next up previous
Up: Neural Network Scheduler for Previous: Conditions of Constraint


Alterated Points on Implementation

The Java program actually differs in some points from the description above.

When the number of the times is tex2html_wrap_inline438 , tex2html_wrap_inline440 comets can be observed at most. So the solution of the neural network should satisfy: the number of the cells which satisfy tex2html_wrap_inline442 is nearly tex2html_wrap_inline444 . Therefore the system counts the number of those cells when it calculates the tex2html_wrap_inline446 , and if the number is larger than tex2html_wrap_inline448 , it decrease the value of tex2html_wrap_inline450 moreover. On the other hand, if the number of those cells is smaller than tex2html_wrap_inline452 , it increase tex2html_wrap_inline454 moreover. In this way the system tries to find the solution more quickly.

Because a neural network method does not assures that it returns the best solution, the result may have some room to be improve. Therefore, the program optimizes the result after the network finished. It definitely makes the solution keep the condition that, when the solution says comet i should be observed at time j, the contition of the comet is worse at any other time, otherwise another comet should be scheduled to those time. If there is a time k when the condition of the comet is better than at than at time j and no other comet is scheduled at that time, the program modifies the comet should be obsreved at time k. It does the check for all comets.



Go back to Neural Network Scheduler for Comets Observation.
Go back to Seiichi Yoshida's Home Page.
Copyright(C) Seiichi Yoshida (comet@aerith.net). All rights reserved.
Tue Apr 15 03:17:05 JST 1997