Method and system for predicting KPI values, plant states and alarms in an industrial process
Predetermined Key Performance Indicators (KPIs) of an industrial process may be predicted. A plurality of tags each identify a corresponding KPI of the industrial process and historical values for the KPIs that are identified by the plurality of tags. A KPI forecast model is trained for each of the KPIs that are identified by the plurality of tags, wherein each of the KPI forecast models is trained based at least in part on the received historical values for at least some of the KPIs that are identified by the plurality of tags. A forecasted KPI value is generated for each of the KPIs identified by the plurality of tags based at least in part on the corresponding KPI forecast model that corresponds to the respective KPI.
1 . A method for predicting predetermined Key Performance Indicators (KPIs) of an industrial process, the method comprising:
receiving, by a controller, a plurality of tags, wherein each tag identifies a corresponding KPI of the industrial process;
receiving, by the controller, historical values for each of the KPIs that are identified by the plurality of tags;
training, by the controller, a KPI forecast model for each of the KPIs that are identified by the plurality of tags, wherein each of the KPI forecast models is trained based at least in part on the received historical values for one or more KPIs of a plurality of the KPIs that are identified by the plurality of tags; and
generating, by the controller, a forecasted KPI value for each of the KPIs identified by the plurality of tags based at least in part on the KPI forecast model that corresponds to the respective KPI;
receiving, by the controller, one or more alarm limits, historical alarms, and/or events associated with the one or more KPIs of the plurality of KPIs;
identifying, by the controller, one or more overlapping episodes associated with the one or more KPIs and related to the historical alarms and/or the events;
excluding, by the controller, the one or more overlapping episodes of the historical alarms and/or the events that are associated with the one or more KPIs of the plurality of KPIs;
training, by the controller, an alarm forecast model based on at least on the events associated with the one or more KPIs exclusive of the historical alarms and/or the events related to the one or more overlapping episodes of the historical alarms and/or the events; and
determining, by the controller, using the trained alarm forecast model whether the forecasted KPI value is outside the corresponding alarm limit;
forecasting, by the controller, an alarm for each of the one or more KPIs based on the trained alarm forecast model, wherein the alarm forecasted by the trained alarm forecast model excludes the one or more overlapping episodes; and
automatically adjusting, by the controller, one or more parameters associated with the industrial process based on the forecasted KPI values and the forecasted alarms, wherein the one or more parameters associated with the industrial process are adjusted to negate the forecasted alarm.
2 . The method of claim 1 , comprising:
receiving a time to forecast KPI value for each of the plurality of KPIs; and
generating the forecasted KPI value for each of the plurality of KPIs identified by the plurality of tags at the time in future that corresponds to the time to forecast the KPI value.
3 . The method of claim 2 , wherein each of the KPI forecast models is trained based at least in part on the received historical values for the one or more KPIs of the plurality of KPIs of the industrial process that are identified by the plurality of tags and one or more of the forecasted KPI values.
4 . The method of claim 1 , wherein the industrial process comprises a plurality of plants, wherein each plant comprising two or more tags of the plurality of tags that each identify a corresponding KPI of the corresponding plant, and each plant has two or more predetermined plant states, and the method further comprising:
training a plant state forecast model for each of the plurality of plants, wherein each of the plant state forecast models is trained to forecast the plant state of the respective plant based at least in part on the received historical values for the corresponding KPIs that are associated with the respective plant and/or the forecasted KPI values for the corresponding KPIs that are associated with the respective plant; and
generating a forecasted plant state for at least one of the plurality of plants of the industrial process based at least in part on the plant state forecast model that corresponds to the respective plant.
5 . The method of claim 4 , wherein each of the plant state forecast models is used to forecast the plant state of the respective plant based at least in part on the received historical values for the KPIs that are associated with the respective plant and one or more of the forecasted KPI values for one or more of the plurality of KPIs that are associated with the respective plant.
6 . The method of claim 4 , further comprising:
receiving the one or more alarm limits for the one or more KPIs of the plurality of KPIs associated with each of the plurality of plants; and
wherein each of the plant state forecast models is trained to forecast the plant state of the respective plant based at least in part on the received historical values for the KPIs that are associated with the respective plant and the one or more alarm limits for the one or more KPIs of the plurality of KPIs associated with the respective plant.
7 . The method of claim 6 , wherein each of the plant state forecast models is trained to forecast the plant state of the respective plant based at least in part on the received historical values for the KPIs that are associated with the respective plant, the one or more alarm limits for the one or more KPIs of the plurality of KPIs associated with the respective plant, and one or more of the forecasted KPI values for one or more of the plurality of KPIs that are associated with the respective plant.
8 . The method of claim 1 , further comprising:
training the alarm forecast model for the one or more KPIs of the plurality of KPIs, wherein the alarm forecast model is trained based on the received historical values for the one or more KPIs of the plurality of KPIs, the historical alarms and/or the events associated with one or more KPIs of the plurality of KPIs, and one or more of the alarm limits for one or more KPIs of the plurality of KPIs.
9 . The method of claim 8 , wherein the alarm forecast model is also trained based on one or more of the forecasted KPI values.
10 . The method of claim 8 , further comprising:
generating a forecasted alarm for each of the one or more KPIs of the plurality of KPIs based the forecasted KPI value for the respective KPI.
11 . The method of claim 8 , wherein each of the historical alarms and/or the events associated with the one or more KPIs of the plurality of KPIs identify a source of the historical alarm and/or the event, a category of the historical alarm and/or the event and a condition of the historical alarm and/or the event, and wherein the alarm forecast model is trained based on the source, the category and/or the condition of one or more of the historical alarms and/or the events associated with the one or more KPIs of the plurality of KPIs of the industrial process.
12 . A system for predicting predetermined Key Performance Indicators (KPIs) of an industrial process, the system comprising:
an I/O port;
a memory;
a controller operatively coupled to the I/O port and the memory, the controller configured to:
receive, via the I/O port, a plurality of tags, wherein each tag identify a corresponding KPI of the industrial process;
receive, via the I/O port, historical values for each of the KPIs that are identified by the plurality of tags;
train a KPI forecast model for each of the KPIs that are identified by the plurality of tags, wherein each of the KPI forecast models is trained based at least in part on the received historical values for one or more KPIs of a plurality of KPIs that are identified by the plurality of tags; and
generate a forecasted KPI value for each of the plurality of KPIs identified by the plurality of tags based at least in part on the corresponding KPI forecast model that corresponds to the respective KPI;
receive one or more alarm limits, historical alarms, and/or events associated with the one or more KPIs of the plurality of KPIs;
identify one or more overlapping episodes associated with the one or more KPIs of the plurality of KPIs and related to the historical alarms and/or the events;
exclude, the one or more overlapping episodes of the historical alarms and/or the events that are associated with the one or more KPIs of the plurality of KPIs;
training, by the controller, an alarm forecast model based on at least on the events associated with the one or more KPIs of the plurality of KPIs exclusive of the historical alarms and/or the events related to the one or more overlapping episodes of the historical alarms and/or the events; and
determine, using the trained alarm forecast model whether the forecasted KPI value is expected to fall outside the corresponding alarm limit;
forecast, an alarm for each of the one or more KPIs based on the trained alarm forecast model, wherein the alarm forecasted by the trained alarm forecast model excludes the one or more overlapping episodes; and
automatically adjust, one or more parameters associated with the industrial process based on the forecasted KPI values and the forecasted alarms, wherein the one or more parameters associated with the industrial process are adjusted to negate the forecasted alarm.
13 . The system of claim 12 , wherein each of the KPI forecast models is trained based at least in part on the received historical values for the one or more KPIs of the plurality of KPIs of the industrial process that are identified by the plurality of tags and one or more of the forecasted KPI values.
14 . The system of claim 12 , wherein the industrial process comprises a plurality of plants, wherein each plant comprising two or more tags of the plurality of tags that each identify a corresponding KPI of the corresponding plant, and each plant has two or more predetermined plant states, and the controller is configured to:
train a plant state forecast model for each of the plurality of plants, wherein each of the plant state forecast models is trained to forecast the plant state of the respective plant based at least in part on the received historical values for the corresponding KPIs that are associated with the respective plant; and
generate a forecasted plant state for at least one of the plurality of plants of the industrial process based at least in part on the plant state forecast model that corresponds to the respective plant.
15 . The system of claim 12 , wherein the controller is configured to:
train the alarm forecast model for the one or more KPIs of the plurality of KPIs, wherein the alarm forecast model is trained based on the received historical values for the one or more KPIs of the plurality of KPIs, the historical alarms and/or the events associated with the one or more KPIs of the plurality of KPIs, and one or more of the alarm limits for the one or more KPIs of the KPIs.
16 . A non-transitory computer readably medium storing instructions that when executed by one or more processors causes the one or more processors to:
receive a plurality of tags wherein each tag identify a corresponding KPI of an industrial process;
receive historical values for each of the KPIs that are identified by the plurality of tags;
train a KPI forecast model for each of the KPIs that are identified by the plurality of tags, wherein each of the KPI forecast models is trained based at least in part on the received historical values for one or more KPIs of a plurality of KPIs that are identified by the plurality of tags; and
generate a forecasted KPI value for each of the KPIs identified by the plurality of tags based at least in part on the corresponding KPI forecast model that corresponds to the respective KPI;
receive one or more alarm limits, historical alarms, and/or events associated with the one or more KPIs of the plurality of KPIs;
identify one or more overlapping episodes associated with the one or more KPIs of the plurality of KPIs and related to the historical alarms and/or the events;
exclude, the one or more overlapping episodes of the historical alarms and/or the events that are associated with the one or more KPIs of the plurality of KPIs;
train, by the controller, an alarm forecast model based on at least on the events associated with the one or more KPIs of the plurality of KPIs exclusive of the historical alarms and/or the events related to the one or more overlapping episodes of the historical alarms and/or the events; and
determine, using the trained alarm forecast model whether the forecasted KPI value is expected to fall outside the corresponding alarm limit;
forecast, an alarm for each of the one or more KPIs based on the trained alarm forecast model, wherein the alarm forecasted by the trained alarm forecast model excludes the one or more overlapping episodes; and
automatically adjust, one or more parameters associated with the industrial process based on the forecasted KPI values and the forecasted alarms, wherein the one or more parameters associated with the industrial process are adjusted to negate the forecasted alarm.