Abstract
Neural network is applied to the knowledge-based prediction rule system of solar flares based on the attributes of sunspot characteristics. The inputs are the magnetic class, modified Zürich class, area and number of sunspots, and the output gives the occurrence probability of flares in the following day. Simple 3-layer network model has been trained by using the if-then rule based patterns, and almost 99% of correct answer rate is obtained which can substitute for collating data base with present sunspot characteristics. Also, some tests for the specification of input units, the necessary number of hidden units and generalization capability have been made. Further sophistication of the network is to be expected for the practical prediction of our sun-earth environment.
