Prediction system and prediction method

Patent metadata

Field	Value
Publication number	US-12462267-B2
Application number	US-202117911155-A
Country	US
Kind code	B2
Filing date	Jan 15, 2021
Priority date	Jun 22, 2020
Publication date	Nov 4, 2025
Grant date	Nov 4, 2025

Abstract

A system performs factor selection processing, which includes performing a factor selection operation for selecting one or two or more factors from one or a plurality of factors; factor filtering processing which includes: determining, about each of the one or the plurality of factors before the factor selection processing is performed, whether a factor value for prediction input to a prediction model is within a range of a plurality of factor values for specimen used to identify the prediction model; and excluding a factor, a result of the determination of which is false, and outputting a factor not excluded; and prediction processing which includes calculating the prediction value of the prediction target by inputting, to the prediction model, a factor value for prediction about each of one or more factors including the factor selected in the factor selection processing and not including the factor excluded in the factor filtering processing.

First claim

The invention claimed is:

1 . A prediction system comprising:
a central processing unit (CPU); a second CPU coupled to the first CPU; and a plurality of generators generating electric power, wherein the first CPU is programmed to execute: a factor selecting unit that performs factor selection processing including performing a factor selection operation, which is an operation for selecting one or two or more factors from one or a plurality of factors based on averaged contribution about each of the one or the plurality of factors, and the factor selection processing including averaging, about each of one or a plurality of factors, a contribution to a prediction model to be identified based on the plurality of factor values for past specimens and a plurality of contributions respectively corresponding to a plurality of points in time in past; a factor filter unit that minimizes a prediction model error by performing factor filtering processing before performing the factor selection processing, the factor filtering processing including: determining, about each of the one or the plurality of factors, whether a factor value for prediction that is input to a prediction model of power demand to calculate a prediction value of a prediction target of power demand is within a range of a plurality of factor values for past specimens used to identify the prediction model; and excluding a factor, a result of the determination of which is false, and outputting a factor not excluded; and a predicting unit that performs prediction processing including calculating the prediction value of the prediction target by inputting, to the prediction model, a factor value for prediction about each of one or more factors including the factor selected in the factor selection processing and not including the factor excluded in the factor filtering processing, wherein the prediction processing includes: adjusting, based on a contribution to prediction of each of the one or more factors, a matching degree of the factor to a prediction model to be identified; and inputting a factor value for prediction of each of the one or more factors to the prediction model based on the matching degree after the adjusting, and wherein at least one of the factors is outdoor temperature, wherein the second CPU is programmed to: based on the calculated prediction value, create an operation schedule including a number of the plurality of generators to be stated and an output distribution of the generators, and execute the created operation schedule to start and operate the number of generators.

2 . The prediction system according to claim 1 ,
wherein the factor filtering processing includes outputting, when an excluded factor is present, at least one factor having an absolutely or relatively strong correlation with the excluded factor among factors, factor values for prediction of which are within a range of a plurality of factor values for past specimens.

3 . The prediction system according to claim 1 ,
wherein the factor selection processing includes identifying, about each of one or a plurality of factors, a prediction model using a plurality of factor values for past specimens of the factor and calculating an evaluation value of the identified prediction model, and wherein the factor selection operation is an operation for selecting the one or two or more factors used in a prediction model having a highest calculated evaluation value.

4 . The prediction system according to claim 1 ,
wherein in the factor filtering processing, the one or the plurality of factors include a prediction target, the prediction target serving as a factor among the one or the plurality of factors, a factor value for prediction is a calculated prediction value and a plurality of factor values for past specimens are a plurality of values obtained in past.

5 . The prediction system according to claim 1 ,
wherein the factor selection processing includes: determining whether a difference between information amounts before and after a plurality of factor values for prediction are added to a plurality of factor values for past specimens of all factors selected in the factor filtering processing exceeds a predetermined threshold; upon determining a result of the determination is false, selecting the one or two or more factors without the factor selection operation as factors used to calculate a prediction value anew; and upon determining the result of the determination is true, performing a selection operation anew in order to select one or two or more factors serving as factors used to calculate a prediction value anew.

6 . The prediction system according to claim 1 , wherein the factor selection processing includes:
determining whether a value conforming to a prediction error calculated using a prediction model based on a factor value for prediction about each of the selected one or two or more factors exceeds a predetermined threshold; upon determining a result of the determination is false, selecting the one or two or more factors without the factor selection operation as factors used to calculate a prediction value anew; and upon determining n the result of the determination is true, performing a selection operation anew in order to select one or two or more factors serving as factors used to calculate a prediction value anew.

7 . The prediction system according to claim 1 ,
wherein the factor selecting unit presents information representing at least one of (A) and (B) described below: (A) about each of one or a plurality of factors, whether a factor value for prediction is within a range of a plurality of factor values for past specimens; and (B) for each of one or a plurality of factors, contributions to prediction before and after the exclusion of the factor.

8 . A prediction method for a system including: a central processing unit (CPU); a second CPU coupled to the first CPU; and a plurality of generators generating electric power, the method comprising:
performing factor selection processing, by the first CPU, including performing a factor selection operation, which is an operation for selecting one or two or more factors from one or a plurality of factors; minimizing a prediction model error, by the first CPU, by performing factor filtering processing before performing the factor selection processing, the factor filtering processing including: determining, about each of the one or the plurality of factors, whether a factor value for prediction input to a prediction model of power demand to calculate a prediction value of a prediction target of power demand is within a range of a plurality of factor values for past specimens used to identify the prediction model; and excluding a factor, a result of the determination of which is false, and outputting a factor not excluded; performing prediction processing, by the first CPU, including calculating the prediction value of the prediction target by inputting, to the prediction model, a factor value for prediction about each of one or more factors including the factor selected in the factor selection processing and not including the factor excluded in the factor filtering processing, wherein the prediction processing includes: adjusting, based on a contribution to prediction of each of the one or more factors, a matching degree of the factor to a prediction model to be identified; and inputting a factor value for prediction of each of the one or more factors to the prediction model based on the matching degree after the adjusting, wherein at least one of the factors is outdoor temperature, and where the method further comprising: by the second CPU, based on the calculated prediction value, creating an operation schedule

Assignees

• Hitachi Ltd

Inventors

• Utsumi Masato
• Watanabe Tohru
• Shigemori Ikuo
• Ito Tetsu
• Hamaba Daisuke
• Yamazaki Jun

Classifications

• G06N20/20

  Ensemble learning · CPC title

• G05B23/02

  Electric testing or monitoring · CPC title

• G06Q10/04

  Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem" (market predictions or forecasting for commercial activities G06Q30/0202) · CPC title

• G06Q30/0202 Primary

  Market predictions or forecasting for commercial activities · CPC title