Information processing device, information processing method, and computer program product

1 . An information processing device comprising:

a memory; and

one or more processors coupled to the memory and configured to:

generate, by machine learning using time-series data of a variable for a phenomenon related to an abnormality in a system to be monitored, a prediction model and a physical model, the prediction model being a model for predicting an indicator used to identify a timing of maintenance of the system, the physical model being a model for predicting the variable; and

perform either one of a first prediction process and a second prediction process, the first prediction process using the physical model that is learned using the indicator predicted by the prediction model, the second prediction process being a process of correcting the indicator predicted by the prediction model, by using the variable predicted using the physical model, wherein

the second prediction process includes predicting the variable by the physical model, predicting the indicator by using relation information indicating a relation between the predicted variable and the indicator, and correcting the indicator predicted by the prediction model by using the indicator predicted by using the relation information.

2 . The device according to claim 1 , wherein

the memory is configured to store therein a plurality of types of sub-libraries including nonlinear basis functions based on a dependent variable or an independent variable and generation probabilities of the nonlinear basis functions included in each of the sub-libraries, wherein

the one or more processors are configured to generate the prediction model and the physical model by combining one or more of the nonlinear basis functions extracted based on the generation probabilities from the sub-libraries.

3 . The device according to claim 2 , wherein

the one or more processors are configured to correct the generation probabilities to increase the generation probability of a nonlinear basis function having a greater influence on the indicator than other nonlinear basis functions among the nonlinear basis functions, and

the first prediction process uses the physical model generated by combining the nonlinear basis functions extracted based on the generation probabilities corrected.

4 . The device according to claim 2 , wherein

the one or more processors are configured to:

generate, for an object model that is either of the prediction model and the physical model, a plurality of the object models by combining one or more of the nonlinear basis functions extracted based on the generation probabilities from the plurality of types of sub-libraries, and calculate loss functions for the respective object models;

correct the generation probabilities and hyperparameters of the machine learning, based on the loss functions, and

regenerate, by machine learning using the hyperparameters corrected, the object models by combining one or more of the nonlinear basis functions extracted based on the generation probabilities corrected from the plurality of types of sub-libraries.

5 . The device according to claim 4 , wherein the one or more processors are configured to output information indicating the object models selected by a rank order based on the loss functions.

6 . The device according to claim 1 , wherein the second prediction process includes

predicting the variable by the physical model,

predicting the indicator by using the relation information indicating the relation between the predicted variable and the indicator,

predicting a prior probability distribution of the indicator by the prediction model, and

calculating a posterior probability distribution based on multiplication of the prior probability distribution by the indicator that is predicted by using the relation information.

7 . The device according to claim 1 , wherein

an object model that is either of the prediction model and the physical model includes a linear regression equation, and

the one or more processors are configured to estimate a coefficient of the linear regression equation by sparse estimation.

8 . An information processing method to be executed by an information processing device, the method comprising:

generating, by machine learning using time-series data of a variable for a phenomenon related to an abnormality in a system to be monitored, a prediction model and a physical model, the prediction model being a model for predicting an indicator used to identify a timing of maintenance of the system, the physical model being a model for predicting the variable; and

performing either one of a first prediction process and a second prediction process, the first prediction process using the physical model that is learned using the indicator predicted by the prediction model, the second prediction process being a process of correcting the indicator predicted by the prediction model, by using the variable predicted using the physical model, wherein

the second prediction process includes predicting the variable by the physical model, predicting the indicator by using relation information indicating a relation between the predicted variable and the indicator, and correcting the indicator predicted by the prediction model by using the indicator predicted by using the relation information.

9 . A computer program product comprising a non-transitory computer-readable medium including programmed instructions, the instructions causing a computer to execute:

generating, by machine learning using time-series data of a variable for a phenomenon related to an abnormality in a system to be monitored, a prediction model and a physical model, the prediction model being a model for predicting an indicator used to identify a timing of maintenance of the system, the physical model being a model for predicting the variable; and

performing either one of a first prediction process and a second prediction process, the first prediction process using the physical model that is learned using the indicator predicted by the prediction model, the second prediction process being a process of correcting the indicator predicted by the prediction model, by using the variable predicted using the physical model, wherein

the second prediction process includes predicting the variable by the physical model, predicting the indicator by using relation information indicating a relation between the predicted variable and the indicator, and correcting the indicator predicted by the prediction model by using the indicator predicted by using the relation information.

10 . The device according to claim 6 , wherein

the one or more processors are configured to perform:

predicting the variable by a second physical model,

predicting the indicator by using the relation information and the variable predicted by the second physical model, and

calculating a second posterior probability distribution based on multiplication of the posterior probability distribution by the indicator that is predicted by using the variable predicted by the second physical model.