Data-driven Perspective Research Articles

Reconstructing 3D ocean subsurface salinity (OSS) from T–S mapping via a data-driven deep learning model

Despite the increasing volume of oceanographic data, the available ocean salinity data remains patently insufficient which limits studying ocean dynamics, climate change and the calculation of salinity-related ocean elements. Considering that traditional salinity reconstruction methods often suffer from various factors such as additional constraints, priori physical assumptions and large of specific regression coefficients, a generative adversarial networks (GANs) based deep learning (DL) framework is proposed to directly construct a near real-time, high-resolution daily three-dimensional ocean subsurface salinity (3D OSS) dataset from a data-driven perspective in this study. Four models with different structural combinations are designed in the China’s marginal seas. Experimental results demonstrate that the models can successfully reconstruct the high-precision and high-resolution 3D OSS on a daily scale in 12 depth levels (from 2 m to 200 m). The asymmetric inception-3DGAN model with all enhanced structures has the highest accuracy, the average root-mean-squared error (RMSE) is 0.135psu, the average coefficient of determination (R2) is 0.5641 and the percent bias is 0.436%. Comparing with model without enhanced structures, the average RMSE is decreased by 22.41% when adding all the enhancement structures. Besides, temporal and spatial error analysis are also conducted to evaluate the models’ performance from different aspects. Finally, the model results are used to analyze common ocean elements, such as water mass properties, dynamic fields and geostrophic velocity fields, demonstrating that the 3D OSS dataset construction approach proposed in this study not only provides some new insights into ocean observations, but also has high application values.

Chained Gaussian processes to estimate battery health degradation with uncertainties

Estimating the average health degradation of a new battery design is a crucial objective for manufacturers to estimate its value. Furthermore, to quantify financial risks, associated uncertainties should be modeled precisely. From a data-driven perspective, Gaussian process regression (GPR) is often a method of choice since it simultaneously learns complex models and naturally includes uncertainties. However, GPR methods generally rely on a stationarity assumption which imposes severe constraints on uncertainties. In this paper, we illustrate the limits of standard GPR and show that the Chained Gaussian processes, a more general framework introduced in the Machine learning community, is a useful alternative, allowing more accurate quantification of uncertainties.

Adaptive Graph-Learning Convolutional Network for Multi-Node Offshore Wind Speed Forecasting

Multi-node wind speed forecasting is greatly important for offshore wind power. It is a challenging task due to unknown complex spatial dependencies. Recently, graph neural networks (GNN) have been applied to wind forecasting because of their capability in modeling dependencies. However, existing methods usually require a pre-defined graph structure, which is not optimal for the downstream task and limits the application scope of GNN. In this paper, we propose adaptive graph-learning convolutional networks (AGLCN) that can automatically infer hidden associations among multi-nodes through a graph-learning module. It simultaneously integrates the temporal and graph convolutional modules to capture temporal and spatial features in the data. Experiments are conducted on real-world multi-node wind speed data from the China Sea. The results show that our model achieves state-of-the-art results in all multi-scale wind speed predictions. Moreover, the learned graph can reveal spatial correlations from a data-driven perspective.

Tourists’ travel mode and length of stay: Application of a fully nested Archimedean copula structure

Tourism decisions are interdependent of one another. This study aims to explore the interrelations among three pertinent tourism choices, i.e., travel season, transport mode and length of stay, from a data-driven perspective. To achieve a joint model of three travel behavioural decisions, this paper adopts a Frank-type fully nested Archimedean copula function, which is capable of efficiently incorporating a logistic regression model, a multinomial logit model and a lognormal accelerated failure time model. Then, we empirically evaluate the proposed model framework using single-destination Sydney-based trip-level data collected in 2017–2018. Gumbel, Clayton and Frank copula results are compared. To illustrate its application, the modelling results are used to build a simulated COVID-19 pandemic scenario according to the social distancing restrictions within New South Wales, Australia, and a three-dimensional elasticity analysis for trip destinations is performed. The findings suggest the model provides nuanced insights into simulating tourist behaviours and transport policy appraisal aimed at tourism recovery or/and development.

Data-Driven Analysis of Risk-Assessment Methods for Cold Food Chains.

The problem of cold-chain food safety is becoming increasingly prominent. Cold food chain risk assessment is an important way to ensure cold-chain food safety. Using CiteSpace, this study analyzes the knowledge map of research hotspots in the field of cold-chain food safety over the past 18 years, identifies the research keywords, presents the centrality statistics, and calculates the cluster values and average cluster contour values. Adopting a data-driven perspective, risk-assessment methods for cold food chains are summarized based on qualitative risk assessment, quantitative risk assessment, and comprehensive qualitative and quantitative risk assessment. The advantages and disadvantages of each are summarized. Finally, the problems and challenges in current cold food chain risk-assessment research are summarized in three aspects: the data credibility of cold food chain traceability systems, cold-chain food safety audit methods, and nontraditional cold food chain risk assessment. Suggestions are given for strengthening the cold food chain risk-assessment system to provide a decision-making reference to help regulatory authorities take risk prevention and control measures.

Recommending on graphs: a comprehensive review from a data perspective

Recent advances in graph-based learning approaches have demonstrated their effectiveness in modelling users’ preferences and items’ characteristics for Recommender Systems (RSs). Most of the data in RSs can be organized into graphs where various objects (e.g. users, items, and attributes) are explicitly or implicitly connected and influence each other via various relations. Such a graph-based organization brings benefits to exploiting potential properties in graph learning (e.g. random walk and network embedding) techniques to enrich the representations of the user and item nodes, which is an essential factor for successful recommendations. In this paper, we provide a comprehensive survey of Graph Learning-based Recommender Systems (GLRSs). Specifically, we start from a data-driven perspective to systematically categorize various graphs in GLRSs and analyse their characteristics. Then, we discuss the state-of-the-art frameworks with a focus on the graph learning module and how they address practical recommendation challenges such as scalability, fairness, diversity, explainability, and so on. Finally, we share some potential research directions in this rapidly growing area.

Detecting Changes in Correlation Networks with Application to Functional Connectivity of fMRI Data.

Research questions in the human sciences often seek to answer if and when a process changes across time. In functional MRI studies, for instance, researchers may seek to assess the onset of a shift in brain state. For daily diary studies, the researcher may seek to identify when a person's psychological process shifts following treatment. The timing and presence of such a change may be meaningful in terms of understanding state changes. Currently, dynamic processes are typically quantified as static networks where edges indicate temporal relations among nodes, which may be variables reflecting emotions, behaviors, or brain activity. Here we describe three methods for detecting changes in such correlation networks from a data-driven perspective. Networks here are quantified using the lag-0 pair-wise correlation (or covariance) estimates as the representation of the dynamic relations among variables. We present three methods for change point detection: dynamic connectivity regression, max-type method, and a PCA-based method. The change point detection methods each include different ways to test if two given correlation network patterns from different segments in time are significantly different. These tests can also be used outside of the change point detection approaches to test any two given blocks of data. We compare the three methods for change point detection as well as the complementary significance testing approaches on simulated and empirical functional connectivity fMRI data examples.

Stock control analytics: a data-driven approach to compute the fill rate considering undershoots

One of the most frequently used inventory policies is the order-point, order-up-to-level (s, S) system. In this system, the inventory is continuously reviewed and a replenishment request is placed whenever the inventory position drops to or below the order point, s. The variable replenishment order quantity and the variable replenishment cycle characterize the system by the use of complex mathematical computations. Different methodological approaches diminish the mathematical complexity by neglecting the undershoots, i.e., the quantity that the inventory position is below the order point when it is reached. In this paper, we conceptually and empirically analyse the bias that neglecting the undershoots introduces into the estimation of the fill rate. After that, we suggest a new methodology developed under a data-driven perspective that uses a state-dependent parameter algorithm to correct such a bias. As a result, we propose two new methods, one parametric and the other nonparametric, to enhance the fill rate estimate. Both methods, named analytics fill rate methods, remove the bias that neglecting the undershoots introduces and are used to illustrate the practical implications of this hypothesis on the performance and design of the (s, S) system. This research is developed in a lost sales context with simulated stochastic and i.i.d. discrete demands as well as actual sales data.

Multiscale Time Series Decomposition for Structural Dynamic Properties: Long-Term Trend and Ambient Interference

In structural health monitoring (SHM), variations of structural dynamic properties are paramount to indicate the health status of structures. Structural dynamic properties are time variant due to various long-term effects (e.g., structural deterioration) and periodic effects (e.g., periodic variations of temperature, humidity, and traffic). Sometimes periodic effects will interfere with the quantification of long-term effects. Though important, there are still limited research studies aiming to distinguish these two effects. Given the amount of SHM data, it is possible to solve the issue from a data-driven perspective. This article proposes a time series decomposition methodology to divide the time series of structural dynamic properties into long-term parts, multiscale periodic parts, holiday parts, and error parts. We extract the 10-year dynamic properties of a long-span bridge using the fast Bayesian FFT identification algorithm and choose two dynamic modes of that bridge as examples to explain how our proposed methodology works. We use the long-term parts to extract the rules of structural deterioration. The multiscale periodic parts are utilized to find the relationships with the periodically varying ambient conditions (i.e., temperature and humidity) in different time scales (i.e., yearly, weekly, and daily). Then, the long-term and periodic effects can be distinguished. For the long-term effects, the modal frequencies tend to decrease but the damping ratios seem to increase. For the periodic effects, we find that the increment of temperature will lead to the decrease of both modal frequencies and damping ratios. The variations of modal frequencies induced by deterioration and temperature are of the same amplitudes. The variations of both modal frequencies and damping ratios are not significantly related with humidity. This article could provide references for damage detection and safety assessment for similar bridges.

On the contagion leakage via incoming flights during China’s aviation policies in the fight against COVID-19

For nearly three years with the COVID-19 pandemic, China has implemented a set of strict policies to control the flux of potential virus carriers in cross-border flights: The so-called Circuit Breaker mechanism. In this study, we review the evolution of this mechanism – a rather unique experiment in the global aviation system — from a data-driven perspective. Specifically, we perform an investigation on the extent of violations and their potential drivers. In total, 183 events are analyzed covering the period from epidemic outbreak in early 2020 to December 2021. In addition to describing the spatial extent and temporal evolution, we develop a regression model which helps us to better understand the universal patterns. By dissecting an under-investigated phenomenon, we believe that our study contributes to the rich literature on aviation and COVID-19, not only in the specific context of China, but also by assessing some of the challenges and potential of containing a global health threat using strict aviation policies.

High-throughput screening of hypothetical metal-organic frameworks for thermal conductivity

Thermal energy management in metal-organic frameworks (MOFs) is an important, yet often neglected, challenge for many adsorption-based applications such as gas storage and separations. Despite its importance, there is insufficient understanding of the structure-property relationships governing thermal transport in MOFs. To provide a data-driven perspective into these relationships, here we perform large-scale computational screening of thermal conductivity k in MOFs, leveraging classical molecular dynamics simulations and 10,194 hypothetical MOFs created using the ToBaCCo 3.0 code. We found that high thermal conductivity in MOFs is favored by high densities (> 1.0 g cm−3), small pores (< 10 Å), and four-connected metal nodes. We also found that 36 MOFs exhibit ultra-low thermal conductivity (< 0.02 W m−1 K−1), which is primarily due to having extremely large pores (~65 Å). Furthermore, we discovered six hypothetical MOFs with very high thermal conductivity (> 10 W m−1 K−1), the structures of which we describe in additional detail.

Galyean appreciation club review: a holistic perspective of the societal relevance of beef production and its impacts on climate change.

This article provides a science-based, data-driven perspective on the relevance of the beef herd in the U.S. to our society and greenhouse gas (GHG) contribution to climate change. Cattle operations are subject to criticism for their environmental burden, often based on incomplete information disseminated about their social, economic, nutritional, and ecological benefits and detriments. The 2019 data published by the U.S. Environmental Protection Agency reported that U.S. beef cattle emitted 22.6% of the total agricultural emissions, representing about 2.2% of the total anthropogenic emissions of CO2 equivalent (CO2e). Simulations from a computer model developed to address global energy and climate challenges, set to use extreme improvements in livestock and crop production systems, indicated a potential reduction in global CO2e emissions of 4.6% but without significant enhancement in the temperature change by 2030. There are many natural and anthropogenic sources of CH4 emissions. Contrary to the increased contribution of peatlands and water reservoirs to atmospheric CO2e, the steady decrease in the U.S. cattle population is estimated to have reduced its methane (CH4) emissions by about 30% from 1975 to 2021. This CH4 emission deacceleration of 2.46 Mt CO2e/yr2 might be even more significant than reported. Many opportunities exist to mitigate CH4 emissions of beef production, leading to a realistic prospect of a 5% to 15% reduction in the short term after considering the overlapping impacts of combined strategies. Reduction strategies include feeding synthetic chemicals that inactivate the methyl-coenzyme M reductase (the enzyme that catalyzes the last step of methanogenesis in the rumen), red seaweed or algae extracts, ionophore antibiotics, phytochemicals (e.g., condensed tannins and essential oils), and other nutritional manipulations. The proposed net-zero concept might not solve the global warming problem because it will only balance future anthropogenic GHG emissions with anthropogenic removals, leaving global warming on a standby state. Recommendations for consuming red meat products should consider human nutrition, health, and disease and remain independent of controversial evidence of causational relationships with perceived negative environmental impacts of beef production that are not based on scientific data.

Curriculum analytics of an Open Distance Learning (ODL) Programme: A data-driven perspective

Curriculum analytics of an Open Distance Learning (ODL) Programme: A data-driven perspective

A Research on the Adaptive Dynamic Scheduling Based on Scenario Deduction

Aiming at the uncertainty of the intelligent manufacturing system, this paper reveals the evolution mechanism of the manufacturing system and its internal and external environment from a data-driven perspective. And it proposes an adaptive dynamic scheduling method based on scenario deduction, which provides the basis for the intelligent manufacturing support in an uncertain environment. Firstly, the paper uses the ontology to describe the scenario status, production activities, market environment and production subject of manufacturing system. And then builds the scenario deduction model based on the dynamic fuzzy cognitive map, establishing the data-driven manufacturing scenario network structure. Through the dynamic fuzzy cognitive map, it presents the market and production scenario evolution and accordingly guides the adaptive choice of dynamic scheduling strategy in uncertain production environment. The results show that the scenario deduction model is basically consistent with the manufacturing system evolution process in terms of time inference and demand prediction, and it verifies the adaptability and effectiveness of the proposed dynamic dispatching method through examples.

Investigation of the structure-odor relationship using a Transformer model

The relationships between molecular structures and their properties are subtle and complex, and the properties of odor are no exception. Molecules with similar structures, such as a molecule and its optical isomer, may have completely different odors, whereas molecules with completely distinct structures may have similar odors. Many works have attempted to explain the molecular structure-odor relationship from chemical and data-driven perspectives. The Transformer model is widely used in natural language processing and computer vision, and the attention mechanism included in the Transformer model can identify relationships between inputs and outputs. In this paper, we describe the construction of a Transformer model for predicting molecular properties and interpreting the prediction results. The SMILES data of 100,000 molecules are collected and used to predict the existence of molecular substructures, and our proposed model achieves an F1 value of 0.98. The attention matrix is visualized to investigate the substructure annotation performance of the attention mechanism, and we find that certain atoms in the target substructures are accurately annotated. Finally, we collect 4462 molecules and their odor descriptors and use the proposed model to infer 98 odor descriptors, obtaining an average F1 value of 0.33. For the 19 odor descriptors that achieved F1 values greater than 0.45, we also attempt to summarize the relationship between the molecular substructures and odor quality through the attention matrix.

Controller Performance Monitoring: A Survey of Problems and a Review of Approaches from a Data-Driven Perspective with a Focus on Oscillations Detection and Diagnosis

Controller Performance Monitoring: A Survey of Problems and a Review of Approaches from a Data-Driven Perspective with a Focus on Oscillations Detection and Diagnosis

A data-driven perspective for sensing urban functional images: Place-based evidence in Hong Kong

Urban life involves a large variety of urban functions and human activities in a dense context due to the inherent nature of cities. Although technical frameworks have been previously proposed to understand urban functions and activities, there are limited studies that concern the individual places within a city and their detailed characteristics at a local scale. Using points of interest (POIs), we present a data-driven analytical framework to explore urban space containing urban functions and relevant activities by focusing on particular urban places. Urban functions are first extracted and induced by leveraging a latent Dirichlet allocation (LDA) topic modeling technique. We next evaluate the thematic functional differences among the selected places using the location quotient (LQ). Furthermore, tourist functions are assumed to occur in places within a city and carry broadly identifiable information; thus, tourist places are studied by comparing their perceptual experiences using the high-frequency keywords retrieved from tourist reviews on TripAdvisor. By adopting Hong Kong as our case study, the findings reveal considerable diversity of urban functions across different places, while each place displays the distinctive trait of urban functions. Tourist impressions reflected online are primarily consistent with the corresponding functional identities of these places but exhibit additional details associated with emotional and temporal aspects. This study uses a bottom-up assessment of local functions, and discusses their practical implications as related to city branding strategies.

Data-Driven Method for Detecting Flight Trajectory Deviation Anomaly

With the rapid growth of air traffic, it often happens that aircraft deviate from the original flight plan during actual flight. This paper proposes an anomaly detection method for aircraft trajectory deviation to realize single-point and successive multipoint anomaly detection from a data-driven perspective. Given the one-to-many relationship between reporting points of planned and real trajectories, a matching algorithm is used to match these points. Four trajectory deviation features (which are the position deviation, distance deviation, altitude deviation, and flight stage) are defined. On this basis, a one-class support vector machine is trained to detect single-point anomalies using the deviation features as input. Furthermore, successive multipoint anomaly detection of the aircraft is realized by considering the deviation of successive segments of the trajectory. Taking the flights taking off and landing at four Chinese hub airports as examples, the proposed method obtained an F score, which is a balance of the precision and recall, over 0.92, indicating it achieves high accuracy for anomaly detection.

Active Defense Research against False Data Injection Attacks of Power CPS Based on Data-Driven Algorithms

The terminal equipment interconnection and the network communication environment are complex in power cyber–physical systems (CPS), and the frequent interaction between the information and energy flows aggravates the risk of false data injection attacks (FDIAs) in the power grid. This paper proposes an active defense framework against FDIAs of power CPS based on data-driven algorithms in order to ensure that FDIAs can be efficiently detected and processed in real-time during power grid operation. First, the data transmission scenario and false data injection forms of power CPS were analyzed, and the FDIA mathematical model was expounded. Then, from a data-driven perspective, the algorithm improvement and process design were carried out for the three key links of data enhancement, attack detection, and data reconstruction. Finally, an active defense framework against FDIAs was proposed. The example analysis verified that the method proposed in this paper could effectively detect FDIAs and perform data reconstruction, providing a new idea for the active defense against FDIAs of power CPS.

Deep learning-based solvability of underdetermined inverse problems in nonlinear ultrasonic characterization of micro damages

Detection and evaluation of micro-damages in the early stages of engineering failure are crucial for various industrial structures to ensure their safety and prevent further catastrophic accidents. The nonlinear ultrasonic technique (NUT) has gained increasing popularity and recognition for breaking through the detection sensitivity limit upon micro-damages that usually are invisible to conventional linear techniques. However, it remains an ongoing challenge to quantitatively characterize micro-damages using NUT due to great difficulties in fully modeling the complicated interaction mechanism between the nonlinear ultrasonic waves and micro-damages. This work presents a data-driven perspective for solving multiparameter underdetermined inverse problems that are at the core of NUT, while allowing by-passing the creation of high-fidelity physics-based models. Nonlinear Lamb wave measurements with group-velocity mismatching are conducted to introduce both size and localization information of damages to the assembled dataset. A nonlinearity-aware discrete wavelet transform-bidirectional long short-term memory network is proposed to directly process nonlinear ultrasonic responses to automatically model latent nonlinear dynamics, thus establishing the complex mapping between the nonlinear ultrasonic signals and the multi-dimensional damage features. In particular, an attempt is made to augment the physical explainability of the proposed deep learning approach through a frequency component importance analysis. The trained network enables accurate and explainable predictions of length and localization of closed cracks and robustness against varying degrees of noise. Our work paves a promising and practical way to promote the transformation of NUT from the qualitative analysis for accurate and efficient quantitative prediction.