SEARCH · Engineering Papers
Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.
Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.
Techniques for solving the feature selection problem are presented. Topics discussed include the reduction of the number of variables in "best b", and the iterative selection of H sub i.
Nuclear power plants collect and store large volumes of heterogeneous data from various components and systems. With recent advances in machine learning (ML) techniques, these data can be leveraged to develop diagnostic and short-term forecasting models to better predict future equipment condition. Maintenance operations can then be planned in advance whenever degraded performance is predicted, thus resulting in fewer unplanned outages and the optimization of maintenance activities. This enables lower maintenance costs and improves the overall economics of nuclear power. This paper focuses on developing a short-term forecasting process that leverages a feature selection process to distill large volumes of heterogeneous data and predict specific equipment parameters. A variety of feature selection methods, including Shapley Additive Explanations (SHAP) and variance inflation factor (VIF), were used to select the optimal features as inputs for three ML methods: long short-term memory (LSTM) networks, support vector regression (SVR), and random forest (RF). Each combination of model and input features was used to predict a pump bearing temperature both 1 and 24 hours in advance, based on actual plant system data. The optimal inputs for the LSTM and SVR were selected using the SHAP values, while the optimal input for the RF consisted solely of the response variable itself. Each model produced similar 1-hour-ahead predictions, with root mean square errors (RMSEs) of roughly 0.006. For the 24-hour-ahead predictions, differences could be seen between LSTM, SVR, and RF, as reflected by model performances of 0.036 +- 0.014, 0.0026 +- 0, and 0.063 +- 0.004 RMSE, respectively. As big data and continuous online monitoring become more widely available, the proposed feature selection process can be used for many applications beyond the prediction of process parameters within nuclear infrastructure.
Distance measures which are useful for feature selection are considered, giving attention to the divergence distance measure and the Jeffreys-Matusita (JM) distance. Experimental studies show that the JM-distance yields more reliable results than other distance measures. A number of questions which are not solved by the experiments are discussed. The investigation provides an explanation for previous observations that the JM-distance and a saturating transform of divergence are highly useful for feature selection in the multiclass case.
A study area near Ribeirao Preto in Sao Paulo state was selected, with predominance in sugar cane. Eight features were extracted from the 4 original bands of LANDSAT image, using low-pass and high-pass filtering to obtain spatial features. There were 5 training sites in order to acquire the necessary parameters. Two groups of four channels were selected from 12 channels using JM-distance and entropy criterions. The number of selected channels was defined by physical restrictions of the image analyzer and computacional costs. The evaluation was performed by extracting the confusion matrix for training and tests areas, with a maximum likelihood classifier, and by defining performance indexes based on those matrixes for each group of channels. Results show that in spatial features and supervised classification, the entropy criterion is better in the sense that allows a more accurate and generalized definition of class signature. On the other hand, JM-distance criterion strongly reduces the misclassification within training areas.
This paper presents the development of a new entropy-based feature selection method for identifying and quantifying impacts. Here, impacts are defined as statistically significant differences in spatio-temporal fields when comparing datasets with and without an external forcing in an Earth system model. Temporal feature selection is performed by first computing the cross-fuzzy entropy to quantify similarity of patterns between two datasets and then applying changepoint detection to identify regions of statistically constant entropy. The method is used to capture temperate north surface cooling from a 9-member simulation ensemble of the Mt. Pinatubo volcanic eruption, which injected 10 Tg of SO 2 into the stratosphere. The results estimate a mean difference decrease in near surface air temperature of -0.560 K with a 99% confidence interval between -0.864 K and -0.257 K between April and November of 1992, one year following the eruption. A sensitivity analysis with decreasing SO 2 injection revealed that the impact is statistically significant at 5 Tg but not at 3 Tg. Using identified features, a dependency graph model based on a 9-day lag had significantly fewer nodes than a graph based on monthly means. Furthermore, this demonstrates our method’s ability to perform dimension reduction while still uncovering source-to-impact pathways.
Orthogonal transforms and feature selection using pattern, Harr, Walsh, Fourier and Karhunen- Loeve spaces for providing efficient classification algorithms in pattern recognition
The problem dealt with concerns feature selection or reducing the dimension of the data to be processed from n to k. By reducing the dimension of the data from n to k, classification time is generally reduced. Yet the dimension reduction should not be so great that classification accuracy is impaired. Thus, the general problem is considered of classifying an n-dimensional observation vector x into one of m-distinct classes where each class is normally distributed with mean and covariance. It is shown that the probability of misclassification is minimized if a maximum likelihood classification procedure is used to classify the data. The dimension of each observation vector to be processed is conveniently reduced by performing the transformation y = Bx, where B is a K by n matrix of rank k. Thus, the n-dimensional classification problem transforms into a k-dimensional classification problem.
Recent years have seen rapid growth in high-dimensional datasets. Most existing machine learning (ML) algorithms fail in high-dimensional settings where many features could be redundant. A critical process of feature selection is thus applied in such a setting that helps in identifying the most relevant features while removing redundant ones. With the increase in high dimensionality, one is also faced with problems of efficiency and interpretation in performing such selection methods. Therefore, this paper proposes a “novel” feature selection framework that uses an ensemble of interpretable ML algorithms to perform feature selection and the ranking of final features. Finally, this framework is applied to a gene expression dataset obtained through collaboration with the National Aeronautics and Space Administration (NASA)’s Biological and Physical Sciences (BPS) team and helps identify important and relevant genes contributing to specific target attributes through classification tasks.
Feature selection software was developed at the Earth Resources Laboratory that is capable of inputting up to 36 channels and selecting channel subsets according to several criteria based on divergence. One of the criterion used is compatible with the table look-up classifier requirements. The software indicates which channel subset best separates (based on average divergence) each class from all other classes. The software employs an exhaustive search technique, and computer time is not prohibitive. A typical task to select the best 4 of 22 channels for 12 classes takes 9 minutes on a Univac 1108 computer.
Walsh functions in image processing, rotational feature selection and pattern recognition, defining set of orthogonal transformations
Mass spectrometry (MS) promises to be a powerful tool for potential biosignature detection during astrobiological missions on ocean worlds in our solar system. Accurate and generalizable machine learning methods could enhance science return on investment by predicting seawater chemistry and classifying isotopic biosignatures, either as a signature consistent with microbial life (biotic) or as a novelty (unclassified/unique). However, machine learning models are likely to be complex and involve interactions between MS features, making biosignatures difficult to interpret. Feature selection methods provide biological and chemical context that help interpret the mechanisms of machine learning models, but these methods also need the ability to detect complex interactions. Previously, we developed a machine learning feature selection algorithm called nearest-neighbor projected distance regression (NPDR) that has the ability to identify important model features that involve complex interactions and automatically reduce correlation and the dimensionality in a high-dimensional variable space. The standard distance metrics used in NPDR – Manhattan and Euclidean – assume the multivariate data are isotropic, which is often violated in real data due to differences in the covariance between variables. Thus, we extend NPDR to include a random forest distance, and other anisotropic distance metrics, for computing nearest neighbors. We also augment the isotope-ratio MS data with time-series features from the raw MS signal to improve biotic classification. We test NPDR on our novel experimental ocean world seawater analog MS data. We measure isotope fractionations of volatile CO 2 that could be measured in exospheres or plumes. Samples include baseline abiotic conditions using a range of possible seawater chemistry consistent with Europa and Enceladus, and biotic samples that include microbes in these seawaters. We use penalized NPDR with random forest proximity to identify interpretable microbial molecular signatures. We compare features with random forest importance, and we train a classifier that discriminates between biotic and abiotic samples with high accuracy. These ML-trained ocean-world analog MS data could be used to assist in identifying biosignatures during future missions.
Because of the continually changing environment of a space station, visual feedback is a vital element of a telerobotic system. A real time visual servoing system would allow a telerobot to track and manipulate randomly moving objects. Methodologies for the automatic selection of image features to be used to visually control the relative position between an eye-in-hand telerobot and a known object are devised. A weighted criteria function with both image recognition and control components is used to select the combination of image features which provides the best control. Simulation and experimental results of a PUMA robot arm visually tracking a randomly moving carburetor gasket with a visual update time of 70 milliseconds are discussed.
Three gaps impede the development of cost-effective and accurate data-driven building energy modeling/models (DBEM) for energy forecasting and predictive control strategies. Gap 1: data bias is common in building operation data, but this topic is hardly studied in DBEM; Gap 2: high data dimensionality is common in DBEM, but a systematic and scalable methodology is lacking to solve the problem; Gap 3: the interactions between data bias and high dimensionality have not been systematically studied for DBEM and predictive control in buildings. In this work, to address the three gaps mentioned above, we develop a framework that integrates active learning and feature selection for DBEM used for whole building data predictive control (or DPC, which is a branch of model predictive control). The framework provides a systematic methodology and automatic workflow that starts with raw data from building automation systems to the establishment of data-driven energy models for DPC controllers. The developed strategies and framework are evaluated in a virtual testbed based on EnergyPlus and BCVTB. Improved performance and reduced computational complexity are observed from the DBEM built with the developed framework, as well as the DPC controller based on that DBEM, indicating the effectiveness of the developed framework.
A criterion for linear feature selection is proposed which is based on mean square apporximation of class density functions. It is shown that for the widest possible class of approximants, the criterion reduces to Devijver's Bayesian distance. For linear approximants the criterion is equivalent to well known generalized Fisher criteria.
Feature selection and the information content of Thematic Mapper Simulator (TMS) data are investigated for a forested region in northern Idaho. The optimal TMS channels for forest structural characteristics are determined, and the capability of TMS data to describe the structural variability within a forest stand is evaluated. The comparative performance of TMS and MSS data to discriminate forest structural factors using per-pixel maximum likelihood classification is examined, and four optimal TMS channels are classified in order to ascertain if the full complement of TM channels provide higher accuracies than the four optimal ones.
Accurate estimation of the State of Health (SOH) for second-life batteries (SLBs) is crucial given their increasing use in energy storage applications. Precise SOH prediction is essential for safe operation and robust battery management systems. A major challenge is the limited availability of datasets for building reliable degradation models. To address this, synthetic data generation through linear interpolation is performed to extend the available data, making it more representative of real-world battery operating conditions. By analyzing feature correlation with SOH, the most relevant features are selected for the model. The proposed approach employs a convolutional neural network (CNN) model trained on this interpolated, feature-selected dataset, using time series data of voltage, temperature, and current over a cycle. By focusing on highly correlated features, the model achieves over 95% accuracy, with mean absolute error and root mean squared error up to 2.27% and 2.64%, respectively, in SOH estimation for two battery datasets tested. These results highlight the potential of combining synthetic data generation and feature selection to enhance SOH predictions, showcasing the superior performance of the proposed CNN model for both new batteries and SLBs.
Results that suggest the possibility of using a sequential monotone process for solving the feature selection problem using Householder transformations are applied to the divergence separability criterion and an expression for the gradient of the divergence with respect to the generator of a single Householder transformation will be developed. This expression for the gradient is used in any number of differential correction schemes (iterators) that attempt to extremize the divergence. Data sets provided by the Earth Observations Division-JSC are used to demonstrate selecting the Householder transformations that generate the kxn matrix defining the best (in the sense of extremizing the divergence) k linear combinations of features. The tests allow initial comparisons to be made with results. In particular, this technique does not appear to require initial guesses for the iterator to be generated without replacement, exhaustive search, or other similar schemes.
The computational procedure and associated computer program for a linear feature selection technique are presented. The technique assumes that: a finite number, m, of classes exists; each class is described by an n-dimensional multivariate normal density function of its measurement vectors; the mean vector and covariance matrix for each density function are known (or can be estimated); and the a priori probability for each class is known. The technique produces a single linear combination of the original measurements which minimizes the one-dimensional probability of misclassification defined by the transformed densities.