Methods used to generate process models

Abstract

A method (100) is proposed for generating a process model for modeling a manual mode program instance of a plant process, the program instance including a related sequence of operating actions, i.e., mining a plant history log to generate a workflow model, comprising the following steps: providing a plurality of log events (S1) of a plurality of operating actions (110) of the plant process; selecting a plurality of related sequences of manual mode operating actions from the plurality of log events (S2); filtering the plurality of related sequences of manual mode operating actions according to individual plant parts (S3); identifying a succession order from the filtered plurality of related sequences of manual mode operating actions (S4); determining statistical properties of the values ​​of the related process variables and / or the statistical properties of the values ​​of the related set point changes for each successively ordered manual mode operating action from the filtered plurality of related sequences of manual mode operating actions (S5); and generating a process model (S6) of the manual mode program instance by arranging the related manual mode operating actions with the succession order of each operating action, each operating action being assigned a statistical property of the value of the related process variable and / or a statistical property of the value of the related set point change.

Description

Technical Field

[0001] The present invention relates to a method for generating a process model, which models a plurality of manual mode program instances of a plant process. Background Art

[0002] In plant operations, certain plant-related procedures, such as starting up or shutting down equipment or emptying tanks, are often performed manually in the operator's office. According to the current state of the art, operators are provided with special operator displays or event batching recipes to enforce reliable execution of manual procedure instances. However, these efforts to guide operators are often insufficient, and the quality of execution depends entirely on the operator's experience, diligence, and current mental state. Summary of the Invention

[0003] The present invention therefore relates to a method for generating a process model, a method for generating recommended set points, a method for generating a warning signal, a support system, a use of a generated process model, a computer program and a computer-readable storage medium, the subject matter of which is as described in the independent claims.

[0004] Advantageous modifications of the invention are set forth in the dependent claims. All combinations of at least two of the features disclosed in the description, the claims, and the drawings fall within the scope of the invention. To avoid repetition, features disclosed with respect to the method should also apply to the system described and can be claimed with respect to that system.

[0005] In other words, the present invention provides a method of generating a workflow model as a process model for a manual mode program instance, wherein knowledge is learned from sequences of manual operation actions performed by an operator by mining historically corresponding past log data.

[0006] The basic concept of the present invention is to record operator actions during multiple operating procedures, or to extract such operator actions and related knowledge from past event logs. The recorded procedures, i.e., procedure instances, are translated into a process model, which can be used to guide the operator through the execution of each corresponding procedure instance according to the specified rules of the process model and / or to monitor or check the execution of the procedure.

[0007] Throughout this description of the present invention, sequences of procedural steps are presented to make the process easier to understand. However, a skilled artisan will recognize that many of the procedural steps can also be performed in a different order and result in the same or corresponding results. In this sense, the order of the procedural steps can be changed accordingly. Some features have counters to improve readability or to make the assignment more explicit, but this does not imply the presence of certain features.

[0008] To achieve these and other advantages and in accordance with the purposes of the present invention, as embodied and broadly described herein, there is provided a method for generating a process model that models an artificial mode program instance of a plant process, the program instance including a sequence of related operational actions, which includes the following steps.

[0009] In a step of the method, a plurality of log events of a plurality of operating actions of the plant process are provided. Another step of the method selects a plurality of related sequences of manual mode operating actions from the plurality of log events. Another step of the method filters the plurality of related sequences of manual mode operating actions according to individual plant parts. Another step of the method identifies a sequential order from the plurality of filtered related sequences of manual mode operating actions. Another step of the method determines statistical properties of the values ​​of the related process variables and / or statistical properties of the values ​​of the related set point changes for each sequentially ordered manual mode operating action from the plurality of filtered related sequences of manual mode operating actions. Another step of the method generates a process model of the manual mode program instance by arranging the related manual mode operating actions in a sequential order, and each operating action is assigned statistical properties of the values ​​of the related process variables and / or statistical properties of the values ​​of the related set point changes.

[0010] In other words, the method is described as recording operator actions during an operational procedure or extracting such operator actions from a historically relevant past event log. The recorded procedures, referred to as procedure instances, are translated into a process model that can be used to guide operators through the execution of the corresponding manual procedures and monitor their compliance with the relevant rules defined by the simple waterfall process model.

[0011] In this context, a manual mode operational action is an operational action performed by a human operator with respect to a control loop of an automated control system.

[0012] Operator actions, such as setpoint changes, can be derived directly from the sequence of operator actions. Setpoints are typically recorded as time series. If the event log does not directly include setpoint changes, they can be derived from the time series data. Whenever the time series changes and the setpoint is not being controlled by the model predictive controller at the time, this corresponds to an operator action.

[0013] A process model extractor can be constructed that performs the described method of analyzing log files and generating a process model with a partial action sequence diagram, and uses a user-provided list of relevant process variables to create preconditions for operator actions. Thus, the method can be viewed as at least semi-automatically generating a process model for an operating procedure by extracting the process model from the control system log files.

[0014] Advantageously, by this method, technical knowledge about the results generated by the operator's manual operating actions is derived and summarized by means of a process model.

[0015] The term setpoint includes any input to a plant or a plant part, respectively any input to a control loop controlling a process value of the plant, which means that the term setpoint includes controller setpoints and / or actuation values, as eg valve settings.

[0016] According to one aspect, filtering of multiple related sequences of manual mode operating actions is performed based on individual plant conditions.

[0017] By this additional filtering with respect to individual plant conditions, more technological knowledge details can be included in the process model.

[0018] According to one aspect, the individual plant conditions are determined by at least threshold values ​​of the relevant process variables at the start of the manual mode program instance.

[0019] To identify individual plant conditions, threshold values ​​for process variables associated with individual manual mode program instances may be used to gain more detailed technical knowledge about the manual mode program instances.

[0020] That is, program instances that reach the same threshold at the start time and / or are within a specified range characterized by the threshold may constitute a group of artificial mode program instances.

[0021] This aspect of the described method is particularly relevant when the exact course of operational action depends on the specific nature of the plant situation.The threshold value may be defined by plant experts.

[0022] For this method, the plant expert will define multiple thresholds for signals and process variables, for example, before starting a manual mode program instance, the temperature threshold is defined as T1245>100°, the pressure threshold is defined as P3214<30 bar, etc. For each program instance, these thresholds will be checked at the start time of the program instance.

[0023] According to one aspect, individual plant conditions are determined by threshold values ​​for a subset of relevant process variables at the start of a manual mode program instance.

[0024] Advantageously, there may be individual plant conditions characterized by a subset of related process values ​​for characterizing the plant condition in more detail.

[0025] According to one aspect, the individual plant conditions are determined by at least one time series of values ​​of a relevant process variable of the plant part during a time period before the start of the manual mode program instance.

[0026] Advantageously, individual plant conditions can be characterized by at least one time severity of the value of the associated process variable for a more detailed characterization of the plant condition. This means that the history of the plant process can be correlated to running manual mode program instances via the time severity.

[0027] According to one aspect, the individual plant conditions are determined by the number of clusters of time series of values ​​of process variables of the plant parts during a period before the start of the manual mode program instance, wherein the clusters are constructed by scoring the similarity of the time series of values ​​of the related process variables.

[0028] By clustering the time series, for example using methods from unsupervised machine learning, such as k-means, DBSCAN, agglomerative clustering, mean-shift clustering, or other methods, the prerequisites for manual mode program instances can advantageously be characterized in a very specific manner, and thus the process model can more accurately reflect the operator's technical knowledge. The scoring similarity can be measured by a minimum distance or another similarity measure of the time series of the values ​​of the relevant process variables involved.

[0029] The clustering can also be performed with a subset of the time series of values ​​of the relevant process variables, which subset is preselected by plant experts or defined by filtering out signals with low information content, for example measured by information entropy.

[0030] After performing piecewise aggregation approximation or symbolic aggregation approximation and / or Chebyshev distance calculation on the characteristics of the manual operation actions of the corresponding manual mode program instance, the scoring of the similarity of at least one plant condition is completed by calculating Euclidean distance and / or dynamic time warping and / or cosine similarity and / or Levenshtein similarity.

[0031] According to one aspect, the individual plant conditions are determined by the values ​​of the process variables and / or the setpoint values ​​and / or the plant parts of the manual operation actions and / or the setpoint values ​​and / or the number of manual operation actions of the manual mode program instance.

[0032] This allows the operator's more specific technical knowledge to be integrated into the process model.

[0033] According to one aspect, annotations associated with manual mode operating actions are assigned to the corresponding manual mode operating actions.

[0034] These comments can later help the operator to run the manual mode program instance taking the comments into account.

[0035] According to one aspect, manual mode operating actions having assigned statistical properties with associated process variable values ​​above a limit value are omitted from the process model.

[0036] Through this aspect of the method, atypical process variations do not interfere with the design of a correct process model.

[0037] According to one aspect of the present invention, among the plurality of related sequences of manual mode operating actions, specific manual mode operating actions that rarely occur within the plurality of related sequences of manual mode operating actions are omitted from the process model.

[0038] The term rarely occurs may be understood as a predefined lower limit on the relative occurrence rate within all sequences of manual mode operating actions considered when generating the process model.

[0039] This helps improve process models and capture operators’ technical knowledge by not taking into account potentially irrelevant operating actions.

[0040] A method according to one of the preceding claims is provided for generating recommended set points for manual mode operating actions for a manual mode program instance, wherein the recommended set points are determined using a machine learning model from a plurality of associated set points associated with individual plant conditions to be assigned to the associated manual mode operating actions with respect to the individual plant conditions.

[0041] Using machine learning to recommend setpoint values ​​to change can improve the use of process models for operator-run instances of manual mode procedures, giving precise recommendations rather than just providing a range of possible values ​​for setpoint changes.

[0042] A method for generating a warning signal related to a plant process is provided, comprising the following steps. In the method steps, log events of the plant process are continuously provided. In a further step, individual plant conditions are determined from the log events of the plant process. In a further step, the warning signal is generated based on the individual plant conditions adapted to the process model according to the preceding claim.

[0043] Such a warning signal may indicate to the operator to initiate a manual mode procedure instance associated with the plant process.

[0044] By monitoring operator actions compared to the process model, the start of a plant-related procedure can be detected and the operator alerted to start a manual mode process instance.

[0045] A support system is provided that includes a controller configured to support an operator in executing manual mode operational actions according to a process model generated according to the method described above by displaying relevant process values ​​and / or recommended setpoint changes to the operator based on a sequence of manual mode operational actions generated by the process model. The support system can support and monitor the execution of a manual mode process instance by using the process model at runtime to guide an operator through the manual mode process instance and monitor its execution.

[0046] The process model is created by using a simple waterfall process model and a dialog system that is able to interpret the waterfall process model and simultaneously monitor conditions in the process plant, map them on pre-set conditions of process steps, notify the operator when the pre-set conditions are met, warn the operator when he takes an action that is not part of the process or takes an action that has not yet met the pre-set conditions, and monitor and create protocols for specific execution procedures.

[0047] In other words, during the execution of an operating procedure, monitoring using the described process model allows for automatic triggering of manual mode operating actions using technical knowledge of previously observed operator actions to avoid operators performing actions without system support.

[0048] According to one aspect, the support system includes a process model editor that enables an operator to fine-tune the process model, for example by using a dialog-based or wizard-like system external to the actual control system.

[0049] The invention provides a use of the process model generated according to the above method to support an operator in performing manual mode operation actions of a manual mode program instance.

[0050] According to another aspect, a computer program is disclosed comprising instructions which, when executed by a computer, cause the computer to perform one of the described methods.

[0051] According to another aspect of the present invention, a computer-readable storage medium is disclosed on which a computer program is stored. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The accompanying drawings are included to provide a further understanding of the present invention and are incorporated in and constitute a part of this application, illustrate embodiments of the present invention, and together with the description serve to explain the principles of the present invention. The drawings show:

[0053] Figure 1 is a flow chart of the method used to generate the process model;

[0054] Figure 2 is a time series of values ​​of process variables and associated log file entries;

[0055] Figure 3 is an example of a program instance and its transformation;

[0056] Figure 4 is a matrix of procedures and resulting process models; and

[0057] Figure 5 It is a continuous display content to support running manual mode process instances. DETAILED DESCRIPTION

[0058] Figure 1 An overview of a method 100 for generating a process model that models a manual mode program instance of a plant process, wherein the program instance includes a related sequence of operating actions, is schematically shown.

[0059] In a first step S1, a plurality of log events of a plurality of operating actions of a plant process may be provided by, for example, the control system log file 110. In another step S2, a plurality of related sequences of manual mode operating actions are selected from the plurality of log events.

[0060] In a further step S3, a plurality of related sequences of manual mode operating actions are filtered according to individual plant parts.

[0061] In a further step S4, a sequential order is identified from the plurality of filtered related sequences of manual mode operating actions.

[0062] In a further step S5, statistical properties of the values ​​of the associated process variables are determined from the plurality of filtered related sequences of manual mode operating actions and / or statistical properties of the values ​​of the associated set point changes for each ordered manual mode operating action.

[0063] This step may comprise a step S5a of determining statistical properties of the values ​​of the relevant process variables, and a step S5b of determining statistical properties of the values ​​of the relevant set point changes, and a method step S5c, wherein in the method step, a set point for recommending a set point to an operator, for example, is determined using a machine learning model from a plurality of relevant set points associated with the individual plant conditions to assign to the relevant manual mode operating action with respect to the individual plant conditions.

[0064] Statistical properties may characterize upper and lower bounds on a record of setpoint selections from a manual mode program instance, as logged by multiple operating actions.

[0065] The process model of the manual mode program instance is generated by arranging the relevant manual mode operating actions in sequential order and in step S6 each operating action is assigned statistical properties of the value of the relevant process variable and / or the value of the relevant setpoint change.

[0066] Figure 2 a shows as an example a first time series 210 and a second time series 220 of values ​​of process variables A (fill level), B (pressure), respectively, and Figure 2 b shows an example of a relevant log file entry.

[0067] Figure 3a shows two examples of manual mode program instances originating from relevant log files or log file events, and Figure 3bShows the sequence from which manual mode operating actions are extracted.

[0068] To generate a process model for the described method, manual-mode program instance start and end times and associated process variables can be obtained from the operator actions recorded by the historical control system log file. For each corresponding start and end time of the logged action, it can be determined whether the control loop of the plant process, represented by the associated setpoint of the plant process, was in manual or automatic mode at the start of the program instance. This is repeated through all log entries from the start to the end of the program instance.

[0069] a) If the setpoint change is determined to be performed in manual mode of the corresponding control loop, this operation action will be added to the list of operation actions of the program instance, including the actual setpoint value and the new setpoint value respectively.

[0070] b) If a setpoint is changed automatically (e.g., the value of an actuator within a control loop or the target value of a PID controller (Proportional-Integral-Derivative) controlled by a higher-level process control) to manual mode, this action is added to the list of actions for the manual mode program instance, including the corresponding actual change in the setpoint. If the setpoint control is changed back to automatic mode, the corresponding setpoint is removed from the list of actions. This means that any entry made by the operator for the plant, particularly any entry related to the manual mode program instance, is entered into the list.

[0071] c) At the start of an action, all relevant, respectively associated process variable values ​​are assigned to the action list of the program instance for determining the range of possible setpoint settings or another statistical property. Actions can also be added from other lists. Examples of data sources are so-called audit trails, which collect operator actions at a more detailed level, particularly recorded in the user interface of a distributed control system, or action actions can be extracted from setpoint signals and added to the list of the program instance each time the setpoint value changes.

[0072] Actions from different sources are merged according to the order with respect to their corresponding timestamps.

[0073] As a result of this step, several lists of related sequences of operating actions are identified during program execution, together with the values ​​of process variables. Figure 3a Two examples of two such operating action lists with associated process values ​​are shown, which together form two examples of program instances.

[0074] Figure 3bThis paper illustrates how to process multiple program instances to extract a sequential order for determining a partial sequence diagram for each time stamp of an operator action. Next, each operator action within the sequential order is assigned an identifier ID and a sequence number (#) that captures the frequency with which a particular setpoint change with a specified direction of change is performed. From this, the sequential order of the operator actions for the program instance is determined to create a partial sequence diagram. The partial sequence diagram captures the identified sequential order relationships for the human operator actions required by the manual mode program instance. This results in a directed acyclic graph (DAG) of the operator actions.

[0075] An identifier (ID) establishes a unique label for an operational action (e.g., the ID of a valve operated by an operator, a pump started by an operator, etc.), which can correspond to a control loop within multiple program instances, and the identifier can be linked to a number that indicates repeated occurrences of the same operational action within separate program instances.

[0076] To generate a process model using a directed acyclic graph (DAG) of operator actions, for each of a plurality of related sequences of manual mode operation actions, it may be determined:

[0077] - whether a setpoint change is being performed for the first time, to initialize a counter number (e.g. with 1 or 0) and assign this counter number, e.g. as a prefix or suffix, to the operating action, e.g. Figure 3b shown.

[0078] If the corresponding setpoint change has occurred previously, the counter number is incremented and assigned to the corresponding operating action as a prefix or suffix, respectively.

[0079] like Figure 3a In the example shown, there is program instance 2, in which the flow rate is reduced twice. To distinguish between these two operating actions, counter numbers 1 and 2 are assigned, respectively. The direction of the set point change can be recorded and can also be linked to the operating action (e.g., lowering vs. raising the set point, such as a fluid flow rate or temperature change, or starting or stopping a device, such as a motor or pump). Figure 3a In the example shown, the identifier includes the direction of change (close, open, increase (U for up in the example), or decrease (D for down in the example).

[0080] Operational actions that do not appear in all of the multiple recorded program instances, or appear in only a very small number (based on a percentage threshold) of the multiple recorded program instances, are removed from the list of operational actions of the program instances for modeling the process.

[0081] The same is true if the variance of the process value when performing an action is above a threshold, in which case the process value may be omitted for that step.

[0082] In order to generate a directed acyclic graph (DAG) from the sequential list of human operation actions of a program instance, a specific matrix is ​​created, which indicates the order of human operation actions within the sequential list of human operation actions, such as Figure 4a The numbers shown at the grid positions of the matrix reflect the frequency with which the action indicated in the first column, ie, the corresponding label of the matrix, precedes the action indicated in the first row, ie, the label of the matrix.

[0083] Some of the actions may change their position in the sequential order without any technical impact on the manual mode program instance. If the numbers in the grid of the matrix indicate that two actions A and B have a prior relationship with any sequential order (i.e., both the values ​​[A->B]>0 and [B->A]>0 are valid, as is the case with CB1 and CA1 in this example), the grid numbers at the corresponding positions of the matrix can be selected or changed in the following way:

[0084] (a) 0 can be chosen for both indicated sequential directions, or (b) the smaller of the two numbers within the grid, which may be caused by a threshold difference, or just a randomly chosen digit of the two numbers (e.g. in the case of a tie) can be set to 0.

[0085] Using this matrix, generate the DAG and the corresponding process model. Perform the following steps until the matrix (or the list of nodes in the directed graph) is processed:

[0086] 1) Search for columns that do not contain values ​​> 0

[0087] 2) Add the corresponding operation action to the DAG / process model by building an edge between the new operation action and the last action(s) added to the graph (if any).

[0088] 3) Set the numbers in the large rows of the matrix corresponding to the operations added to the DAG / process model to 0

[0089] Figure 4b An example of a constructed DAG is given. The process model includes the assignment of ranges of process values ​​to operational actions, which are sequentially ordered according to the process model. Alternative procedures for creating such a flowchart can be constructed for any other graphical representation besides a matrix, for example using an adjacency list.

[0090] To annotate the process model with process value ranges for each action in a flow chart (DAG), the method extracts the process value ranges (minimum and maximum) when executing the recorded individual action. Alternatively, the program determines a confidence interval for the corresponding value of the process variable (in 90% of cases, the value of the process variable lies between values ​​A and B when the step is executed).

[0091] The distribution range of process values ​​used to recommend a setpoint change or other statistical properties of the process values ​​can be learned using a machine learning model, for example, a regression that estimates the recommended setpoint change depending on the relevant process values ​​or all process values ​​of the plant and / or plant part, respectively. Alternatively, a confidence interval is derived from the recorded setpoint changes (e.g., in 90% of the cases, the operator selected a setpoint value between A and B) or a machine learning model is trained to correlate the setpoint value with a key performance indicator (KPI), such as execution time, quality measurement, throughput, etc.

[0092] The order in which the described procedures generate the DAG can be modified.

[0093] The method for generating a process model may include an operator adding annotations during the execution and recording of manual operator actions, such as by voice recording: "A prerequisite for the next setpoint change to open valve C is that the level value is X." Such annotations and / or prerequisites that may be added to the process model may define required prerequisites in addition to or in lieu of the values ​​of process variables in the process model.

[0094] A method for generating a process model may include a procedure for editing an initial process model, including:

[0095] -Remove operation actions;

[0096] - Adding operational actions recorded by the logging system (e.g. checking external data or interacting with a field operator);

[0097] - define the maximum and minimum values ​​of the set point, or define the formula for calculating the set point value;

[0098] - Select the value of the relevant process variable as the prerequisite for the operation action and specify the upper and lower limits of the process variable value.

[0099] The method for generating a process model may include extracting constraints on process variables and set points from documents by text mining, including operating actions and set point changes and values ​​of process variables.

[0100] Additionally or alternatively, this may be done automatically.

[0101] Procedure execution support systems can record how operators execute procedures.

[0102] The program execution system also guides the operator through the program steps and informs the operator whether the prerequisites for the next manual action are met and which setpoint the operator should change, including the recommended specific value. If the program execution system used as an auxiliary is integrated with the operator interface of the process control system, it can prevent the execution of manual actions that are not included in the process model. ("Are you sure you want to...?")

[0103] This program execution support system can display the manual mode program instance and manual mode operation actions on the screen, for example, Figure 5 Dialog flow shown. After a program instance starts, the program execution system displays all actions without predecessors, including the minimum and maximum process values ​​for their execution. After a manual action is executed, the corresponding action and the corresponding edge of the model are removed from the display, and the next action that can be executed (i.e., without predecessors) is added to the display.

[0104] Management of open manual actions can be implemented in different ways (working on a copy of the process model and deleting rows of executed actions, or calculating until all predecessors are executed, maintaining a list of predecessors and updating it by deleting executed actions, etc.).

[0105] The workflow of the procedure execution support system can be automatically triggered (e.g., when the operator performs the first one or two actions): "Are you currently executing procedure X?"

[0106] The program execution support system can monitor the time elapsed between operating actions and compare it with the recorded historical values ​​in order to remind the operator to perform and anticipate the next operating action.

[0107] The program execution support system can be configured to support operator training, including simulations of plant processes that can simulate the relationship between set point changes and process variables, such as through machine learning. For example, it can simulate and train what happens if a valve is opened at a low tank level.

Claims

1. A method (100) for generating a process model, the process model modeling an artificial mode program instance of a plant process, the program instance comprising a related sequence of operating actions, comprising the following steps: Providing a plurality of log events (S1) of a plurality of operation actions (110) of the factory process; selecting a plurality of related sequences of manual mode operation actions from the plurality of log events (S2); filtering the plurality of related sequences of manual mode operating actions according to individual plant parts (S3); identifying a sequential order from among the filtered plurality of related sequences of the manual mode operating actions (S4); determining, based on the filtered plurality of correlated sequences of manual mode operating actions, statistical properties of the values ​​of the associated process variables and / or statistical properties of the values ​​of the associated setpoint changes for each successively ordered manual mode operating action (S5); The process model (S6) of the manual mode program instance is generated by arranging related manual mode operating actions in the sequential order of each operating action, each operating action being assigned the statistical properties of the value of the related process variable and / or being assigned the statistical properties of the value of the related set point change.

2. The method (100) of claim 1, wherein said filtering of said plurality of related sequences of manual mode operating actions is performed according to individual plant conditions.

3. The method (100) of claim 2, wherein the individual plant condition is determined by at least one threshold value of a relevant process variable at the start of the manual mode program instance.

4. The method (100) of claim 3, wherein the individual plant conditions are determined by threshold values ​​for a subset of the related process variables at the start of the manual mode program instance.

5. Method (100) according to any one of claims 2 to 4, wherein the individual plant conditions are determined by at least one time series of values ​​of relevant process variables of the plant part during a time period before the start of the manual mode program instance.

6. A method (100) according to any one of claims 2 to 4, wherein the individual plant conditions are determined by a time series of a number of clusters of values ​​of process variables of the plant part during a time period before the start of the manual mode program instance, wherein the clusters are constructed by scoring the similarity of the time series of values ​​of related process variables.

7. The method (100) according to any one of claims 2 to 4, wherein comments related to manual mode operating actions are assigned to the respective manual mode operating actions.

8. The method (100) according to any one of claims 2 to 4, wherein manual mode operating actions having assigned statistical properties of the value of the associated process variable above a limit value are omitted from the process model.

9. The method (100) according to any one of claims 2 to 4, wherein specific manual mode operating actions in the multiple related sequences of manual mode operating actions that rarely occur within the multiple related sequences of manual mode operating actions are omitted from the process model.

10. A method for generating a recommended set point for a manual mode operating action for a manual mode program instance in a method according to any one of the preceding claims, wherein the recommended set point is determined using a machine learning model using the multiple related set points related to individual plant conditions to be assigned to the related manual mode operating action with respect to the individual plant conditions.

11. A method for generating a warning signal associated with a plant process, comprising the steps of: providing a continuous log of events of said plant processes; According to any one of claims 3 to 7, individual plant conditions are determined by log events of said plant process; According to the preceding claim, the warning signal is generated based on the individual plant condition.

12. A support system comprising a controller, the support system being configured to support an operator in performing manual mode operating actions according to a process model by displaying relevant process values ​​and / or recommended set point changes to the operator according to a sequence of manual mode operating actions generated by the process model, the process model being generated according to the method of any one of claims 1 to 11.

13. Use of the process model generated according to the method according to any one of claims 1 to 10 to support an operator in executing the manual mode operation actions of a manual mode program instance.

14. A computer program product comprising instructions which, when executed by a computer, cause the computer to perform the method according to any one of claims 1 to 11.

15. A computer-readable storage medium having stored thereon the computer program product according to claim 14.

Images