allowing unsupervised temperature predictions that can support real-time adaptive control in microgrids. Results demonstrate that LSTM models can effectively capture temporal dependencies and yield accurate short-term temperature forecasts, providing a foundation for ML-driven microgrid control systems.
The second component centers on solar irradiance forecasting using Transformer-based models. A decade-long dataset from the Texas Mesonet Archive was used to train and test the Transformer, which proved successful in capturing irradiance trends and generalizing well to unseen data. A novel rolling Transformer architecture was developed to extrapolate irradiance values even when measurements are unavailable. While the model excelled in short-term predictions, some performance limitations were identified in long-range forecasting, emphasizing the importance of future hyperparameter tuning and architectural refinements. These insights contribute directly to the integration of ML forecasting into PV system control.
In a complementary effort, a critical review of machine learning applications in solar power forecasting was conducted to identify widespread methodological errors in the literature. This review emphasized the importance of proper data preprocessing, cross-validation, and realistic evaluation techniques. Many prior studies were found to be overfitting or relying on limited datasets, leading to inflated claims. By analyzing a wide range of forecasting models, including statistical, deep learning, and hybrid approaches, this work offers a clear set of best practices, urging the research community to adopt more rigorous and transparent model evaluation protocols to ensure real-world applicability and reproducibility.
Building upon this foundation, a comparative study was performed between classical statistical and ensemble learning algorithms (e.g., Random Forests and Gradient Boosting Machines) and deep learning models (LSTMs, sequence-to-sequence, and Transformer networks) for solar power prediction. The findings reveal that well tuned ensemble models, particularly Random Forest and GBM, significantly outperformed deep learning models, achieving a cross-validated of 0.9641. This challenges the conventional focus on deep learning and underscores the potential of hybrid and ensemble methods for more interpretable and robust forecasting solutions.
In addition to solar forecasting, this dissertation presents a novel battery storage prediction framework that utilizes advanced machine learning models to estimate key battery parameters such as State of Health (SoH), State of Charge (SoC), and remaining useful life (RUL). The approach is validated using a diverse set of over 200 battery datasets from NASA, MIT, and Sandia National Laboratories. A distinctive advantage of this framework is its generalizable models trained on one type of battery dataset demonstrate strong predictive performance when applied to entirely different type of battery and use cases. This cross dataset applicability enables the development of reliable battery management systems, especially in contexts where data is sparse or inconsistent, such as remote energy storage systems integrated with solar or wind power. The method offers both economic and operational benefits, allowing for proactive energy storage planning and increased system reliability.
Overall, the integration of forecasting models for temperature, irradiance, solar power, and battery performance offers a unified, machine learning enhanced approach to renewable energy reliability. The presented techniques not only outperform existing methods but also lay a strong foundation for real-time control, planning, and optimization of future smart grids and decentralized energy systems
Preconditioning for Implicit Runge-Kutta Methods Applied to Nonlinear Time-Dependent PDEs
No full textIn this work, we adapt block Jacobi, block Gauss-Seidel, and block LD preconditioners to be effective on the systems of linear equations that arise from applying Implicit Runge Kutta (IRK) methods to time-dependent and nonlinear parabolic partial differential equations (PDEs). When PDEs are linear with time-independent coefficients, the system of algebraic equations that arise from IRK methods has a tensor product structure, which previous authors have used to show order-optimality for the aforementioned preconditioning methods. Nonlinear PDEs or linear PDEs with time-dependent coefficients, however, do not produce this nice structure. We conduct a series of numerical experiments that suggest the adaptations derived in this work are still order-optimal and more efficient at solving time-dependent and nonlinear problems than constructing a time-independent approximation at each step. Additionally, there is a tradeoff when selecting how many stages to use in an IRK method. High stage counts result in larger algebraic systems and larger (fewer) time steps. Low stage counts result in smaller (more) time steps and smaller algebraic systems. Further experiments in this work suggest that 4-stage IRK methods yield the lowest total computation time
Black Women and the Luxury of Choice, Purchase or Protest: A Qualitative Phenomenological Exploration of Identity, Resistance, and the Complex Choices of Consuming Luxury Fashion
No full textThis dissertation, Black Women and the Luxury of Choice, Purchase or Protest: A Qualitative Phenomenological Exploration of Identity, Resistance, and the Complex Choices of Consuming Luxury Fashion, investigates how Black women navigate luxury retail spaces and make purchasing decisions shaped by their lived experiences with racial bias, appearance-based discrimination, and historical exclusion. Utilizing a qualitative phenomenological approach, this study centers the voices of Black women to explore how their identities, cultural consciousness, and experiences of marginalization inform whether they choose to purchase luxury fashion or engage in strategic economic resistance.
Guided by the Stimulus-Organism-Response (SOR) Theory, Black Feminist Theory and the Triple Consciousness Theory, this research examines how environmental stimuli; such as discriminatory service, exclusionary brand practices, and historical legacies like Jim Crow; trigger emotional and cognitive responses that directly impact consumer behavior. Through in-depth interviews with Black women who have engaged with luxury retail spaces, four major themes emerged: (a) Service Quality and Microaggressions, (b) Identity and Consciousness Development, (c) Fashioning Safety: Appearance as Protection and Risk, and (d) Protest and Power: The Cost of Strategic Resistance.
Findings reveal that for Black women, luxury consumption is not simply a matter of fashion or status but a complex negotiation of self-worth, cultural identity, and social justice. Their decisions to purchase or protest are deeply rooted in personal dignity and a commitment to economic empowerment.
This study contributes to both theoretical and practical discussions on consumer behavior, calling for meaningful industry reforms that foster more inclusive, equitable, and affirming retail environments. Ultimately, it affirms that the luxury of choice for Black women is a profound exercise of autonomy, resistance, and power in a marketplace that has historically overlooked their presence and influence
Special Education Teacher Perceptions of Behavior Interventions Used to Support Academic Progress in Students With an Emotional Disturbance in Economically-Disadvantaged Urban North Texas Schools
No full textThis study focused on providing behavior interventions to support students identified with an
emotional disturbance (ED). Students identified with an ED are more likely to perform poorly
in academics given their inability to concentrate or maintain interpersonal relationships with
teachers and peers. Understanding the effectiveness of different behavioral interventions for
students with emotional disturbances will help teachers improve their classroom
management. The researcher targeted middle schools in North Texas with an economically-
disadvantaged designation to provide insight into needed resources and teacher preparation to
provide services to students with an ED. The purpose of this qualitative descriptive study was
to examine the perceptions of special education teachers in inclusion classrooms of the
behavior interventions to support special education students with emotional disturbance in
selected urban economically-disadvantaged North Texas middle schools. Social
constructivism and social constructivist disability theories served as the theoretical
framework of the study. Purposive sampling was used to recruit eight participants of Special
Education teachers in North Texas for the study. Data collection was conducted using semi-
structured interviews and a focus group. A qualitative descriptive study design was used, and
the collected data was analyzed thematically. The findings of this study could be used to
determine the resources needed by teachers to identify effective strategies to implement BIPs
in inclusion and special education classrooms for students with an ED in middle schools in
North Texas
The Range is Hot: Studies in Terminal Ballistics
No full textThe high dynamic loading of a ballistic impact event induces projectile fragmentation. Replacing inert with reactive material (RM) projectiles also adds chemical energy to the kinetic energy induced impact event. The chemical energy conversion process that occurs upon impact, ignition, and fragmentation is not well understood and is the focus of this dissertation. Reactive material projectiles on a macroscale appear as bulk metal, but on the microscale RMs are composed of fine granular powders that have been processed into highly consolidated projectiles. These powders often include metal fuels such as aluminum (Al) and titanium (Ti) and can also be combined with metal oxide such as molybdenum trioxide (MoO3). The measurement of energy conversion of RMs under ballistic impact loading is a challenge addressed here by advancing diagnostics and introducing new analytical techniques.
Results reveal processes fundamental to advancing RM technologies. Increasing the concentration of MoO3 improves gas phase energy conversion. Varying the chamber’s gas environment shows the importance of MoO3 on ignition induced by kinetic energy during ballistic impact events. With a focus on energy partitioning, radiative energy loss is significant during ballistic impact events, however an increase in radiant emission did not correlate to an increase in peak pressure. In fact, greater radiant losses are associated with lower peak pressures, confirming that emitted energy is not deposited into the gas phase. From a measurement science perspective, the Texas Tech University High-Velocity Impact-Ignition Testing System (HITS) is a versatile diagnostic that can be used in many different configurations to gather repeatable and reliable data. By using pressure and temperature measurements to analyze energy exchange, new insight into the dynamic events of ballistic testing have been introduced. With the introduction of novel performance metrics and improved understanding of the experimental setup, more accurate investigations of projectile behaviors will be realized and inform technological developments
Evaluating Opportunity Culture Improving Student Achievement through Strategic Staffing in West Texas
No full textTreatment Responsibility in Racial and Ethnic COVID-19 Disparities: How do Physicians Matter?
No full textThis dissertation examines how the roles and responsibilities of physicians were framed within the dual context of the COVID-19 pandemic and persistent health disparities in the United States. Grounded in framing theory, with particular emphasis on problem definition and responsibility attribution, the study explores how communication strategies shape public understanding of physicians’ contributions during a public health crisis. A multimodal methodological approach including a visual-verbal analysis of the American Medical Association’s (AMA) YouTube videos, analyzing how the organization used both imagery and language to frame physicians' roles, values, and identities. It also employed qualitative content analysis to examine how mainstream news outlets reported on physicians during the same period. This comparative framework assessed differences in narrative construction, emotional appeal, and the use of storytelling between a health advocacy organization and traditional media institutions. Findings shed light on how strategic framing by professional organizations may influence perceptions of professional responsibility, particularly when health organizations use visual and narrative tools to assert authority and shape public discourse. By highlighting the ways emotions, imagery, and messaging are used to construct issue narratives, this research contributes to ongoing conversations about media framing, public health communication, and organizational storytelling. It further shows how digital, discursive and organizational convergences are contributing to the development of organizational framing
RECUERDOS: Storymaking in Mexican American Spaces
No full textThis article explores the collaboration of the Southwest Collection/Special Collections (SWC) Library at Texas Tech University with the Fort Stockton, Texas, Barrio Fest, a community celebration of the legacy of Mexican American culture and history in that West Texas town. Barrio Fest represents an act of community storymaking: the composition and preservation of community memories, histories, and identities, that simultaneously decenter the academic archive and integrate the resources of research universities into achieving the goals of Fort Stockton’s Mexican American community. The SWC’s partnership with Fort Stockton highlights the need for academic archives to foster relationships that make communities aware of their resources without displacing community leadership of such projects
