Post on 20-Dec-2015
PREDICTING INTERPLANETARY PREDICTING INTERPLANETARY SHOCKS USING NEURALSHOCKS USING NEURAL
NETWORKSNETWORKS
A.İbrahim PALAA.İbrahim PALA
METU Aerospace Engineering DepartmentMETU Aerospace Engineering Department
Senior StudentSenior Student
21.12.200521.12.2005
OUTLINEOUTLINE
WHAT IS AN INTERPLANETARY SHOCK?WHAT IS AN INTERPLANETARY SHOCK?
WHAT IS A NEURAL NETWORK?WHAT IS A NEURAL NETWORK?
HOW THE NEURAL NETWORKS ARE HOW THE NEURAL NETWORKS ARE USED TO PREDICT THE USED TO PREDICT THE INTERPLANETARY SHOCKS?INTERPLANETARY SHOCKS?
CONCLUSIONCONCLUSION
INTERPLANETARYINTERPLANETARY SHOCKSSHOCKS
Charged particles areCharged particles are
constantly streamingconstantly streaming
outward from the Sun.outward from the Sun.
When events such asWhen events such as
Coronal Mass EjectionsCoronal Mass Ejections
(CMEs) occur on the Sun,(CMEs) occur on the Sun,
an Interplanetary (IP) shockan Interplanetary (IP) shock
may be generated. may be generated.
INTERPLANETARYINTERPLANETARY SHOCKSSHOCKS
TThe abrupt boundary he abrupt boundary formed at the front formed at the front of a plasma cloud of a plasma cloud moving much faster moving much faster than the rest of the than the rest of the solar wind, as it solar wind, as it pushes its way pushes its way through through interplanetary space. interplanetary space.
NEURAL NETWORKSNEURAL NETWORKS
NEURONSNEURONS
SYNAPSESYNAPSE
(INPUT->WEIGHTS) (INPUT->WEIGHTS)
COLLECTORCOLLECTOR
ACTIVATION ACTIVATION FUNCTION->OUTPUTFUNCTION->OUTPUT
NEURAL NETWORKSNEURAL NETWORKS CONSTRUCTING A NEURAL NETWORK BY CONSTRUCTING A NEURAL NETWORK BY
BRINGING LOTS OF NEURONS TOGETHER AND BRINGING LOTS OF NEURONS TOGETHER AND INTERCONNECTING AND TRAINING THEMINTERCONNECTING AND TRAINING THEM
SYSTEM PARALLELISM AND STRUCTURAL SYSTEM PARALLELISM AND STRUCTURAL DISTRIBUTION OF OVERALL FUNCTIONDISTRIBUTION OF OVERALL FUNCTION
->compensating the errors due to a ->compensating the errors due to a nonfunctional neuronnonfunctional neuron
GENERALIZATION AND LEARNING ABILITYGENERALIZATION AND LEARNING ABILITY
->giving meaningful outputs for also the ->giving meaningful outputs for also the non-trained inputsnon-trained inputs
PREDICTING THE ARRIVAL OFPREDICTING THE ARRIVAL OFIP SHOCKS BY NEURAL IP SHOCKS BY NEURAL
NETWORKSNETWORKSIP Shocks causes the incoming particle IP Shocks causes the incoming particle
flux observed at or near Earth to flux observed at or near Earth to increase by orders of magnitude.increase by orders of magnitude.
Sharp increases in particle intensity Sharp increases in particle intensity are called Energetic Storm Particle are called Energetic Storm Particle (ESP) events(ESP) events
PREDICTING THE ARRIVAL OFPREDICTING THE ARRIVAL OFIP SHOCKS BY NEURAL IP SHOCKS BY NEURAL
NETWORKSNETWORKSINPUTSINPUTS:: Observations Observations
from two ACE instruments:from two ACE instruments:
TheThe ElectronElectron, Proton, , Proton, andand AlphaAlpha MonitorMonitor (EPAM) (EPAM): : particle intensity at five particle intensity at five different energy levels. different energy levels. This data is used to This data is used to detect detect the onsetthe onset of an event, of an event, several days in advance, several days in advance, and to and to predict the ESP predict the ESP event / IP shock arrivalevent / IP shock arrival. .
PREDICTING THE ARRIVAL OFPREDICTING THE ARRIVAL OFIP SHOCKS BY NEURAL IP SHOCKS BY NEURAL
NETWORKSNETWORKSINPUTSINPUTS
TheThe Solar Solar WindWind ElectronElectron, Proton, , Proton, andand AlphaAlpha MonitorMonitor (SWEPAM) (SWEPAM): solar wind : solar wind plasma temperature, plasma temperature, density, and speed. density, and speed. This data is used to This data is used to detect the actual IP detect the actual IP shock arrivalshock arrival. .
PREDICTING THE ARRIVAL OFPREDICTING THE ARRIVAL OFIP SHOCKS BY NEURAL NETWORKSIP SHOCKS BY NEURAL NETWORKS((RealReal-time -time InterplanetaryInterplanetary ShockShock PredictionPrediction
(RISP) (RISP) SystemSystem ))TRAINING THE ARTIFICIAL NEURAL TRAINING THE ARTIFICIAL NEURAL
NETWORK TO PREDICT THE TIME NETWORK TO PREDICT THE TIME REMAINING UNTIL THE SHOCK ARRIVESREMAINING UNTIL THE SHOCK ARRIVES
TRAINED BY 37 OF THE 56 SELECTED TRAINED BY 37 OF THE 56 SELECTED SHOCKSSHOCKS
PERFORMANCE ASSESSED BY FORECASTING PERFORMANCE ASSESSED BY FORECASTING ARRIVAL TIME COUNTDOWNS FOR 19 ARRIVAL TIME COUNTDOWNS FOR 19 SHOCKS NOT USED DURING TRAININGSHOCKS NOT USED DURING TRAINING
PREDICTING THE ARRIVAL OFPREDICTING THE ARRIVAL OFIP SHOCKS BY NEURAL IP SHOCKS BY NEURAL NETWORKSNETWORKS
Particle intensities for test sequence 9. Predicted time until shock arrival.
REFERENCESREFERENCES www.physics.usyd.edu.au/~cairns/teaching/lecture11/www.physics.usyd.edu.au/~cairns/teaching/lecture11/
node5.html+interplanetary+shocks&hl=trnode5.html+interplanetary+shocks&hl=tr
http://www-spof.gsfc.nasa.gov/Education/wgloss.htmlhttp://www-spof.gsfc.nasa.gov/Education/wgloss.html
http://sd-www.jhuapl.edu/ACE/EPAM/RISP/
Vandegriff , J., Wagstaff, K., Ho , G., Plauger, J. Forecasting space weather: Predicting interplanetary shocks using neural networks. Adv.Space Res. 36, 2323-2327, 2005.
www.srl.caltech.edu/ACE/
Efe, M.Ö., Kaynak, O., Yapay sinir ağları ve uygulamaları, Bogazici University,2005