Model-driven Methods Research Articles

Data-Supported Prediction of Surface Settlement Behavior on Opencast Mine Dumps Using Satellite-Based Radar Interferometry Observations

To ensure the safe repurposing of post-mining landscapes, understanding and managing geotechnical risks, particularly ground movements such as settlements on opencast mining dump surfaces, is critical. Satellite-based radar interferometry (InSAR) technology offers highly detailed data on vertical ground movements with a high spatial and temporal resolution. By combining a data-driven approach, using InSAR-generated high-resolution datasets, with model-driven methods such as inverse modeling and classic time–settlement models, the efficient monitoring and prediction of opencast mine dump settlements can be achieved. This dual approach—leveraging advanced data analysis tools and precise modeling—yields valuable insights into spatial settlement behavior. In particular, classic time–settlement models are applied to the InSAR data through least square regression and Taylor approximation. The integration of both approaches enables the more robust, data-validated forecasts of key geotechnical indicators, such as the time to settlement stabilization and the expected maximum settlement over large areas. An application at a mine in central Germany illustrates the method by generating spatial predictions of the settlement behavior over more than 200 ha. In general, the results provide a comprehensive dataset for investigating other factors influencing the settlement behavior of opencast mine dumps.

A support vector regression-based interval power flow prediction method for distribution networks with DGs integration

In distribution networks with distributed generators (DGs), power generation and load demand exhibit increased randomness and volatility, and the line parameters also suffer more frequent fluctuations, which may result in significant state shifts. Existing model-driven methods face challenges in efficiently solving uncertain power flow, especially as the size of the system increases, making it difficult to meet the demand for rapid power flow analysis. To address these issues, this paper proposes an SVR-based interval power flow (IPF) prediction method for distribution networks with DGs integration. The method utilizes intervals to describe system uncertainty and employs Support Vector Regression (SVR) for model training. The input feature vector consists of the intervals of active power generation, load demand, and line parameters, while the output feature vector represents the intervals of voltage or line transmission power. Ultimately, the SVR-based IPF prediction model is established, capturing the linear mapping relationship between input data and output IPF variables. Simulation results demonstrate that the proposed method exhibits high prediction accuracy, strong adaptability, and optimal computation efficiency, meeting the requirements for rapid and real-time power flow analysis while considering the uncertainty in distribution networks with DGs integration.

Sparse regularized graph pooling network optimal sensor placement method for diesel engine vibration fault perception system

Vibration sensor network optimization increases monitoring effectiveness and reduces sensor quantity and transmission burden. However, the traditional model-driven methods depend on precise finite element models, challenging for complex machinery. This paper presents a data-driven approach using the Sparse Regularized Graph Pooling Network (SRGPN), which conceptualizes sensor networks as graphs and uses graph pooling to identify optimal sensor combinations. A sparsity regularization term related to the number of sensors is included in the loss function, aiming for the minimal yet effective sensor combination. Additionally, a monitoring capability metric suited for diesel engines with multi-source impulse signals is proposed, reflecting the sensors’ monitoring capacity. Validated through simulations and tests on a diesel engine test bench, the results show that SRGPN optimally places sensors near excitation sources, balancing sensor count and monitoring needs. This approach shows potential for optimizing sensor placements in condition monitoring.

An improved spectral clustering method for accurate detection of brain resting-state networks

This paper proposes a data-driven analysis method to accurately partition large-scale resting-state functional brain networks from fMRI data. The method is based on a spectral clustering algorithm and combines eigenvector direction selection with Pearson correlation clustering in the spectral space. The method is an improvement on available spectral clustering methods, capable of robustly identifying active brain networks consistent with those from model-driven methods at different noise levels, even at the noise level of real fMRI data.

A Combined Data-Driven and Model-Based Algorithm for Accurate Battery Thermal Runaway Warning.

With the increasingly widespread application of large-scale energy storage battery systems, the demand for battery safety is rising. Research on how to detect battery anomalies early and reduce the occurrence of thermal runaway (TR) accidents has become particularly important. Existing research on battery TR warning algorithms can be mainly divided into two categories: model-driven and data-driven methods. However, the common model-driven methods are often of high complexity, with poor versatility and low early warning capability; and the common data-driven methods are mostly based on neural networks, requiring substantial training costs, with better early warning capabilities but higher false alarm probabilities. To address the limitations of existing works, this paper proposes a combined data-driven and model-based algorithm for accurate battery TR warnings. Specifically, the K-Means algorithm serves as the data-driven module, capturing outliers in battery data, and the Bernardi equation serves as the model-driven module used to evaluate battery temperature. Ultimately, the outputs of the weighted model-driven module and data-driven module are combined to comprehensively assess whether the battery is abnormal. The proposed algorithm combines the advantages of model-driven and data-driven approaches, achieving a 25 min advance warning for thermal runaway, with a significantly reduced probability of false alarms.

Multi-task learning mixture density network for interval estimation of the remaining useful life of rolling element bearings

Existing remaining useful life (RUL) predictions of rolling element bearings have the following shortcomings. 1) Model-driven methods typically employ a sole model for processing the data of an individual, making it challenging to accommodate the variety of degradation behaviors and susceptible to abnormal interference. 2) Data-driven methods place greater emphasis on training data, and in reality, it can be challenging to acquire comprehensive data covering the lifecycle. 3) Many studies fail to give adequate attention to the assessment of RUL uncertainty. This paper proposes a multi-task learning mixture density network (MTL-MDN) method to address these issues. Firstly, the peak-of-Histogram (PoHG) is extracted and served as the novel health indicators. Secondly, multi-task learning dictionaries are constructed based on the evolution law of PoHG, thus combining both model-driven and data-driven strategies. Finally, a multi-task learning strategy is proposed with mixture density networks. It effectively accomplishes the collaborative learning objective of numerous degradation samples in the regression problem and accomplishes the uncertainty assessment of RUL. After analyzing the experimental and real-world degradation data of rolling element bearings throughout their lifecycle, and comparing it to other modern RUL prediction methods, it becomes evident that the proposed MTL-MDN method offers superior prediction accuracy and robustness.

Reservoir properties inversion using attention-based parallel hybrid network integrating feature selection and transfer learning

The inversion of subsurface reservoir properties is of profound significance to the oil and gas energy development and utilization. The strong heterogeneity and complex pore structure of underground reservoirs pose a challenge to efficient oil and gas energy development and utilization across the world, which increases the necessity of developing an efficient reservoir properties inversion method. However, traditional model-driven methods are confronted with the challenges of strong nonlinearity and geological heterogeneity. Moreover, previous studies rarely emphasized the importance of nonlinear feature selection and transfer learning (TL). Aiming to address the research gaps, a reservoir properties inversion method was proposed by combining random forest (RF) feature selection, bidirectional temporal convolutional network (BiTCN), bidirectional gated recurrent units (BiGRU) network, multi-head attention (MHA) mechanism and TL strategy. First, the RF was used to screen the features with significant correlation with the target reservoir properties. Thereafter, combining BiTCN and BiGRU network to leverage their complementary strengths, a parallel dual branch feature learning network was constructed to learn richer geological information from logging data. Meanwhile, MHA was introduced to fuse the output features of the dual network structure. Finally, the fused features were passed through the fully connected module to output the inversion results. TL was used to associate the correlation between reservoir properties and model inversion to improve the inversion performance. The application results with actual field data showed that the proposed method was accurate and robust in reservoir properties inversion. This study can provide a new way for reliable reservoir properties inversion and promote the application of artificial intelligence in data-driven energy science.

Direction of Arrival Estimation of Coherent Sources via a Signal Space Deep Convolution Network

In the field of direction of arrival (DOA) estimation for coherent sources, subspace-based model-driven methods exhibit increased computational complexity due to the requirement for eigenvalue decomposition. In this paper, we propose a new neural network, i.e., the signal space deep convolution (SSDC) network, which employs the signal space covariance matrix as the input and performs independent two-dimensional convolution operations on the symmetric real and imaginary parts of the input signal space covariance matrix. The proposed SSDC network is designed to address the challenging task of DOA estimation for coherent sources. Furthermore, we leverage the spatial sparsity of the output from the proposed SSDC network to conduct a spectral peak search for obtaining the associated DOAs. Simulations demonstrate that, compared to existing state-of-the-art deep learning-based DOA estimation methods for coherent sources, the proposed SSDC network achieves excellent results in both matching and mismatching scenarios between the training and test sets.

Joint Gravity and Magnetic Inversion Using CNNs’ Deep Learning

Enhancing the reliability of inversion results has always been a prominent issue in the field of geophysics. In recent years, data-driven inversion methods leveraging deep neural networks (DNNs) have gained prominence for their ability to address non-uniqueness issues and reduce computational costs compared to traditional physically model-driven methods. In this study, we propose a GMNet machine learning method, i.e., a CNN-based inversion method for gravity and magnetic field data. This method relies more on data-driven training, and in the prediction phase after the model is trained, it does not heavily depend on a priori assumptions, unlike traditional methods. By forward modeling gravity and magnetic fields, we obtain a substantial dataset to train the CNN model, enabling the direct mapping from field data to subsurface property distribution. Applying this method to synthetic data and one-field data yields promising inversion results.

Prediction of permeability in a tight sandstone reservoir using a gated network stacking model driven by data and physical models

Introduction: Permeability is one of the most important parameters for reservoir evaluation. It is commonly measured in laboratories using underground core samples. However, it cannot describe the entire reservoir because of the limited number of cores. Therefore, petrophysicists use well logs to establish empirical equations to estimate permeability. This method has been widely used in conventional sandstone reservoirs, but it is not applicable to tight sandstone reservoirs with low porosity, extremely low permeability, and complex pore structures.Methods: Machine learning models can identify potential relationships between input features and sample labels, making them a good choice for establishing permeability prediction models. A stacking model is an ensemble learning method that aims to train a meta-learner to learn an optimal combination of expert models. However, the meta-learner does not evaluate or control the experts, making it difficult to interpret the contribution of each model. In this study, we design a gate network stacking (GNS) model, which is an algorithm that combines data and model-driven methods. First, an input log combination is selected for each expert model to ensure the best performance of the expert model and selfoptimization of the hyperparameters. Petrophysical constraints are then added to the inputs of the expert model and meta-learner, and weights are dynamically assigned to the output of the expert model. Finally, the overall performance of the model is evaluated iteratively to enhance its interpretability and robustness.Results and discussion: The GNS model is then used to predict the permeability of a tight sandstone reservoir in the Jurassic Ahe Formation in the Tarim Basin. The case study shows that the permeability predicted by the GNS model is more accurate than that of other ensemble models. This study provides a new approach for predicting the parameters of tight sandstone reservoirs.

A novel data-driven sensor placement optimization method for unsupervised damage detection using noise-assisted neural networks with attention mechanism

Optimization of sensor placement (OSP) is one of the important steps in structural health monitoring to reduce the instrumentation cost, and improve damage detection. Although modal analysis is an optional intermediate process in terms of damage detection, conventional OSP methods for damage detection are mostly dependent on mode shapes. In this case, the sensor placement and damage detection are highly relying on the accuracy of modal analysis, and the results may not be adapted to different type of excitations. In this article, a novel noise-assisted neural network with attention mechanism is proposed based on the above challenges. This method which can be used to optimize sensor placement in an unsupervised and data-driven manner, is verified using a dataset simulated from the ASCE benchmark and an experimental dataset obtained from shake table tests. The results from the simulated dataset show that the percentage of the sensors that could be removed are higher than the conventional effective independence (EFI) method, with a highest of 62.5% for cases with low noise levels. In the meantime, the occurrence and the level of damage can still be well detected with a reduced number of sensors. Most importantly, in the proposed approach the optimal sensor placement configurations can be determined adaptively to account for different forms of excitations and noise levels, which is impossible for conventional model-driven methods. The results obtained from the experimental dataset show that the proposed method is also effective for real-world applications. As a result, the proposed data-driven OSP method skips the conventional model analysis process and directly focuses on the sensor arrangement that enables accurate detection of damage. It also has the potential for application in related fields, such as the monitoring of aerospace and mechanical infrastructures.

Dynamic Motion Transition: A Hybrid Data-driven and Model-driven Method for Human Pose Transitions.

The rapid, accurate, and robust computation of virtual human figures' "in-between" pose transitions from available and sometimes sparse inputs is of fundamental significance to 3D interactive graphics and computer animation. Various methods have been proposed to produce natural lifelike transitions of human pose automatically in recent decades. Nevertheless, conventional pure model-driven methods require heuristic knowledge (e.g., least motion guided by physics laws) and ad-hoc clues (e.g., splines with non-uniform time warp) that are difficult to obtain, learn, and infer. With the fast emergence of large-scale datasets readily available to animators in the most recent years, deep models afford a powerful alternative to tackle the aforementioned challenges. However, pure data-driven methods still suffer from the remaining challenges such as unseen data in practice and less generative power in model/domain/data transfer, and the measurement of the generative power has always been omitted in these works. In essence, data-driven methods solely rely on the qualities and quantities of training datasets. In this paper, we propose a hybrid approach built upon the seamless integration of data-driven and model-driven methods, called Dynamic Motion Transition (DMT), with the following salient modeling advantages: (1) The data augmentation capability based on the limited human locomotion data capture and the concept of force-derived directly from physical laws; (2) Force learning by which skeleton joints are driven to move, and the Conditional Temporal Transformer (CTT) being trained to learn the force change in the local range, both at the fine level; and (3) At the coarse level, the effective and flexible creation of the subsequent step motion using Dynamic Movement Primitives (DMP) until the target is reached. Our extensive experiments have confirmed that our model can outperform the state-of-the-art methods under the newly devised metric by virtue of the least action loss function. In addition, our novel method and system are of immediate benefit to many other animation tasks such as motion synthesis and control, and motion tracking and prediction in this bigdata graphics era.

TMELand: An end-to-end Pipeline for Quantification and Visualization of Waddington's Epigenetic Landscape based on Gene Regulatory Network.

Waddington's epigenetic landscape is a framework depicting the processes of cell differentiation and reprogramming under the control of a gene regulatory network (GRN). Traditional model-driven methods for landscape quantification focus on the Boolean network or differential equation-based models of GRN, which need sophisticated prior knowledge and hence hamper their practical applications. To resolve this problem, we combine data-driven methods for inferring GRNs from gene expression data with model-driven approach to the landscape mapping. Specifically, we build an end-to-end pipeline to link data-driven and model-driven methods and develop a software tool named TMELand for GRN inference, visualizing Waddington's epigenetic landscape, and calculating state transition paths between attractors to uncover the intrinsic mechanism of cellular transition dynamics. By integrating GRN inference from real transcriptomic data with landscape modeling, TMELand can facilitate studies of computational systems biology, such as predicting cellular states and visualizing the dynamical trends of cell fate determination and transition dynamics from single-cell transcriptomic data. The source code of TMELand, a user manual, and model files of case studies can be downloaded freely from https://github.com/JieZheng-ShanghaiTech/TMELand.

Magnetic Resonance Electrical Properties Tomography Based on Modified Physics-Informed Neural Network and Multiconstraints.

This paper presents a novel method based on leveraging physics-informed neural networks for magnetic resonance electrical property tomography (MREPT). MREPT is a noninvasive technique that can retrieve the spatial distribution of electrical properties (EPs) of scanned tissues from measured transmit radiofrequency (RF) in magnetic resonance imaging (MRI) systems. The reconstruction of EP values in MREPT is achieved by solving a partial differential equation derived from Maxwell's equations that lacks a direct solution. Most conventional MREPT methods suffer from artifacts caused by the invalidation of the assumption applied for simplification of the problem and numerical errors caused by numerical differentiation. Existing deep learning-based (DL-based) MREPT methods comprise data-driven methods that need to collect massive datasets for training or model-driven methods that are only validated in trivial cases. Hence we proposed a model-driven method that learns mapping from a measured RF, its spatial gradient and Laplacian to EPs using fully connected networks (FCNNs). The spatial gradient of EP can be computed through the automatic differentiation of FCNNs and the chain rule. FCNNs are optimized using the residual of the central physical equation of convection-reaction MREPT as the loss function (L). To alleviate the ill condition of the problem, we added multiconstraints, including the similarity constraint between permittivity and conductivity and the ℓ1 norm of spatial gradients of permittivity and conductivity, to the L. We demonstrate the proposed method with a three-dimensional realistic head model, a digital phantom simulation, and a practical phantom experiment at a 9.4T animal MRI system.

Learning Robust Predictive Control: A Spatial-Temporal Game Theoretic Approach.

This article investigates robust predictive control problem for unknown dynamical systems. Since the dynamics unavailability restricts feasibility of model-driven methods, learning robust predictive control (LRPC) framework is developed from the aspect of time consistency. Under feedback-like control causality, the robust predictive control is then reconstructed as spatialbKKtemporal games, and we guarantee stability through time-consistent Nash equilibrium. For gradation clarity, our framework is specified as four-follow contents. First, multistep feedback-like control causality is drawn from time series analysis, and Takens' theorem provides theoretical support from steady-state property. Second, control problem is reconstructed as games, while performance and robustness partition the game into temporal nonzero-sum subgames and spatial zero-sum ones, respectively. Next, multistep reinforcement learning (RL) is designed to solve robust predictive control without system model. Convergence is proven through bounds analysis of oscillatory value functions, and properties of receding horizon are derived from time consistency. Finally, data-driven implementation is given with function approximation, and neural networks are chosen to approximate value functions and feedback-like causality. Weights are estimated with least squares errors. Numerical results verify the effectiveness.

Nonparametric Dynamic Granger Causality based on Multi-Space Spectrum Fusion for Time-varying Directed Brain Network Construction.

Nonparametric estimation of time-varying directed networks can unveil the intricate transient organization of directed brain communication while circumventing constraints imposed by prescribed model-driven methods. A robust time-frequency representation - the foundation of its causality inference - is critical for enhancing its reliability. This study proposed a novel method, i.e., nonparametric dynamic Granger causality based on Multi-space Spectrum Fusion (ndGCMSF), which integrates complementary spectrum information from different spaces to generate reliable spectral representations to estimate dynamic causalities across brain regions. Systematic simulations and validations demonstrate that ndGCMSF exhibits superior noise resistance and a powerful ability to capture subtle dynamic changes in directed brain networks. Particularly, ndGCMSF revealed that during instruction response movements, the laterality in the hemisphere ipsilateral to the hemiplegic limb emerges upon instruction onset and diminishes upon task accomplishment. These intrinsic variations further provide reliable features for distinguishing two types of hemiplegia (left vs. right) and assessing motor functions. The ndGCMSF offers powerful functional patterns to derive effective brain networks in dynamically changing operational settings and contributes to extensive areas involving dynamical and directed communications.

Characterizing High Rate GNSS Velocity Noise for Synthesizing a GNSS Strong Motion Learning Catalog

Data-driven approaches to identify geophysical signals have proven beneficial in high dimensional environments where model-driven methods fall short. GNSS offers a source of unsaturated ground motion observations that are the data currency of ground motion forecasting and rapid seismic hazard assessment and alerting. However, these GNSS-sourced signals are superposed onto hardware-, location- and time-dependent noise signatures influenced by the Earth’s atmosphere, low-cost or spaceborne oscillators, and complex radio frequency environments. Eschewing heuristic or physics based models for a data-driven approach in this context is a step forward in autonomous signal discrimination. However, the performance of a data-driven approach depends upon substantial representative samples with accurate classifications, and more complex algorithm architectures for deeper scientific insights compound this need. The existing catalogs of high-rate (≥1Hz) GNSS ground motions are relatively limited. In this work, we model and evaluate the probabilistic noise of GNSS velocity measurements over a hemispheric network. We generate stochastic noise time series to augment transferred low-noise strong motion signals from within 70 kilometers of strong events (≥ MW 5.0) from an existing inertial catalog. We leverage known signal and noise information to assess feature extraction strategies and quantify augmentation benefits. We find a classifier model trained on this expanded pseudo-synthetic catalog improves generalization compared to a model trained solely on a real-GNSS velocity catalog, and offers a framework for future enhanced data driven approaches.

Comparison of model-driven soft measurement methods for compressor air flow in gas-steam combined cycle power units

The determination of the air flowrate at the compressor inlet is meaningful for the performance diagnosis of heavy-duty gas turbine combined cycle (GTCC) power units, whereas the compressor air flow generally cannot be directly obtained from the field data. In this paper, the procedures of three model-driven methods for soft measurement of compressor air flow were presented, including energy balance of the coupled HRSG (M1), static pressure at the compressor manifold (M2) and gas turbine joint operation line method (M3). In the case of a M701F4 GTCC power unit, the methods M1 and M3 were validated by the field and testing data; while the method M2 was calibrated and the effective area at the air manifold was decided as 12.606 m2 using M1 at the full load of gas turbine. The compressor air mass flow at overall loads was computed with the various methods. The comparative investigation to these methods shows that, the compressor inlet air flow is determined as 682.75 kg/s by M1 and 664.07 kg/s by M3, respectively, at the full load and at the ambient temperature 29.912 °C. The air flow obtained by M1 and M2 appears relatively higher than that by M3 at typical various loads of gas turbine; however, the method M3 shows a good accordance with the actual value of the air flow for turbine rotor cooling. Both M2 and M3 agree in the theoretical demonstration that compressor air flow is strongly related to the gas turbine load at certain conditions. The method M1 can be deemed as a reference for another two methods in the case of the steady operating state at full load; whereas it is not recommendable for the real-time measurement of compressor air flow, because of the large amounts of required input variables and the fluctuation in the flow meters.

P/S wavefield separation using modified U-Net based on ConvNeXt architecture

P/S wavefields separation is an essential step in multi-component seismic data processing, significantly impacting the quality of subsequent processing workflows if not performed accurately. Conventional wavefield separation methods, called model-driven methods, primarily rely on the polarization directions of P- and S-wavefields, which are obtained depending on accurate model parameters such as velocity and density. However, seismic model building is a well-known challenge for real data. As a result, the conventional methods rarely produce ideal pure P- or S-wavefields. Deep learning (DL) techniques can solve many real-world problems by constructing complex nonlinear mappings between inputs and desired outputs based on large-scale datasets, such as image classification, semantic segmentation, and object detection. Inspired by the advantages of DL techniques, we construct a improved U-ConvNeXt network architecture by combining the U-Net with the ConvNeXt, and apply it to implement P/S wavefields separation. Compared with the conventional U-Net, the ConvNeXt has better feature extraction capabilities because it introduces the idea of transformer architecture design into CNN. Consequently, the proposed U-ConvNeXt-based P/S wavefields separation method produces cleaner P- and S-wavefields with less computational costs. In addition, the proposed method can achieve nearly the same accuracy as the conventional methods without relying on model parameters, and its computational efficiency is much higher than conventional methods. Finally, we verify the proposed method with several numerical examples, and numerical results validate the effectiveness and accuracy of the proposed P/S wavefields separation method.

Dual-Domain Prior-Driven Deep Network for Infrared Small-Target Detection

In recent years, data-driven deep networks have demonstrated remarkable detection performance for infrared small targets. However, continuously increasing the depth of neural networks to enhance performance has proven impractical. Consequently, the integration of prior physical knowledge related to infrared small targets within deep neural networks has become crucial. It aims to improve the models’ awareness of inherent physical characteristics. In this paper, we propose a novel dual-domain prior-driven deep network (DPDNet) for infrared small-target detection. Our method integrates the advantages of both data-driven and model-driven methods by leveraging the prior physical characteristics as the driving force. Initially, we utilize the sparse characteristics of infrared small targets to boost their saliency at the input level of the network. Subsequently, a high-frequency feature extraction module, seamlessly integrated into the network’s backbone, is employed to excavate feature information. DPDNet simultaneously emphasizes the prior sparse characteristics of infrared small targets in the spatial domain and their prior high-frequency characteristics in the frequency domain. Compared with previous CNN-based methods, our method achieves superior performance while utilizing fewer convolutional layers. It has a performance of 78.64% IoU, 95.56 Pd, and 2.15 × 10−6 Fa on the SIRST dataset.