Prediction model generation device, prediction model generation method, and recording medium

A prediction model generation method comprising:

• storing a plurality of explanatory variables that affect demand of power in a first storage unit;
• storing a plurality of objective variables in a second storage unit;
• inputting instruction information on classification based on at least one of a date and time condition and an environmental condition via an input unit;
• generating a plurality of classes based on the instruction information;
• and
• calculating a plurality of prediction models corresponding to the plurality of classes,
• when calculating the plurality of prediction models corresponding to the plurality of classes,
• extracting a learning data set corresponding to each of the plurality of classes from among the plurality of explanatory variables stored in the first storage unit and the plurality of objective variables stored in the second storage unit;
• extracting an evaluation data set corresponding to each of the plurality of classes from among the plurality of explanatory variables stored in the first storage unit and the plurality of objective variables stored in the second storage unit;
• calculating a parameter included in each of the plurality of prediction models based on the learning data set;
• evaluating prediction performance of the plurality of prediction models based on the evaluation data set;
• calculating a plurality of prediction submodels, based on a corresponding parameter for each of the plurality of prediction models;
• and
• setting weights based on prediction errors to be given to each of the plurality of prediction submodels,
• wherein prediction values based on the plurality of prediction models are values obtained by adding values obtained by multiplying the prediction values of the plurality of prediction submodels by corresponding weights.

A non-transitory computer-readable recording medium that stores a program, the program that causes a computer to execute:

• storing a plurality of explanatory variables that affect demand of power;
• storing a plurality of objective variables;
• inputting instruction information on classification based on at least one of a date and time condition and an environmental condition;
• generating a plurality of classes based on the instruction information;
• and
• calculating a plurality of prediction models corresponding to the plurality of classes,
• the calculating the plurality of prediction models comprises:
• extracting a learning data set corresponding to each of the plurality of classes from among the plurality of explanatory variables and the plurality of objective variables;
• extracting an evaluation data set corresponding to each of the plurality of classes from among the plurality of explanatory variables and the plurality of objective variables;
• calculating a parameter included in each of the plurality of prediction models based on the learning data set;
• evaluating prediction performance of the plurality of prediction models based on the evaluation data set;
• calculating a plurality of prediction submodels, based on a corresponding parameter for each of the plurality of prediction models;
• and
• setting weights based on prediction errors to be given to each of the plurality of prediction submodels,
• wherein prediction values based on the plurality of prediction models are values obtained by adding values obtained by multiplying the prediction values of the plurality of prediction submodels by corresponding weights.