Data-driven Statistical Models Research Articles

Enhanced Estimation of Traffic Noise Levels Using Minute-Level Traffic Flow Data through Convolutional Neural Network

The advent of high-resolution minute-level traffic flow data from video surveillance on roads has opened up new opportunities for enhancing the estimation of traffic noise levels. In this study, we propose an innovative method that utilizes time series traffic flow data (TSTFD) to estimate traffic noise levels using a deep learning Convolutional Neural Network (CNN). Unlike traditional traffic flow data, TSTFD offer a unique structure and composition suitable for multidimensional data analysis. Our method was evaluated in a pilot study conducted in Foshan City, China, utilizing traffic flow information obtained from roadside video surveillance systems. Our results indicated that the CNN-based model surpassed traditional data-driven statistical models in estimating traffic noise levels, achieving a reduction in mean squared error (MSE) by 10.16%, mean absolute error (MAE) by 4.48%, and an improvement in the coefficient of determination (R²) by 1.73%. The model demonstrated robust generalization capabilities throughout the test period, exhibiting mean errors ranging from 0.790 to 1.007 dBA. However, the model’s applicability is constrained by the acoustic propagation environment, demonstrating effectiveness on roads with similar surroundings while showing limited applicability to those with different surroundings. Overall, this method is cost-effective and offers enhanced accuracy for the estimation of traffic noise level.

A multidimensional AI-trained correction to the 1D approximate model for Airborne TDEM sensing

The computational resources required to solve the full 3D inversion of time-domain electromagnetic data are immense. To overcome the time-consuming 3D simulations, we construct a surrogate model, more precisely, a data-driven statistical model to replace the 3D simulations. It is trained on 3D data and predicts the approximate output much faster. We construct a surrogate model that predicts the discrepancy between a 1D subsurface model and a deviation of the 1D assumption. The latter response is fastly computable with a semi-analytical 1D forward model. We exemplify the approach on a two-layered case. The results are encouraging even with few training samples. Given the computational cost related to the 3D simulations, there are limitations in the number of training samples that can be generated. In addition, certain applications require a wide range of parameters to be sampled, such as the electrical conductivity parameters in a saltwater intrusion case. The challenge of this work is achieving the best possible accuracy with only a few thousand samples. We propose to view the performance in terms of learning gain, representing the gain from the surrogate model whilst still acknowledging a residual discrepancy. Our works open new avenues for effectively simulating 3D TDEM data.

Hydrogen station prognostics and health monitoring model

Hydrogen fuel has shown promise as a clean, alternative fuel aiding in the reduction of fossil fuel dependence within the transportation sector. However, hydrogen refueling stations and infrastructure remains a barrier and are a prerequisite for consumer adoption of low-cost and low-emission fuel cell electric vehicles (FCEVs). The costs for FCEV fueling include both station capital costs and operation and maintenance (O&M) costs. Contributing to these O&M costs, unscheduled maintenance is presently more costly and more frequent than for similar gasoline fueling infrastructure and is asserted to be a limiting factor in achieving FCEV customer acceptance and cost parity. Unscheduled maintenance leads to longer station downtime, therefore, causing an increase in missed fueling opportunities, which forces customers to seek refueling at other operable stations that may be significantly farther away. This research proposes a framework for a hydrogen station prognostics health monitoring (H2S PHM) model that can minimize unexpected downtime by predicting the remaining useful life for primary hydrogen station components within the major station subsystems. The H2S PHM model is a data-driven statistical model, based on O&M data collected from 34 retail hydrogen stations located in the U.S. The primary subcomponents studied are the dispenser, compressor, and chiller. The remaining useful life calculations are used to decide whether or not maintenance should be completed based on the prediction and expected future station use. This paper presents the background, method, and results for the H2S PHM model as for a means for improving station availability and customer confidence in FCEVs and hydrogen infrastructure.

Deep ensemble inverse model for image-based estimation of solar cell parameters

Physical models can help improve solar cell efficiency during the design phase and for quality control after the fabrication process. We present a data-driven approach to inverse modeling that can predict the underlying parameters of a finite element method solar cell model based on an electroluminescence (EL) image of a solar cell with known cell geometry and laser scribed defects. For training the inverse model, 75 000 synthetic EL images were generated with randomized parameters of the physical cell model. We combine 17 deep convolutional neural networks based on a modified VGG19 architecture into a deep ensemble to add uncertainty estimates. Using the silicon solar cell model, we show that such a novel approach to data-driven statistical inverse modeling can help apply recent developments in deep learning to new engineering applications that require real-time parameterizations of physical models augmented by confidence intervals. The trained network was tested on four different physical solar cell samples, and the estimated parameters were used to create the corresponding model representations. Resimulations of the measurements yielded relative deviations of the calculated and the measured junction voltage values of 0.2% on average with a maximum of 10%, demonstrating the validity of the approach.

Data-driven statistical reduced-order modeling and quantification of polycrystal mechanics leading to porosity-based ductile damage

Predicting the process of porosity-based ductile damage in polycrystalline metallic materials is an essential practical topic. Ductile damage and its precursors are represented by extreme values in stress and material state quantities, the spatial probability density function (PDF) of which are highly non-Gaussian with strong fat tails. Traditional deterministic forecasts utilizing sophisticated continuum-based physical models generally lack in representing the statistics of structural evolution during material deformation. Computational tools which do represent complex structural evolution are typically expensive. The inevitable model error and the lack of uncertainty quantification may also induce significant forecast biases, especially in predicting the extreme events associated with ductile damage. In this paper, a data-driven statistical reduced-order modeling framework is developed to provide a probabilistic forecast of the deformation process of a polycrystal aggregate leading to porosity-based ductile damage with uncertainty quantification. The framework starts with computing the time evolution of the leading few moments of specific state variables from the spatiotemporal solution of full-field polycrystal simulations. Then a sparse model identification algorithm based on causation entropy, including essential physical constraints, is utilized to discover the governing equations of these moments. An approximate solution of the time evolution of the PDF is obtained from the predicted moments exploiting the maximum entropy principle. Numerical experiments based on polycrystal realizations of a representative body-centered cubic (BCC) tantalum illustrate a skillful reduced-order model in characterizing the time evolution of the non-Gaussian PDF of the von Mises stress and quantifying the probability of extreme events. The learning process also reveals that the mean stress is not simply an additive forcing to drive the higher-order moments and extreme events. Instead, it interacts with the latter in a strongly nonlinear and multiplicative fashion. In addition, the calibrated moment equations provide a reasonably accurate forecast when applied to the realizations outside the training data set, indicating the robustness of the model and the skill for extrapolation. Finally, an information-based measurement is employed to quantitatively justify that the leading four moments are sufficient to characterize the crucial highly non-Gaussian features throughout the entire deformation history considered.

A coupled experimental and statistical approach for an assessment of SARS-CoV-2 infection risk at indoor event locations

Indoor event locations are particularly affected by the SARS-CoV-2 pandemic. At large venues, only incomplete risk assessments exist, whereby no suitable measures can be derived. In this study, a physical and data-driven statistical model for a comprehensive infection risk assessment has been developed. At venues displacement ventilation concepts are often implemented. Here simplified theoretical assumptions fail for the prediction of relevant airflows for airborne transmission processes. Thus, with locally resolving trace gas measurements infection risks are computed more detailed. Coupled with epidemiological data such as incidences, vaccination rates, test sensitivities, and audience characteristics such as masks and age distribution, predictions of new infections (mean), situational R-values (mean), and individual risks on- and off-seat can be achieved for the first time. Using the Stuttgart State Opera as an example, the functioning of the model and its plausibility are tested and a sensitivity analysis is performed with regard to masks and tests. Besides a reference scenario on 2022-11-29, a maximum safety scenario with an obligation of FFP2 masks and rapid antigen tests as well as a minimum safety scenario without masks and tests are investigated. For these scenarios the new infections (mean) are 10.6, 0.25 and 13.0, respectively. The situational R-values (mean) – number of new infections caused by a single infectious person in a certain situation – are 2.75, 0.32 and 3.39, respectively. Besides these results a clustered consideration divided by age, masks and whether infections occur on-seat or off-seat are presented. In conclusion this provides an instrument that can enable policymakers and operators to take appropriate measures to control pandemics despite ongoing mass gathering events.

Physics-Guided Deep Learning for Improved Production Forecasting in Unconventional Reservoirs

Summary The complexity of physics-based modeling of fluid flow in hydraulically fractured unconventional reservoirs, together with the abundant data from repeated factory-style drilling and completion of these resources, has prompted the development and application of data-driven statistical models for predicting hydrocarbon production performance. More recently, machine learning algorithms have been widely studied in developing data-driven prediction models for unconventional reservoirs. These models often require a large amount of high-quality training data with sufficient range to avoid excessive extrapolation and produce reliable predictions. Unlike statistical models, physics-based models represent causal relations between input and output variables to provide predictions beyond available data. While a detailed physics-based description of fluid flow in unconventional reservoirs is not yet available, approximate physical flow functions have been proposed to capture the general production behavior of unconventional wells. These physical functions can be augmented with the available data to improve data-driven methods by constraining the models to adhere to the general production trends. In this paper, we develop a physics-constrained data-driven model by embedding physical flow functions into neural network models. Since the performance of the physics-constrained model depends on the relevance of the embedded physics, a sizeable residual prediction error is expected if the collected data do not sufficiently match the embedded model. The residual model typically represents errors in the description of inputs or any missing physical phenomenon. We compensate for such errors through residual learning, where an auxiliary neural network is designed to learn the complex relationship between the input parameters (such as formation and completion properties) and the expected prediction residuals. The new physics-guided deep learning (PGDL) model augments any physics-constrained prediction model with residual learning to increase its prediction accuracy. Several synthetic and actual field data sets, from the Bakken play, are used to demonstrate the performance of the PGDL model.

AI’s 10 to Watch, 2022

AI’s 10 to Watch, 2022

Assessing the Influence of Anthropogenic Causal Factors on Landslide Susceptibility in Bukit Antarabangsa, Selangor

This study sought to assess the influence of causal factors related to anthropogenic activities on landslide occurrence in Bukit Antarabangsa, a township northeast of Kuala Lumpur in Ampang Jaya Municipal Council. Anthropogenic factors were chosen based on the township’s rapid growth, numerous landslide records and intensity of hillside development. The study used a data-driven statistical model to identify factors most predictive of landslide occurrence based on an inventory of 20 landslides, and to evaluate the extent to which susceptibility was driven by factors related to urban development. A total of 17 factors were categorized into four clusters (geological, geomorphological, hydro-tographical and anthropogenic). Factor maps were classified to derive factor classes for each parameter. The dataset was then processed using a weight-of-evidence statistical model to determine the contrast value of each factor class. Contrast value (C) reflects the extent to which each factor class predicts landslide occurrence. The C-weighted factor maps were then combined to derive the landslide susceptibility index (LSI). The LSI enabled visualization of the spatial distribution of susceptibility based on a given combination of factors. Susceptibility maps were prepared for combinations containing only non-anthropogenic parameters and all landslide parameters. The study compared these combinations to determine the influence of anthropogenic factors on total LSI. Similar analyses were conducted to determine the effect of each anthropogenic factor on LSI. The results indicated that lineament density, distance to lineament and distance to road had a significant influence on landslide occurrence. A strong correlation with landslide occurrence was observed for land use/land cover, especially in high susceptibility zones, followed closely by the influence of one distance to road factor class. The results could be useful in planning infrastructure corridors in densely built-up landslide-prone areas.

Formal Safety Envelopes for Provably Accurate State Classification by Data-Driven Flight Models

Aerospace systems are inherently stochastic and increasingly data-driven, thus hard to formally verify. Data-driven statistical models can be used to estimate the state and classify potentially anomalous conditions of aerospace systems from multiple heterogeneous sensors with high accuracy. In this paper, we consider the problem of precisely bounding the regions in the sensor input space of a stochastic system in which safe state classification can be formally proven. As an archetypal application, we consider a statistical model created to detect aerodynamic stall in a prototype wing retrofitted with piezoelectric sensors and used to generate data in a wind tunnel for different flight states. We formally define safety envelopes as regions parameterized by and , to respectively capture how model-predictable observed sensor values are, and given these values, how likely the model’s accurate state classification is. Safety envelopes are formalized in the Agda proof assistant, used to also generate formally verified runtime monitors for sensor data stream analyses in the Haskell programming language. We further propose a new metric for model classification quality, evaluate it on our wing prototype model, and compare it to the model restricted to two different fixed airspeeds, and enhanced to a continuous Gaussian process regression model. Safety envelopes are an important step in formally verifying precise probabilistic properties of data-driven models used in stochastic aerospace systems and could be used by advanced control algorithms to maintain these systems well within safe operation boundaries.

Integrated approach of remote sensing and machine learning to simulate and predict petroleum pollution and algal blooms along Aqaba Gulf

To simulate and predict the appropriate indices for algal blooms and petroleum pollution, this study combined remote sensing data and models of Machine Learning ML for the Aqaba Gulf’s condition. For algal blooms indication; floating algal index (FAI) was selected as the best index with 0.937 and for petroleum indices; ratio index (RI) was selected with 0.984. The collected data within the number of samples were separated into two parts: 80% for calibration to train and adjust the back propagation in neural network BPNN and partial least squares regression PLSR, and 20% for the external validation. Therefore, and based on the RI data, FAI was predicted using MLP, the obtained results showed that the ML algorithms gave models with high quality performance with R2= 0.955 and RMSE = 10.90. The PLSR and multilayer perceptron (MLP) were used to predict petroleum pollution using the extracted values to bands. The results showed that both models were obtained excellent models for predicting petroleum pollution. In general, MLP outperforms PLSR; within R²=0.941. Accordingly, the ML model was able to estimate the algal blooms and petroleum contamination with good accuracy. In the validation process, the determination coefficient R2 was 0.84 with an average square error equal to 0.076. As demonstrated, MLP could be a powerful mathematical tool for environmental analysis and prediction. The integration of remote sensing indices and data-driven statistical modeling were highly recommended for further similar studies.

Model-based design for self-sustainable sensor nodes

Long-term and maintenance-free operation is a critical feature for large-scale deployed battery-operated sensor nodes. Energy harvesting (EH) is the most promising technology to overcome the energy bottleneck of today’s sensors and to enable the vision of perpetual operation. However, relying on fluctuating environmental energy requires an application-specific analysis of the energy statistics combined with an in-depth characterization of circuits and algorithms, making design and verification complex. This article presents a model-based design (MBD) approach for EH-enabled devices accounting for the dynamic behavior of components in the power generation, conversion, storage, and discharge paths. The extension of existing compact models combined with data-driven statistical modeling of harvesting circuits allows accurate offline analysis, verification, and validation. The presented approach facilitates application-specific optimization during the development phase and reliable long-term evaluation combined with environmental datasets. Experimental results demonstrate the accuracy and flexibility of this approach: the model verification of a solar-powered wireless sensor node shows a determination coefficient (R2) of 0.992, resulting in an energy error of only -1.57% between measurement and simulation. Compared to state-of-practice methods, the MBD approach attains a reduction of the estimated state-of-charge error of up to 10.2% in a real-world scenario. MBD offers non-trivial insights on critical design choices: the analysis of the storage element selection reveals a 2–3 times too high self-discharge per capacity ratio for supercapacitors and a peak current constrain for lithium-ion polymer batteries.

Reconstructing GRACE-like time series of high mountain glacier mass anomalies

High mountain glaciers (HMGs), called the water towers of the world, are vulnerable to the effects of climate change and thus are rapidly shrinking. Monitoring and evaluating large-scale glacier and snow (GS) mass changes are critical for humans and ecosystems. Although modern gravity satellites monitor GS on a global scale, the contemporary Gravity Recovery and Climate Experiment (GRACE) and GRACE Follow−On (GRACE−FO) missions are only able to do so at coarse spatial resolutions and are restricted by their limited observation periods. Moreover, because of the complex mechanism behind a glacier's mass balance, existing reconstruction methods for GRACE-like mass anomalies, including machine learning approaches and statistical models, cannot be applied to HMGs. Here, we propose a precipitation and temperature data-driven statistical model combining hydrology and GS processes to reconstruct GRACE-like mass anomalies of HMGs, from which the hydrological and GS signals can be further separated. The prediction and reconstruction performance of this method was consistent with the GRACE/GRACE−FO observations (the median correlation coefficient/Nash-Sutcliffe efficiency: ~0.92/0.85). Additionally, the separated GS signals agreed with the independent external data used for comparison. Compared with machine learning methods, this method better reconstructed the long-term trend of GRACE-like mass anomalies. Our study provides an acceptable method for expanding mass anomaly time series on HMGs, thereby assisting in the sustainable management and protection of water resources in their downstream areas.

Strangers in a Strange Land: New Experimental System for Understanding Driving Culture Using VR

In order for autonomous vehicles to adapt to local norms in human driving, it is critical to profile how human driving differs across geographical locations. While ethnographers have qualitatively described regional differences in driving style, data-driven statistical models might help computer-driven cars drive like locals and recognize how local drivers might be signaling through hand/body movement and motion of their vehicles. To this end, we have created an experimental system and method to profile driving behavior and interaction using a multi-participant virtual reality (VR) driving simulation environment. The system was designed to be portable and to support cross-cultural experimental deployments. We aim to make sure the system is operational and functional, can model diverse scenarios, generates data fit for analysis, and captures expected behaviors. We describe the system, test scenarios, and findings of the proof-of-concept study conducted in the U.S. and Israel.

Spatial-temporal dynamics of a microbial cooperative behavior resistant to cheating

Much of our understanding of bacterial behavior stems from studies in liquid culture. In nature, however, bacteria frequently live in densely packed spatially-structured communities. How does spatial structure affect bacterial cooperative behaviors? In this work, we examine rhamnolipid production—a cooperative and virulent behavior of Pseudomonas aeruginosa. Here we show that, in striking contrast to well-mixed liquid culture, rhamnolipid gene expression in spatially-structured colonies is strongly associated with colony specific growth rate, and is impacted by perturbation with diffusible quorum signals. To interpret these findings, we construct a data-driven statistical inference model which captures a length-scale of bacterial interaction that develops over time. Finally, we find that perturbation of P. aeruginosa swarms with quorum signals preserves the cooperating genotype in competition, rather than creating opportunities for cheaters. Overall, our data demonstrate that the complex response to spatial localization is key to preserving bacterial cooperative behaviors.

Independent features form integrated objects: Using a novel shape-color “conjunction task” to reconstruct memory resolution for multiple object features simultaneously

Though much progress has been made to understand feature integration, debate remains regarding how objects are represented in mind based on their constituent features. Here, we advance this debate by introducing a novel shape-color “conjunction task” to reconstruct memory resolution for multiple object features simultaneously. In a first experiment, we replicate and extend a classic paradigm originally tested using a change detection task. Replicating previous work, memory resolution for individual features was reduced when the number of objects increased, regardless of the number of to-be-remembered features. Extending previous work, we found that high resolution memory near perfect in resemblance to the target was selectively impacted by the number of to-be-remembered features. Applying a data-driven statistical model of stochastic dependence, we found robust evidence of integration for lower-resolution feature memories, but less evidence for integration of high-resolution feature memories. These results suggest that memory resolution for individual features can be higher than memory resolution for their integration. In a second experiment which manipulated the nature of distracting information, we examined whether object features were directly bound to each other or by virtue of shared spatial location. Feature integration was disrupted by distractors sharing visual features of target objects but not when distractors shared spatial location – suggesting that feature integration can be driven by direct binding between shape and color features in memory. Our results constrain theoretical models of object representation, providing empirical support for hierarchical representations of both integrated and independent features.

Transdiagnostic phenotypes of compulsive behavior and associations with psychological, cognitive, and neurobiological affective processing

Compulsivity is a poorly understood transdiagnostic construct thought to underlie multiple disorders, including obsessive-compulsive disorder, addictions, and binge eating. Our current understanding of the causes of compulsive behavior remains primarily based on investigations into specific diagnostic categories or findings relying on one or two laboratory measures to explain complex phenotypic variance. This proof-of-concept study drew on a heterogeneous sample of community-based individuals (N = 45; 18–45 years; 25 female) exhibiting compulsive behavioral patterns in alcohol use, eating, cleaning, checking, or symmetry. Data-driven statistical modeling of multidimensional markers was utilized to identify homogeneous subtypes that were independent of traditional clinical phenomenology. Markers were based on well-defined measures of affective processing and included psychological assessment of compulsivity, behavioral avoidance, and stress, neurocognitive assessment of reward vs. punishment learning, and biological assessment of the cortisol awakening response. The neurobiological validity of the subtypes was assessed using functional magnetic resonance imaging. Statistical modeling identified three stable, distinct subtypes of compulsivity and affective processing, which we labeled “Compulsive Non-Avoidant”, “Compulsive Reactive” and “Compulsive Stressed”. They differed meaningfully on validation measures of mood, intolerance of uncertainty, and urgency. Most importantly, subtypes captured neurobiological variance on amygdala-based resting-state functional connectivity, suggesting they were valid representations of underlying neurobiology and highlighting the relevance of emotion-related brain networks in compulsive behavior. Although independent larger samples are needed to confirm the stability of subtypes, these data offer an integrated understanding of how different systems may interact in compulsive behavior and provide new considerations for guiding tailored intervention decisions.

Ubiquitous proximity to a critical state for collective neural activity in the CA1 region of freely moving mice

Using miniscope recordings of calcium fluorescence signals in the CA1 region of the hippocampus of mice, we monitor the neural activity of hippocampal regions while the animals are freely moving in an open chamber. Using a data-driven statistical modeling approach, the statistical properties of the recorded data are mapped to spin-glass models with pairwise interactions. Considering the parameter space of the model, the observed system is generally near a critical state between two distinct phases. The close proximity to the criticality is found to be robust against different ways of sampling and segmentation of the measured data. By independently altering the coupling distribution and the network structure of the statistical model, the network structures are found to be vital to maintain the proximity to the critical state. We further find the observed assignment of the coupling strengths makes the net coupling at each site more balanced with slight variation, which likely helps the maintenance of the critical state. Network analysis on the connectivity obtained by thresholding the coupling strengths find the connectivity of the networks to be well described by a random network model. These results are consistent across different experiments, sampling and segmentation choices in our analysis. A new result of our analysis is that the proximity to critical state and all the network properties are largely maintained even if random subsamples with a fraction of neurons are selected from the dataset as long as the number of neurons is more than 30 to 40. Thus the relevant degrees of freedom of CA1 region in the collective state we studied is not as large as one would expect.

ESTIMATING PEDESTRIAN SUSCEPTIBILITY TO TRAFFIC ACCIDENTS IN CURITIBA, BRAZIL

Even though pedestrians represented 40% of all urban displacements in Brazil in 2017, they are still highly vulnerable to traffic accidents, with a mortality rate of 2.89 per 100 thousand inhabitants in 2018. The literature suggests a relationship between the occurrence of traffic accidents and demographic, socioeconomic, and urban structure variables. In this study, this relationship was investigated through a data-driven statistical model (logistic regression) combined with GIS spatial analysis, applied to estimate the pedestrian susceptibility to traffic accidents in the City of Curitiba, in Southern Brazil. By adopting broadly available spatial information, the proposed methods were robust in estimating the events, presenting an area under the ROC curve of 0.82 in the cross-validation. Additionally, the results highlighted a strong and statistically significant correlation between the pedestrian crashes and the analyzed variables of road system hierarchy, presence of BRT routes, land-use, population density and per capita income.

Quantitative phosphoproteomics uncovers dysregulated kinase networks in Alzheimer's disease.

Alzheimer's disease (AD) is a form of dementia characterized by amyloid-β plaques and tau neurofibrillary tangles that progressively disrupt neural circuits in the brain. The signaling networks underlying AD pathological changes are poorly characterized at the phosphoproteome level. Using mass spectrometry, we analyzed the proteome and tyrosine, serine and threonine phosphoproteomes of temporal cortex tissue from patients with AD and aged-matched controls. We identified cocorrelated peptide clusters that were linked to varying levels of phospho-tau, oligodendrocyte, astrocyte, microglia and neuron pathologies. We found that neuronal synaptic protein abundances were strongly anti-correlated with markers of microglial reactivity. We also observed that phosphorylation sites on kinases targeting tau and other new signaling factors were correlated with these peptide modules. Finally, we used data-driven statistical modeling to identify individual peptides and peptide clusters that were predictive of AD histopathologies. Together, these results build a map of pathology-associated phosphorylation signaling events occurring in AD.