发明名称 Prediction method, prediction system and program
摘要 A method for predicting an output variable from explanatory values provided as sets of combinations of discrete variables and continuous variables includes receiving input data that contains the explanatory variables to predict the output variable; searching for each element in the combinations for elements in a plurality of sets with matching discrete variables using training data which the output variable has been observed; applying a function giving the degree of similarity between two sets weighed by a scale variable to each element in the input data, and to one or more elements found in the elements of the input data to calculate function values, and calculating the sum of the function values for all of the elements in the input data; and applying the calculated sum for each element to a prediction equation for predicting the output variable to calculate a prediction value of the output variable for each element.
申请公布号 US9087294(B2) 申请公布日期 2015.07.21
申请号 US201313732693 申请日期 2013.01.02
申请人 International Business Machines Corporation 发明人 Kojaku Sadamori;Morimura Tetsuro;Osogami Takayuki;Takahaski Rikiya
分类号 G06F17/00;G06N5/02;G06N99/00;G06F17/18;G06Q50/06 主分类号 G06F17/00
代理机构 Cantor Colburn LLP 代理人 Cantor Colburn LLP
主权项 1. A method for predicting an output variable from explanatory values provided as sets of combinations of discrete variables and continuous variables, wherein the method comprises: receiving, with a processing device, input data that contains the explanatory variables to predict the output variable; searching for each element in the combinations of discrete variables and continuous variables in the sets included in the input data for elements in a plurality of sets with matching discrete variables using training data related to a plurality of sets in which the output variable has been observed; applying a function giving the degree of similarity between two sets weighed by a scale variable that represents a scalar corresponding to the output variable and is obtained from a continuous variable to each element in the input data, and to one or more elements found in the elements of the input data to calculate function values, and for calculating the sum of the function values for all of the elements in the input data; and applying the calculated sum for each element to a prediction equation for predicting the output variable to calculate a prediction value of the output variable for each element; wherein a discrete variable is a variable representing the type of raw material included in a unit in which the output variable has been observed, continuous variables are represented as a vector summarizing continuous quantities related to the raw material included in the unit, a scale variable is a value corresponding to the output variable in the calculation results of an equation including the volume, mass or other variable of the raw material included in the unit and a plurality of these variables, and the output variable is the amount of power consumed by the unit.
地址 Armonk NY US