State estimating apparatus
First Claim
1. A state estimating apparatus for inferring a value of single output data from a plurality of input data given as input factors and estimating a state of a system, the plurality of input data and the single output data being time series data which continuously change, and the plurality of input data and the single output data having an implicit relationship which continuously changes, comprising:
- input space quantization means for quantizing an input space having the input data from case data indicating a plurality of past input data and past single output data stored in advance, in accordance with a required precision of the single output data;
storage means for calculating a number of times of occurrence of the single output data corresponding to each input event in the input space quantized by said input space quantization means, a mean value of the single output data, and a mean value of change amounts of the single output data, and for storing a set of the number of times of occurrence and the mean values as a causal relationship model;
means for, on the basis of a concept of continuous mapping of a topology, calculating a neighborhood of the input space which satisfies the required precision of the single output data as a statistical amount of all input data of the case data, and for expressing the calculated neighborhood of the input space as a quantization number by which the input space is quantized;
similar case extraction means for, when a new input event is input, enlarging the calculated neighborhood of the input space to a predetermined degree, and extracting an input case similar to the calculated neighborhood from an input case in the input space;
similarity determination means for determining a similarity between the new input event and the extracted similar input case on the basis of the predetermined degree of enlarging; and
estimating means for selecting an input case corresponding to the similarity determined by said similarity determination means from the extracted similar input case, and estimating a value of output data corresponding to the new input event on the basis of the mean value of the single output data corresponding to the selected input case and the mean value of change amounts of the single output data.
1 Assignment
0 Petitions
Accused Products
Abstract
In a state estimating apparatus, an input space quantization section quantizes an input space according to a required precision of output data. A storage section calculates and stores the number of times of occurrence of output data corresponding to each input event in the quantized input space, the mean values of the output data, and the change amounts of the output data. A neighborhood definition section calculates an input space neighborhood satisfying the required precision of the output. A similar case extraction section zooms up the predetermined neighborhood of the input space to the predetermined degree and extracts input case similar to the neighborhood from input case in the input space. A similarity determination section determines the similarity between the new input event and the extracted similar input cases on the basis of the degree of zooming. An estimating section selects input cases corresponding to the similarity from the similar input case, and estimates a value of output data corresponding to the new input event on the basis of the mean value of output data and the mean value of change amounts of the output data.
97 Citations
6 Claims
-
1. A state estimating apparatus for inferring a value of single output data from a plurality of input data given as input factors and estimating a state of a system, the plurality of input data and the single output data being time series data which continuously change, and the plurality of input data and the single output data having an implicit relationship which continuously changes, comprising:
-
input space quantization means for quantizing an input space having the input data from case data indicating a plurality of past input data and past single output data stored in advance, in accordance with a required precision of the single output data; storage means for calculating a number of times of occurrence of the single output data corresponding to each input event in the input space quantized by said input space quantization means, a mean value of the single output data, and a mean value of change amounts of the single output data, and for storing a set of the number of times of occurrence and the mean values as a causal relationship model; means for, on the basis of a concept of continuous mapping of a topology, calculating a neighborhood of the input space which satisfies the required precision of the single output data as a statistical amount of all input data of the case data, and for expressing the calculated neighborhood of the input space as a quantization number by which the input space is quantized; similar case extraction means for, when a new input event is input, enlarging the calculated neighborhood of the input space to a predetermined degree, and extracting an input case similar to the calculated neighborhood from an input case in the input space; similarity determination means for determining a similarity between the new input event and the extracted similar input case on the basis of the predetermined degree of enlarging; and estimating means for selecting an input case corresponding to the similarity determined by said similarity determination means from the extracted similar input case, and estimating a value of output data corresponding to the new input event on the basis of the mean value of the single output data corresponding to the selected input case and the mean value of change amounts of the single output data. - View Dependent Claims (2, 3, 5, 6)
-
-
4. A state estimating apparatus for generating one output data as an output result from a plurality of input data and estimating a state of a system from the plurality of input data given as input factors, the plurality of input data and the output data being time series data which continuously change, and the plurality of input data and the output data having a relationship which continuously changes, comprising:
-
input space quantization means for quantizing an input space having the plurality of input data from case data indicating past input and output data stored in advance, in accordance with a required precision of the output data; classification means for dividing the case data into clusters according to a predetermined criterion when a quantization number for quantization performed by said input space quantization means exceeds a predetermined value; counting means for quantizing an overall range of the past input and output data stored in advance and counting a number of times of occurrence of classified case data for each quantization number for each case data classified by said classification means; possibility distribution generation means for generating a possibility distribution by performing normalization using a maximum value of the counted number of times of occurrence; determination means for displaying the possibility distribution generated for each classified case data, identifying a generated possibility on the basis of a predetermined criterion, and determining whether to use the possibility distribution; and possibility-of-attribution determination means for, when a new input event is input, determining a possibility of attribution of the new input event with respect to each cluster, generated by said classification means, by using the possibility distribution.
-
Specification