摘要 |
PROBLEM TO BE SOLVED: To provide a highly precise prediction method with a high adaptation function through the use of the learning of a model used for prediction and various information. SOLUTION: The demanded prediction device is provided with the input means 1 of time sequential data, a data base 2 being a storage means, a data selection means 6, a base model formation means 7, an auxiliary model formation means 8, a prediction means 3, a model evaluation means 5 and a predicted result output means 4. The data selection means 6 improves prediction precision by selecting data used for forming the model and the base model creation means 7 forms a model structure showing a trend, circulation, a season, month/day, an unusual day characteristic on a prediction object and the knowledge of a prediction expert. The auxiliary model creation means 8 makes a difference between prediction data and past time sequential data only by a base model into the model structure in terms of learning. The model evaluation means 5 evaluates the error of prediction by the model and selects optimum data. |