Data-driven Learning Algorithm Research Articles

Nonparametric data-driven learning algorithms for multilocation inventory systems

We study a multilocation inventory system with unknown demand distribution using a nonparametric approach. The system consists of multiple distribution centers and customer locations, where products are shipped from the distribution centers to fulfill customer demands. We propose a novel algorithm, DMLI, for adaptive inventory management. Under specific conditions, we establish that the average expected T-period regret of DMLI converges to the optimal rate of O(1/T).

Data-driven learning algorithm to predict full-field aerodynamics of large structures subject to crosswinds

Quick and high-fidelity updates about aerodynamic loads of large-scale structures, from trains, planes, and automobiles to many civil infrastructures, serving under the influence of a broad range of crosswinds are of practical significance for their design and in-use safety assessment. Herein, we demonstrate that data-driven machine learning (ML) modeling, in combination with conventional computational methods, can fulfill the goal of fast yet faithful aerodynamic prediction for moving objects subject to crosswinds. Taking a full-scale high-speed train, we illustrate that our data-driven model, trained with a small amount of data from simulations, can readily predict with high fidelity pressure and viscous stress distributions on the train surface in a wide span of operating speed and crosswind velocity. By exploring the dependence of aerodynamic coefficients on yaw angles from ML-based predictions, a rapid update of aerodynamic forces is realized, which can be effectively generalized to trains operating at higher speed levels and subject to harsher crosswinds. The method introduced here paves the way for high-fidelity yet efficient predictions to capture the aerodynamics of engineering structures and facilitates their safety assessment with enormous economic and social significance.

A data-driven acceleration-level scheme for image-based visual servoing of manipulators with unknown structure.

The research on acceleration-level visual servoing of manipulators is crucial yet insufficient, which restricts the potential application range of visual servoing. To address this issue, this paper proposes a quadratic programming-based acceleration-level image-based visual servoing (AIVS) scheme, which considers joint constraints. Besides, aiming to address the unknown problems in visual servoing systems, a data-driven learning algorithm is proposed to facilitate estimating structural information. Building upon this foundation, a data-driven acceleration-level image-based visual servoing (DAIVS) scheme is proposed, integrating learning and control capabilities. Subsequently, a recurrent neural network (RNN) is developed to tackle the DAIVS scheme, followed by theoretical analyses substantiating its stability. Afterwards, simulations and experiments on a Franka Emika Panda manipulator with eye-in-hand structure and comparisons among the existing methods are provided. The obtained results demonstrate the feasibility and practicality of the proposed schemes and highlight the superior learning and control ability of the proposed RNN. This method is particularly well-suited for visual servoing applications of manipulators with unknown structure.

Identifying Prominent Environmental Covariates Using Variable Selection Methodologies for Digital Soil Mapping of Tamil Nadu, India

High dimensional datasets that depict intricate spatial variations are necessary to predict complex landscape structures and the corresponding soil properties taking into account the size of the research region in addition to the data attributes. The number and quality of the input datasets taken into consideration essentially determine the quantity and quality of the soil properties that may be predicted thanks to data-driven learning algorithms. The use of variable selection strategies both before and after the prediction can have a significant impact on the outcome and can lower the related computing load. The majority of commonly used variable selection techniques such as correlation analysis, stepwise regression and recursive feature elimination, among others perform recursive statistical/mathematical comparison to identify the significant covariates that improve the effectiveness of the algorithm proposed. In order to identify the effective environmental variables in predicting the soil attribute, this article investigated a widely used recursive ranking method called recursive feature elimination. The covariate layer that produced the lowest RMSE was placed first according to the rankings of the covariates provided by recursive feature elimination. The findings showed that among other factors physiography, mean rainfall, rock outcrop difference ratio, elevation and mean temperature will be effective in predicting the soil properties required for digital soil mapping.

Human-in-the-loop distributed cooperative tracking control with applications to autonomous ground vehicles: A data-driven mixed iteration approach

This paper proposes a data-driven learning algorithm for the human-in-the-loop cooperative tracking control of multi-agent systems with completely unknown dynamics. The core role of the human is to supervise a team of agents via sending a control command to only the leader vehicle, where the leader is a non-autonomous agent that has a nonzero but bounded control input receiving the control command from the human operator. To effectively reduce communication frequency among vehicles, a distributed observer only using the relative input–output information is projected to approximate the relative state information of agents. Then, taking the transient state into account instead of the steady-state, a value iteration (VI) algorithm is proposed to solve the optimal output trajectory tracking control without an initial stabilizing control policy but sacrificing the convergence rate. Under the VI algorithm, a mixed iteration (MI) algorithm by making the full advantages of both VI and the traditional policy iteration algorithms is developed, which not only guarantees the convergence rate but also removes the requirement of an initial stabilizing control policy. Finally, a practical multi-robot network is provided to verify and compare the advantages of the proposed theoretical algorithm.

A combined experimental and analytical method to determine the EHL friction force distribution between rollers and outer raceway in a cylindrical roller bearing

Friction force is a crucial factor causing power loss and fatigue spalling of rolling element bearings. A combined experimental and analytical method is proposed to quantitatively determine the elastohydrodynamic lubrication (EHL) friction force distribution between rollers and outer raceway in a cylindrical roller bearing (CRB). An experimental system with the instrumented bearing and housing was developed for measuring radial load distribution and friction torque of bearings. A simplified model of friction force expressed by dimensionless speed, load, and material parameters was given. An inequality constrained optimization problem was established and solved by using an experimental data-driven learning algorithm for determining the uncertain parameters in the model. The effect of speed, load, and lubricant property on friction force and friction coefficient was discussed.

An iterative data-driven learning algorithm for calibration of the internal model in advanced wind turbine controllers

Modern industrial wind turbine controllers for partial-load region control are becoming increasingly complex by progressively relying on modeled aerodynamic characteristics. These advanced turbine controllers generally consist of a combined wind speed estimator and tracking controller, allowing for a granular trade-off between energy capture maximization and (fatigue) load minimization. Because of the limited measurements available to the controller, the control scheme's internal model quality is of utmost importance in satisfying performance and stability requirements. Therefore, the calibration thereof is of particular interest. To date, little work has been performed on the direct calibration of the model information. This work proposes a data-driven iterative learning algorithm for calibrating the internal physical model parameters. The learning algorithm uses generally available closed-loop turbine measurements, complemented with an external measurement of the rotor effective wind speed (REWS), and is thereby largely nondisruptive. The algorithm is based on steady-state assumptions and performs iterative batch-wise updates of the internal control model toward convergence. As the algorithm corrects at the actual turbine operating point, short-term relocations of the turbine's operating point can be used to calibrate in a broader operational domain. Results show outstanding learning capabilities for an aerodynamically degraded wind turbine under realistic turbulent wind conditions. Moreover, a sensitivity study is performed to expose the algorithm's susceptibility to measurement errors, algorithm tuning, and the size of the data set.

A computational method for the load spectra of large-scale structures with a data-driven learning algorithm

For complex engineering systems, such as trains, planes, and offshore oil platforms, load spectra are cornerstone of their safety designs and fault diagnoses. We demonstrate in this study that well-orchestrated machine learning modeling, in combination with limited experimental data, can effectively reproduce the high-fidelity, history-dependent load spectra in critical sites of complex engineering systems, such as high-speed trains. To meet the need for in-service monitoring, we propose a segmentation and randomization strategy for long-duration historical data processing to improve the accuracy of our data-driven model for long-term load-time history prediction. Results showed the existence of an optimal length of subsequence, which is associated with the characteristic dissipation time of the dynamic system. Moreover, the data-driven model exhibits an excellent generalization capability to accurately predict the load spectra for different levels of passenger-dedicated lines. In brief, we pave the way, from data preprocessing, hyperparameter selection, to learning strategy, on how to capture the nonlinear responses of such a dynamic system, which may then provide a unifying framework that could enable the synergy of computation and in-field experiments to save orders of magnitude of expenses for the load spectrum monitoring of complex engineering structures in service and prevent catastrophic fatigue and fracture in those solids.

Guest Editorial: Learning, optimisation and control of cyber‐physical systems

Guest Editorial: Learning, optimisation and control of cyber‐physical systems

Adaptive Modulation and Coding for Underwater Acoustic Communications Based on Data-Driven Learning Algorithm

With the development of the underwater acoustic (UWA) adaptive communication system, energy-efficient transmission has become a critical topic in underwater acoustic (UWA) communications. Due to the unique characteristics of the underwater environment, the transmitter node will almost always have outdated channel state information (CSI), which results in low energy efficiency. In this paper, we take full advantage of bidirectional links and propose an adaptive modulation and coding (AMC) scheme that aims to maximize the long-term energy efficiency of a single link by jointly scheduling the coding rate, modulation order, and transmission power. Considering the complexity characteristics of UWA channels, we proposed a bit error ratio (BER) estimation method based on deep neural networks (DNN). The proposed network could realize channel estimation, feature extraction, and BER estimation by using a fixed pilot of the feedback link. Then, we design a channel classification method based on the estimated BERs of the modulation and coding scheme (MCS) and further model the UWA channels as a finite-state Markov chain (FSMC) with an unknown transition probability. Thus, we formulate the AMC problem as a Markov Decision Process (MDP) and solve it through a reinforcement learning framework. Considering the large state-action pairs, a double deep Q-network (DDQN) based scheme is proposed. Simulation results demonstrate that the proposed AMC scheme outperforms the fixed MCS with a perfect channel information state, and achieves near-optimal energy efficiency.

Characteristics of Artificial Intelligence Clinical Trials in the Field of Healthcare: A Cross-Sectional Study on ClinicalTrials.gov.

Artificial intelligence (AI) has driven innovative transformation in healthcare service patterns, despite a lack of understanding of its performance in clinical practice. We conducted a cross-sectional analysis of AI-related trials in healthcare based on ClinicalTrials.gov, intending to investigate the trial characteristics and AI’s development status. Additionally, the Neo4j graph database and visualization technology were employed to construct an AI technology application graph, achieving a visual representation and analysis of research hotspots in healthcare AI. A total of 1725 eligible trials that were registered in ClinicalTrials.gov up to 31 March 2022 were included in this study. The number of trial registrations has dramatically grown each year since 2016. However, the AI-related trials had some design drawbacks and problems with poor-quality result reporting. The proportion of trials with prospective and randomized designs was insufficient, and most studies did not report results upon completion. Currently, most healthcare AI application studies are based on data-driven learning algorithms, covering various disease areas and healthcare scenarios. As few studies have publicly reported results on ClinicalTrials.gov, there is not enough evidence to support an assessment of AI’s actual performance. The widespread implementation of AI technology in healthcare still faces many challenges and requires more high-quality prospective clinical validation.

Optimal output regulation for unknown continuous-time linear systems by internal model and adaptive dynamic programming

This paper addresses an optimal output regulation problem for linear time-invariant systems with unknown dynamics. First, a new augmented virtual system is designed to replace the original system and the internal model. Then, by incorporating the data from the augmented virtual system with adaptive dynamic programming (ADP) method, a new iterative learning equation without requiring integral operations or constructing internal model in advance is proposed for establishing the data-driven learning algorithm. Compared with existing ADP-based learning algorithms for linear continuous-time systems, the proposed learning algorithm relaxes both requirements on recording the complete continuous data and setting an initial stabilizing control policy. Finally, the effectiveness of the proposed algorithm is illustrated by an example.

Data-driven decision support for rail traffic control: A predictive approach

Advanced decision support for rail traffic control is significant for enhancing the safety and quality of railway transport service. Data-driven methods have shown powerful learning ability and wide extensibility for prediction, classification, and decision-making problems. In this paper, we propose a hybrid prediction model based on deep forest (DF) ensemble learning to analyze the two common rail traffic control actions, i.e., changing the dwelling times and running times. This basic concept is to mimic the decision-making of rail traffic controllers, providing them with advanced decisions/control actions using data-driven learning algorithms. According to the decision-making approach of rail traffic controllers, the learning process of the model is split into two stages, i.e., learning the type of action (ToA) and the number of changes (NoC) in the dwelling times and running times (i.e., how many dwelling times or running times have been changed compared with the planned ones). The first stage is a classification problem; thus, DF classifier with the synthetic minority oversampling technique (SMOTE) is employed to deal with imbalanced data. In the second stage, the DF regressor treats the NoC in the dwelling and running times as numerical variables and utilizes the information from stage one, i.e., the prediction results of the classification model, to make predictions. The proposed model is calibrated on train operation data from two high-speed railway lines in China. The experimental results and comparative analyses show that the proposed method provides advanced and timely decision support for controllers. These characteristics of the proposed model are imperative for supporting the dynamic decision-making of controllers to manage railway traffic.

A data-driven algorithm to detect false data injections targeting both frequency regulation and market operation in power systems

This paper focuses on detecting cyber-attacks targeting the Automatic Generation Control (AGC) loop and market operation. To achieve this, a new data-driven learning algorithm is proposed that ensembles various learning tools such as K-Means clustering, Synthetic Minority Oversampling Technique oversampling, and Support Vector Data Description models as the base learners. Next, this paper devises a new feature variable, EF, which generates a relatively higher value for attack scenarios than normal grid states, thus aiding the classifier in predictions with low false alarms. The proposed approach can detect a wide range of cyber-attacks, including unseen attack cases. This paper further modelled and validated the detection of profit-oriented intelligent market operation attacks, absent in most of the previous works. Such attacks project themselves within the proximity of nominal grid states, making them difficult to predict. The algorithm's performance is finally compared with other learning models, and it is found that the proposed technique has superior prediction with True Positive Rate, True Negative Rate, and Geometric Accuracy as 98.13%, 99.85%, and 98.98%, respectively.

Data-driven adaptive-critic optimal output regulation towards water level control of boiler-turbine systems

This paper considers a new adaptive-critic optimal output regulation scheme for continuous-time general linear systems with unknown dynamics and unmeasurable disturbance. The objective of researching the optimal output regulation problem is pursuing the stability of the closed-loop system while enabling the output to optimally track the reference signal with disturbance rejection. The problem is solved by employing an approximate optimal regulator that consist of optimal feedback control obtained by a data-driven learning algorithm and optimal feedforward control achieved by solving regulator equation. The missing system matrices of the plant are exactly figured out by employing state/input data. Moreover, by employing the minimal polynomial of the exosystem matrix, the parameter of interference in the output equation is available. Then, a predefined cost function is presented based on the regulator equation that aims to design the optimal control law. The stability analysis demonstrate that the presented control scheme can stabilize the closed-loop system, meanwhile the output asymptotically tracks the reference signal. Finally, the effectiveness of the developed optimal control scheme is validated through an application towards the drum water level control of boiler-turbine systems.

A methodology for thermal simulation of interconnects enabled by model reduction with material property variation

A thermal simulation methodology is developed for interconnects enabled by a data-driven learning algorithm accounting for variations of material properties, heat sources and boundary conditions (BCs). The methodology is based on the concepts of model order reduction and domain decomposition to construct a multi-block approach. A generic block model is built to represent a group of interconnect blocks that are used to wire standard cells in the integrated circuits (ICs). The blocks in this group possess identical geometry with various metal/via routings. The data-driven model reduction method is thus applied to learn material property variations induced by different metal/via routings in the blocks, in addition to the variations of heat sources and BCs. The approach is investigated in two very different settings. It is first applied to thermal simulation of a single interconnect block with similar BCs to those in the training of the generic block. It is then implemented in multi-block thermal simulation of a FinFET IC, where the interconnect structure is partitioned into several blocks each modeled by the generic block model. Accuracy of the generic block model is examined in terms of the metal/via routings, BCs and thermal discontinuities at the block interfaces.

Resilient practical cooperative output regulation for MASs with unknown switching exosystem dynamics under DoS attacks

In this paper, the resilient practical cooperative output regulation problem (CORP) is addressed for heterogeneous linear multi-agent systems with unknown switching exosystem dynamics under denial-of-service (DoS) attacks. Different from most existing cooperative output regulation works, the exosystem dynamics considered in this paper are unknown to all agents and switching in different time intervals. A novel resilient practical cooperative output regulation control scheme is developed. This scheme consists of (i) a data-driven learning algorithm to learn the unknown switching exosystem matrices; (ii) an auxiliary observer to estimate the state of the exosystem; (iii) distributed resilient observers to estimate the state of the auxiliary system; and (iv) distributed controllers for individual agents. By using the dwell-time method, rigorous stability analysis shows that the resilient observation errors are globally ultimately bounded in the presence of DoS attacks and the resilient practical CORP is solved. Finally, a simulation example is given to illustrate the effectiveness of the proposed control scheme.

Data-driven learning of Mori–Zwanzig operators for isotropic turbulence

Developing reduced-order models for turbulent flows, which contain dynamics over a wide range of scales, is an extremely challenging problem. In statistical mechanics, the Mori–Zwanzig (MZ) formalism provides a mathematically exact procedure for constructing reduced-order representations of high-dimensional dynamical systems, where the effects due to the unresolved dynamics are captured in the memory kernel and orthogonal dynamics. Turbulence models based on MZ formalism have been scarce due to the limited knowledge of the MZ operators, which originates from the difficulty in deriving MZ kernels for complex nonlinear dynamical systems. In this work, we apply a recently developed data-driven learning algorithm, which is based on Koopman's description of dynamical systems and Mori's linear projection operator, on a set of fully resolved isotropic turbulence datasets to extract the Mori–Zwanzig operators. With data augmentation using known turbulence symmetries, the extracted Markov term, memory kernel, and orthogonal dynamics are statistically converged and the generalized fluctuation–dissipation relation can be verified. The properties of the memory kernel and orthogonal dynamics, and their dependence on the choices of observables are investigated to address the modeling assumptions that are commonly used in MZ-based models. A series of numerical experiments are then constructed using the extracted kernels to evaluate the memory effects on prediction. The results show that the prediction errors are strongly affected by the choice of observables and can be further reduced by including the past history of the observables in the memory kernel.

Quantum element method for quantum eigenvalue problems derived from projection-based model order reduction

An effective multi-element simulation methodology for quantum eigenvalue problems is investigated. The approach is derived from a reduced-order model based on a data-driven learning algorithm, together with the concept of domain decomposition. The approach partitions the simulation domain of a quantum eigenvalue problem into smaller subdomains that, referred to as elements, could be the building blocks for quantum structures of interest. In this quantum element method (QEM), each element is projected onto a functional space represented by a set of basis functions (or modes) that are generated from proper orthogonal decomposition (POD). To construct a POD model for a large domain, these projected elements can be combined together, and the interior penalty discontinuous Galerkin method is applied to achieve the interface continuity and stabilize the numerical solution. The POD is able to optimize the basis functions specifically tailored to the geometry and parametric variations of the problem and can therefore substantially reduce the degree of freedom (DoF) needed to solve the Schrödinger equation. To understand the fundamental issues of the QEM, demonstrations in this study focus on examining the accuracy and DoF of the QEM influenced by the training settings for generation of POD modes, selection of the penalty number, suppression of interface discontinuities, structure size and complexity, etc. It has been shown that the QEM is able to achieve a substantial reduction in the DoF with a high accuracy even beyond the training conditions for the POD modes if the penalty number is selected within an appropriate range.

Data-driven algorithm for real-time fatigue life prediction of structures with stochastic parameters

Fatigue crack growth analysis using extended finite element method (XFEM) is an efficient way to predict the residual life of structures; however, when the structure parameters vary stochastically, it will be very hard to make accurate predictions. To bridge this research gap, this work proposed a data-driven learning algorithm to improve the prediction capacity of fatigue life by considering stochastic parameters of structures. In this new algorithm, the XFEM was firstly employed to generate a large amount of datasets that pair the structural responses with remaining fatigue life. Then, the back propagation neural network (BPNN) was employed to construct a fatigue life prediction model based on the XFEM datasets. Real-time prediction for the structural fatigue life was achieved using the constructed BPNN model without knowing the exact distribution functions of stochastic parameters. Several numerical examples were performed to evaluate the performance of the proposed algorithm. The analysis results demonstrate that the proposed data-driven algorithm can accurately predict the fatigue life of the structures with stochastic parameters.