| 摘要 |
PURPOSE:To adjust a correction amount for the logical expression of a learning term corresponding to an error amount and to converge a theoretical value to an actual result value by variably setting a smooth gain based on the relation of the error amount of this time and the statistical amount of distribution and the error amount between the actual result value upto present and the theoretical value. CONSTITUTION:The actual temperature of a steel plate is measured for every fixed period of time T and an actual value A is sent out to an error calculation part 13. On the other hand, a theoretical value calculation part 12 calculates the theoretical value C and sends it out to the error calculation part 13. The error calculation part 13 calculates the error amount (x) of the actual result value A and the theoretical value, sends it out to an error distribution function deciding part 15 and also stores it in a learning data storage part 14. The error distribution function deciding part 15 reads the distribution value sigma<2> and the average value (m) of the error amount from the storage part 14, calculates the distribution function of the error amount (x) of this time and sends it out to a smooth gain calculation part 16. The smooth gain calculation part 16 calculates the smooth gain in the period T of this time and a learning term calculation part 18 reads the learning term of the period T of a previous time from the storage part 14 and calculates the learning term of the period T of this time. |
