Abstract: There is disclosed a method of prediction time-series continuous data in which the data is accurately predicted based on measuring points and measured values at the measuring points of the data. From past measuring points t0 to t3 before a prediction point t4, a closest measuring point t3 which is closest to the prediction point t4 is sampled. Past n measuring points before the closest measuring point t3 are also sampled. Subsequently, a predicted value Q4 is determined from an interpolation formula f(t), which can derive a measured value Q3 at the closest measuring point t3 from the closest measuring point t3, i.e., Q3=f(t3) which is prepared from the past n measuring points, the measured values, the prediction point t4 and the predicted value Q4. The method can be implemented in a controller.