Generation of a representation of a process network comprising at least two interconnected chemical plants

Abstract

SUMMARY OF THE INVENTION A computer-implemented method is proposed for generating a problem-specific representation of a process network to enable control or monitoring of a process network having at least two interconnected chemical plants, the method comprising the steps of: providing a first digital representation of the process network, the first digital representation including a digital process representation of each plant, connections of each plant to other plants and sensor elements disposed in the process network; generating a graph structure based on the first digital representation, the graph structure including vertices representing unit operations and edges linking the unit operations representing at least physico-chemical quantities, the edges including edge metadata representing at least the physico-chemical quantities and measurable tags; generating a folded graph structure based on the graph structure, the folded graph structure including vertices representing virtual unit operations and edges linking the virtual unit operations representing at least the physico-chemical quantities, the edges including edge metadata representing observable physico-chemical quantities and their relationships to the vertices; deriving a set of equilibrium equations from the folded graph structure; and providing the set of equilibrium equations and the physico-chemical quantities for monitoring and / or controlling the operation of the process network.

Description

Technical Field

[0001] Field The present invention relates to a system and a computer-implemented method for generating problem-specific representations of a process network to enable monitoring and / or control of a process network having at least two plants. The system further relates to use cases of the problem-specific representations.

Background Art

[0002] Background Chemical production is a very complex environment, especially when two or more production plants are involved. A chemical plant typically includes multiple assets for producing chemical products. There are multiple feeds of pure components or mixtures, and at various stages, energy is provided or withdrawn. Multiple sensors are distributed within such a plant for monitoring and control purposes to collect data. Thus, chemical production is a data-heavy environment. However, until now, it has been a challenge to monitor and control interconnected plants. In a process engineering flowsheet, a simulator includes a graph structure that simulates a chemical plant. Models in such a flowsheet simulator are usually constructed to solve a given problem and cannot be easily migrated to other problems. In particular, such simulators are static and cannot be easily adjusted. Furthermore, model design is cumbersome and time-consuming. In particular, each block / node used in a flowsheet model has a large number of unknowns that are mostly fixed but perhaps parameterized, and thus requires a specific number of specifications or additional constraints / equations.

[0003] The use of graph theory to describe a single plant has been proposed by Preisig (Copmuter and Chemical Engineering, 33(2009), 598-604).

[0004] Preisig et al., SIMS 2004, Copenhagen, Denmark, 23 - 24 September 2004, pp 413 - 420 and Computers and Chemical Engineering 33(2009)598 - 604 describe a modeler constructed with respect to implementing a physical view of the world. The modeler constitutes an abstract process representation in the form of a topology with two levels of refinement. First, a physical view of the space occupied by the process and its associated environment, which is the physical topology, is defined. The first refinement is seen as the coloring of the topology by adding the species present in the plant. Finally, the second refinement is to add variables and equations that describe the behavior of the individual components of the topology. The modeler described by Preisig aims to guarantee a structurally solvable simulation problem, i.e., differential - algebraic equations of index 1. The modeler enables the generation of models for problems including dynamic simulation, optimization, and control design. Still, constructing the physical topology of a process is a design process, not an automated process, which requires a deep understanding of the process being modeled.

[0005] McCabe et al. describe the concept of unit operations in chemical engineering (「unit operations of chemical engineering」Mc Graw Hill, Oct.2004,7th edition). Existing modelers are static and generate a set of equations that are adjusted for user - specific problems. These models are not dynamic in the sense that they are not automatically adaptable to different optimization problems, for example. Furthermore, the interdependencies between process components or between multiple plants are difficult to capture and lead to non - robust results. The object of the present invention relates to a system and method for generating a problem - specific representation of two or more interconnected plants to enable the control of a process network having at least two plants. The system further relates to the use cases of the problem - specific representation.

Summary of the Invention

Means for Solving the Problems

[0006] Overview The proposed solution provides a more flexible approach for process simulation. In particular, the process is mapped and prepared such that the model or optimization can be easily customized to the specific needs of the process user. Further, the accuracy is improved by combining a classical equilibrium-based model with a data-driven model, such as a data-driven model that captures environmental effects not represented by a rigorous model based on physical laws. The proposed solution is particularly suitable for monitoring, planning, or controlling processes within a plant network, such as a chemical production park including downstream and upstream plants, an energy generation complex including a refinery, etc. The proposed solution enables a more flexible approach to generate network nodes and adapt the network model according to current conditions. For example, if one plant within the network fails to function, the network model can be correspondingly adapted and still provide accurate predictions for monitoring or further controlling the plant network.

[0007] A computer-implemented method is proposed for generating a model representation of a process network having at least two interconnected chemical plants to enable controlling or monitoring the process network. The method includes: - providing a digital representation of the process network, the digital representation including: 〇 a digital process representation of each plant, 〇 the connection of each plant to other plants (realized by mass or energy flow in the description) and sensor elements arranged within the process network ; - generating a graph structure based on the first digital representation, the graph structure including: Vertices representing unit operations Edges representing physicochemical quantities, where the physicochemical quantities include mass flow, energy flow, and component flow, the edges are linked to vertices, the edges include measurable tags for each of the represented physicochemical quantities, and the measurable tags indicate whether the physicochemical quantity can be measured within the process network or cannot be measured, the edge including the step of generating - The step of classifying physicochemical quantities that can be measured within the process network as observable physicochemical quantities - The step of classifying physicochemical quantities that can be calculated from the equilibrium equations around the vertices as observable physicochemical quantities - The step of generating a folded graph structure based on the graph structure by folding the edges with unobservable physicochemical quantities into folded vertices - The folded graph structure · Folded vertices representing virtual unit operations · Vertices representing unit operations 〇Edges that link to the folded vertices and / or vertices and represent only observable physicochemical quantities including the step of generating - The step of deriving a set of equilibrium equations for each mass flow, energy flow, or component flow around each vertex, where the set of equilibrium equations models the plant network and the set of equilibrium equations can be used for monitoring, control, production planning, and prediction models, the step of deriving - The step of providing the set of equilibrium equations to a control device, a monitoring device, a production planning device, and / or a prediction model generator including

[0008] The process network can be understood as a network of at least two chemical plants. A chemical plant is a plant in which at least one chemical reaction takes place. At least two plants can be interconnected. This interconnection can be realized by energy exchange from one plant to another, by mass transport from one plant to another, and, in rare cases, by exchange of information between plants or their control systems captured by expression. The process network can be located at one site. Alternatively, the process network can be located across at least two sites, each site containing one or more interconnected chemical plants.

[0009] The digital process representation of each plant can be a P&ID (Piping and instrumentation diagram) representation derived from the plant's or piping and instrumentation diagram.

[0010] The digital representation of the process network can further include the types and locations of offline measurements taken from the process, known correlations between online and / or offline data and other variables in the graph - that is, expressions / models.

[0011] The digital representation can also include signals that can be provided in three main forms of realization: · Online signals can be provided by sensors installed directly in the plant. · The digital representation can also include offline signals, such as laboratory data and expert knowledge. Laboratory data can be related, for example, to concentrations determined offline in a laboratory. · The digital representation can also include expressions. An expression is a relationship that links one or more signals to another physical quantity. · Digital representations can also include models. The models can be made available via an API and can link any number of available signals to any number of new signals that can be related to physical quantities.

[0012] More specifically, a first digital representation of a process network can include plant components, component characteristics such as physical dimensions and layout, operating conditions such as operating parameters, total mass flow connections between plant components, sensor components including location and measured quantities, and chemical data related to chemical properties, for example, molecular weight and reactions for each plant, expert knowledge related to thermodynamics, or known separations of components within a liquid-liquid separator such as a reduced thermodynamic relationship. It can include a smart piping and instrumentation diagram for each plant. Unit operations can represent columns, reactors, pumps, heat exchangers, crystallers (cristaller), and other known pieces of equipment that can be installed within a plant. Unit operations can further include transport, where transport defines connections between plants and can comprise pipes, ships, trucks, trains, forklifts, or any means of moving substances between unit operations. Models reflecting the process steps of a particular unit operation can be reflected at the corresponding vertices. These models can be rigorous models, simple functions, or data-driven or hybrid models. These models can require physical quantities linked to each unit operation as input parameters. In particular, chemical reactions can occur in unit operations and can be modeled. In particular, the process network of two connected plants can include at least one unit operation where a reaction occurs, and the corresponding vertex can include a model that converts physicochemical quantities from one connected edge to physicochemical quantities on another connected edge.

[0013] Vertices can further represent vertex metadata, which includes physical quantities linked to each unit operation, such as volume, diameter, or data for internal structures - for example, the number and shape of pipes in a multitube reactor.

[0014] By providing vertices that represent vertex metadata including physical quantities linked to each unit operation, input parameters for models related to the process steps of each unit operation are provided in a very efficient way. Access is accelerated - because the graph structure is self - containing in that it contains its related information. Accelerated access leads to faster execution of method steps, which in turn enables a set of equilibrium equations to be provided more quickly. This leads to a reduction in the time to generate a model of the process network. This also reduces the time when monitoring and / or controlling. A short lag time is extremely important for monitoring and / or controlling a process network having at least two interconnected plants.

[0015] In the graph structure, an edge is to link unit operations. An edge can represent at least physicochemical quantities. Representing at least physicochemical quantities can include the flow of conserved physicochemical quantities input to / from the unit operation. The conserved physicochemical quantities are physicochemical quantities that follow conservation laws. Another term can be conserved quantities. Physicochemical quantities can also include variables and constraints. The conserved quantities can be related to quantities conserved within a self - contained system. These can be conserved physical quantities such as total mass flow, energy flow, and component flow, for example. The conserved quantities can be related to quantities for which conservation can be derived from physical laws. Further, the constitutive equations are related to the conserved quantities and the knowledge that these equations do not change along an edge, for example, the sum of concentrations being equal to 1. Thus, an edge can further include constitutive rules.

[0016] Component flow is related to the mass flow of specific components within a process network. For example, when a reaction involving two components x and y yields a new component z, the total mass flow entering the reactor is mass_total = mass_x + mass_y, while the component flow for x is mass_x.

[0017] The total mass flow can be understood, for example, as the sum of all component flows. Edge metadata can also be included on each edge. Edge metadata can include physicochemical quantities and can be provided as signals. Physicochemical quantities can further include the physicochemical quantities required to fully describe the conserved physicochemical quantities. Physicochemical quantities can include temperature, pressure, weight, mass, energy, concentration, concentration, activity.

[0018] Measurable tags can be provided for each physicochemical quantity on an edge. When the physicochemical quantity can be measured within the process network by a sensor, the tag can be "measured", and when the physicochemical quantity cannot be measured, the tag can either not be associated or the tag "un-known" can be associated. Measured physicochemical quantities can be provided directly from sensor signals or can be derived from sensor signals by expert knowledge or expressions. Physicochemical quantities can be provided offline or online. In other words, being measured means that the physicochemical quantity is available in the process network from in-line measurements as data from sensors, offline data, such as laboratory data, expert knowledge, or expressions.

[0019] Expressions can be simple mathematical relationships between online or offline data and physicochemical quantities, such as unit conversions.

[0020] A physical chemical quantity can be considered observable when the selected physical chemical quantity can be measured in a process network, can be calculated from an equilibrium equation around a vertex, or can be measured and calculated. A physical quantity is considered to be calculated if its value can be derived from other measured physical chemical quantities by physical laws. The equilibrium equations are based on the conservation laws of the respective physical chemical quantities.

[0021] Generating, based on a graph structure, a folded graph structure, vertices representing virtual unit operations, and edges linking at least virtual unit operations representing physical chemical quantities, wherein the edges contain edge metadata representing observable physical chemical quantities and their relationships to the vertices, ensures that all remaining physical chemical quantities within the graph structure are observable. Observability is essential for deriving a set of solvable equilibrium equations. The relationship for a vertex where a physical chemical quantity is an output parameter of one vertex and an input parameter of the next vertex can be understood.

[0022] Generating, based on a graph structure, a folded graph structure having only observable physical chemical quantities on the edges and their relationships to the vertices can include the step of generating new vertices, which are folded vertices representing virtual unit operations, can include several unit operations, and all selected physical quantities connected to those vertices are observable. This new vertex can then be understood as a virtual unit operation. A new model reflecting the process steps of that new virtual unit operation can be generated and reflected at the corresponding vertex. This model can be an exact model, a simple function, or a data-driven model or a hybrid model. These models can require the physical quantities linked to the respective unit operations as input parameters. Figuratively, these folded vertices are generated by combining unit removing edges with unobservable physical chemical quantities and folding the corresponding vertices to form one folded vertex. This can be repeated.

[0023] Here, the set of balance equations derived from the graph structure can be solved, and thus the graph structure enables the extraction of balance equations for any particular problem. This significantly reduces the complexity of the digital representation. This reduced representation is problem-specific - the reason being that the folded graph only contains problem-specific physicochemical quantities. This enables the control or monitoring of a process network having at least two interconnected chemical plants. Without resorting to the set of balance equations that fully describe the plant network and providing these equations and physicochemical quantities, it would not be possible to control and monitor. The computation time is significantly reduced - the reason being that only the balance equations need to be solved. This further enables the description of complex plant networks in an easily digestible way such that prediction and production planning can be used.

[0024] Physical quantities that can be measured and determined can be labeled as redundant. It can be considered beneficial to define expressions at the first level of digital representation. This increases the amount of physicochemical quantities that can be measured and labeled. This increases the speed for generating the folded graph. This then enables the set of balance equations and physicochemical quantities to be provided quickly. This leads to a reduction in the lag time when monitoring and controlling. A short lag time is extremely important for monitoring and / or controlling a process network having at least two interconnected plants.

[0025] The step of generating a graph structure based on the first digital representation can further include generating a converged graph structure by assigning labels to all selected physicochemical quantities depending on whether all selected physicochemical quantities are measured physicochemical quantities, determined physicochemical quantities, measured and determined physicochemical quantities, or physicochemical quantities that are neither measured nor determined physicochemical quantities. The latter can be labeled as unobservable.

[0026] The determined and measurable physical quantities can be labeled as redundant. Providing such labels enables consistency checks.

[0027] When physicochemical quantities are redundant, the measured physicochemical quantities can be compared with the determined ones. Comparing the measured physicochemical quantities with each of the determined ones can be described as a consistency check.

[0028] Consistency is only confirmed when both physicochemical quantities are the same. By the same, it means that the physicochemical quantities are the same within the error margin, or the residual between each selected physicochemical quantity is less than the threshold, or the residual between each selected quantity only shows random noise.

[0029] Labels can be provided as attributes for a folded graph structure. Providing these labels at the level of the folded graph is a very efficient way to provide information. This accelerates the accessibility of the labeled information. The accelerated access leads to faster execution of the method steps, which in turn enables the set of equilibrium equations and the physicochemical quantities to be provided more quickly. This leads to a reduction in the lag time when monitoring and / or controlling. A short lag time is extremely important for monitoring and / or controlling a process network having at least two interconnected plants.

[0030] Labels can be provided in a balanced set. Providing these labels in a balanced set further accelerates the accessibility of information. The accelerated access leads to a faster execution of the method steps, which in turn enables the balanced set to be provided faster. This leads to a reduction in the lag time when monitoring and / or controlling. A short lag time is extremely important for monitoring and / or controlling a process network having at least two interconnected plants. Essentially, the information does not need to be looked up in a separate database. Today's databases are often in a cloud environment, which results in slower access compared to on-site retrieval of data.

[0031] For labeling, observability and redundancy analysis can be performed. The redundancy and observability analysis can be performed by applying the algorithms disclosed in Kretsovalis et al. (Comput. Chem. Engng, Vol12, No7, pp 671-687, and 689-703, 1988). The step of generating a converged graph can include defining an expression between physicochemical quantities that can be stored in the graph. Storing the expression in the graph reduces the time for retrieving information from the converged graph.

[0032] The step of generating a folded graph structure based on the graph structure can be preceded by folding edges containing unobservable physicochemical quantities into vertices. By folding the graph structure, new vertices can be generated. These new vertices are related to unit operations that do not need to reflect the unit operations derived from the first digital representation. The virtual unit operation can be a unit operation that does not need to reflect the unit operations derived from the first digital representation or a unit operation that reflects the unit operations derived from the first digital representation.

[0033] This leads to a graph where all stored quantities can be derived based on the measured and selected physical properties. The folded graph represents the maximum information available from all known data. The graph can be further folded to reduce the complexity of any set of derived equilibrium equations while still ensuring the structural resolvability of the derived system. This significantly reduces the complexity of the digital representation. This reduced representation is problem-specific - because the folded graph only contains problem-specific physico-chemical quantities. This enables the control or monitoring of a process network having at least two interconnected chemical plants. Without resorting to the set of equilibrium equations that completely describe the plant network and without providing these equations, it would not be possible to control and monitor. The computation time is significantly reduced - because only the equilibrium equations need to be solved. This further enables the description of complex plant networks in an easy-to-digest way such that prediction and production planning are possible.

[0034] Generating a folded graph structure can include generating a folded graph structure for each stored quantity, e.g., a folded graph structure for total mass flow and a separate folded graph structure for energy flow. This may be required when the observables for each stored quantity are not on the same edge. Generating a folded graph structure for each stored quantity preserves the maximum possible information within the graph. It is beneficial to maintain the maximum possible information for control and / or monitoring.

[0035] The system of equilibrium equations can be stored in the graph or in a separate database.

[0036] This has the advantage that a new system of equilibrium equations does not need to be generated every time a requirement for monitoring and / or controlling the plant network is provided.

[0037] The step of generating a folded graph structure based on a graph structure may be preceded by folding edges including unobservable physicochemical quantities so as to become vertices, and may be followed by further folding edges including observable physicochemical quantities so as to become vertices based on an objective.

[0038] For a certain specific problem, a reduced set of information is sufficient. These specific problems may be defined as objectives based on the objective that the graph can be further folded. Folding can be implemented by aggregation of vertices and edges to become new folded vertices. This new folded vertex can then be understood as a new unit operation of the folded form. A new folded model reflecting the process steps of the new specific unit operation can be generated and reflected at the corresponding folded vertex. This model can be an exact model, a simple function, or a data-driven model or a hybrid model. These models can require physical quantities linked to their respective unit operations as input parameters.

[0039] By further folding, unnecessary information is removed. This results in an uncomplicated graph. As a result, the set of equilibrium equations is likewise uncomplicated but sufficient for its specific problem.

[0040] Therefore, the set of equilibrium equations provided for monitoring and / or control is fast to solve. This leads to a reduction in the lag time when monitoring and / or controlling. A short lag time is extremely important for monitoring and / or controlling a process network having at least two interconnected plants. Such specific problems can occur when the edge between two vertices is not important for control and / or monitoring and only the equilibrium equations of the inflow to the first vertex and the outflow from the second vertex are important.

[0041] The method steps for generating a converged graph structure can further include receiving a trigger signal, and the method steps for generating a converged graph structure are initiated by the evaluation of the trigger signal.

[0042] The trigger signal can be provided by a watchdog device. The watchdog device can monitor the process network to detect changes within the process network. Detecting a change within the process network can initiate generating a converged graph structure.

[0043] Changes within the process network can lead to changes in the observability of physicochemical quantities, and some observed quantities may become unobservable, which in turn will lead to an unsolvable system of equations. Therefore, monitoring and / or control will be applicable.

[0044] By the method steps for generating a converged graph structure receiving a trigger signal initiated by the evaluation of the trigger signal, all physicochemical quantities are labeled to all selected physicochemical quantities according to whether they are measured physicochemical quantities, determined physicochemical quantities, measured and determined physicochemical quantities, or neither measured nor determined physicochemical quantities, and these changes will be reflected in the step of generating a converged graph structure.

[0045] Changes in the observability of physicochemical quantities will then be reflected in the generated folded graph structure. The step of deriving a set of equilibrium equations from the folded graph structure will also reflect the changes in the process network. Therefore, the changes in the process network will also be reflected in the provided set of equilibrium equations. This dynamic approach leads to a more robust way of generating problem-specific representations to enable controlling or monitoring a process network having at least two interconnected chemical plants.

[0046] These changes in the process network can be sensor failures or other errors.

[0047] The first aspect is reflected in the following items: 1. A computer-implemented method of generating a model representation of a process network having at least two interconnected chemical plants to enable controlling or monitoring the process network, - providing a digital representation of the process network, the digital representation comprising 〇 a digital process representation of each plant, 〇 the connection of each plant to other plants (realized by mass or energy flow in the description) and sensor elements arranged within the process network including; - generating a graph structure based on the first digital representation, the graph structure comprising 〇 vertices representing unit operations, 〇 edges representing physicochemical quantities, the physicochemical quantities including mass flow, energy flow, and component flow, the edges linking to vertices, the edges including measurable tags for each of the represented physicochemical quantities, the measurable tags indicating whether the physicochemical quantity can be measured within the process network or cannot be measured including; - classifying physicochemical quantities measurable within the process network as observable physicochemical quantities; - classifying physicochemical quantities calculable from the equilibrium equations around the vertices as observable physicochemical quantities; - generating a folded graph structure based on the graph structure by folding edges having physicochemical quantities not classified as observable into folded vertices, - the folded graph structure comprising · A collapsible vertex representing a virtual unit operation, · A vertex representing a unit operation 〇 A collapsible vertex and / or an edge that links to the vertex and represents only observable physicochemical quantities including a generating step, - A step of deriving a set of equilibrium equations for each mass flow, energy flow, or component flow around each vertex, where the set of equilibrium equations models the plant network and the set of equilibrium equations is usable for monitoring, control, production planning, and prediction models; a deriving step, - A step of providing the set of equilibrium equations to a control device, a monitoring device, a production planner device, and / or a prediction model generator A computer-implemented method including the above steps. 2. The method according to item 1, wherein the vertex further represents vertex metadata including physical quantities linked to respective unit operations. 3. The method according to any one of the preceding items, wherein generating a collapsible graph structure includes generating a collapsible graph for each physicochemical quantity. 4. The method according to item 3, wherein providing a set of equilibrium equations from a collapsible graph structure includes providing a set of equilibrium equations for each conserved quantity. 5. The step of folding an edge is - Selecting at least two vertices connected via the edge, - Folding the edge between at least two vertices, thereby creating a virtual vertex followed by the method according to any one of the preceding items. 6. The step of generating a graph structure based on the first digital representation further includes generating a converged graph structure by assigning labels to all physicochemical quantities according to whether all physicochemical quantities are measured physicochemical quantities, determined physicochemical quantities, measured and determined physicochemical quantities, or physicochemical quantities that are neither measured nor determined physicochemical quantities, according to any one of the preceding items. 7. The method further includes receiving a trigger signal, and the method step of generating a converged graph structure is initiated by evaluating the trigger signal indicating that, according to item 7. 8. A system for generating a problem-specific representation of a process network to enable controlling or monitoring a process network having at least two interconnected chemical plants, - A processor configured to perform the method steps according to any one of items 1 to 7, - A set of equilibrium equations for monitoring and / or controlling the operation of the process network Comprising a system. 9. A computer program product that, when executed on a computer, performs the method steps according to any one of method items 1 to 7. 10. A modeling system for generating a model representation of a process network having at least two interconnected chemical plants to enable controlling or monitoring the process network, comprising a processor and a communication interface, the processor - Providing to the processor via the communication interface a digital representation of the process network, 〇 The digital process representation of each plant, 〇 The connection of each plant to other plants (realized by mass or energy flow in the description) and sensor elements arranged within the process network Including the step of providing a digital representation, - A step of generating a graph structure in a processor based on a first digital representation, the graph structure comprising: 〇Vertices representing unit operations, 〇Edges representing physicochemical quantities, the physicochemical quantities including mass flow, energy flow, and component flow, the edges linking to vertices, the edges including measurable tags for each of the represented physicochemical quantities, the measurable tags indicating whether the physicochemical quantity can be measured within the process network or cannot be measured, the edges including; the step of generating; - A step of classifying, by the processor, physicochemical quantities that can be measured within the process network as observable physicochemical quantities; - A step of classifying, by the processor, physicochemical quantities that can be calculated from the equilibrium equations around the vertices as observable physicochemical quantities; - A step of generating, by the processor, a folded graph structure based on the graph structure by folding edges having physicochemical quantities not classified as observable into folded vertices; - wherein the folded graph structure · Folded vertices representing virtual unit operations, · Vertices representing unit operations 〇Edges linking to the folded vertices and / or vertices and representing only observable physicochemical quantities including; - A step of deriving, by the processor, a set of equilibrium equations for each mass flow, energy flow, or component flow around each vertex, the set of equilibrium equations modeling the plant network, the set of equilibrium equations being usable for monitoring, control, production planning, prediction models; the step of deriving; - A step of providing the set of equilibrium equations to a control device, a monitoring device, a production planner device, and / or a prediction model generator via a communication interface A modeling system configured to perform.

[0048] In a second aspect, a method for monitoring a process network having at least two plants, comprising: - receiving, via an input interface, a request for at least one process network operating parameter; - extracting, via the input interface, a system of equilibrium equations from a folded graph structure of the process network; - retrieving, from a database, historical data related to measured physicochemical quantities and metadata related to at least one process network operating parameter; - receiving current data related to observable physicochemical quantities and metadata about the observable physicochemical quantities; - determining values for at least one process network operating parameter by solving the system of equilibrium equations based on the historical data and the current data; - providing, via an output interface, the values of the at least one process network operating parameter A system of equilibrium equations from a folded graph structure of a process network is derived by the method according to any one of items 1 to 7 of the first aspect.

[0049] The input interface can be a physical interface (e.g., keyboard, mouse, touch screen, touch pad) or a non-physical interface (e.g., function call, API), and the input interface can also be a combination of a physical interface and a non-physical interface.

[0050]

[0051] ​The output interface can be a physical interface (e.g., a screen, a monitor) or a non - physical interface (e.g., a function call, an API), and the output interface can also be a combination of a physical interface and a non - physical interface.

[0052] Metadata about an observable physicochemical quantity can refer to additional information related to the observable physicochemical quantity (e.g., the location of the sensor, the timestamp).

[0053] At least one process network operation parameter refers to an operation parameter intended to be monitored. This process network operation parameter can be any selected observable physical quantity in a folded - graph structure or any performance metric derivable from these selected observable physicochemical quantities. At least one process network operation parameter can reflect the operation parameter at a specific point in time, which can be the current operation parameter and, in the process network at a specific point in time, for example, a unique unit operation. At least one process network operation parameter can be temperature, concentration, total mass flow, or any performance metric derivable therefrom.

[0054] The proposed method provides a fast and reliable way to monitor a process network having at least two plants that would otherwise not be possible. The provided set of balance equations greatly reduces the complexity when generating an appropriate model. By specifying the amount of the top concern, the model can be reduced to a practical model with less complexity for observing these amounts. Therefore, the calculation time is greatly reduced. The reason is that only the balance equations need to be solved. This leads to a reduction in the lag time when monitoring. A short lag time is extremely important for monitoring and / or controlling a process network having at least two interconnected plants. Historical data can refer to data from the recent past to the history and means sufficient data to conduct a substantial steady-state test.

[0055] The step of providing a system of balance equations can also include providing a folded graph structure. Thereby, the system of balance equations can be generated from the current folded graph structure. This can be beneficial in an environment where the folded graph structure changes.

[0056] Providing a system of balance equations from a folded graph can only be performed when the value of at least one process network operation parameter cannot be directly retrieved from the database as an observed quantity. This increases the speed of determining the value of at least one process network operation parameter. The reason is that the value can be direct. The observed quantities and metadata related to at least one process network operation parameter are the observed quantities and metadata required to determine the value of at least one process network operation parameter.

[0057] The step of retrieving historical data related to observable physicochemical quantities and metadata related to at least one process network operation parameter from the database can include retrieving time-series data.

[0058] Determining values for at least one process network operating parameter by solving a balanced system can be carried out by solving a set of balance equations as an optimization problem. Solving a set of balance equations as an optimization problem includes minimizing an error that reflects the deviation from zero in the respective balance equations. A set of balance equations can be provided for each conserved quantity. Thereafter, the error can be defined for each set of balance equations for each conserved quantity. The optimization problem is then to minimize all errors from all sets of balance equations.

[0059] Solving a set of balance equations as an optimization problem has the advantage that uncertainties in observable physicochemical quantities are taken into account. These uncertainties can result, for example, from sensor noise, sensor failures, incorrect metadata. As a result, the method is more robust.

[0060] The method can further include performing a stationary test on the selected observable physicochemical quantity and the metadata related to at least one process network operating parameter from the database.

[0061] An algorithm for performing a stationary test on time series data is described by Levente et al. (AIChE Journal, 2018, Vol. 00, No. 00, p 1-12).

[0062] The concept of describing a process network in a balanced form is only applicable when the process network is in a steady state. Applying a steady-state test has the advantage that only the steady state is considered. Applying a steady-state test further has the advantage of ensuring that the system being monitored is currently in a steady state. If the steady-state test reveals that the current state of the process network is not static, a signal can be generated. This signal can be an alarm signal and can be provided to the plant network control center. The alarm signal can trigger the shutdown of one plant or the shutdown of the process network. The steady-state test includes time-series analysis based on volatility (a typical model from economics) or activity (a custom metric derived from a nearly zero-dispersion test scaled by a hyper log-log algorithm). Further tests are performed to detect outliers and anomalies based on models derived from historical data sets.

[0063] The step of extracting observables and metadata related to at least one process network operating parameter can further include steps of data reconciliation and / or gross error detection. Methods for data reconciliation and / or gross error detection are described, for example, by Yuan Yuan et al. (AICHE Journal, Vol. 61, No. 10, p. 3232 - 3248).

[0064] Data reconciliation addresses random noise on observable physicochemical quantities that can result from variations or noise on sensor signals used to determine or measure each observable physicochemical quantity.

[0065] The use of data reconciliation has the advantage that accurate and reliable information about the state of the process network is extracted and a single consistent data set representing the most likely state of the process network is extracted.

[0066] The use of gross error detection also has the advantage that accurate and reliable information regarding the state of the process network is extracted and a single consistent dataset representing the most likely state of the process network is extracted.

[0067] Gross error detection has the further advantage that gross errors can become apparent and can therefore be detected. The detection of gross errors can generate a gross error signal. The gross error signal can be provided to the control center of the process network. The gross error signal can further trigger the generation of a converged graph structure. The detection of gross errors can be used as a direct trigger signal.

[0068] Alternatively, the gross error signal can be provided to a watchdog device, which then generates a trigger signal.

[0069] The step of performing the consistency check enables the extracted historical data related to observable physicochemical quantities and the metadata related to at least one process network operating parameter to evaluate whether they are consistent.

[0070] Consistency is only confirmed when both physicochemical quantities are identical. Identical means that the physicochemical quantities are identical within the error margin, or when the residual between each selected physicochemical quantity is less than the threshold value or the residual between each selected quantity only exhibits random noise. A consistency signal can be generated according to the result of the consistency check.

[0071] The consistency check can be performed by a watchdog device, and the consistency signal can be used as a trigger signal, which can then be used to trigger the generation of a converged graph structure.

[0072] Alternatively, the consistency check signal can be provided to a watchdog device, which then generates a trigger signal.

[0073] The trigger signal can be provided by a watchdog device. The watchdog device can monitor the process network to detect changes in the process network. Detecting a change in the process network can initiate generating a converged graph structure.

[0074] The second aspect is also reflected in the following items: 1. A method of monitoring a process network having at least two plants, - receiving a request for at least one process network operating parameter via an input interface; - retrieving a system of equilibrium equations from a folded graph structure of a process network having observable physicochemical quantities on edges and its relationships to vertices via an input interface; - retrieving historical data related to observable physicochemical quantities and metadata related to at least one process network operating parameter from a database; - determining values for at least one process network operating parameter by solving the system of equilibrium equations based on the historical data and current data; - providing the values of at least one process network operating parameter via an output interface and including the method. 2. The method according to item 1, including the step of performing a steady-state test. 3. The method according to any one of the preceding items, including the steps of data reconciliation and / or gross error detection. 4. The method according to any one of the preceding items, including the step of consistency checking. 5. The method according to any one of the preceding items, wherein at least one process network operation parameter includes at least two or more process network operation parameters. 6. The method according to item 5, including the step of providing values of at least two or more process network operation parameters via an output interface. 7. The method according to any one of the preceding items, further comprising the step of receiving current data related to observable physicochemical quantities and metadata related to at least one process network operation parameter. 8. A system for monitoring a process network having at least two plants, - An input interface, 〇Receiving a request for at least one process network operation parameter, 〇Extracting a system of equilibrium equations from a folded graph structure of a process network having observable physicochemical quantities on edges and their relationships to vertices, 〇Retrieving historical data related to observable physicochemical quantities and metadata related to at least one process network operation parameter from a database for the input interface, - An output interface, 〇Providing values of at least one process network operation parameter for the output interface, - A processor that performs the method steps according to any one of items 1 to 7 A system comprising. 9. A computer program product that, when executed on a computer, performs the method steps according to any one of method items 1 to 7.

[0075] In a third aspect, a method for controlling a process network having at least two plants, - Receiving requirements for at least one optimization objective via an input interface by specifying at least one process parameter to be optimized; - Retrieving a system of equilibrium equations from a folded graph structure of a process network having observable physicochemical quantities on edges and their relationships to vertices via an input interface; - Retrieving historical data related to observable physicochemical quantities and metadata related to at least one process network operation parameter to be optimized from a database; - Determining values for at least one process network operation parameter to be optimized by solving the system of equilibrium equations; - Providing values of at least one process network operation parameter via an output interface A method comprising:

[0076] The system of equilibrium equations from the folded graph structure of the process network is derived by the method according to any one of items 1 to 7 in the first aspect.

[0077] The step of providing the system of equilibrium equations may also include providing a folded graph structure. Thereby, the system of equilibrium equations can be generated from the current folded graph structure. This can be beneficial in an environment where the folded graph structure changes frequently.

[0078] The step of determining values for at least one process network operation parameter to be optimized by solving the system of equilibrium equations may be preceded by defining an optimization objective function. The optimization objective function can be the value for at least one process network operation parameter to be optimized or the deviation from the target value of the value for at least one process network operation parameter to be optimized.

[0079] By solving a system of equilibrium equations, the step of determining values for at least one optimized process network operation parameter can then be reduced to solving a system of equilibrium equations from the folded graph structure of the process network having observable physicochemical quantities on the edges and its relationships to the vertices, using as a constraint minimizing the objective function and best achieving the equilibrium equations. Solutions to the optimization problem are described in a book (e.g., https: / / www.springer.com / de / book / 9780387303031).

[0080] The step of receiving current data related to observable physicochemical quantities and metadata related to at least one optimized process network operation parameter enables process control of the plant network.

[0081] The observable quantities and metadata related to at least one optimized process network operation parameter are the observable quantities and metadata required to determine values for at least one optimized process network operation parameter by solving a system of equilibrium equations.

[0082] The step of retrieving historical data related to observable physicochemical quantities and metadata related to at least one process network operation parameter from a database can include retrieving time series data.

[0083] Determining values for at least one process network operating parameter that is optimized by solving a set of balance equations can be carried out by solving the set of balance equations as an optimization problem. Solving the set of balance equations as an optimization problem includes minimizing an error that reflects a deviation from zero in the respective balance equations. The set of balance equations can be provided for each conserved quantity. Thereafter, the error can be defined for each set of balance equations for each conserved quantity. The optimization problem is then to minimize all errors from all sets of balance equations, i.e., constraint violations.

[0084] Solving the set of balance equations as an optimization problem has the advantage that uncertainties in observable physicochemical quantities can be taken into account. These uncertainties can result, for example, from sensor noise, sensor failure, incorrect metadata - for example, incorrect units of measurement. As a result, the method is more robust.

[0085] The method can further include performing a steady-state test on the selected observable physicochemical quantities and metadata related to at least one process network operating parameter to be optimized from a database.

[0086] An algorithm for performing a steady-state test on time-series data is described by Levente et al. (AIChE Journal, 2018, Vol. 00, No. 00, p 1-12).

[0087] The concept of describing a process network in a balanced form is only applicable when the process network is in a steady state. Applying a steady-state test has the advantage that only the steady state is considered. Applying a steady-state test further has the advantage of ensuring that the system being monitored is currently in a steady state. If the steady-state test reveals that the current state of the process network is not steady, a signal can be generated. This signal can be an alarm signal and can be provided to the plant network control center. The alarm signal can trigger the shutdown of one plant or the shutdown of the process network. The steady-state test includes time-series analysis based on volatility (a typical model from economics) or activity, and detects outliers and anomalies based on the dynamics of time constants derived from a historical dataset.

[0088] The step of extracting metadata related to the measured quantity and at least one process network operation parameter to be optimized can further include the step of data reconciliation and / or gross error detection.

[0089] Methods for data reconciliation and / or gross error detection are described, for example, by Yuan Yuan et al. (AICHE Journal, Vol. 61, No. 10, p. 3232 - 3248).

[0090] Data reconciliation addresses random noise on observable physicochemical quantities, which can result from variations or noise on the sensor signals used to determine or measure each observable physicochemical quantity.

[0091] The use of data reconciliation has the advantage that accurate and reliable information about the state of the process network is extracted, and a single consistent dataset representing the most likely state of the process network is extracted.

[0092] The use of gross error detection also has the advantage that accurate and reliable information about the state of the process network is extracted and a single consistent dataset representing the most likely state of the process network is extracted.

[0093] Gross error detection has the further advantage that gross errors can be made apparent and thus detected. Detection of gross errors can generate a gross error signal.

[0094] The gross error signal can be provided to the control center of the process network. The gross error signal can further trigger the generation of a converged graph structure.

[0095] The gross error signal can be used as a direct trigger signal. Alternatively, the gross error signal can be provided to a watchdog device, which then generates a trigger signal.

[0096] The step of performing the consistency check enables the extracted historical data related to observable physicochemical quantities and the metadata related to at least one process network operation parameter to evaluate whether they are consistent. Consistency is only confirmed when the measured and determined physicochemical quantities are the same. Same means that the physicochemical quantities are the same within the error margin, or the residual between each selected physicochemical quantity is less than the threshold, or the residual between each selected quantity only shows random noise. A consistency signal can be generated according to the result of the consistency check.

[0097] The consistency check can be performed by a watchdog device, and the consistency signal can be used as a trigger signal, which can then be used to trigger the generation of a converged graph structure.

[0098] Alternatively, the consistency check signal can be provided to a watchdog device, which then generates a trigger signal.

[0099] The trigger signal can be provided by a watchdog device. The watchdog device can monitor the process network to detect changes in the process network. When a change in the process network is detected, generating a converged graph structure can be initiated. The process parameters to be optimized can include plant output, energy consumption, and CO2 emissions.

[0100] The step of specifying further constraint objectives for a balanced system has the advantage that further available information can be used in the optimization problem. Adding further information will increase the reliability of control. When the further constraint objectives for the balanced system include at least one plant of the plant network, this has the further advantage that determining the optimized operating parameters for the optimization objective from the extracted observables by solving the balanced system under the evaluation of the constraints can further include further constraints provided by a data-driven model or a hybrid model. The data-driven model can be generated according to the method of the fourth aspect. The process network can be restricted by physical limitations. Adding further constraint objectives for the balanced system is related to the physical limitations of the process, which can be reflected in the optimization step. The physical limitations of the process network can include feed capacity, storage capacity, cooling capacity, and safety constraints.

[0101] At least one optimization objective by specifying at least one process parameter to be optimized can include at least two or more process parameters to be optimized.

[0102] The third aspect is also reflected in the following items: 1. A method for controlling a process network having at least two plants, comprising: - receiving, via an input interface, a request for at least one optimization objective by specifying at least one process parameter to be optimized; - extracting, via the input interface, a system of equilibrium equations from a folded graph structure of the process network having observable physicochemical quantities on the edges and their relationships to the vertices; - retrieving from a database historical data related to the observable physicochemical quantities and metadata related to at least one process network operating parameter to be optimized; - determining values for at least one process network operating parameter to be optimized by solving the system of equilibrium equations; - providing, via an output interface, the values of at least one process network operating parameter to be optimized. A method as described above. 2. The method according to item 1, further comprising receiving current data related to the observable physicochemical quantities and metadata related to at least one process network operating parameter to be optimized. 3. The method according to any one of the preceding items, wherein retrieving the observable quantities and metadata is preceded by performing a steady-state test. 4. The method according to any one of the preceding items, wherein retrieving the observable quantities and metadata includes steps of data reconciliation and / or gross error detection. 5. The method according to any one of the preceding items, further comprising specifying additional constraint objectives for the system of equilibrium equations. 6. The method according to any one of the preceding items, wherein the additional constraint objectives for the system of equilibrium equations include a model of at least one plant of the plant network. 7. The model of at least one plant of the plant network includes a data-driven model, a data-driven model based on a rigorous model, and a hybrid model, and is the method according to item 6. 8. The model is the method according to item 7, which associates the input of the plant or process network with the output of the plant or process network. 9. The further restrictive purpose for the balanced system is the method according to item 5, which is related to the physical limitations of the process. 10. At least one optimization objective by specifying at least one process parameter to be optimized includes at least two or more process parameters to be optimized, and is the method according to any one of the preceding items. 11. A system for controlling a process network having at least two plants, - An input interface, 〇Receiving a request for at least one optimization objective by specifying at least one process parameter to be optimized, 〇Extracting a system of equilibrium equations from the folded graph structure of the process network having observable physicochemical quantities on the edges and their relationships to the vertices, An input interface for this purpose, - An output interface, 〇Providing values for at least one process network operation parameter to be optimized, An output interface for this purpose, - A processor configured to perform the method steps according to any one of items 1 to 10 A system comprising. 12. A computer program product that, when executed on a computer, performs the method steps according to any one of items 1 to 10 of the method.

[0103] A method for generating a hybrid model for monitoring and / or controlling a process network having at least two plants connected to each other, comprising: - providing a system of equilibrium equations from a folded graph structure of a process network having observable physicochemical quantities on the edges and their relationships to the vertices; - receiving, via an input interface, at least one objective specifying at least one process parameter dependency to be trained; - retrieving, via an input interface, historical data of a process network having at least two plants connected to each other; - training a hybrid model including the system of equilibrium equations and a data-driven model based on the historical data and at least one objective specifying at least one process parameter dependency to be trained; - providing the trained hybrid model via an output interface and including the method.

[0104] The system of equilibrium equations from the folded graph structure of the process network is derived by the method according to any one of items 1 to 7 of the first aspect.

[0105] Retrieving the historical data of the process network having at least two plants connected to each other can include retrieving only the historical data related to at least one objective specifying at least one process parameter dependency to be trained. This leads to a reduction of the data set compared to the full set of historical data. This reduces the time to retrieve the data because only a subset of the available data is retrieved. This then leads to a faster training process.

[0106] In a production process environment, it is difficult to provide a large dataset that enables reliable training of data-driven models. The size of the required dataset increases with the number of dependencies being learned. The step of receiving, via an input interface, at least one objective that specifies at least one process parameter dependency to be trained addresses that problem. By receiving at least one objective that specifies at least one process parameter dependency to be trained, the scope of the training process is restricted and thus it is possible to work with a reduced training dataset and thus train a hybrid model more quickly.

[0107] One way to extract or determine relevant data would be to start with a folded graph structure that fully describes the system using all observables. In the next step, the measured signals can be removed. In the next step, all selected physical quantities can be labeled. This can be repeated as long as all physical chemical quantities remain observable. This results in the minimum dataset being required to train a hybrid model according to at least one objective that specifies at least one process parameter dependency to be trained.

[0108] In the case of machine learning, historical data needs to be provided. This data can be retrieved from a database. The data for each plant can be stored separately in a corresponding database for each plant. Alternatively, the data for all plants can be stored in an enterprise database that can be provided as a cloud service.

[0109] Historical data can include a time series of measured values and / or observables. In general, historical data includes the various states of a production plant or a network of production plants. These states can include, among other things, steady state, startup state, shutdown state, and error state.

[0110] The method can further include performing a steady-state test on the observable physicochemical quantity and the metadata related to at least one process network operation parameter from the database.

[0111] The concept of describing a process network in the form of an equilibrium equation is only applicable when the process network is in a steady state. Applying a steady-state test to historical data has the advantage that only the steady state is considered.

[0112] An algorithm for performing a steady-state test on time series data is described by Levente et al. (AIChE Journal, 2018, Vol. 00, No. 00, p 1-12), which discloses a statistical framework for systematically determining mean stationary in the context of a continuous manufacturing process.

[0113] For monitoring and / or controlling a plant or a plant network, the steady state is a major concern. As a result, it is necessary to determine the time series data related to the steady state of the production process.

[0114] This can be done by steady-state or event analysis. Steady-state data is related to the steady state of the production process. This enables inspection of time-series data and classification of the data into stationary and non-stationary, where non-stationary data is related to a ramp-up state, a ramp-down state, an on / off state, or an error state and can be correspondingly labeled. In particular, some segments of the process are analyzed independently and the labels are tied to the plant level. Some parts of the process can actually be made stationary, while others are not, so that constraints on stationarity can be relaxed for specific parts of the process depending on the targeted at least one process network operating parameter. The time-series data will then be separated according to their labels. The dataset for each label can then be further separated into respective training and test datasets. Since the balance equation requires a stationary system, only stationary data points are useful. Here, all data clusters recorded under stationary operating conditions are filtered. Any abnormal shutdown periods or other non-stationary segments within the dataset are removed. Stationarity tests can detect outliers and anomalies based on the dynamics of time constants derived from the historical dataset, including time-series analysis based on volatility (a typical model from economics) or activity. Such stationarity analysis makes it possible to reduce the historical dataset to data representing normal operating conditions in the process network. This restricts the historical data to stationary and / or cyclic stationary operating conditions.

[0115] Following the stationarity test, the historical dataset can be further improved by data validation and data reconciliation. This further strengthens the historical dataset. Finally, only the strengthened historical dataset can be used to train the hybrid model.

[0116] Preparing the historical data in this way provides a clean dataset that reflects the true process conditions in the process network. Any effects of unsteady operating conditions, gross errors, or random errors are reduced. As a result, any hybrid model trained based on such clean data will not be affected by such effects.

[0117] Providing a trained hybrid model can include providing a rigorous model based on physicochemical laws and providing a data-driven model based on historical data.

[0118] The main advantages of combining a flexible model from a graph-structured database with a data-driven model into a hybrid model: - The flexible graph structure allows for the adaptation of the rigorous model to the current state of the process network, including taking into account errors while maintaining the trained model. As a result, retraining the model remains largely unnecessary even if there are errors in the process network.

[0119] - Since the rigorous model cannot account for environmental effects such as seasonal ones, the data-driven part adds accuracy to the rigorous model and thus corrects any deficiencies.

[0120] - Depending on the input / output structure of the hybrid model, the graph database and the extracted model can be used to calculate missing data points on the input part of the hybrid model.

[0121] - The modeling approach is further scalable and can be extended from multiple plants through a full fairbunt (Verbund) or plant complex to the value chain.

[0122] - When a hybrid model is generated, the hybrid model enables tight control of the entire process network, for example, on a daily basis. Dependencies between components of the process network can be easily captured by a base graph structure that enables more accurate prediction.

[0123] The use of the hybrid model is to provide recommendations on how to operate the process network. For example, the hybrid model can provide recommendations regarding plant flow for a particular purpose. From such a flow, specific operating parameters at the plant level can be determined. Further, the hybrid model can be used to detect anomalies based on real-time sensor data. In such cases, the output of the hybrid model can be compared with the real-time sensor data. If there is a significant difference, a notification or alarm can be triggered within the affected plant or used for root cause analysis.

[0124] Conversely, any differences detected over time between the hybrid model output and the real-time sensor data can provide an indicator regarding model drift. If such drift is detected, the model can be retrained based on more recent historical data, or the elements of the base graph structure can be updated to build a new rigorous model that takes such changes into account.

[0125] The following description relates to the systems, methods, computer programs, and computer-readable storage media outlined above. In particular, the systems, computer programs, and computer-readable storage media are configured to perform the method steps described above and further described below.

[0126] In the context of the present invention, a plant can refer to any facility where a particular product is made or power is generated.

[0127] In the context of the present invention, a chemical plant refers to any manufacturing facility based on a chemical process, for example, using a chemical process to convert feedstock into a product. In contrast to discrete manufacturing, chemical manufacturing is based on continuous or batch processes. Thus, such monitoring and / or control of a chemical plant is time-dependent and thus based on large time series datasets. A chemical plant can include more than 1,000 sensors that generate measurement data points every few seconds. Such dimensions result in several terabytes of data being handled within a system that controls and / or monitors a chemical plant. A small-scale chemical plant can include two or three thousand sensors that generate data points every 1 s to 10 s. For comparison, a large-scale chemical plant can include twenty or thirty thousand, for example, 10,000 to 30,000 sensors that generate data points every 1 s to 10 s. Explaining the situation of such data results in handling from several hundred gigabytes to several terabytes.

[0128] A chemical plant can produce products by one or more chemical processes that convert feedstock into products via one or more intermediate products. Preferably, the chemical plant provides an encapsulated facility that produces products that can be used as feedstock for the next step in the value chain. The chemical plant can be a large-scale plant such as an oil and gas facility, a gas purification plant, a carbon dioxide capture facility, a liquefied natural gas (LNG) plant, an oil refinery, a petrochemical facility, or a chemical facility. For example, an upstream chemical plant in petrochemical process production includes a steamcracker where naphtha is processed into ethylene and propylene. These upstream products can then be provided to further chemical plants to derive downstream products such as polyethylene or polypropylene, and the downstream products can again serve as feedstock for chemical plants that derive further downstream products. A chemical plant can be used to manufacture discrete products. In one example, one chemical plant can be used to manufacture a precursor for polyurethane foam. Such a precursor can be provided to a second chemical plant for the manufacture of discrete products such as separation plates containing polyurethane foam.

[0129] Value chain production leading to a final product via various intermediate products can be dispersed at various locations or integrated within a fairbunt site or chemical park. Such a fairbunt site or chemical park comprises a network of interconnected chemical plants where products manufactured in one plant can serve as feedstock for another plant.

[0130] A chemical plant can include a plurality of assets such as, to name a few, heat exchangers, reactors, pumps, pipes, distillation or absorption columns, etc. In a chemical plant, some assets can be considered extremely important. Extremely important assets are those whose interruption would have a significant impact on plant operation. This can lead to the deterioration of the manufacturing process. A decrease in product quality or even a halt in production can be the result. In the worst-case scenario, a fire, explosion, or toxic gas release can be the result of such an interruption. Therefore, such extremely important assets can require more precise monitoring and / or control, while other assets depend on the chemical processes and substances involved. To monitor and / or control chemical processes and assets, multiple actors and sensors can be embedded in the chemical plant. Such actors or sensors can provide process- or asset-specific data related to, for example, the state of individual assets, the state of individual actors, the composition of chemical substances, or the state of the chemical process. In particular, process- or asset-specific data can include one or more of the following data categories: - Process operation data such as the composition of feedstock or intermediate products - Process monitoring data such as flow, material temperature, etc. - Asset operation data such as current, voltage, etc., and - Asset monitoring data such as asset temperature, asset pressure, vibration, etc. including one or more of these.

[0131] The fourth aspect is also reflected in the following items: 1. A method for generating a hybrid model for monitoring and / or controlling a process network having at least two plants connected to each other, comprising - providing a system of equilibrium equations from a folded graph structure of a process network having observable physicochemical quantities on edges and their relationships to vertices; - Receiving, via an input interface, at least one objective specifying at least one process parameter dependency to be trained; - Retrieving, via the input interface, historical data of a process network having at least two plants connected to each other; - Training a hybrid model including a balanced system and a data-driven model based on the historical data and at least one objective specifying at least one process parameter dependency to be trained; - Providing, via an output interface, the trained hybrid model A method comprising the above steps. 2. The method according to item 1, further comprising the step of performing a steady-state test on the historical data of the process network. 3. The method according to any one of the preceding items, further comprising the steps of data reconciliation and / or gross error detection. 4. The method according to any one of the preceding items, wherein the hybrid model further includes a rigorous model reflecting physicochemical laws. 5. A system for generating a hybrid model for monitoring and / or controlling a process network having at least two plants connected to each other, - An input interface for receiving one objective specifying at least one process parameter dependency to be trained; - An output interface for providing the trained hybrid model; - A processor configured to perform the method according to any one of items 1 to 4 A system comprising the above components. 6. A computer program product that, when executed on a computer, performs the method according to any one of items 1 to 4.

[0132] This disclosure applies to the systems, methods, computer programs, computer-readable non-transitory media, and computer program products disclosed herein as well. Therefore, no distinction is made among systems, methods, computer programs, computer-readable non-volatile memory media, or computer program products. All features are disclosed in connection with the systems, methods, computer programs, computer-readable non-transitory storage media, and computer program products disclosed herein. BRIEF DESCRIPTION OF THE DRAWINGS

[0133] DETAILED DESCRIPTION OF THE INVENTION

[0134] Detailed Description FIG. 1 shows a process network of two plants having a first plant A 10 and a second plant B 20, and the two plants are interconnected by a product transport system 30.

[0135] A simplified flowchart of plant A10 is shown in FIG. 2. This simple flowchart is a digital process representation of the first plant A.

[0136] In this case, the plant is a simplified ammonia production plant 100. The product supply unit 110 provides educts to the mixer 120, and the pipe system 130 then transports the stream of mixed educts to the reactor 140 where the reaction takes place. The heat exchanger 140 liquefies the reaction product before separation. After the reaction, ammonia is separated from the residue in the separator 150. The product is provided to the transport system 30 via the product pipe 150. The residue is provided to the splitter via the residue pipe 170, and the splitter returns a portion of the residue to the mixer, and another portion of the residue is provided to a further location. In this example, the temperature sensor 180, the pressure sensor 190, and the volumetric flow rate sensor 195 are provided on the residue pipe 170.

[0137] Figure 3 shows the graph structure 200 of the first plant. Each vertex 2xx in this graph structure represents a unit operation. Vertex 210 represents the unit operation of a mixer, vertex 220 represents the unit operation of a reactor, vertex 320 represents the unit operation of a heat exchanger, vertex 240 represents the unit operation of a separator, and vertex 250 represents the unit operation of a splitter. A further vertex 260 - the environment vertex - is added to the graph structure. This vertex serves as a sink and a source, ensuring that the graph structure represents a self - contained system. Describing the plant as a self - contained system has the advantage that the conservation laws of physics apply.

[0138] Edges link vertices. Edges represent at least physicochemical quantities and metadata representing at least physicochemical quantities and measurable tags.

[0139] In the case of edges 415, 425, 435 around the unit vertex 240, these physicochemical quantities include that the total mass flow entering vertex 240, represented as the selected physicochemical quantity on edge 415, is equal to the sum of the mass flows represented by edges 425 and 435 exiting vertex 240.

[0140] Edges 415, 425, 435 also include metadata representing at least the selected physical quantities.

[0141] One physicochemical quantity represented in the metadata of edge 415 is the total mass flow entering the unit operation 240. One physicochemical quantity represented in the metadata of 435 is the mass of NH3 exiting the unit operation 240. One physicochemical quantity represented in the metadata of edge 425 is the combined residue, in this example, the mass of N2 and H2. Further physicochemical quantities represented in the metadata of edge 425 are the values from the temperature sensor, pressure sensor 190, and flow sensor 195, namely, the pressure P, temperature T, and volumetric flow rate F of the residue.

[0142] Metadata also includes measurable tags. On edge 425, P, T, and F will be tagged as measurable.

[0143] A further physicochemical relationship represented by edge 425 is a relationship in which the mass flow for the residue can be determined from P, T, and F.

[0144] Under this condition, only one of the mass flows represented by edge 415 and edge 435 has to be measured to determine the other mass flows.

[0145] By using the metadata on the edge to evaluate all physicochemical quantities, a new graph can be generated. An example of such a new graph structure is shown in Figure 4a.

[0146] 300 represents a graph having vertices 310 to 350. In this example, one physicochemical quantity (assuming total mass flow) on all edges is measured and / or determined using a plurality of physicochemical quantities.

[0147] This means that this physicochemical quantity is observable. The generated folded graph structure 400 does not change as seen in Figure 4b.

[0148] Figures 5a and 5b show an alternative graph structure in which one physicochemical quantity is only measured and / or determined on edges 515, 555 and remains unknown on edges 525, 535, 545, and then a folded graph structure 600 is generated that only includes the edges on which the physicochemical quantity is observable.

[0149] Depending on the physicochemical quantity starting with the same graph structures 400 and 500, the folded graph structures can be different. Therefore, a folded graph is generated for each physicochemical quantity.

[0150] Figure 6 shows the network of plants, where each plant is shown as graph structures 2000 and 3000. Feeds for the first plant are shown as 2100 and 2200. River 4000 serves as a water supply for cooling. The cooling water is provided to vertex 2300 that represents a unit operation. Product 2900 is produced in the first plant. Waste 2800 is also produced in the first plant. The waste of the first plant serves as a feed for the second plant. The waste is provided to vertex 3200, and the second feed 3500 is provided to the second plant. The second plant provides an output product at 3800. The distribution from product 2900 to waste 2800 can depend on various process parameters, which in turn affect the product output at 3800. In this example, the process parameters to be trained are related to the mass flow at output 3800. The mass flow at output 3800 is a function of the temperature of the river water. Generally, this is not a relationship that can be solved by an exact model. In this case, the hybrid model can be trained based on the historical data of the first plant.

[0151] Figure 7 shows method 5000 of the first aspect. In the first method step 5100, a first digital representation of a process network is provided that includes the digital process representation of each plant and the connections of each plant to other plants and sensor elements arranged within the process network. The digital process representation of each plant can be according to Figure 2.

[0152] In step 5200, a graph structure is generated based on the first digital representation. The graph structure 〇Vertices representing unit operations 〇Edges linking unit operations representing stored quantities, including edge metadata representing physicochemical quantities, and measurable tags including.

[0153] In step 5300, the folded graph structure Vertices representing virtual unit operations Edges linking to at least virtual unit operations representing physicochemical relationships, including edge metadata representing observable physicochemical quantities, and the relationships thereof to vertices generated based on the graph structure generated in step 5200 including

[0154] In step 5400, a set of equilibrium equations is derived from the folded graph structure In step 5500, a set of equilibrium equations and physicochemical quantities for monitoring and / or controlling the operation of the process network are provided

[0155] Figure 8 shows the method 6000 of the second aspect In step 6100, a request for at least one process network operation parameter is received via the input interface

[0156] In step 6200, a set of equilibrium equations and physicochemical quantities from the folded graph structure are retrieved via the input interface

[0157] In step 6300, historical data related to observable physicochemical quantities and metadata related to at least one process network operation parameter are retrieved from the database

[0158] In step 6400, values for at least one process network operation parameter are determined by solving a system of equilibrium equations based on the historical data and current data

[0159] In step 6500, values of at least one process network operation parameter are provided via the output interface

[0160] Figure 9 shows the method 7000 of the third aspect In step 7100, requirements for at least one optimization objective are received via an input interface by specifying at least one process parameter to be optimized.

[0161] In step 7200, a set of equilibrium equations and physicochemical quantities are received via an input interface from a folded graph structure.

[0162] In step 7300, historical data related to observable physicochemical quantities and metadata related to at least one process network operation parameter to be optimized are retrieved from a database.

[0163] In step 7400, values for at least one process network operation parameter to be optimized are determined by solving a system of equilibrium equations.

[0164] In step 7500, values of at least one process network operation parameter to be optimized are provided via an output interface.

[0165] FIG. 10 shows a method 8000 of a fourth aspect. In step 8100, a set of equilibrium equations and physicochemical quantities are received via an input interface from a folded graph structure.

[0166] In step 8200, at least one objective for specifying at least one process parameter dependency to be trained is received via an input interface.

[0167] In step 8300, historical data of a process network having at least two plants connected to each other are retrieved via an input interface.

[0168] At step 8400, training is performed on a hybrid model that includes a balanced system and a data-driven model based on at least one purpose specifying historical data and at least one process parameter dependency to be trained.

[0169] At step 8500, the trained hybrid model is provided via an output interface.

Claims

1. A computer-implemented method for generating a model representation of a process network having at least two interconnected chemical plants to enable controlling or monitoring said process network, comprising: - providing a digital representation of said process network, said digital representation comprising: 〇 a digital process representation of each plant, 〇 connections of each plant to other plants (realized by mass or energy flows in the description) and sensor elements arranged within said process network; - generating a graph structure based on said first digital representation, said graph structure comprising: 〇 vertices representing unit operations, 〇 edges representing physicochemical quantities, said physicochemical quantities including mass flow, energy flow, and component flow, said edges linking vertices, said edges including measurable tags for each of the represented physicochemical quantities, said measurable tags indicating whether the physicochemical quantity can be measured within said process network or not; - classifying said physicochemical quantities that can be measured within said process network as observable physicochemical quantities; - classifying physicochemical quantities that can be calculated from equilibrium equations around vertices as observable physicochemical quantities; - deleting edges having physicochemical quantities not classified as observable, and integrating a plurality of vertices adjacent to the deleted edges into collapsed vertices to generate a collapsed graph structure based on said graph structure, - said collapsed graph structure comprising: ・ collapsed vertices representing virtual unit operations, ・ vertices representing unit operations; A step of generating a foldable vertex and / or an edge that links to the vertex and represents only observable physicochemical quantities, - A step of deriving a set of equilibrium equations for each mass flow, energy flow, or component flow around each vertex, wherein the set of equilibrium equations models the network of the plant and the set of equilibrium equations is usable for monitoring, control, production planning, and prediction models, the step of deriving, - A computer-implemented method comprising providing the set of equilibrium equations to a control device, a monitoring device, a production planner device, and / or a prediction model generator. **Claim 2** The method according to claim 1, wherein the vertex further represents vertex metadata including physical quantities linked to the respective unit operations. **Claim 3** The method according to claim 1 or claim 2, wherein generating the foldable graph structure includes generating a foldable graph for each physicochemical quantity. **Claim 4** The method according to claim 3, wherein providing a set of equilibrium equations from the foldable graph structure includes providing a set of equilibrium equations for each physicochemical quantity. **Claim 5** The step of folding the edge - Selecting at least two vertices connected via the edge, - Folding the edge between the at least two vertices, thereby creating a virtual vertex, followed by the method according to any one of claims 1 to 4. **Claim 6** The step of generating a graph structure based on the first digital representation further includes generating a converged graph structure by assigning labels to all the physicochemical quantities according to whether all the physicochemical quantities are measured physicochemical quantities, determined physicochemical quantities, measured and determined physicochemical quantities, or physicochemical quantities that are neither measured nor determined physicochemical quantities, according to any one of claims 1 to 5.

7. The method according to claim 6, further comprising receiving a trigger signal, wherein the step of generating a converged graph structure in the method is initiated by evaluating the trigger signal indicating that a physicochemical quantity cannot be measured.

8. A system for generating a problem-specific representation of a process network to enable controlling or monitoring a process network having at least two interconnected chemical plants, - A processor configured to perform the method steps according to any one of claims 1 to 7, - A system comprising an output interface for providing the set of equilibrium equations for monitoring and / or controlling the operation of the process network.

9. A computer program product that, when executed on a computer, performs the steps of the method according to any one of claims 1 to 7.

10. A method for monitoring a process network having at least two plants, - Receiving a request for at least one process network operation parameter via an input interface; - Retrieving a set of equilibrium equations and physicochemical quantities from a folded graph structure via the input interface; - Retrieving from a database historical data related to observable physicochemical quantities and metadata related to the at least one process network operation parameter, wherein the set of equilibrium equations is derived according to the method described in any one of claims 1 to 7; the retrieving step, - Determining values for the at least one process network operation parameter by solving the system of equilibrium equations based on the historical data and current data; - Providing the values for the at least one process network operation parameter via an output interface. A method comprising the steps.

11. A method for controlling a process network having at least two plants, - Receiving, via an input interface, a request for at least one optimization objective by specifying at least one process network operation parameter to be optimized; - Retrieving, via the input interface, a set of equilibrium equations and physicochemical quantities from a folded graph structure, wherein the set of equilibrium equations is derived according to the method described in any one of claims 1 to 7; the retrieving step, - Retrieving from database historical data historical data related to observable physicochemical quantities and metadata related to the at least one process network operation parameter to be optimized; - Determining values for the at least one process network operation parameter to be optimized by solving the system of equilibrium equations; - Providing the values for the at least one process network operation parameter to be optimized via an output interface. A method comprising the steps.

12. A method for generating a hybrid model for monitoring and / or controlling a process network having at least two plants connected to each other, - extracting, via an input interface, a set of equilibrium equations and physicochemical quantities from a folded graph structure, wherein the set of equilibrium equations is derived according to the method of any one of claims 1 to 7, the step of extracting; - receiving, via an input interface, at least one objective specifying at least one process parameter dependency; - extracting, via an input interface, historical data of the process network having at least two plants connected to each other; - training a hybrid model including the system of equilibrium equations and a data-driven model based on the historical data and the at least one objective; - providing the trained hybrid model via an output interface. A method comprising.

Images