摘要 |
PROBLEM TO BE SOLVED: To detect a certain change point from time series data online with high accuracy at a point in time when a sharp change occurs. SOLUTION: The time series data are sequentially read, and a parameter of a time series model is learned by a time series model learning device 11 and is stored in a complexity and parameter storage device 14. To each the piece of the input data, the parameters are sequentially read from the storage device 14 to calculate values of a loss function, and they are sequentially added. Thereby, complexity as an application error of the model to the input time series data is sequentially calculated, and is stored in the storage device 14. A change point search device 15 reads the complexity to the time series data before and after the change point to all of change point candidates from the storage device 14, compares the sum thereof with the complexity to all the time series data when the change point is not assumed, calculates a change point degree on the basis of a difference value thereof, and detects the change point by magnitude of the change point degree to output it. COPYRIGHT: (C)2005,JPO&NCIPI
|