Modifying a forecasting model based on qualitative information
Techniques regarding modifying a forecasting model are provided. For example, one or more embodiments described herein can comprise a modeling system, which can comprise a memory that can store computer executable components. The modeling system can also comprise a processor, operably coupled to the memory, and that can execute the computer executable components stored in the memory. The computer executable components can include an identifying component that can identify qualitative information related to a forecasting model of an operation of a system, wherein the forecasting model was generated through data obtained from the operation of the system. The computer executable components can include a modifying component that can modify the forecasting model based on intervention information generated based on the qualitative information, wherein the intervention information is associated with a prediction of the forecasting model.
1 . A system comprising:
a memory that stores computer-executable components; and
a processor, operatively coupled to the memory, that executes at least one of the computer-executable components that:
identifies qualitative information associated with predictions generated by a forecasting neural network model during historical operations of an industrial process, wherein the qualitative information is specified in a descriptive language, and wherein the forecasting neural network model was trained using a training dataset comprising historical quantitative data obtained from equipment used in the industrial process during the historical operations of the industrial process,
identifies a subset of the training dataset associated with a portion of the forecasting neural network model associated with erroneous predictions of the predictions based on a defined criterion associated with at least one of sensitivity of prediction or quantity of training data;
identifies a subset of the qualitative information associated with the subset of the training dataset;
generates intervention information based on the subset of the qualitative information and the subset of the training dataset, wherein the generating comprises converting the subset of the qualitative information from the descriptive language into a declarative language;
retrains the forecasting neural network model using the intervention information to mitigate the erroneous predictions associated with the portion of the forecasting neural network model; and
controls, using the retrained forecasting neural network model, operations of the industrial process.
2 . The system of claim 1 , wherein the qualitative information comprises an indication that a portion of the industrial process operates differently than has been modeled by the forecasting neural network model.
3 . The system of claim 2 , wherein the indication comprises qualitative monotonic information regarding a prediction and an operation of the industrial process.
4 . The system of claim 1 , wherein identifying the subset of the training dataset comprises identifying the subset of the training dataset associated with a portion of the industrial process where collecting certain data is dangerous.
5 . The system of claim 1 , wherein identifying the subset of the training dataset comprises identifying the subset of the training dataset associated with a portion of the industrial process associated with a lower quantity of training data than other portions of the industrial process.
6 . The system of claim 1 , wherein identifying the subset of the training dataset comprises identifying the subset of the training dataset associated with a portion of the industrial process wherein collecting certain data would be destructive to the industrial process.
7 . The system of claim 1 , wherein the intervention information comprises at least one intervention.
8 . The system of claim 7 , wherein generating the intervention information further comprises generating at least one metric to evaluate the at least one intervention.
9 . A computer-implemented method, comprising:
identifying, by a system operatively coupled to a processor, qualitative information associated with predictions generated by a forecasting neural network model during historical operations of an industrial process, wherein the qualitative information is specified in a descriptive language, and wherein the forecasting neural network model was trained using a training dataset comprising historical quantitative data obtained from equipment used in the industrial process during the historical operations of the industrial process;
identifying, by the system, a subset of the training dataset associated with a portion of the forecasting neural network model associated with erroneous predictions of the predictions based on a defined criterion associated with at least one of sensitivity of prediction or quantity of training data;
identifying, by the system, a subset of the qualitative information associated with the subset of the training dataset;
generating, by the system, intervention information based on the subset of the qualitative information and the subset of the training dataset, wherein the generating comprises converting the subset of the qualitative information from the descriptive language into a declarative language;
retraining, by the system, the forecasting neural network model using the intervention information to mitigate the erroneous predictions associated with the portion of the forecasting neural network model; and
controlling, by the system, using the retrained forecasting neural network model, operations of the industrial process.
10 . The computer-implemented method of claim 9 , wherein the qualitative information comprises an indication that a portion of the industrial process operates differently than has been modeled by the forecasting neural network model.
11 . The computer-implemented method of claim 10 , wherein the indication comprises qualitative monotonic information regarding a prediction and an operation of the industrial process.
12 . The computer-implemented method of claim 9 , wherein identifying the subset of the training dataset comprises identifying the subset of the training dataset associated with a portion of the industrial process where collecting certain data is dangerous.
13 . The computer-implemented method of claim 9 , wherein identifying the subset of the training dataset comprises identifying the subset of the training dataset associated with a portion of the industrial process associated with a lower quantity of training data than other portions of the industrial process.
14 . The computer-implemented method of claim 9 , wherein identifying the subset of the training dataset comprises identifying the subset of the training dataset associated with a portion of the industrial process wherein collecting certain data would be destructive to the industrial process.
15 . A computer program product comprising a computer readable storage medium having program instructions embodied therewith, wherein the program instructions are executable by a processor to cause the processor to:
identify qualitative information associated with predictions generated by a forecasting neural network model during historical operations of an industrial process, wherein the qualitative information is specified in a descriptive language, and wherein the forecasting neural network model was trained using a training dataset comprising historical quantitative data obtained from equipment used in the industrial process during the historical operations of the industrial process;
identify a subset of the training dataset associated with a portion of the forecasting neural network model associated with erroneous predictions of the predictions based on a defined criterion associated with at least one of sensitivity of prediction or quantity of training data;
identify a subset of the qualitative information associated with the subset of the training dataset;
generate intervention information based on the subset of the qualitative information and the subset of the training dataset, wherein the generating comprises converting the subset of the qualitative information from the descriptive language into a declarative language;
retrain the forecasting neural network model using the intervention information to mitigate the erroneous predictions associated with the portion of the forecasting neural network model; and
control, using the retrained forecasting neural network model, operations of the industrial process.
16 . The computer program product of claim 15 , wherein the qualitative information comprises an indication that a portion of the industrial process operates differently than has been modeled by the forecasting neural network model.
17 . The computer program product of claim 16 , wherein the indication comprises qualitative monotonic information regarding a prediction and an operation of the industrial process.
18 . The computer program product of claim 15 , wherein identifying the subset of the training dataset comprises identifying the subset of the training dataset associated with a portion of the industrial process where collecting certain data is dangerous.
19 . The computer program product of claim 15 , wherein identifying the subset of the training dataset comprises identifying the subset of the training dataset associated with a portion of the industrial process associated with a lower quantity of training data than other portions of the industrial process.
20 . The computer program product of claim 15 , wherein identifying the subset of the training dataset comprises identifying the subset of the training dataset associated with a portion of the industrial process wherein collecting certain data would be destructive to the industrial process.