The high uncertainty of renewables introduces more dynamics to power systems. The conventional way of monitoring and controlling power systems is no longer reliable. New strategies are needed to ensure the stability and reliability of power systems. This work aims to assess the use of machine learning methods in analyzing data from renewable integrated power systems to aid the decisionmaking of electricity market participants. Specifically, the work studies the cases of electricity price forecast, solar panel detection, and how to constrain the machine learning methods to obey domain knowledge.Chapter 2 proposes to diversify the data source to ensure a more accurate electricity price forecast. Specifically, the proposed two-stage method, namely the rerouted method, learns two types of mapping rules: the mapping between the historical wind power and the historical price and the forecasting rule for wind generation. Based on the two rules, we forecast the price via the forecasted generation and the learned mapping between power and price. The massive numerical comparison gives guidance for choosing proper machine learning methods and proves the effectiveness of the proposed method. Chapter 3 proposes to integrate advanced data compression techniques into machine learning algorithms to either improve the predicting accuracy or accelerate the computation speed. New semi-supervised learning and one-class classification methods are proposed based on autoencoders to compress the data while refining the nonlinear data representation of human behavior and solar behavior. The numerical results show robust detection accuracy, laying down the foundation for managing distributed energy resources in distribution grids. Guidance is also provided to determine the proper machine learning methods for the solar detection problem. Chapter 4 proposes to integrate different types of domain knowledge-based constraints into basic neural networks to guide the model selection and enhance interpretability. A hybrid model is proposed to penalize derivatives and alter the structure to improve the performance of a neural network. We verify the performance improvement of introducing prior knowledge-based constraints on both synthetic and real data sets.

Recent experimental and mathematical work has shown the interdependence of the rod and cone photoreceptors with the retinal pigment epithelium in maintaining sight. Accelerated intake of glucose into the cones via the theoredoxin-like rod-derived cone viability factor (RdCVF) is needed as aerobic glycolysis is the primary source of energy production. Reactive oxidative species (ROS) result from the rod and cone metabolism and recent experimental work has shown that the long form of RdCVF (RdCVFL) helps mitigate the negative effects of ROS. In this work I investigate the role of RdCVFL in maintaining the health of the photoreceptors. The results of this mathematical model show the necessity of RdCVFL and also demonstrate additional stable modes that are present in this system. The sensitivity analysis shows the importance of glucose uptake, nutrient levels, and ROS mitigation in maintaining rod and cone health in light-damaged mouse models. Together, these suggest areas on which to focus treatment in order to prolong the photoreceptors, especially in situations where ROS is a contributing factor to their death such as retinitis pigmentosa (RP). A potential treatment with RdCVFL and its effects has never been studied in mathematical models. In this work, I examine an optimal control with the treatment of RdCVFL and mathematically illustrate the potential that this treatment might have for treating degenerative retinal diseases such as RP. Further, I examine optimal controls with the treatment of both RdCVF and RdCVFL in order to mathematically understand the potential that a dual treatment might have for treating degenerative retinal diseases such as RP. The RdCVFL control terms are nonlinear for biological accuracy but this results in the standard general theorems for existence of optimal controls failing to apply. I then linearize these models to have proof of existence of an optimal control. Both nonlinear and linearized control results are compared and reveal similarly substantial savings rates for rods and cones.

Inside cells, axonal and dendritic transport by motor proteins is a process that is responsible for supplying cargo, such as vesicles and organelles, to support neuronal function. Motor proteins achieve transport through a cycle of chemical and mechanical processes. Particle tracking experiments are used to study this intracellular cargo transport by recording multi-dimensional, discrete cargo position trajectories over time. However, due to experimental limitations, much of the mechanochemical process cannot be directly observed, making mathematical modeling and statistical inference an essential tool for identifying the underlying mechanisms. The cargo movement during transport is modeled using a switching stochastic differential equation framework that involves classification into one of three proposed hidden regimes. Each regime is characterized by different levels of velocity and stochasticity. The equations are presented as a state-space model with Markovian properties. Through a stochastic expectation-maximization algorithm, statistical inference can be made based on the observed trajectory. Regime predictions and particle location predictions are calculated through an auxiliary particle filter and particle smoother. Based on these predictions, parameters are estimated through maximum likelihood. Diagnostics are proposed that can assess model performance and therefore also be a form of model selection criteria. Model selection is used to find the most accurate regime models and the optimal number of regimes for a certain motor-cargo system. A method for incorporating a second positional dimension is also introduced. These methods are tested on both simulated data and different types of experimental data.

Patients suffering from Retinitis Pigmentosa (RP), the most common type of inherited retinal degeneration, experience irreversible vision loss due to photoreceptor degeneration. The preservation of cone photoreceptors has been deemed medically relevant as a therapy aimed at preventing blindness in patients with RP. Cones rely on aerobic glycolysis to supply the metabolites necessary for outer segment (OS) renewal and maintenance. The rod-derived cone viability factor (RdCVF), a protein secreted by the rod photoreceptors that preserves the cones, accelerates the flow of glucose into the cone cell stimulating aerobic glycolysis. This dissertation presents and analyzes ordinary differential equation (ODE) models of cellular and molecular level photoreceptor interactions in health and disease to examine mechanisms leading to blindness in patients with RP.

First, a mathematical model composed of four ODEs is formulated to investigate the progression of RP, accounting for the new understanding of RdCVF’s role in enhancing cone survival. A mathematical analysis is performed, and stability and bifurcation analyses are used to explore various pathways to blindness. Experimental data are used for parameter estimation and model validation. The numerical results are framed in terms of four stages in the progression of RP. Sensitivity analysis is used to determine mechanisms that have a significant affect on the cones at each stage of RP. Utilizing a non-dimensional form of the RP model, a numerical bifurcation analysis via MATCONT revealed the existence of stable limit cycles at two stages of RP.

Next, a novel eleven dimensional ODE model of molecular and cellular level interactions is described. The subsequent analysis is used to uncover mechanisms that affect cone photoreceptor functionality and vitality. Preliminary simulations show the existence of oscillatory behavior which is anticipated when all processes are functioning properly. Additional simulations are carried out to explore the impact of a reduction in the concentration of RdCVF coupled with disruption in the metabolism associated with cone OS shedding, and confirms cone-on-rod reliance. The simulation results are compared with experimental data. Finally, four cases are considered, and a sensitivity analysis is performed to reveal mechanisms that significantly impact the cones in each case.

The cell is a dense environment composes of proteins, nucleic acids, as well as other small molecules, which are constantly bombarding each other and interacting. These interactions and the diffusive motions are driven by internal thermal fluctuations. Upon collision, molecules can interact and form complexes. It is of interest to learn kinetic parameters such as reaction rates of one molecule converting to different species or two molecules colliding and form a new species as well as to learn diffusion coefficients.

Several experimental measurements can probe diffusion coefficients at the single-molecule and bulk level. The target of this thesis is on single-molecule methods, which can assess diffusion coefficients at the individual molecular level. For instance, super resolution methods like stochastic optical reconstruction microscopy (STORM) and photo activated localization microscopy (PALM), have a high spatial resolution with the cost of lower temporal resolution. Also, there is a different group of methods, such as MINFLUX, multi-detector tracking, which can track a single molecule with high spatio-temporal resolution. The problem with these methods is that they are only applicable to very diluted samples since they need to ensure existence of a single molecule in the region of interest (ROI).

In this thesis, the goal is to have the best of both worlds by achieving high spatio-temporal resolutions without being limited to a few molecules. To do so, one needs to refocus on fluorescence correlation spectroscopy (FCS) as a method that applies to both in vivo and in vitro systems with a high temporal resolution and relies on multiple molecules traversing a confocal volume for an extended period of time. The difficulty here is that the interpretation of the signal leads to different estimates for the kinetic parameters such as diffusion coefficients based on a different number of molecules we consider in the model. It is for this reason that the focus of this thesis is now on using Bayesian nonparametrics (BNPs) as a way to solve this model selection problem and extract kinetic parameters such as diffusion coefficients at the single-molecule level from a few photons, and thus with the highest temporal resolution as possible.

In this dissertation two research questions in the field of applied experimental design were explored. First, methods for augmenting the three-level screening designs called Definitive Screening Designs (DSDs) were investigated. Second, schemes for strategic subdata selection for nonparametric predictive modeling with big data were developed.

Under sparsity, the structure of DSDs can allow for the screening and optimization of a system in one step, but in non-sparse situations estimation of second-order models requires augmentation of the DSD. In this work, augmentation strategies for DSDs were considered, given the assumption that the correct form of the model for the response of interest is quadratic. Series of augmented designs were constructed and explored, and power calculations, model-robustness criteria, model-discrimination criteria, and simulation study results were used to identify the number of augmented runs necessary for (1) effectively identifying active model effects, and (2) precisely predicting a response of interest. When the goal is identification of active effects, it is shown that supersaturated designs are sufficient; when the goal is prediction, it is shown that little is gained by augmenting beyond the design that is saturated for the full quadratic model. Surprisingly, augmentation strategies based on the I-optimality criterion do not lead to better predictions than strategies based on the D-optimality criterion.

Computational limitations can render standard statistical methods infeasible in the face of massive datasets, necessitating subsampling strategies. In the big data context, the primary objective is often prediction but the correct form of the model for the response of interest is likely unknown. Here, two new methods of subdata selection were proposed. The first is based on clustering, the second is based on space-filling designs, and both are free from model assumptions. The performance of the proposed methods was explored visually via low-dimensional simulated examples; via real data applications; and via large simulation studies. In all cases the proposed methods were compared to existing, widely used subdata selection methods. The conditions under which the proposed methods provide advantages over standard subdata selection strategies were identified.

Cancer is a disease involving abnormal growth of cells. Its growth dynamics is perplexing. Mathematical modeling is a way to shed light on this progress and its medical treatments. This dissertation is to study cancer invasion in time and space using a mathematical approach. Chapter 1 presents a detailed review of literature on cancer modeling.

Chapter 2 focuses sorely on time where the escape of a generic cancer out of immune control is described by stochastic delayed differential equations (SDDEs). Without time delay and noise, this system demonstrates bistability. The effects of response time of the immune system and stochasticity in the tumor proliferation rate are studied by including delay and noise in the model. Stability, persistence and extinction of the tumor are analyzed. The result shows that both time delay and noise can induce the transition from low tumor burden equilibrium to high tumor equilibrium. The aforementioned work has been published (Han et al., 2019b).

In Chapter 3, Glioblastoma multiforme (GBM) is studied using a partial differential equation (PDE) model. GBM is an aggressive brain cancer with a grim prognosis. A mathematical model of GBM growth with explicit motility, birth, and death processes is proposed. A novel method is developed to approximate key characteristics of the wave profile, which can be compared with MRI data. Several test cases of MRI data of GBM patients are used to yield personalized parameterizations of the model. The aforementioned work has been published (Han et al., 2019a).

Chapter 4 presents an innovative way of forecasting spatial cancer invasion. Most mathematical models, including the ones described in previous chapters, are formulated based on strong assumptions, which are hard, if not impossible, to verify due to complexity of biological processes and lack of quality data. Instead, a nonparametric forecasting method using Gaussian processes is proposed. By exploiting the local nature of the spatio-temporal process, sparse (in terms of time) data is sufficient for forecasting. Desirable properties of Gaussian processes facilitate selection of the size of the local neighborhood and computationally efficient propagation of uncertainty. The method is tested on synthetic data and demonstrates promising results.

Acoustic emission (AE) signals have been widely employed for tracking material properties and structural characteristics. In this study, the aim is to analyze the AE signals gathered during a scanning probe lithography process to classify the known microstructure types and discover unknown surface microstructures/anomalies. To achieve this, a Hidden Markov Model is developed to consider the temporal dependency of the high-resolution AE data. Furthermore, the posterior classification probability and the negative likelihood score for microstructure classification and discovery are computed. Subsequently, a diagnostic procedure to identify the dominant AE frequencies that were used to track the microstructural characteristics is presented. In addition, machine learning methods such as KNN, Naive Bayes, and Logistic Regression classifiers are applied. Finally, the proposed approach applied to identify the surface microstructures of additively manufactured Ti-6Al-4V and show that it not only achieved a high classification accuracy (e.g., more than 90\%) but also correctly identified the microstructural anomalies that may be subjected to further investigation to discover new material phases/properties.

Functional brain imaging experiments are widely conducted in many fields for study- ing the underlying brain activity in response to mental stimuli. For such experiments, it is crucial to select a good sequence of mental stimuli that allow researchers to collect informative data for making precise and valid statistical inferences at minimum cost. In contrast to most existing studies, the aim of this study is to obtain optimal designs for brain mapping technology with an ultra-high temporal resolution with respect to some common statistical optimality criteria. The first topic of this work is on finding optimal designs when the primary interest is in estimating the Hemodynamic Response Function (HRF), a function of time describing the effect of a mental stimulus to the brain. A major challenge here is that the design matrix of the statistical model is greatly enlarged. As a result, it is very difficult, if not infeasible, to compute and compare the statistical efficiencies of competing designs. For tackling this issue, an efficient approach is built on subsampling the design matrix and the use of an efficient computer algorithm is proposed. It is demonstrated through the analytical and simulation results that the proposed approach can outperform the existing methods in terms of computing time, and the quality of the obtained designs. The second topic of this work is to find optimal designs when another set of popularly used basis functions is considered for modeling the HRF, e.g., to detect brain activations. Although the statistical model for analyzing the data remains linear, the parametric functions of interest under this setting are often nonlinear. The quality of the de- sign will then depend on the true value of some unknown parameters. To address this issue, the maximin approach is considered to identify designs that maximize the relative efficiencies over the parameter space. As shown in the case studies, these maximin designs yield high performance for detecting brain activation compared to the traditional designs that are widely used in practice.