Data-driven Algorithm Research Articles

A Generalized Deep Reinforcement Learning Model for Distribution Network Reconfiguration with Power Flow-Based Action-Space Sampling

Distribution network reconfiguration (DNR) is used by utilities to enhance power system performance in various ways, such as reducing line losses. Conventional DNR algorithms rely on accurate values of network parameters and lack scalability and optimality. To tackle these issues, a new data-driven algorithm based on reinforcement learning is developed for DNR in this paper. The proposed algorithm comprises two main parts. The first part, named action-space sampling, aims at analyzing the network structure, finding all feasible reconfiguration actions, and reducing the size of the action space to only the most optimal actions. In the second part, deep Q-learning (DQN) and dueling DQN methods are used to train an agent to take the best switching actions according to the switch states and loads of the system. The results show that both DQN and dueling DQN are effective in reducing system losses through grid reconfiguration. The proposed methods have faster execution time compared to the conventional methods and are more scalable.

Hybrid WARIMA-WANN algorithm for data prediction in bridge health monitoring system

Application of data-driven algorithm to model and predict the future trend of the structural health monitoring (SHM) data serves as important references for the future safety evaluation and decision-making of the bridges. An innovative hybrid WARIMA-WANN algorithm combining the Wavelet model, the autoregressive moving average (ARIMA) model and the artificial neural network (ANN) model is proposed to predict massive monitoring data in the bridge SHM system. The WPT serves as the decomposition function for the linear and nonlinear components of the hybrid model. The separated linear/nonlinear components are individually predicted with the ARIMA/ANN models and then combined to obtain the prediction values of the future data. The hybrid method is applied to predict the deflection values of a single-tower cable-stayed bridge. Multiple parameters in individual models are discussed and selected to increase the credibility of the prediction results for the example bridge. The results show that the proposed hybrid algorithm successfully captures the linear and nonlinear coupling relationships inside the SHM data from the frequency perspective of view and shows the best prediction performance when compared with the other four existing models. Different prediction steps obviously influence prediction performance due to the periodic nature of the bridge SHM data.

Data-Driven Model Predictive Control for Redundant Manipulators With Unknown Model.

The tracking control of redundant manipulators plays a crucial role in robotics research and generally requires accurate knowledge of models of redundant manipulators. When the model information of a redundant manipulator is unknown, the trajectory-tracking control with model-based methods may fail to complete a given task. To this end, this article proposes a data-driven neural dynamics-based model predictive control (NDMPC) algorithm, which consists of a model predictive control (MPC) scheme, a neural dynamics (ND) solver, and a discrete-time Jacobian matrix (DTJM) updating law. With the help of the DTJM updating law, the future output of the model-unknown redundant manipulator is predicted, and the MPC scheme for trajectory tracking is constructed. The ND solver is designed to solve the MPC scheme to generate control input driving the redundant manipulator. The convergence of the proposed data-driven NDMPC algorithm is proven via theoretical analyses, and its feasibility and superiority are demonstrated via simulations and experiments on a redundant manipulator. Under the drive of the proposed algorithm, the redundant manipulator successfully carries out the trajectory-tracking task without the need for its kinematics model.

Two-dimensional reinforcement learning model-free fault-tolerant control for batch processes against multi- faults

Aiming at the characteristics of batch process changing along with time and batch directions, the existence of unmodeled dynamics, and the partial failure of actuators or/and sensors, we propose a novel 2D reinforcement learning (RL) fault tolerant control strategy without considering model parameters. Firstly, a two-Dimensional (2D) augmented state space model and 2D Q function-based fault tolerant control (FTC) framework is established. The 2D Bellman equation can be acquired by analyzing the relationship between the 2D value function and the 2D Q function. Based on the extended model and Q-learning concept of RL, a data-driven FTTC independent of model parameters is designed, and a 2D data-driven Q-learning algorithm is proposed. Finally, taking the pressure holding phase in the injection process as the experimental object, the control effect is compared with that of the traditional model-based FTC, and better tracking performance and unbiasedness to the probing noise can be proved.

Modeling solvation dynamics of transition metal redox ion through on-the-fly multi-objective Bayesian-optimized force field.

Modeling solvation dynamics and properties is crucial for developing electrolytes for electrochemical energy storage and conversion devices. This work reports an on-the-fly multi-objective Bayesian optimization (OTF-MOBO) method to parameterize force fields for modeling ionic solvation structures, thermodynamics, and transport properties using molecular dynamics simulations. By leveraging solvation-free energy and solvation radii as training data, we employ the data-driven OTF-MOBO algorithm to actively optimize the force field parameters. The modeling accuracy was evaluated in molecular dynamics simulations until the Pareto front in the parameter space was reached through minimized prediction errors in both solvation-free energy and solvation radii. Using transition metal redox ions (Fe3+/Fe2+, Cr3+/Cr2+, and Cu2+/Cu+) in aqueous solution as examples, we demonstrate that simple force fields combining the Lenard-Jones potential and Coulombic potential can achieve relative error below 2% in both solvation free energy and solvation radii. The optimized force fields can be further extrapolated to predict solvation entropy and diffusivities with relative error below 10% compared with experiments.

Predicting indoor personalized heat stress using wearable sensors and data-driven models

Rising global temperatures pose heightened heat stress challenges, especially impacting vulnerable populations such as older adults in low-income urban areas of color, who disproportionately face energy inefficiencies and environmental burdens. Traditional research often overlooks individual metabolic rates and behaviors in heat stress assessments, relying instead on isolated environmental measures in laboratory settings. This paper presents a novel, occupant-centric approach that integrates wearable sensor data, building information, and air-conditioning usage to predict extreme heat events and assess their impact on occupants through calibrated thermal building simulations for seven older adults in Houston, Texas. We developed a personalized heat stress model based on existing heat index equations and dynamic predicted mean votes, enhanced by measured metabolic rates. A data-driven machine learning algorithm was then used to predict extreme heat events with this model serving as the ground truth. Our findings reveal that elevated activity levels significantly contribute to discomfort during extreme heat events, often undetected by the heat index equation—an environmental metric—alone. Furthermore, we observe that combining metabolic rates with measured indoor temperatures consistently predicts extreme heat events with over 95 % accuracy, demonstrating almost 13 % improvement compared to using indoor temperature in isolation. This study provides a valuable proof of concept that underscores the merits of integrating multiple data sources to raise the accuracy and predictability of indoor extreme heat events. Our research not only offers potential to enhance early warning systems and guide architectural decisions but also influences building material choices, significantly mitigating the adverse effects of extreme heat.

Flexible Modeling of Nonstationary Extremal Dependence using Spatially Fused LASSO and Ridge Penalties

Statistical modeling of a nonstationary spatial extremal dependence structure is challenging. Parametric max-stable processes (MSPs) are common choices for modeling spatially indexed block maxima, where an assumption of stationarity is usual to make inference feasible. However, this assumption is unrealistic for data observed over a large or complex domain. We develop a computationally efficient method for estimating extremal dependence using a globally nonstationary but locally stationary MSP construction, with the spatial domain divided into a fine grid of subregions, each with its own dependence parameters. We use LASSO ( L 1 ) or ridge ( L 2 ) penalties to obtain spatially smooth parameter estimates. We then develop a novel data-driven algorithm to merge homogeneous neighboring subregions. The algorithm facilitates model parsimony and interpretability. To make our model suitable for high-dimensional data, we exploit a pairwise likelihood to perform inference and discuss its computational and statistical efficiency. We apply our proposed method to model monthly maximum temperature data at over 1400 sites in Nepal and the surrounding Himalayan and sub-Himalayan regions; we show significant improvements in model fit compared to a stationary model. Furthermore, we demonstrate that the estimated merged partition is interpretable from a geographic perspective and leads to better model diagnostics by adequately reducing the number of parameters.

Comparative analysis of data-driven electric vehicle battery health models across different operating conditions

The work covers the development of a data-driven algorithm and computes the performance of learning models for lithium-ion battery state of health (SOH) estimation. A wide range of environmental and temperature conditions (15 °C, 25 °C, and 35 °C) at different charging and discharging rates of 1C and 2C are used for electric vehicle battery health estimation. The result of the tested data of cell ‘a’ is validated with a different set of cell ‘b’ on identical test parameters, and the results are tabulated and compared. At 25 °C, the mean absolute errors for the regression algorithms decision tree (DT), k-nearest neighbor (KNN), and random forest (RF) are 3.78641E-03, 3.62524E-03, and 6.16931E-03. The mean absolute percent error for regression algorithms DT, KNN, and RF is 1.48921E-03, 1.40631E-03, and 2.40260E-03. The root mean square error for regression algorithms DT, KNN, and RF is 1.26813E-02, 9.73320E-03, and 1.17238E-02, and the mean squared error for regression algorithms DT, KNN, and RF is 1.60816E-04, 9.47351E-05, and 1.37448E-04. The results show that the KNN and DT methods accurately estimate the SOH under diversified operating conditions in comparison with RF methods and can foster advanced battery health monitoring systems.

Data-driven optimal shrinkage of singular values under high-dimensional noise with separable covariance structure with application

We develop a data-driven optimal shrinkage algorithm, named extended OptShrink (eOptShrink), for matrix denoising with high-dimensional noise and a separable covariance structure. This noise is colored and dependent across samples. The algorithm leverages the asymptotic behavior of singular values and vectors of the noisy data's random matrix. Our theory includes the sticking property of non-outlier singular values, delocalization of weak signal singular vectors, and the spectral behavior of outlier singular values and vectors. We introduce three estimators: a novel rank estimator, an estimator for the spectral distribution of the pure noise matrix, and the optimal shrinker eOptShrink. Notably, eOptShrink does not require estimating the noise's separable covariance structure. We provide a theoretical guarantee for these estimators with a convergence rate. Through numerical simulations and comparisons with state-of-the-art optimal shrinkage algorithms, we demonstrate eOptShrink's application in extracting maternal and fetal electrocardiograms from single-channel trans-abdominal maternal electrocardiograms.

Object classification through heterogeneous fog with a fast data-driven algorithm using a low-cost single-photon avalanche diode array

This study presents a framework for classifying a wooden mannequin’s poses using a single-photon avalanche diode (SPAD) array in dynamic and heterogeneous fog conditions. The target and fog generator are situated within an enclosed fog chamber. Training datasets are continuously collected by configuring the temporal and spatial resolutions on the sensor's firmware, utilizing a low-cost SPAD array sensor priced below $5, consisting of an embedded SPAD array and diffused VCSEL laser. An extreme learning machine (ELM) is trained for rapid pose classification, as a benchmark against CNN. We quantitatively justify the selection of nodes in the hidden layer to balance the computing speed and accuracy. Results demonstrate that ELM can accurately classify mannequin poses when obscured by dynamic heavy fog to 35 cm away from the sensor, enabling real-time applications in consumer electronics. The proposed ELM achieves 90.65% and 89.58% accuracy in training and testing, respectively. Additionally, we demonstrate the robustness of both ELM and CNN as the fog density increases. Our study also discusses the sensor’s current optical limitations and lays the groundwork for future advancements in sensor technology.

Getting away from the cutoff in regression discontinuity designs

Regression discontinuity (RD) designs are highly popular in economic research because of their strong internal validity and straightforward intuition. While RD estimates are local in nature, several recent articles propose methods that generalize RD estimates to units outside a small neighborhood of the cutoff. In this article, I introduce the getaway package, which implements the method proposed by Angrist and Rokkanen (2015, Journal of the American Statistical Association 110: 1331-1344) to extrapolate treatment-effect estimates “away from the cutoff”, relying on a classical unconfoundedness condition. Additionally, the package features a data-driven algorithm designed to identify a set of covariates that fulfills the unconfoundedness assumption. It also incorporates a toolkit intended for testing and visualization purposes.

Data-driven insights into cavitation phenomena: From spatiotemporal features to physical state transitions

The application of data-driven methods to study cavitation flow provides insights into the underlying mechanisms and richer physical details of cavitation phenomena. This paper aims to analyze the physically interpretable multi-state cavitation behavior. Initially, the spatiotemporal features of the cavitation flow are represented as network trajectories using principal component analysis. The k-means++ algorithm is then employed to obtain coarse-grained flow field states, and the centroid of each cluster served as a representative for the attributes of that state. Subsequently, the Markov state model is constructed to capture the dynamic transitions in the cavitation flow field. Through a detailed analysis of the dynamic transition model, the cavitation flow field states with genuine physical mechanisms are refined. Finally, proper orthogonal decomposition (POD) is utilized to extract the flow patterns corresponding to different states. The distribution characteristics of the flow field modes in different states correspond to their physical properties. These data-driven algorithm enables a detailed analysis of the typical states in periodic cavitation processes, such as cavity growth, development, shedding, and collapse, providing a deeper understanding of the cavitation flow characteristics in different typical states.

An optimization framework for achieving optimal hydrocyclone's performance aligning with decision-makers' preferences

Previous hydrocyclone optimization often neglected interactions among key performance objectives, which limits hydrocyclone wide applications to meet increasingly diverse industry demands. This study proposes an optimization framework to identify the most suitable hydrocyclone design and operating conditions with conflicting key performance objectives. An advanced multi-objective evolutionary algorithm (PICEA-g) is employed to generate Pareto-optimal solutions that capture trade-offs among multiple conflicting objectives. A novel data-driven predictive algorithm, INFO-ELM, is introduced to establish nonlinear relationships between key variables and performance objectives, thereby accelerating the search for Pareto-optimal solutions by PICEA-g. Furthermore, a multi-criteria decision-making method (TOPSIS) is utilized to determine the optimal solution based on decision-makers' preferences, ensuring consistency between preference information and decision outcomes. The framework's effectiveness is validated across various separation scenarios using two decision-making strategies. This study offers a comprehensive approach to address trade-offs in hydrocyclone optimization, widening its application in diverse separation scenarios.

Solving optimal predictor-feedback control using approximate dynamic programming

This paper is concerned with approximately solving the optimal predictor-feedback control problem of multiplicative-noise systems with input delay in infinite horizon. The optimal predictor-feedback control, provided by the analytical method, is determined by Riccati–ZXL equations and is hard to obtain in the case of unknown system dynamics. We aim to propose a policy iteration (PI) algorithm for solving the optimal solution by approximate dynamic programming. For convergence analysis of the algorithm, we first develop a necessary and sufficient stabilizing condition, in the form of several new Lyapunov-type equations, which parameterizes all predictor-feedback controllers and can be seen as an important addition to Lyapunov stability theory. We then propose an iterative scheme for the Riccati–ZXL equations computations, along with convergence analysis, based on the condition. Inspired by this scheme, a data-driven online PI algorithm, convergence implied in that of the iterative scheme, is proposed for the optimal predictor-feedback control problem without full system dynamics. Finally, a numerical example is used to evaluate the proposed PI algorithm.

2575: Title: Evaluation of data-driven motion algorithm OncoFreeze AI in a Siemens Biograph PET-CT

2575: Title: Evaluation of data-driven motion algorithm OncoFreeze AI in a Siemens Biograph PET-CT

Algorithm prediction of single particle irradiation effect based on novel TFETs

In order to predict the single particle irradiation of tunnel field effect transistor (TFET) devices, a deep learning algorithm network model was built to predict the key characterization parameters of the single particle transient. Computer aided design (TCAD) technique is used to study the influence of single particle effect on the novel stacked source trench gate TFET device. The results show that with the increase of drain voltage, incident width of heavy ions (less than 0.04μm), and linear energy transfer, the drain transient current and collected charge also increase. The prediction results of deep learning algorithm show that the relative error percentage of drain current pulse peak (IDMAX) and collected charge (Qc) is less than 10%, and the relative error percentage of most predicted values is less than 1%. Comparison experiments with five traditional machine learning methods (support vector machine, decision tree, K-nearest algorithm, ridge regression, linear regression) show that the deep learning algorithm has the best performance and has the smallest average error percentage. This data-driven deep learning algorithm model not only enables researchers who are not familiar with semiconductor devices to quickly obtain the transient data of a single particle under any conditions; at the same time, it can be applied to digital circuit design as a data-driven device model reflecting the reliability of single particle transient. The application of deep learning in the field of device irradiation prediction has a highly promising prospect in the future.

Epileptic network identification: insights from Dynamic Mode Decomposition of sEEG data.

For medically-refractory epilepsy patients, stereoelectroencephalography (sEEG) is a surgical method using intracranial recordings to identify brain networks participating in early seizure organization and propagation (i.e., the epileptogenic zone, EZ). If identified, surgical EZ treatment via resection, ablation or neuromodulation can lead to seizure-freedom. To date, quantification of sEEG data, including its visualization and interpretation, remains a clinical and computational challenge. Given elusiveness of physical laws or governing equations modelling complex brain dynamics, data science offers unique insight into identifying unknown patterns within high dimensional sEEG data. We apply here an unsupervised data-driven algorithm, Dynamic Mode Decomposition (DMD), to sEEG recordings from five focal epilepsy patients (three with temporal lobe, and two with cingulate epilepsy), who underwent subsequent resective or ablative surgery and became seizure free. DMD obtains a linear approximation of nonlinear data dynamics, generating coherent structures ("modes") defining important signal features, used to extract frequencies, growth rates and spatial structures. DMD was adapted to produce Dynamic Modal Maps (DMMs) across frequency sub-bands, capturing onset and evolution of epileptiform dynamics in sEEG data. Additionally, we developed a static estimate of EZ-localized electrode contacts, termed the Higher-Frequency Mode-based Norm Index (MNI). DMM and MNI maps for representative patient seizures were validated against clinical sEEG results and seizure-free outcomes following surgery. DMD was most informative at higher frequencies, i.e. gamma (including high-gamma) and beta range, successfully identifying EZ contacts. Combined interpretation of DMM/MNI plots best identified spatiotemporal evolution of mode-specific network changes, with strong concordance to sEEG results and outcomes across all five patients. The method identified network attenuation in other contacts not implicated in the EZ. This is the first application of DMD to sEEG data analysis, supporting integration of neuroengineering, mathematical and machine learning methods into traditional workflows for sEEG review and epilepsy surgical decision-making.

Crustal-mantle layering deposits in the Von Kármán crater, South Pole-Aitken basin

Comprehending the properties of the Moon's crust and mantle materials is essential for decoding its interior. Whether the South Pole-Aitken (SPA) basin potentially excavated the upper mantle materials remains mysterious. The Chang'E Four Lunar Penetrating Radar (LPR) has provided an unprecedented opportunity to investigate the SPA basin in the Von Kármán (VK) crater. Utilizing a data-driven generative deep learning inversion algorithm, the subsurface physical properties along the Yutu-2 path were determined. 4 distinct deposit zones were discovered beneath the surface. All zones feature an upper low-permittivity (paleo-)regolith layer overlaying on a lower high-permittivity layer of rockier ejecta or basal basalt. Zone 1, an ejecta layer from the Finsen crater within the SPA Compositional Anomaly, has a bulk density (2.81±0.22g/cm3) denser than that of highland crust represented by materials originating from VKL and VKL' craters in the Mg-Pyroxene Annulus, which are present in Zones 2 and 3. This suggests the possible presence of mantle materials in Zone 1, providing the first diagnostic evidence that the SPA impact may have penetrated the upper lunar mantle.

Explainable data-driven Q-learning control for a class of discrete-time linear autonomous systems

Explaining what a reinforcement learning (RL) control agent learns play a crucial role in the safety critical control domain. Most of the approaches in the state-of-the-art focused on imitation learning methods that uncover the hidden reward function of a given control policy. However, these approaches do not uncover what the RL agent learns effectively from the agent-environment interaction. The policy learned by the RL agent depends in how good the state transition mapping is inferred from the data. When the state transition mapping is wrongly inferred implies that the RL agent is not learning properly. This can compromise the safety of the surrounding environment and the agent itself. In this paper, we aim to uncover the elements learned by data-driven RL control agents in a special class of discrete-time linear autonomous systems. Here, the approach aims to add a new explainable dimension to data-driven control approaches to increase their trust and safe deployment. We focus on the classical data-driven Q-learning algorithm and propose an explainable Q-learning (XQL) algorithm that can be further expanded to other data-driven RL control agents. Simulation experiments are conducted to observe the effectiveness of the proposed approach under different scenarios using several discrete-time models of autonomous platforms.

Discovering Ancestral Instrumental Variables for Causal Inference From Observational Data.

Instrumental variable (IV) is a powerful approach to inferring the causal effect of a treatment on an outcome of interest from observational data even when there exist latent confounders between the treatment and the outcome. However, existing IV methods require that an IV is selected and justified with domain knowledge. An invalid IV may lead to biased estimates. Hence, discovering a valid IV is critical to the applications of IV methods. In this article, we study and design a data-driven algorithm to discover valid IVs from data under mild assumptions. We develop the theory based on partial ancestral graphs (PAGs) to support the search for a set of candidate ancestral IVs (AIVs), and for each possible AIV, the identification of its conditioning set. Based on the theory, we propose a data-driven algorithm to discover a pair of IVs from data. The experiments on synthetic and real-world datasets show that the developed IV discovery algorithm estimates accurate estimates of causal effects in comparison with the state-of-the-art IV-based causal effect estimators.