Search NASA⌕ Search

SEARCH · Search NASA

Results for “convolution”

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.

At least 1,099 records · Page 61

Deep Neural Network Algorithm for CMC Microstructure Characterization and Variability Quantification

Microstructure characterization and variability quantification are crucial for understanding ceramic matrix composites (CMCs) mechanical behavior and deformation mechanisms across length scales. Traditionally, analyses of the micrographs obtained from microscopy are labor-intensive. However, with the vast improvement in computer vision (CV) and deep learning (DL), an automated algorithm can be designed to extract essential microstructure variability from micrographs which can then be used to construct a statistically representative volume element (SRVE). The DL-based algorithm spans the taxonomy of microstructure analyses, including semantic segmentation of microstructure constituents, secondary phases, matrix/fiber interface, and defects, and quantifying the microstructure variability in terms of probability distributions. In this work, C/SiNC and SiC/SiNC CMCs microstructures are semantically segmented through a deep convolutional neural network, followed by variability quantification through the implementation of a fully connected regression layer, hence forming a deep regression network. The deep regression network operates in a feedforward regime, in which the neuron output signal traverses through the network in a unidirectional manner. The weight tensor associated with each layer is updated through a backpropagation stochastic gradient descent approach. The input gray-scale image obtained through in-house scanning electron microscope and confocal microscope micrographs is augmented through affine transformations to increase the training set size, which is then processed through four strided convolutional layers. This compresses the image resolution by half at each layer while increasing the image depth by applying different filters (image encoding). The class activation maps (CAMs) corresponding to the applied filters highlight the key architectural features and assist with the semantic segmentation of the microstructure.

Hamza, Mohamed H.↗

Inverse Analysis with Variational Autoencoders: A Comparison of Shallow and Deep Networks

Inverse problems are applied to determine unknown properties by matching observational data with a physical model that often takes many parameters as input. To overcome the underconstrained nature of inverse problems and achieve good performance, an approach is presented involving regularization with a technique known as a variational autoencoder (VAE), which is trained to map a high-dimensional parameter space with a complex structure to a low-dimensional latent space with a simple structure. We apply this approach to unconditioned realizations of the parameters (heterogeneous hydraulic fields) for a hydrogeological inverse problem. Two types of hydraulic conductivity fields are used to evaluate the characterization for the different levels of heterogeneity complexity of the physical inputs. This approach keeps the computational cost of generating the training data low. The reason is unconditioned realizations neither rely on the observational data used to perform the inverse analysis nor require any groundwater flow model forward runs. In addition, this approach applies regularization on a low-dimensional latent space from the VAE and increases optimization efficiency through automatic differentiation. Furthermore, two different neural network (NN) structures are tested for their utility in using the VAE for inverse analysis. The performance of a deep, convolutional neural network strongly depends on the dimensionality of the latent space, which requires tuning. In contrast, a shallow, dense neural network provides consistently accurate characterization without tuning. Furthermore, our approach evaluates the advantages of the shallow, dense neural network over the deep, convolutional one and enables future application to a wide range of inverse problems.

97 MATHEMATICS AND COMPUTING↗

Applying Machine Learning and Bayesian Inference to Identify and Locate Moving Anthropogenic Sources Using Distributed Acoustic Sensing Data

Distributed acoustic sensing (DAS) systems, which use existing telecommunication fibers, offer high‐resolution capabilities ideal for recording anthropogenic sources. However, the complexity of urban environments and the large amount of data recorded by DAS require automated methods to efficiently detect and categorize anthropogenic sources. Here, we evaluate how well three machine learning models (k‐nearest neighbor [k‐NN], convolutional neural networks, and recurrent‐convolutional neural networks) can identify various anthropogenic sources recorded by DAS. Our findings reveal that both k‐NN and neural network methods perform well in high signal‐to‐noise ratio (SNR) settings. However, their accuracy decreases at SNRs <4. We also use Kalman filtering, a form of Bayesian inference, on backprojected locations of these sources to recover locations that generally fall within standard smartphone Global Positioning System errors. By combining machine learning and Kalman filter results, we calculate a multidimensional model of moving anthropogenic sources. These results demonstrate the potential of DAS data in urban seismology for accurately identifying and locating such sources. Depending on the research objectives, these sources can be further studied or filtered out to improve the quality of seismic data for earthquake studies. Such methods provide a valuable tool for urban seismology and seismic hazard analysis.

Luckie, Thomas William [Sandia National Laboratori↗

Kimberlina 1.2 Velocity Models and Seismic Data

Kimberlina 1.2 Velocity model and synthetic seismic data, produced in collaboration of teams at the National Energy Technology Laboratory, Los Alamos National Laboratory, and Lawrence Livermore National Laboratory through the National Risk Assessment Partnership. Data is associated with the following publication: Zheng Zhou, Youzuo Lin, Zhongping Zhang, Yue Wu, Zan Wang, Robert Dilmore, and George Guthrie, "A Data-Driven CO2 Leakage Detection Using Seismic Data and Spatial-Temporal Densely Connected Convolutional Neural Networks," International Journal of Greenhouse Gas Control, Vol 90, 2019. The Kimberlina 1.2 Velocity models were produced by Zan Wang, Robert Dilmore, William Harbert, and Lianjie Huang at NETL. The following citations are directly related to the creation of the velocity models: Wang, Z. Harbert, W., Dilmore, R., Huang, L. Modeling of time-lapse seismic monitoring using CO2 leakage simulations for a model CO2 storage site with realistic geology: Application in assessment of early leak-detection capabilities. International Journal of Greenhouse Gas Control. V. 76, September 2018, Pages 39-52. https://doi.org/10.1016/j.ijggc.2018.06.011 Wang, Z., Dilmore, R., Harbert, W. Inferring CO2 saturation from synthetic surface seismic and downhole monitoring data using machine learning for leakage detection at CO2 sequestration sites. International Journal of Greenhouse Gas Control, V. 100, September 2020. https://doi.org/10.1016/j.ijggc.2020.103115 The velocity models were built based on the Kimberlina 1.2 aquifer impact data which is associated with the following publications: Buscheck, T.A., Mansoor, K., Yang, X., Wainwright, H., and Carroll, S. (2019). Downhole pressure and chemical monitoring for CO2 and brine leak detection in aquifers above a CO2 storage reservoir. International Journal of Greenhouse Gas Control. 91. 102812. 10.1016/j.ijggc.2019.102812. Xianjin Yang, Thomas A. Buscheck, Kayyum Mansoor, Zan Wang, Kai Gao, Lianjie Huang, Delphine Appriou, Susan A. Carroll, Assessment of geophysical monitoring methods for detection of brine and CO2 leakage in drinking water aquifers, International Journal of Greenhouse Gas Control, Volume 90, 2019, 102803, ISSN 1750-5836, https://doi.org/10.1016/j.ijggc.2019.102803 The synthetic seismic data was produced by Youzuo Lin and team at LANL, and are associated with the following citations: Jordan, P. D., and J. L. Wagoner. Characterizing Construction of Existing Wells to a CO2 Storage Target: The Kimberlina Site, California. Zheng Zhou, Youzuo Lin, Zhongping Zhang, Yue Wu, Zan Wang, Robert Dilmore, and George Guthrie, "A Data-Driven CO2 Leakage Detection Using Seismic Data and Spatial-Temporal Densely Connected Convolutional Neural Networks," International Journal of Greenhouse Gas Control, Vol 90, 2019.

CO2 Leakage↗

Digital twin framework for PIP-II linac: AI-driven multi-scale modeling from ion source to 800 MeV

The PIP-II superconducting linac at Fermilab is designed to deliver multi-megawatt proton beams for neutrino physics and other high-intensity applications. To expedite commissioning and enhance operational reliability, we have developed an EPICS-based data flow framework that seamlessly integrates digital twins (DT) with physical twins (PT). These digital twins comprise high-fidelity beam dynamics models or data-driven surrogate models connected to their physical counterparts through real-time diagnostics and advanced machine-learning algorithms.Central to this framework is Linac_Gen, an accelerated simulation tool that incorporates convolutional neural networks, random forests, and genetic algorithms to provide up to a tenfold speedup in optimizing the accelerator geometry model. An EPICS translator layer ensures interoperability by efficiently mapping lattice parameters across diverse simulation platforms.Our EPICS-based framework supports multiple operational modes—monitoring, passive learning, closed-loop control, and online learning—covering the entire machine lifecycle. By leveraging HPC resources and multi-objective optimization techniques, the digital twin enables adaptive trajectory correction, real-time fault detection, and predictive modeling of beam stability. This comprehensive approach paves the way for robust, high-intensity operation and data-driven accelerator R&D at Fermilab.

Pathak, Abhishek [Fermilab]↗

Hybrid Enrichment Assay Methods for a UF6 Cylinder Verification Station: FY10 Progress Report

Pacific Northwest National Laboratory (PNNL) is developing the concept of an automated UF6 cylinder verification station that would be located at key measurement points to positively identify each cylinder, measure its mass and enrichment, store the collected data in a secure database, and maintain continuity of knowledge on measured cylinders until the arrival of International Atomic Energy Agency (IAEA) inspectors. At the center of this unattended system is a hybrid enrichment assay technique that combines the traditional enrichment-meter method (based on the 186 keV peak from 235U) with non-traditional neutron-induced high-energy gamma-ray signatures (spawned primarily by 234U alpha emissions and 19F(alpha, neutron) reactions). Previous work by PNNL provided proof-of-principle for the non-traditional signatures to support accurate, full-volume interrogation of the cylinder enrichment, thereby reducing the systematic uncertainties in enrichment assay due to UF6 heterogeneity and providing greater sensitivity to material substitution scenarios. The work described here builds on that preliminary evaluation of the non-traditional signatures, but focuses on a prototype field system utilizing NaI(Tl) and LaBr3(Ce) spectrometers, and enrichment analysis algorithms that integrate the traditional and non-traditional signatures. Results for the assay of Type-30B cylinders ranging from 0.2 to 4.95 wt% 235U, at an AREVA fuel fabrication plant in Richland, WA, are described for the following enrichment analysis methods: 1) traditional enrichment meter signature (186 keV peak) as calculated using a square-wave convolute (SWC) algorithm; 2) non-traditional high-energy gamma-ray signature that provides neutron detection without neutron detectors and 3) hybrid algorithm that merges the traditional and non-traditional signatures. Uncertainties for each method, relative to the declared enrichment for each cylinder, are calculated and compared to the uncertainties from an attended HPGe verification station at AREVA, and the IAEA’s uncertainty target values for feed, tail and product cylinders. A summary of the major findings from the field measurements and subsequent analysis follows: • Traditional enrichment-meter assay using specially collimated NaI spectrometers and a Square-Wave-Convolute algorithm can achieve uncertainties comparable to HPGe and LaBr for product, natural and depleted cylinders. • Non-traditional signatures measured using NaI spectrometers enable interrogation of the entire cylinder volume and accurate measurement of absolute 235U mass in product, natural and depleted cylinders. • A hybrid enrichment assay method can achieve lower uncertainties than either the traditional or non-traditional methods acting independently because there is a low degree of correlation in the systematic errors of the two individual methods (wall thickness variation and 234U/235U variation, respectively). This work has indicated that the hybrid NDA method has the potential to serve as the foundation for an unattended cylinder verification station. When compared to today’s handheld cylinder-verification approach, such a station would have the following advantages: 1) improved enrichment assay accuracy for product, tail and feed cylinders; 2) full-volume assay of absolute 235U mass; 3) assay of minor isotopes (234U and 232U) important to verification of feedstock origin; single instrumentation design for both Type 30B and Type 48 cylinders; and 4) substantial reduction in the inspector manpower associated with cylinder verification.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

Integration of MHD load models with circuit representations the Z generator.

MHD models of imploding loads fielded on the Z accelerator are typically driven by reduced or simplified circuit representations of the generator. The performance of many of the imploding loads is critically dependent on the current and power delivered to them, so may be strongly influenced by the generators response to their implosion. Current losses diagnosed in the transmission lines approaching the load are further known to limit the energy delivery, while exhibiting some load dependence. Through comparing the convolute performance of a wide variety of short pulse Z loads we parameterize a convolute loss resistance applicable between different experiments. We incorporate this, and other current loss terms into a transmission line representation of the Z vacuum section. We then apply this model to study the current delivery to a wide variety of wire array and MagLif style liner loads.

Jennings, Christopher A.↗

Delayed Gamma-Ray Spectroscopy for Non-Destructive Assay of Nuclear Materials

High-­energy, beta-delayed gamma-­ray spectroscopy is a potential, non-­destructive assay techniques for the independent verification of declared quantities of special nuclear materials at key stages of the fuel cycle and for directly assaying nuclear material inventories for spent fuel handling, interim storage, reprocessing facilities, repository sites, and final disposal. Other potential applications include determination of MOX fuel composition, characterization of nuclear waste packages, and challenges in homeland security and arms control verification. Experimental measurements were performed to evaluate fission fragment yields, to test methods for determining isotopic fractions, and to benchmark the modeling code package. Experimental measurement campaigns were carried out at the IAC using a photo-­neutron source and at OSU using a thermal neutron beam from the TRIGA reactor to characterize the emission of high-­energy delayed gamma rays from 235 U, 239 Pu, and 241 Pu targets following neutron induced fission. Data were collected for pure and combined targets for several irradiation/spectroscopy cycle times ranging from 10/10 seconds to 15/30 minutes.The delayed gamma-ray signature of 241 Pu, a significant fissile constituent in spent fuel, was measured and compared to 239 Pu. The 241 Pu/ 239 Pu ratios varied between 0.5 and 1.2 for ten prominent lines in the 2700-­3600 keV energy range. Such significant differences in relative peak intensities make it possible to determine relative fractions of these isotopes in a mixed sample. A method for determining fission product yields by fitting the energy and time dependence of the delayed gamma-­ray emission was developed and demonstrated on a limited 235 U data set. De-­convolution methods for determining fissile fractions were developed and tested on the experimental data. The use of high count-­rate LaBr 3 detectors was investigated as a potential alternative to HPGe detectors. Modeling capabilities were added to an existing framework and codes were adapted as needed for analyzing experiments and assessing application-­specific assay concepts. A de-­convolution analysis of the delayed gamma-­ray response spectra modeled for spent fuel assemblies was performed using the same method that was applied to the experimental spectra.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Scalable Track Detection in SAR CCD Images

Existing methods to detect vehicle tracks in coherent change detection images, a product of combining two synthetic aperture radar images ta ken at different times of the same scene, rely on simple, fast models to label track pixels. These models, however, are often too simple to capture natural track features such as continuity and parallelism. We present a simple convolutional network architecture consisting of a series of 3-by-3 convolutions to detect tracks. The network is trained end-to-end to learn natural track features entirely from data. The network is computationally efficient and improves the F-score on a standard dataset to 0.988, up fr om 0.907 obtained by the current state-of-the-art method.

47 OTHER INSTRUMENTATION↗

Establishing Malware Attribution and Binary Provenance Using Multicompilation Techniques

Malware is a serious problem for computer systems and costs businesses and customers billions of dollars a year in addition to compromising their private information. Detecting malware is particularly difficult because malware source code can be compiled in many different ways and generate many different digital signatures, which causes problems for most anti-malware programs that rely on static signature detection. Our project uses a convolutional neural network to identify malware programs but these require large amounts of data to be effective. Towards that end, we gather thousands of source code files from publicly available programming contest sites and compile them with several different compilers and flags. Building upon current research, we then transform these binary files into image representations and use them to train a long-term recurrent convolutional neural network that will eventually be used to identify how a malware binary was compiled. This information will include the compiler, version of the compiler and the options used in compilation, information which can be critical in determining where a malware program came from and even who authored it.

97 MATHEMATICS AND COMPUTING↗

Seismic classification through sparse filter dictionaries

We tackle a multi-label classi cation problem involving the relation between acoustic- pro le features and the measured seismogram. To isolate components of the seismo- grams unique to each class of acoustic pro le we build dictionaries of convolutional lters. The convolutional- lter dictionaries for the individual classes are then combined into a large dictionary for the entire seismogram set. A given seismogram is classi ed by computing its representation in the large dictionary and then comparing reconstruction accuracy with this representation using each of the sub-dictionaries. The sub-dictionary with the minimal reconstruction error identi es the seismogram class.

58 GEOSCIENCES↗

Automated Threat Recognition For Aviation Security Applications

We have developed a framework for automated threat recognition (ATR) of explosive threat materials for both single-energy and dual-energy X-ray CT systems. Under this framework, two different types of ATR have been developed. The first type of ATR employs supervised machine learning with statistical characterization of target materials for threat identification training. The reliance only on statistical characterization information for threat training uniquely enables this style of ATR to adapt quickly to evolving threats. The second type of ATR employs deep learning through convolutional neural networks. Convolutional neural networks are attractive due to their human-like capacity for learning and strong ability to identify trends and patterns. Although this method is more powerful than the first type, it requires large amounts of training data and therefore is less agile. Each ATR performs threat characterization at the voxel neighborhood level. This approach avoids the use of threat shape as a detection criterion, which is prohibited by DHS and TSA guidelines.

97 MATHEMATICS AND COMPUTING↗

Developing inductively driven diagnostic X-ray sources to enable transformative radiography and diffraction capabilities on Z

Penetrating X-rays are one of the most effective tools for diagnosing high energy density experiments, whether through radiographic imaging or X-ray diffraction. To expand the X-ray diagnostic capabilities at the 26-MA Z Pulsed Power Facility, we have developed a new diagnostic X-ray source called the inductively driven X-pinch (IDXP). This X-ray source is powered by a miniature transmission line that is inductively coupled to fringe magnetic fields in the final power feed. The transmission line redirects a small amount of Zs magnetic energy into a secondary cavity where 150+ kA of current is delivered to a hybrid X-pinch. In this report, we describe the multi-stage development of the IDXP concept through experiments both on Z and in a surrogate setup on the 1 MA Mykonos facility. Initial short-circuit experiments to verify power ow on Z are followed by short-circuit and X-ray source development experiments on Mykonos. The creation of a radiography-quality X-pinch hot spot is verified through a combination of X-ray diode traces, laser shadowgraphy, and source radiography. The success of the IDXP experiments on Mykonos has resulted in the design and fabrication of an IDXP for an upcoming Z experiment that will be the first-ever X-pinch fielded on Z. We have also pursued the development of two additional technologies. First, the extended convolute post (XCP) has been developed as an alternate method for powering diagnostic X-pinches on Z. This concept, which directly couples the current owing in one of the twelve Z convolute posts to an X-pinch, greatly increases the amount of available current relative to an IDXP (900 kA versus 150 kA). Initial short-circuit XCP experiments have demonstrated the efficacy of power ow in this geometry. The second technology pursued here is the inductively driven transmission line (IDTL) current monitor. These low-current IDTLs seek to measure the current in the final power feed with high fidelity. After three generations of development, IDTL current monitors frequently return cleaner current measurements than the standard B-dot sensors that are fielded on Z. This is especially true on high-inductance experiments where the harshest conditions are created in the nal power feed.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Concentric Spherical GNN for 3D Representation Learning

Learning 3D representations that generalize well to arbitrarily oriented inputs is a challenge of practical importance in applications varying from computer vision to physics and chemistry. We propose a novel multi-resolution convolutional architecture for learning over concentric spherical feature maps, of which the single sphere representation is a special case. Our hierarchical architecture is based on alternatively learning to incorporate both intra-sphere and inter-sphere information. We show the applicability of our method for two different types of 3D inputs, mesh objects, which can be regularly sampled, and point clouds, which are irregularly distributed. We also propose an efficient mapping of point clouds to concentric spherical images, thereby bridging spherical convolutions on grids with general point clouds. We demonstrate the effectiveness of our approach in improving state-of-the-art performance on 3D classification tasks with rotated data.

97 MATHEMATICS AND COMPUTING↗

Unifying Combinatorial and Graphical Methods in Artificial Intelligence

Recently, a new graph Laplacian, called the inner product Laplacian, was introduced which generalizes many existing Laplacians, including the normalized and combinatorial Laplacian and their weighted variants. The key observation behind the inner product Laplacian is that by defining appropriate inner product spaces on the vertices and edges, the standard Laplacians can be recovered as Hodge Laplacians over the simplicial complex formed by the edges and vertices. These inner product spaces form a natural way to incorporate non-combinatorial information into the definition of a domain-specific Laplacian. In particular, in contrast to current domain-specific weighting schemes which rely solely on edge weights, information regarding the similarity of non-adjacent vertices and arbitrary pairs of edges can be effectively incorporated into the Laplacian. In order to illustrate this approach we consider the problem of calculating the potential energy of an atomistic configuration using Graph Neural Networks. In comparison with start-of-the-art approaches, such as SchNet, our approach replaces a learned (via auto-encoder) representation of the atom types with an inner product space on atoms based on scientific knowledge (e.g., electronegativity). We will illustrate how this approach captures key chemical properties of the molecules and compare the energy calculations with state-of-the-art neural network approaches. However, to compute the resulting Laplacian involves a mixture of sparse and dense matrix computation and yields a dense matrix as the basis for the graph convolution. This dense convolutional kernel necessitates moving away from the standard message passing framework for graph neural networks and increases the computational cost of applying the kernel. In order to mitigate these costs we investigate means of leveraging the mixed sparse and dense computations to reduce the overall computational cost and how these approaches can be automatically transferred to energy efficient hardware (e.g., field programmable gate arrays (FPGAs)).

97 MATHEMATICS AND COMPUTING↗

Rethinking skip connections in Spiking Neural Networks with Time-To-First-Spike coding

Time-To-First-Spike (TTFS) coding in Spiking Neural Networks (SNNs) offers significant advantages in terms of energy efficiency, closely mimicking the behavior of biological neurons. In this work, we delve into the role of skip connections, a widely used concept in Artificial Neural Networks (ANNs), within the domain of SNNs with TTFS coding. Our focus is on two distinct types of skip connection architectures: (1) addition-based skip connections, and (2) concatenation-based skip connections. We find that addition-based skip connections introduce an additional delay in terms of spike timing. On the other hand, concatenation-based skip connections circumvent this delay but produce time gaps between after-convolution and skip connection paths, thereby restricting the effective mixing of information from these two paths. To mitigate these issues, we propose a novel approach involving a learnable delay for skip connections in the concatenation-based skip connection architecture. This approach successfully bridges the time gap between the convolutional and skip branches, facilitating improved information mixing. We conduct experiments on public datasets including MNIST and Fashion-MNIST, illustrating the advantage of the skip connection in TTFS coding architectures. Additionally, we demonstrate the applicability of TTFS coding on beyond image recognition tasks and extend it to scientific machine-learning tasks, broadening the potential uses of SNNs.

97 MATHEMATICS AND COMPUTING↗

cTULIP: application of a human-based RNA-seq primary tumor classification tool for cross-species primary tumor classification in canine

The domestic dog, Canis familiaris, is quickly gaining traction as an advantageous model for use in the study of cancer, one of the leading causes of death worldwide. Naturally occurring canine cancers share clinical, histological, and molecular characteristics with the corresponding human diseases. In this study, we take a deep-learning approach to test how similar the gene expression profile of canine glioma and bladder cancer (BLCA) tumors are to the corresponding human tumors. We likewise develop a tool for identifying misclassified or outlier samples in large canine oncological datasets, analogous to that which was developed for human datasets. We test a number of machine learning algorithms and found that a convolutional neural network outperformed logistic regression and random forest approaches. We use a recently developed RNA-seq-based convolutional neural network, TULIP, to test the robustness of a human-data-trained primary tumor classification tool on cross-species primary tumor prediction. Our study ultimately highlights the molecular similarities between canine and human BLCA and glioma tumors, showing that protein-coding one-to-one homologs shared between humans and canines, are sufficient to distinguish between BLCA and gliomas. The results of this study indicate that using protein-coding one-to-one homologs as the features in the input layer of TULIP performs good primary tumor prediction in both humans and canines. Furthermore, our analysis shows that our selected features also contain the majority of features with known clinical relevance in BLCA and gliomas. Our success in using a human-data-trained model for cross-species primary tumor prediction also sheds light on the conservation of oncological pathways in humans and canines, further underscoring the importance of the canine model system in the study of human disease.

60 APPLIED LIFE SCIENCES↗

Wind Speed Prediction with Spatio–Temporal Correlation: A Deep Learning Approach

Wind speed prediction with spatio–temporal correlation is among the most challenging tasks in wind speed prediction. In this paper, the problem of predicting wind speed for multiple sites simultaneously is investigated by using spatio–temporal correlation. This paper proposes a model for wind speed prediction with spatio–temporal correlation, i.e., the predictive deep convolutional neural network (PDCNN). The model is a unified framework, integrating convolutional neural networks (CNNs) and a multi-layer perceptron (MLP). Firstly, the spatial features are extracted by CNNs located at the bottom of the model. Then, the temporal dependencies among these extracted spatial features are captured by the MLP. In this way, the spatial and temporal correlations are captured by PDCNN intrinsically. Finally, PDCNN generates the predicted wind speed by using the learnt spatio–temporal correlations. In addition, three error indices are defined to evaluate the prediction accuracy of the model on the wind turbine array. Experiment results on real-world data show that PDCNN can capture the spatio–temporal correlation effectively, and it outperforms the conventional machine learning models, including multi-layer perceptron, support vector regressor, decision tree, etc.

17 WIND ENERGY↗