Deep Learning Technology For Predicting Solar Flares From .
(IJACSA) International Journal of Advanced Computer Science and Applications,Vol. 9, No. 1, 2018Deep Learning Technology for Predicting SolarFlares from (Geostationary OperationalEnvironmental Satellite) DataTarek A M Hamad Nagem, RamiQahwaji, Stan IpsonSchool of Electrical Engineering andComputer ScienceUniversity of BradfordBradford, United KingdomZhiguang WangAlaa S. Al-WaisyGE Global ResearchSan Ramon, CA, United States ofAmericaSchool of Electrical Engineering andComputer ScienceUniversity of BradfordBradford, United KingdomAbstract—Solar activity, particularly solar flares can havesignificant detrimental effects on both space-borne and groundsbased systems and industries leading to subsequent impacts onour lives. As a consequence, there is much current interest increating systems which can make accurate solar flarepredictions. This paper aims to develop a novel framework topredict solar flares by making use of the GeostationaryOperational Environmental Satellite (GOES) X-ray flux 1minute time series data. This data is fed to three integratedneural networks to deliver these predictions. The first neuralnetwork (NN) is used to convert GOES X-ray flux 1-minute datato Markov Transition Field (MTF) images. The second neuralnetwork uses an unsupervised feature learning algorithm to learnthe MTF image features. The third neural network uses both thelearned features and the MTF images, which are then processedusing a Deep Convolutional Neural Network to generate theflares predictions. To the best of our knowledge, this work is thefirst flare prediction system that is based entirely on the analysisof pre-flare GOES X-ray flux data. The results are evaluatedusing several performance measurement criteria that arepresented in this paper.Keywords—Convolutional; neural; network; deep; learning;solar; flare; prediction; space; weather insertI.INTRODUCTIONThe concept of space weather has been defined by the USNational Space Weather Program as “Conditions on the Sunand in the solar wind, magnetosphere, ionosphere andthermosphere that can influence the performance and reliabilityof space-borne and ground-based technological systems andcan endanger human life or health” [1]. There are severalinfluences, originating from space weather phenomena thatdetrimentally affect important industries relying on avionics,satellites, mobile communication networks, and electricitydistribution [2]. All these industries touch our daily lives andthis means that space weather can impact our livesdramatically.Painstaking efforts are currently being made in a number ofinternational centres to create accurate solar flare predictionsystems. This is because many infrastructures could be affectedby significant flares and the cost of building an accurate solarflare prediction system would be much cheaper than the cost ofrepairing damage caused by such a flare. In this work, theproposed prediction system generates two probabilities forEvent and No-event. Event predictions cover significant X andM class flares that might be harmful, while No-eventpredictions cover no-flares and the non-harmful A, B and Cclass flares.Although scientific progress has increased enormously therate of generation of data monitoring solar activity, scientistsare not yet able to fully understand all the detailed causes ofsolar flares. Consequently, efforts are being made to developmethods to predict solar storms, making direct use of the datausing advances in data analysis.Since 1987, there have been many approaches thatattempted to predict solar flares. The first solar flare predictionsystem (called THEOPHRASTUS) was launched by the SpaceEnvironment Services Centre at NOAA, and it predicts X-rayflares with a time window of 24 hours [3]. More recently, threesolar flare prediction systems, ASSA (Automatic SolarSynoptic Analyser), MAG4 (Magnetic Forecast system) [7]and ASAP (Automated Solar Activity Prediction), havebecome a part of the NASA Integrated Space Weather Analysis(ISWA) system [5] and these three systems are brieflydescribed below.The first system, ASSA, is based on an artificial neuralnetwork technique and the ASSA coronal hole data archive,from the period 1997 till 2013, including SDO solar images, topredict solar flares, solar radiation storms and geomagneticstorms. ASSA predicts C, M and X flares. ASSA predictionsare based on statistical analysis of the ASSA sunspot catalogue[6]. The second system, MAG4 was developed at theUniversity of Alabama in Huntsville, to assist NASA SpaceRadiation Analysis Group (SRAG) at the Johnson Space FlightCentre. MAG4 is using Magnetogram data for the Sun. MAG4forecasts X and M class flares, CMEs, and Solar Proton Events(SPE) using McIntosh active-region (AR) classes as the basisof their forecasts [7]. The University of Bradford developed aforecasting model, the Automated Solar Activity Prediction(ASAP) system in 2009. ASAP uses McIntosh classes andother sunspots features which it generates from the solar data.492 P a g ewww.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,Vol. 9, No. 1, 2018ASAP uses SDO/HMI Continuum and Magnetogram imagesas an input to the system, also it uses two neural networks topredict solar flares [3].Recently, the new field of deep learning neural networkresearch has achieved remarkable successes compared withprevious artificial intelligence methods [5]. These includecomplex tasks like medical diagnoses, dealing with hugeamounts of data, pattern recognition and numerous others, suchas the virtualization frameworks for big data reported in [8].Using the deep learning technology for space weatherprediction is still a novel area of research, which needs to beinvestigated to help analyse the huge amount of solar activitydata that are publically available.UFCORIN (Universal Forecast Constructor by OptimizedRegression of Inputs) is open-source software available onlinewhich has been used to predict general time series and solarflares. This system uses HMI image data and GOSE X-ray dataas input to predict X, M, and C solar flare class. In 2016,UFCORIN was extended to use deep learning, and provides24-hour-ahead predictions of solar flares, every 12 minutes byusing a deep learning approach.In this paper, we introduce a solar flare prediction system,summarised in the following subsection, working solely withGOES X-ray flux data that integrates three neural networks todeliver these predictions and provides an automated predictionof solar flares by utilising deep learning techniques.GOES data are available in real-time (available everyminute) and they provide a general indication of flaring acrossthe solar disk. These data come in soft and hard x-ray and areavailable from 2002. However, GOES data provide anindication of flaring without much info about the exact locationof flaring on the solar disk. This could be one of the reasonswhy it is not used heavily for space weather prediction. Theformat of GOES Data is also challenging as it is represented asa time-series signal, which makes it challenging for machinelearning based prediction (Deep learning in particular).A. Overview of the SystemFig. 1 shows the system model which consists of threeunits. Starting from the input (GOES X-ray flux time seriesdata) to the output (solar flare prediction) and including theevaluation of the predictive performance.Unit 1in Fig. 1 converts a sequence of GOES X-ray flux 1minute data time series data to a 64 64 MTF image in twostages. Firstly, it converts the original text data to a MarkovTransition Matrix. Then it encodes the Markov TransitionMatrix as a 64 64 Markov Transition Field (MTF) image asillustrated in Fig. 6. Unit 2 in Fig. 1 learns the features withinthe MTF images. Unit 2 pre-processes and normalizes theimages and then divides the 64 64 images into 64 8 8 patches.These patches are encoded using a Back-propagation Autoencoder to obtain learned feature mappings as indicated in Fig.1. Unit 3 in Fig. 1 provides predictions for solar flares using aCNN. This unit starts by utilising the historical knowledge andlinking the MFT images with the Flare or No-Flare labels.Subsequently, datasets are created for training and testing theneural networks. After training on the associated dataset iscarried out, the trained CNN is run on the test dataset togenerate prediction results, which are evaluated using spaceweather verification metrics.The rest of this paper is organized as follows. Section 2describes the operation of Unit 1 which converts GOES X-rayflux time series data to 64 64 MTF images. Section 3describes Unit 2, which learns features within MTF imagesusing an unsupervised learning algorithm by applying backpropagation. Section 4 describes Unit 3, which makes solarflare predictions using a Deep Convolutional Neural Network.Section 5 discusses the evaluation and performance of thewhole system and Section 6 presents concluding remarks andsuggestions for future work.Fig. 2. A sample 6 hour plot of GOES X-ray flux 1-minute data.Fig. 1. The diagram showing the internal procedures of the system.493 P a g ewww.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,Vol. 9, No. 1, 2018II. PREPARATION OF THE DATAA. The Source X-Ray DataIn this work, 1-minute X-ray flux data from the AmericanGeostationary Operational Environmental Satellites (GOES)are used. The data used are provided from four GOESsatellites, GOES-10, GOES-11, GOES-14, and GOES-15. Allthe data produced are archived and available, and it can befound online at [9]. Two X-ray channels are available as shownin Fig. 2; a harder X-ray channel (0.05-0.4 nm), and a softer Xray channel (0.1-0.8 nm) [10]. For this work, the soft channel isused because provides information about the intensity of solarflares and is used in this work to investigate its suitability forinvestigating the temporal evolution of flares [10].B. Extraction of Relevant X-Ray Flux DataThe temporal evolution of solar flares generally occurs inthree phases [4]. Pre-flare phase: This is the region shown in Fig. 3which consists of fluctuations and a slow increase of Xray flux before the start of the flare event. Impulsive phase: Here the X-ray flux increase quicklyand the main flare energy release occurs during thisphase. Gradual phase: In this phase, the X-ray flux graduallydecreases to the background level.Fig. 4 shows the cropped AIA images of a flaring regioncorresponding to the GOES X-ray data regions in Fig. 3. Theleft image in Fig. 4, captured in the pre-flare phase, shows twosets of nested loops. The middle image in Fig. 4, capturedduring the main phase, shows inner loops becomingsignificantly brighter. In the right-hand image, the flarelaunches a CME. There are many relationships which havebeen recognized between the pre-flare activities and flaring,and these appear as loop brightening activities [15]. However,the method introduced here bases its prediction solely onchanges in the overall X-ray flux during the pre-flare phase.Fig. 4. Cropped AIA images showing three phases of the solar flare whichcontributes to the GOES data shown in Fig. 3– From NASA [11].C. Prediction Optimization for Different Time WindowsThe Time windows of 20, 30, 60 and 120 minutes betweenthe end of a data sample and the start of a flare/no-flare areinvestigated, using the Quadratic score QR, to determine thetime window with the best prediction performance. QR iswidely used as a verification measure to evaluate the accuracyof prediction. The prediction accuracy is calculated by findingthe mean square error between the predictions and theobservations as given by [2].QR (1)where ot are the binary observation outcomes where 1means that flare occurred and 0 means that a flare did nothappen, N is the sample size, and ft is the predictionprobability. QR ranges from 0 (perfect prediction) to 1 (worstpossible prediction) [18].The result for each time window is shown in Table I. It isclearly seen that the best QR is when the time window equals20 minutes. To find the prediction window duration that wouldprovide the best QR value, we followed the method presentedin [2] and applied QR to determine the best prediction windowduration.TABLE I.THE QUADRATIC SCORE (QR) RESULTS FOR 20, 30, 60 AND 120MINUTESFig. 3. The solar flare phases on C8.8 flare that occured on 5th May 2010 –From NASA [11].Sample Size20 minutes21240.13630 minutes0.15360 minutes0.249120 minutes0.590D. Data PresentationFig. 5 shows a sub-system that has been created to generatedatasets by selecting specific data from GOES X-ray flux 1minute data using three steps. The first step identifies a flare.Then selects 120 minutes of data, starting 140 minutes beforethe beginning of the flaring event. Finally, the selected data issaved in a matrix as described in the next subsection.494 P a g ewww.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,Vol. 9, No. 1, 2018Fig. 7. Learning the features within MTF images.III. LEARNING THE FEATURES WITHIN MTF IMAGESFig. 5. Creating dataset of a time series of X-ray flux data with a 20-minutedata window before the flare occurs.E. Conversion of Time Series Data to MTF ImagesTemporal and frequency correlations are majordependencies embedded in time series data. To build acomprehensive but intuitive visualization, the extractedfeatures of the designed data transformation framework shouldbe able to represent the dynamics in both time and frequencywhile there should exist a reverse operation to map theinformation back to the raw GOES time series. The followingsub-sections describe how to encode the dynamical frequencyinformation in the temporal ordering, illustrated in Fig. 6, stepby step.The main idea of this stage is to use GOES time series datato generate Markov transition field while maintaining the timeseries properties. The method applied in this research is takenfrom [14]. MTF images were generated by applying the codeused in [14] to GOES data.Fig. 6. Conversion of GOES X-ray data time series data to MTF images.The Auto-encoder is an unsupervised back-propagationneural network which tries to learn a function hW,b(x) x, andis adjusted so that the input values correspond to the target y(i)̂ (i) [12]. In this work, we assume x is the inputcorresponding to the pixel intensity values for an 8 8 MTFimage patch with 64 pixels so x 64, and there are s2 32hidden units in layer L2. The network is required to learn acompressed representation of the input, because there existonly 32 hidden units. Therefore the auto-encoder shouldattempt to reconstruct the input to 8 8 images (64 pixels) [16]as illustrated in Fig. 7.IV. PREDICTION OF SOLAR FLARES USING A DEEPCONVOLUTIONAL NEURAL NETWORKAs you can see in Fig. 8 the Convolutional Neural Network(CNN) consists of convolutional layers and sub-samplinglayers followed by fully connected layers.A. The Convolutional LayerThe input to this layer is a d d ch MTF image where dis the height and the width of the image (d 64 in this case) andch is the number of channels. Since the MTF images are RGBimages, ch 3. As illustrated in Fig. 9 the convolutional layeruses Kf filters (also called Kernels) of size n n ch where nis the dimension of the filter and n 8 to produce feature maps.The Kf filters are convolved over the MTF image to create Kffeature maps of size d n 1 [16].Fig. 8. Convolutional neural network designed to predict solar flares.495 P a g ewww.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,Vol. 9, No. 1, 2018A. System EvaluationThe performance evaluation was done by comparing thegenerated predictions with the actual flare occurrences asreported by 1-minute GOES data. The data were taken fromfour satellites, GOES-10 data covering (03 Dec 2002 -22 Jun2006) and (11 Apr 2007-30 Dec 2009); GOSE-11 datacovering (23 Jun 2006-10 Apr 2007); GOSE-14 data covering(01 Nov 2009 -26 Oct 2010); and finally GOSE-15 datacovering (27 Oct 2010 -30 Jan 2017). The number of flaringand No-flaring events for each satellite is detailed in Table II.All GOES X-ray data were taken from [9].Fig. 9. Convolving filter over an input image in convolutional layer.Fig. 10. An example of Max pooling.B. The Pooling LayerAfter the generation of the feature maps by theconvolutional layer, the features are then used forclassification. Fig. 10 shows each feature map is downsampled by max-pooling to size p p. Typically, p ranges from2 to 5, for small to big images respectively, and in this workp 4 [16].C. The Fully Connected LayerThis layer takes the outputs from the previous layers whichwere reduced to a one-dimensional feature vector. This layer isfully connected and there is just one output for each class label.The high-level inference in the CNN is performed by this fullyconnected layer.As noted earlier in this paper, the data is classified asflaring if they produced at least one M or X class flare in thefollowing 20 min period and No-flare if they did not cause anyM or X class flares during that period. To determine the flareprediction capability we carried out experiments with 1-minuteGOES data covering (Dec 2002-Dec 2005, Jun 2009- Dec2012) to train the deep learning algorithm. The data covering(Jun 2006 - Dec 2008, Jun 2013 - Jan 2017) are used to test thesystem as shown in Table III. Table IV details the number offlare and no-flare data that were used in these experiments. Thetime coverage of the training set was chosen so that theremaining testing set would contain flare activity from periodsaround the maximum and minimum levels of solar activity.TABLE II.THE NUMBER OF FLARING AND NO-FLARING FOR GOES-10DATA COVERING (03 DEC 2002 -22 JUN 2006), (11 APR 2007-30 DEC 2009);GOSE-11 DATA COVERING (23 JUN 2006-10 APR 2007); GOSE-14 DATACOVERING (01 NOV 2009 -26 OCT 2010); GOSE-15 DATA COVERING (27 OCT2010 -30 JAN 2017) USED IN THIS EXPERIMENTGOES-10From03 Dec 2002To 22 Jun 2006andFrom11 Apr 2007To 30 Dec 2009FlareeventsThe first part of the system, which encodes the GOES datato MTF images, is implemented in Python and the rest of thesystem is implemented in Matlab [17]. The system makesflares predictions based on embedded learning rules. Thesystem was trained using training sets covering data from 3rdDec 2002 till 30th Jan 2017, to ensure this covered a range ofactivity including both solar Maximum and solar Minimum ofthe solar cycle.5181592TABLE III.GOES-14From01 Nov 2009To26 Oct 2010GOES-15From27 Oct 2010To30 Jan . IMPLEMENTATION AND EVALUATION OF THE SYSTEMThree neural networks are integrated into the system topredict solar flares. Fig. 4 shows the integrated system startingfrom the input (GOSE data) to the output of the system(Flare/No-Flare prediction).NoFlareGOES-11from23 Jun 2006To10 Apr 2007NUMBER OF FLARE AND NO-FLARE DATA COVERING (03 DEC2002-30 JAN 2017)03 Dec 2002-30 Jan 2017FlareNo-flareTotal132739815308496 P a g ewww.ijacsa.thesai.org
(IJACSA) International Journal of Advanced Computer Science and Applications,Vol. 9, No. 1, 2018TABLE V.TABLE IV. NUMBER OF FLARE AND NO-FLARE DATA IN TIMEINDEPENDENT TRAINING AND TESTING SETSTraining set(Dec 2002-Dec 2005)(Jun 2009- Dec 2012)Flare793No-Flare2391Testing set( Jun 2006- Des 2008)(Jun 2013-30 Jan 2017)Total3184Flare534No-Flare1590CONTINGENCY TABLE FOR PERFORMANCE MEASUREMENTSCONTAINING THE FOLLOWING ABBREVIATIONS FOR THE NUMBERS OFPREDICTED TRUE POSITIVES A, FALSE POSITIVES B, FALSE NEGATIVES C, ANDTRUE NEGATIVES DFlare observationsFlare predictionFlareNo- FlareFlareabNo- FlarecdTotal2124B. Machine Learning using Cross-ValidationCross-validation is a method that partitions the input datainto subsets so that the learning algorithm can be trained on asubset and internally tested on a different subset. Crossvalidation is a useful approach for analysing the pre
Deep Learning Technology for Predicting Solar Flares from (Geostationary Operational Environmental Satellite) Data . National Space Weather Program as “Conditions on the Sun and in the solar wind, magnetosphere, ionosphere and . rate of generation of data monitoring solar activity, scientists
Bruksanvisning för bilstereo . Bruksanvisning for bilstereo . Instrukcja obsługi samochodowego odtwarzacza stereo . Operating Instructions for Car Stereo . 610-104 . SV . Bruksanvisning i original
10 tips och tricks för att lyckas med ert sap-projekt 20 SAPSANYTT 2/2015 De flesta projektledare känner säkert till Cobb’s paradox. Martin Cobb verkade som CIO för sekretariatet för Treasury Board of Canada 1995 då han ställde frågan
service i Norge och Finland drivs inom ramen för ett enskilt företag (NRK. 1 och Yleisradio), fin ns det i Sverige tre: Ett för tv (Sveriges Television , SVT ), ett för radio (Sveriges Radio , SR ) och ett för utbildnings program (Sveriges Utbildningsradio, UR, vilket till följd av sin begränsade storlek inte återfinns bland de 25 största
Hotell För hotell anges de tre klasserna A/B, C och D. Det betyder att den "normala" standarden C är acceptabel men att motiven för en högre standard är starka. Ljudklass C motsvarar de tidigare normkraven för hotell, ljudklass A/B motsvarar kraven för moderna hotell med hög standard och ljudklass D kan användas vid
LÄS NOGGRANT FÖLJANDE VILLKOR FÖR APPLE DEVELOPER PROGRAM LICENCE . Apple Developer Program License Agreement Syfte Du vill använda Apple-mjukvara (enligt definitionen nedan) för att utveckla en eller flera Applikationer (enligt definitionen nedan) för Apple-märkta produkter. . Applikationer som utvecklas för iOS-produkter, Apple .
Deep Learning: Top 7 Ways to Get Started with MATLAB Deep Learning with MATLAB: Quick-Start Videos Start Deep Learning Faster Using Transfer Learning Transfer Learning Using AlexNet Introduction to Convolutional Neural Networks Create a Simple Deep Learning Network for Classification Deep Learning for Computer Vision with MATLAB
2.3 Deep Reinforcement Learning: Deep Q-Network 7 that the output computed is consistent with the training labels in the training set for a given image. [1] 2.3 Deep Reinforcement Learning: Deep Q-Network Deep Reinforcement Learning are implementations of Reinforcement Learning methods that use Deep Neural Networks to calculate the optimal policy.
och krav. Maskinerna skriver ut upp till fyra tum breda etiketter med direkt termoteknik och termotransferteknik och är lämpliga för en lång rad användningsområden på vertikala marknader. TD-seriens professionella etikettskrivare för . skrivbordet. Brothers nya avancerade 4-tums etikettskrivare för skrivbordet är effektiva och enkla att
