Prediction apparatus, prediction method, and program
A regression model reflecting causality between variation of an explanatory variable and variation of a response variable is constructed. A prediction apparatus predicts characteristic values of a product by using process data obtained from a production facility. The prediction apparatus includes a process data acquisition unit that reads the process data from a storage device that stores the process data obtained from the production facility, and a prediction model generation unit that generates a prediction model on the basis of causality information that defines a combination of first process data and second process data or a value corresponding to the second process data. The first process data and the second process data or the value corresponding to the second process data are included in the read process data. The first process data is used as a predetermined explanatory variable. The second process data or the value corresponding to the second process data is used as response variable. The prediction model has learned features of the process data obtained from the production facility. The prediction model generation unit generates the prediction model and determines a positive/negative variation direction of the response variable in accordance with a positive/negative variation direction of the predetermined explanatory variable.
1. A prediction apparatus that predicts a characteristic value of a product by using process data obtained from a production facility, the prediction apparatus comprising:
processing circuitry configured to:
read the process data from a storage device configured to store the process data obtained from the production facility; and
generate a prediction model on the basis of causality information that defines a combination of first process data and second process data or a value corresponding to the second process data, the first process data and the second process data or the value corresponding to the second process data being included in the read process data, the first process data being used as a predetermined explanatory variable, the second process data or the value corresponding to the second process data being used as a response variable, the prediction model having learned features of the process data obtained from the production facility, wherein the prediction model for determining a positive/negative variation direction of the response variable is generated in accordance with a positive/negative variation direction of the explanatory variable by performing regression analysis using a penalty function that increases a penalty when the positive/negative variation direction of the response variable and the positive/negative variation direction of the explanatory variable are opposite to a positive/negative variation direction indicated by the causality information; and
calculate a predicted value by using the prediction model and to output the predicted value.
2. The prediction apparatus according to claim 1 , wherein the predictive model is an autoregressive model in which output of a first time point depends at least on output of a second time point earlier than the first time point.
3. The prediction apparatus according to claim 1 , wherein in the causality information, causality is generated by using hazard and operability study (HAZOP), failure mode and effect analysis (FMEA), fault tree analysis (FTA), or event tree analysis (ETA), or by using an analysis method based on any one of the HAZOP, the FMEA, the FTA, or the ETA.
4. The prediction apparatus according to claim 1 , wherein the prediction model has a hierarchical structure including a plurality of prediction equations, and includes a second prediction equation in which a predicted value calculated by a first prediction equation is included as an explanatory variable.
5. The prediction apparatus according to claim 1 , wherein the value corresponding to the second process data is a value obtained by sampling a plurality of second process data according to a sample size reduction method, and the prediction model is generated by correlating a range of acquisition timing of the first process data with calculation timing of the value corresponding to the second process data in the production facility, on the basis of a residence time of a processing target in the production facility.
6. The prediction apparatus according to claim 1 , wherein
the production facility performs
a batch stage of sequentially processing a processing target on a transaction-to-transaction basis using a predetermined transaction, and
a subsequent continuous stage of continuously processing the processing target, and
the processing circuitry is further configured to generate the prediction model by correlating a range of completion timing of the batch stage and the calculation timing of the value corresponding to the second process data on the basis of the residence time of the processing target in the production facility.
7. The prediction apparatus according to claim 1 , the processing circuitry is further configured to predict the characteristic values by using the prediction model and data based on the process data obtained from the production facility or an operating condition.
8. The prediction apparatus according to claim 7 , wherein the processing circuitry is further configured to obtain an error variance for the predicted characteristic values in a predetermined period, and causes an output device to output a confidence interval and the predicted characteristic value, the confidence interval being determined by the error variance and an average value of the predicted characteristic values or measured values of the process data.
9. A prediction method comprising:
reading, by a prediction apparatus that predicts a characteristic value of a product by using process data obtained from a production facility, the process data from a storage device configured to store the process data obtained from the production facility;
generating, by the prediction apparatus, a prediction model on the basis of causality information that defines a combination of first process data and second process data or a value corresponding to the second process data, the first process data and the second process data or the value corresponding to the second process data being included in the read process data, the first process data being used as a predetermined explanatory variable, the second process data or the value corresponding to the second process data being used as a response variable, the prediction model having learned features of the process data obtained from the production facility, wherein the generating of the prediction model includes generating the prediction model for determining a positive/negative variation direction of the response variable in accordance with a positive/negative variation direction of the explanatory variable by performing regression analysis using a penalty function that increases a penalty when the positive/negative variation direction of the response variable and the positive/negative variation direction of the explanatory variable are opposite to a positive/negative variation direction indicated by the causality information;
calculating a predicted value by using the prediction model, and
outputting the predicted value.
10. A non-transitory computer readable medium including a instruction stored thereon causing a prediction apparatus that predicts a characteristic value of a product by using process data obtained from a production facility, when executing the instruction, to perform:
reading the process data from a storage device configured to store the process data obtained from the production facility;
generating a prediction model on the basis of causality information that defines a combination of first process data and second process data or a value corresponding the second process data, the first process data and the second process data or the value corresponding to the second process data being included in the read process data, the first process data being used as a predetermined explanatory variable, the second process data or the value corresponding to the second process data being used as a response variable, the prediction model having learned features of the process data obtained from the production facility, wherein the generating of the prediction model includes generating the prediction model for determining a positive/negative variation direction of the response variable in accordance with a positive/negative variation direction of the predetermined explanatory variable by performing regression analysis using a penalty function that increases a penalty when the positive/negative variation direction of the response variable and the positive/negative variation direction of the explanatory variable are opposite to a positive/negative variation direction indicated by the causality information;
calculating a predicted value by using the prediction model, and
outputting the predicted value.