| 摘要 |
PURPOSE: A method for measuring an importance of a variable using a neural network is provided to reduce a number of monitoring objective variables and perform concentrative monitoring for a part of variables affecting a relatively important influence by presenting a standard to judge the importance of respective variables inputted to the neural network with a learning result of the neural network. CONSTITUTION: Learning for the neural network is performed by a plurality of I/O(Input/Output) data pairs(x,y). The learning result is stored as the first connection strength(w(1)) expressing a connection relation between each input node and each hidden node, and the second connection strength(w(2)) expressing the connection relation between each hidden node and an output node. A plurality of indexes is decided by using the stored learning result. The importance of respective nodes is decided by a combination of partial or all decided indexes.
|
