To control an optimum operation state even at transitive time by holding learning data even after the learning of a neuro part ends, correcting a set of the held data, and performing the relearning of the neuro part.

Even after the learning of the neuro part 4 ends, operation state data and learning data on a target manipulated variable are held in a learning data set memory 7 and when learning data is obtained in an operation state, the held learning data set is corrected to perform the relearning of the neuro part 4. The learning data in this case is a pair of a neuro entrance and a neuro exit. Further, the learning data set memory 7 is stored with throttle opening extent data and engine rotating speed data constituting the neuro input and also stored with a manipulated variable as the neuro exit. Then this manipulated variable is obtained by adding a feedback value to the neuro output.