Training Data Samples Research Articles

Semi-supervised CWGAN-GP modeling for AHU AFDD with high-quality synthetic data filtering mechanism

Supervised learning methods demonstrated high classification accuracy for air handling unit (AHU) automated fault detection and diagnosis (FDD) scenarios with well-shaped training datasets. However, for imbalanced training datasets, i.e., much less real-world fault training data samples against an enormous amount of normal data samples, the supervised learning-based methods failed to produce satisfactory FDD results. To address the above-mentioned issue, this study proposes a semi-supervised conditional Wasserstein generative adversarial network with gradient penalty (CWGAN-GP) to generate high-quality synthetic fault training samples. The semi-supervised learning-based AHU AFDD framework is completed by identifying high-quality synthetic fault samples and inserting them into the training pool iteratively. With different numbers of real-world fault samples, comparative experiments are conducted on datasets collected by ASHRAE project RP-1312 in the summer and winter seasons. The experimental results show that the proposed AFDD method has obvious advantages over the traditional method with limited numbers of real-world fault samples. Moreover, the proposed CWGAN-GP-SSL framework achieves superior AFDD performance compared to the existing GAN-based AHU AFDD method.

Technology Empowers Finance: Boundaries and Risks

BigTech credit has enhanced financial inclusion, but it also poses concerns with its boundaries. This article uses theoretical frameworks and numerical simulations to examine the risks and inclusiveness of technology-empowered credit services for “long-tail” clients. This research discovered that the discrepancy between the commercial boundaries of BigTech credit and the technical limitations of risk management poses a risk in BigTech credit. The expanding boundaries of BigTech’s credit business may mitigate the representativeness of the data, resulting in a systematic deviation of unclear characteristics from the training sample data, which reduces the risk-control model’s ability to identify long-tail customers and raises the risk of credit defaults. Further computer simulations validate these results and demonstrate that competition among various companies would expedite the market’s transition over the boundary in case of a capital shortage. Finally, this article proposes setting up a joint-stock social unified credit technology company with data assets as an investment to facilitate the healthy and orderly development of financial technology institutions.

Subspace‐based distributed target detection method with small training data samples

AbstractDetecting distributed targets precisely in homogeneous environments has been a hot topic in radar signal processing. Generally, distributed targets are often modelled with subspace models of unknown coordinates, and clutter is modelled as the complex Gaussian distribution with zero mean and unknown covariance matrix, while covariance matrix is estimated with a set of training data without the target signal. However, in practice, the complexity of the external environment makes the training data that satisfy the condition of independent homogeneous distribution less available. Therefore, it is assumed that the covariance matrix of the clutter is persymmetric structure and the approach of dimensionality reduction using subspace transformations is introduced, two detectors based upon generalised likelihood ratio test criterion and Wald test criterion in homogeneous environments are proposed. Theoretical analyses indicate the constant false alarm rate characteristics of the two proposed detectors for unknown clutter covariance matrices. Simulation analyses indicate that the proposed detector works well even with fewer training data samples, and its detection performance outperforms that of existing contrast detectors.

Research on Status Assessment and Operation and Maintenance of Electric Vehicle DC Charging Stations Based on XGboost

With the rapid development of electric vehicles, the infrastructure for charging stations is also expanding quickly, and the failure rate of charging piles is increasing. To address the effective operation and maintenance of charging stations, a method based on the XGBoost algorithm for electric vehicle DC charging stations is proposed. An operation and maintenance system is constructed based on state analysis, considering the operational status of the charging stations and users’ charging habits. Factors such as driving and charging habits, road traffic, and charging station equipment are taken into account. The training sample data are established using historical data, online monitoring data, and external environmental data, and the charging station status evaluation model is trained using the XGBoost algorithm. Based on the condition assessment results, a risk assessment model is established in combination with fault parameters. Risk tracking of the charging stations is conducted using the energy not charged (ENC), evaluating the risk level of each station and determining the operation and maintenance order. The optimal operation and maintenance model for DC charging stations, aimed at achieving both economic and reliability goals, is constructed to determine the operation and maintenance schedule for each station. The results of the case study demonstrate that the state evaluation and operation and maintenance strategy can significantly improve the reliability of the system and the overall benefits of operation and maintenance while meeting the required standards.

MRI recovery with self-calibrated denoisers without fully-sampled data.

Acquiring fully sampled training data is challenging for many MRI applications. We present a self-supervised image reconstruction method, termed ReSiDe, capable of recovering images solely from undersampled data. ReSiDe is inspired by plug-and-play (PnP) methods, but unlike traditional PnP approaches that utilize pre-trained denoisers, ReSiDe iteratively trains the denoiser on the image or images that are being reconstructed. We introduce two variations of our method: ReSiDe-S and ReSiDe-M. ReSiDe-S is scan-specific and works with a single set of undersampled measurements, while ReSiDe-M operates on multiple sets of undersampled measurements and provides faster inference. Studies I, II, and III compare ReSiDe-S and ReSiDe-M against other self-supervised or unsupervised methods using data from T1- and T2-weighted brain MRI, MRXCAT digital perfusion phantom, and first-pass cardiac perfusion, respectively. ReSiDe-S and ReSiDe-M outperform other methods in terms of peak signal-to-noise ratio and structural similarity index measure for Studies I and II, and in terms of expert scoring for Study III. We present a self-supervised image reconstruction method and validate it in both static and dynamic MRI applications. These developments can benefit MRI applications where the availability of fully sampled training data is limited.

Effective zero-shot learning method for event classification in Φ-OTDR sensing systems

Despite various Φ-OTDR intrusion event recognition methods having achieved high average accuracy rates (over 90%), these methods usually rely on a large amount of training sample data (80% of the data). When faced with certain intrusion events that are difficult to simulate or have few samples available, the model tends to overfit common types of intrusion events. To address this issue, this paper proposes a zero-sample learning one-dimensional residual model based on attribute point loss (APL-ZSL-1DResNet) to recognize one-dimensional intrusion event signals when training samples are insufficient. The proposed method is validated on two datasets, including a self-made dataset and an open dataset. In the experiments, each category of samples was set as zero-sample intrusion events, achieving an average recall rate of 75% and 66% respectively for zero-sample events, and an average recall rate of 94.6% and 83.5% respectively for common intrusion events.

Accurate and efficient feature classification of urban public open spaces: A deep learning-based multivariate time-series approach

Urban public open spaces (POS) are pivotal in sustainable urban planning, recognized for their positive impacts on the health of residents and environments. However, understanding their physical features in detail via remote sensing remains challenging due to the small and complex area characteristics. Advanced approaches struggle with the requirement for extensive training datasets, which are difficult to obtain and potentially inaccurate in certain settings. Sparse sampling of training data may offer a solution to these challenges, but its inability to consider useful contextual characteristics from objects remains as a significant barrier.To address these challenges, we propose an innovative methodology utilizing the Multivariate Long Short Term Memory–Fully Convolutional Network (MLSTM-FCN). This approach is adept at capturing subtle temporal and spectral variations from sparsely annotated data, making it particularly suited for optical remote sensing in urban POS. Our novel approach has achieved 97.86% accuracy in classifying POS features into 11 classes using only 1,870 annotated sample points and maintained over 90% of accuracy even with 187 samples or 10% of total samples, significantly outperform comparison models. This methodology allows for detailed and efficient monitoring of POS and enhances management strategies with minimal resources and effort. Our research opens new pathways for precise urban landscape study.

Analysis of Heartbeat Signals to Detect Sleep Disorders Using Artificial Neural Network Methods

A human sleep disorder detection system has been designed using an AD8232 Electrocardiogram sensor module integrated with a microcontroller and internet connection through ESP 32. The heartbeat signals from the sensor are analyzed using Artificial Neural Network (ANN) methods to determine normal conditions, Obstructive Sleep Apnea (OSA), or Central Sleep Apnea (CSA). The sensor's accuracy was measured over 10 measurements, resulting in 96.85%. Testing with 30 training data samples achieved an accuracy of 93.33%, and testing with 20 training data samples achieved an accuracy of 80%. The system displays output values through the Internet of Things (IoT) with an average computation time of around 7.6 ms.

Prediction of Financial Distress in Transportation and Logistics Companies before, During and after the Covid-19 Pandemic Listed on the Indonesia Stock Exchange

Purpose: This study aims to predict financial distress in transportation and logistics companies before, during, and after the Covid-19 pandemic. Research Methodology: The research subjects were 23 companies, and the study utilized an artificial neural network model. This study uses financial ratios, including the debt-to-asset ratio (DAR), Current Ratio (CR), and Return on Assets (ROA) as input variables in the artificial neural network architecture. The objectives of this study are to calculate the three ratios used as test data, determine the differences in financial ratios between companies reported as financially distressed and those not experiencing financial distress in the training data, identify the artificial neural network model architecture that produces good performance on the training data sample for use in testing predictions, and predict financial distress using an artificial neural network on transportation and logistics companies listed on the Indonesia Stock Exchange, which are part of the research sample. The research sample consisted of 20 transportation and logistics companies listed on the Indonesia Stock Exchange from 2015 to 2019. Results: The results reveal that companies reported as financially distressed have lower average values for the three ratios compared to companies not experiencing financial distress, making them suitable input variables. The best artificial neural network architecture in this study included an input layer with 60 neurons, a hidden layer with 15 neurons, and an output layer with one neuron. This architecture achieved a training performance mean square error (MSE) of 0.125004 and an R value of 50.00%. The study's findings suggest that 12 companies are predicted to experience financial distress.

Zero-shot classification of small target on sea bottom using model-agnostic meta-learning.

A model-agnostic meta-learning (MAML)-based active target classifier to identify small targets (e.g., mines) on the sea bottom in different ocean environments from those present in the training data is proposed. To better classify the targets deviating from those in the training set, MAML is applied to the out-of-distribution samples. Frequency-domain target and clutter scattering signals from various tasks with varying bottom types (silt/clay) and incident angles (low/moderate/high) are utilized as training data samples. MAML significantly outperforms conventional neural networks during the test. The improved generalization of MAML is explained using loss landscape in the form of a smooth convex curve.

A scalable adaptive sampling approach for surrogate modeling of rigid pavements using machine learning

Rigid pavement design is a high-dimensional optimization problem, involving several variables and design considerations. The existing machine learning (ML) design models are either low-dimensional or less accurate than computational simulations. Surrogate modeling is a powerful tool for approximating the results of high-fidelity computational simulations, which is commonly used to approximate the results of these simulations by sampling their solutions to train models. However, conventional and adaptive sampling methods face the challenge of “curse of dimensionality”, due to the exponential increase in required sample size and computations with the number of variables. To overcome this, we propose a scalable adaptive sampling (SAS) method that uses random samples for testing the model's performance but generates new training samples as a subset of a full factorial design of experiments (DoE). The factorial level increases with each iteration, allowing the algorithm to sample training data at a progressively finer scale, and updating the ML models each time adaptively. The proposed method was tested on several 2D benchmark examples, as well as practical pavement design problems. Our results demonstrate that the proposed technique can develop accurate surrogate models for both low- and high-dimensional inference spaces. For a 4D inference space, the surrogate models derived from the proposed method had an order of magnitude lower error than those derived from “one-shot” conventional sampling with the same sample size. For a 6D inference space, the sample size required by the proposed method was only 5 % of that required by conventional sampling for comparable performance.

The Method of Restoring Lost Information from Sensors Based on Auto-Associative Neural Networks

The research aims to develop a neural network-based lost information restoration method when the complex nonlinear technical object (using the example of helicopter turboshaft engines) sensors fail during operation. The basis of the research is an auto-associative neural network (autoencoder), which makes it possible to restore lost information due to the sensor failure with an accuracy of more than 99%. An auto-associative neural network (autoencoder)-modified training method is proposed. It uses regularization coefficients that consist of the loss function to create a more stable and common model. It works well on the training sample of data and can produce good results on new data. Also, it reduces its overtraining risk when it adapts too much to the training data sample and loses its ability to generalize new data. This is especially important for small amounts of data or complex models. It has been determined based on the computational experiment results (the example of the TV3-117 turboshaft engine) that lost information restoration based on an auto-associative neural network provides a data restoring error of no more than 0.45% in the case of single failures and no more than 0.6% in case of double failures of the engine parameter registration sensor event.

Imbalanced generative sampling of training data for improving quality of machine learning model

Design exploration in engineering applications often requires a meticulous experimental or numerical study to evaluate performance (Y) of each design, which may require great effort, time or resources. Reducing the number of these tests for finding a good design is of paramount importance in all engineering fields. This study aims at computing a machine learning (ML) model using less number of designs as training data. Uniform sampling (US) in the design space (based on predefined design parameters) to obtain a training data is a promising approach. We further extend this sampling concept to obtain designs in the design space by also employing the ML model.The designs are selected via two non-uniform (imbalanced) sampling methods (namely, height-based sampling - HBS and gradient-based sampling - GBS) while considering their Y and gradient, dY, values. These values are divided into uniform intervals, and we aim at equalizing the number of designs in the training data at each interval as much as possible. This can force designs to have minimum or maximum Y or dY values, which, in fact, lie on small portion of the design space, in general. Therefore, capturing designs from all design space portions can be enabled.Results of the proposed methods are compared against US along with two well studied non-uniform sampling strategies, Stratified Over Sampling (SOS) and Gaussian-Process Based Sampling (GPBS). To reliably investigate quality of ML models obtained using designs sampled via US, SOS, GPBS, HBS and GBS, we utilize standard test (known) functions (such as Easom and Beale) as substitutes for engineering problems. According to the results presented, ML models using HBS and GBS have either better prediction accuracy or wider applicability compared to all other tested sampling methods.

Bias Detection and Mitigation in AI-Driven Target Marketing: Exploring Fairness in Automated Consumer Profiling

In this work, bias identification and mitigation in AI-driven target marketing are examined with an emphasis on guaranteeing fairness in automated consumer profiling. Significant biases in AI models were found by preliminary investigation, especially impacted by characteristics like purchasing history and geographic location, which closely correspond with protected characteristics like race and socioeconomic position. With a Disparate Impact (DI) of 0.60, a Statistical Parity Difference (SPD) of -0.25, and an Equal Opportunity Difference (EOD) of -0.30, the fairness measures computed for the original models revealed significant biases against certain population groups. We used three main mitigating strategies: pre-processing, in-processing, and post-processing, to counteract these biases. Re- sampling and balancing of training data as part of pre- processing raised the DI to 0.85, SPD to -0.10, and EOD to -0.15. The measures were much better with in- processing, which adds fairness restrictions straight into the learning algorithms, with a DI of 0.90, an SPD of -0.05, and an EOD of -0.10. The most successful were post- processing modifications, which changed model outputs to guarantee fairness; they produced a DI of 0.95, an SPD of -0.02, and an EOD of -0.05. These results support the research already in publication and demonstrate that bias in AI is a complicated and enduring problem that calls for a multidimensional strategy. The paper highlights how crucial ongoing audits, openness, and multidisciplinary cooperation are to reducing prejudice. Marketers, AI practitioners, and legislators will find the ramifications profound, which emphasizes the requirement of moral AI methods to preserve customer confidence and follow laws. This approach advances the larger discussion on AI ethics, promotes justice, and reduces prejudice in AI- driven marketing systems.

A new hybrid reduced order modeling for parametrized Navier–Stokes equations in stream-vorticity formulation

In the context of traditional reduced order modeling methods (ROMs), time and parameter extrapolation tasks remain a formidable challenge. To this end, we propose a hybrid projection/data-driven framework that leverages two subspaces to improve the prediction accuracy of traditional ROMs. We first obtain inaccurate mode coefficients from traditional ROMs in the reduced order subspace. Then, in the prior dimensionality reduced subspace, we correct the inaccurate mode coefficients and restore the discarded mode coefficients through neural network. Finally, we approximate the solutions with these mode coefficients in the prior dimensionality reduced subspace. To reduce the computational cost during the offline training stage, we propose a training data sampling strategy based on dynamic mode decomposition (DMD). The effectiveness of the proposed method is investigated with the parameterized Navier–Stokes equations in stream-vorticity formulation. In addition, two additional time extrapolation methods based on DMD are also proposed and compared.

Hybridizing traditional and next-generation reservoir computing to accurately and efficiently forecast dynamical systems.

Reservoir computers (RCs) are powerful machine learning architectures for time series prediction. Recently, next generation reservoir computers (NGRCs) have been introduced, offering distinct advantages over RCs, such as reduced computational expense and lower training data requirements. However, NGRCs have their own practical difficulties, including sensitivity to sampling time and type of nonlinearities in the data. Here, we introduce a hybrid RC-NGRC approach for time series forecasting of dynamical systems. We show that our hybrid approach can produce accurate short-term predictions and capture the long-term statistics of chaotic dynamical systems in situations where the RC and NGRC components alone are insufficient, e.g., due to constraints from limited computational resources, sub-optimal hyperparameters, sparsely sampled training data, etc. Under these conditions, we show for multiple model chaotic systems that the hybrid RC-NGRC method with a small reservoir can achieve prediction performance approaching that of a traditional RC with a much larger reservoir, illustrating that the hybrid approach can offer significant gains in computational efficiency over traditional RCs while simultaneously addressing some of the limitations of NGRCs. Our results suggest that the hybrid RC-NGRC approach may be particularly beneficial in cases when computational efficiency is a high priority and an NGRC alone is not adequate.

Temperature prediction based on long short‐term memory convolutional neural network Bragg grating sensing

AbstractTo address the constraints associated with conventional fitting techniques for temperature demodulation in the context of subway tunnel fires, a new method of demodulation grating sensing spectrum using long short‐term memory convolutional neural network (LSTM‐CNN) is proposed in this paper. Build the monitoring platform based on LSTM‐CNN ultra‐weak fiber grating temperature measurement system, predict its sensing signals by LSTM‐CNN algorithm, select 18000 spectra as sample data for training, use AdamW stochastic optimization algorithm for training, and carry out the temperature calibration and demodulation error analysis of the Fiber Bragg Grating within the temperature range of 25–75°C. Compared with GRU algorithm, LSTM algorithm and traditional maximum peak method, the algorithm of this paper is good and can effectively improve the measurement accuracy, the experimental results show that: the demodulation accuracy of temperature wavelength prediction in this paper can be up to 99.27%, and the root mean square deviation is 0.08528°C, through the experiments, it is verified that the method proposed in this paper has a certain reference and support in terms of theories and technology significance. It is suitable for the identification and monitoring of fire hazards in underground tunnels, and also has application value in the signal processing of grating array sensing demodulation system.

Neural network for predicting operating modes and calculating energy characteristics of steam turbine plants

The article describes the results of work on the creation of neural networks calculating the technical and economic parameters of all possible modes of operation of a thermal turbine unit with a steam turbine of type PT-60-130/13. In accordance with the objectives set, recommendations for the preparation of training data samples are formulated. The input and output parameters of the condensing and heating modes of operation of the steam turbine are determined. The results of research on determining the most optimal architecture of neural networks for calculating the energy characteristics of steam turbine plants of the heating type are presented. The results of calculations of the mean squared errors of neural network predictions from the results of calculations performed using a verified object-oriented model of a PT-60-130/13 turbine unit are tabulated.Graphs of the dependence of the specific heat consumption for the generation of electrical energy on the power of a PT-60-130/13 turbine unit for condensation and heating modes of operation using neural networks are plotted. The conclusion is formulated about the possibility of using neural networks for the development of energy characteristics and regulatory documentation on fuel use of equipment of thermal power plants.

Semi-supervised process monitoring based on self-training PCA model

Data-centric engineering has recently become as a hot spot in both areas of artificial intelligence and data science, which shifts the focus of engineering application from model to data itself. The model for data-driven process monitoring is usually developed upon a large number of normal data samples, which have been assumed to be well evaluated in advance. However, the actual quality of those training data samples has a great impact to the monitoring performance. If those samples with low quality are included for modeling, the monitoring performance could be severely deteriorated. In practice, due to expensive and time-consuming data quality evaluation procedures, we may only have a quite limited number of evaluated data samples, and hold a large number of unevaluated data samples. This paper aims to develop a semi-supervised monitoring model, which can simultaneously incorporate the evaluated dataset and the original unevaluated dataset. As a result, the new semi-supervised monitoring scheme can save a lot of human efforts, and could be particularly useful in those processes which can only provide a small portion of evaluated data samples in time. The feasibility and efficiency of the new proposed monitoring scheme are examined through case studies of a numerical example and the TE benchmark process.

Predicting wetland soil properties using machine learning, geophysics, and soil measurement data

PurposeMachine learning models can improve the prediction of spatial variation of wetland soil properties, such as soil moisture content (SMC) and soil organic matter (SOM). Their performance, however, relies on the quantity of data used to train the model, limiting their use with insufficient data. In this study, we assessed the use of synthetic data constrained by limited field data for training an eXtreme Gradient Boosting (XGBoost) algorithm used to predict the distribution of soil properties based on geophysical measurements constrained by soil samples.Materials and methodsA spatial distribution of soil apparent electrical conductivity (ECa) and laboratory measurements of SOM and SMC from twenty-two core samples were acquired at the St. Michael restored wetland near Defiance, Ohio. The correlations between ECa, SOM, and SMC were explored for predicting the spatial distribution of SOM and SMC. We used a Beta Variational AutoEncoder (β-VAE) approach to synthetically generate over 70,000 training data from the original twenty-two data from soil cores. The training data samples were taken from the latent space. The XGBoost algorithm was then trained on the β-VAE generated data and used to predict the spatial distribution of SOM and SMC at the site. We also validated the accuracy of the XGBoost predictions using an original holdout model validation technique.Results and discussionsThe generated synthetic data using the β-VAE include both soil attributes and ECa, which are larger and more diverse than the original training set with an absolute mean reconstructed error for SMC and SOM ranging from 0.018 to 0.022 and 0.026 to 0.041, respectively. This indicates that the β-VAE successfully generated a realistic synthetic dataset and overcame the technical barrier of using limited datasets. In addition, using generated data to expand the original training data helps the XGBoost model make more accurate predictions compared to training on the original data. The XGBoost prediction performance yielded average Lin’s concordance correlation coefficient (LCCC) values of 0.82 and 0.85 for SOM and SMC and a ratio of performance to deviation (RPD) values of 1.92 and 2.22 respectively, indicating a good performance.ConclusionsThis study validated the use of β-VAE to successfully generate synthetic wetland soil datasets with attributes of the original field data that can be effectively used to train the machine learning XGBoost model. The proposed framework offers an efficient solution for mapping the spatial variability of soil property in data-scarce wetland soil environments.