PURPOSE: To correct a fuzzy rule into a proper rule by putting together time- series data obtained as inference values of a 1st and a 2nd fuzzy inference part and generating the new fuzzy rule from the fuzzy rule stored in a memory on the basis of time-series data obtained as the arithmetic result.

CONSTITUTION: The time-series data stored in time-series data storage parts 16 and 17 are sent to a data arithmetic part 18. The data arithmetic part 18 puts the two inputted time-series data together to generate the new time-series data. The composite inference value time-series data are combined with input- value time-series data and sent as time-series data to a storage part 19 and stored. Then the time-series data stored in the time-series data storage part 19 are sent to a rule generating means 21. At the same time, a fuzzy rule group 1 is sent from a memory 11 to a rule generating means 21. The rule generating means 21 generates a new fuzzy rule group 3 from the fuzzy rule group 1 on the basis of the inputted time-series data.