Time-series signal predicting apparatus

Abstract

Disclosed is a time series signal predicting apparatus for predicting time series data of the future based on past time series data and current time series data. Vector generating means generates n-dimensional vector data based on a natural number of n inputted time series data and stores them into vector storage means, while further vector generating means generates (n+1)-dimensional vector data based on inputted past time series data, classifies them into a plurality of L vector sets in units of similar vectors by using up to n-dimensional vector data, and stores them into vector set storage means. Subsequently, vector classifying means classifies input vector data according to which vector set they belong to, and vector selecting means selects a plurality of k vector data in a descending order of similarity to the vector data of newly inputted time series data, from a vector set belonging to vector sets to which the vector data have been decided to belong. Then, past data extracting means extracts a plurality of k (n+1)-th-dimensional data from a plurality of k (n+1)-dimensional vectors belonging to the selected vector set, and outputting the extracted data together with their similarities, and data predicting means predicts and outputs time series data of the next time succeeding to the inputted time series data based on the plurality of k extracted data, the input vector data and the similarities.

Description

technical field [0001] The invention relates to a time series signal predicting device for predicting future time series data according to past and current time series data. Background technique [0002] For example, Japanese Laid-Open Patent JP-A-7-84983 discloses a prior art timing signal prediction device. The structure of this prior art timing signal predicting device is shown in FIG. 15 . The prior art timing signal prediction device shown in Figure 15 includes: [0003] (a) a first timing prediction device 1501 designed with a feed-forward neural network; [0004] (b) a second time series forecasting device 1502 designed with an ARMA (autoregressive moving average) model; [0005] (c) The application characteristic measurement device 1504, for each application condition 1507, according to the application condition 1507, the prediction results 1505 and 1506 of the first and second timing prediction devices 1501, 1502 and the actual measurement result 1508, each predi...

AI Technical Summary

Problems solved by technology

[0008] However, the time series signal prediction device of the prior art needs to use multiple prediction devices to improve the prediction accuracy, which will make the system structure complicated

Also, once statistical properties, such as changes in the mean value and variance of the input time-series signal occur, the time-series signal predicting device without a device for dealing with such a change in the statistical property of the time-series signal will experience a large decrease in the accuracy of the predicted time-series signal worsening problem

Images