WO2004040386A1

WO2004040386A1 - Method for controlling the mode of operation of the equipment comprised in a refinery or a petrochemical industrial complex

Info

Publication number: WO2004040386A1
Application number: PCT/DE2003/003426
Authority: WO
Inventors: Günter GRUHN; Sven Orlowski; Thomas Peuker
Original assignee: Siemens Aktiengesellschaft
Priority date: 2002-10-28
Filing date: 2003-10-15
Publication date: 2004-05-13
Also published as: DE10250325A1; DE10250325B4

Abstract

The invention relates to a method and a tool for controlling the mode of operation of the equipment comprised in a refinery or a petrochemical industrial complex, according to which a distinction is made between a planning level, a scheduling level, and a process control level. The data that is generated on the scheduling level and is supplied to the process control level as reference variables is based on a decomposition of a system of equipment into at least two individual systems and a numerical optimization of variables, which is done in each individual system. The optimization process is coordinated by using factors that have been determined in one of the individual systems for optimization purposes in the other individual system.

Description

description

Process for controlling the operation of plants in a refinery or petrochemical industrial complex

The invention relates to a method for controlling the operation of the systems provided in a refinery or a petrochemical industrial complex.

Refineries are multi-product plants that produce a large number of products from different starting materials. The continuous operation of the process plants is opposed by the delivery or delivery of the educts and products, which is generally carried out discontinuously in so-called campaigns.

The task of production planning and control in refineries is timely purchasing and the selection of suitable starting materials, as well as the determination of the operating modes of the process plants with the aim of maximizing the profit achieved in the plant. Furthermore, we also have to ensure that customers are supplied with products that meet specifications on time.

Due to the complexity of the resulting optimization problem and the resulting uncertainties in the data required at the time of production planning, refineries generally have several hierarchical levels for production planning and control.

A planning level belongs to these levels. The aim of this planning level is essentially to generate rough specifications for the educt and product quantities to be purchased. What is needed from the outside is essentially information about educt and product prices as well as possible sales of certain sales estimated by marketing Products. The duration of the planning periods considered typically varies from around two weeks to one year. Often, a more detailed subdivision is made in the sense of short-term planning (two to four weeks) and long-term planning (one to six months).

The entire refinery is generally considered at the planning level. The entire complexity of this plant cannot be taken into account in a numerical solution to the optimization problems that arise at the planning level. This is greatly simplified when mapping the equipment and process system. This applies in particular to the modeling of non-linear system and blending behavior as well as the modeling of tanks.

Exact times and quantities for certain purchases and sales are often not known at the time of planning. For this reason, the relevant information is estimated and viewed in agglomerated form over the planning period or over larger sections of this planning period. The agglomerated view of this data is a characteristic feature of the planning level.

Linear programming (LP) methods are used to formulate and solve the optimization problems at the planning level.

The rough specifications for purchasing and production generated at the planning level are too imprecise and not sufficient to guarantee a smooth operational production process. Therefore, a scheduling level is still provided. This is used to generate time sequences of all the management variables required at the process control level that are sufficiently precise for a smooth operational process. In particular, the assignment of orders to equipment takes place when several suitable equipment are available at the same time. Also on the duling level is attempted to create a profit-optimized schedule. However, other criteria such as the robustness of a created schedule are often taken into account.

The duration of the periods considered in the scheduling is generally between three and fourteen days. Here too, a subdivision is made into short-term scheduling and long-term scheduling, although an exact assignment to the duration of the planning periods under consideration is imprecise.

If the decisions regarding delivery times and delivery quantities of starting materials such as crude oils are not made within the refinery in consultation with the suppliers, an exact quantity and schedule structure for starting materials is required for scheduling. Otherwise, the starting materials determined at the planning level are required for the scheduling. The determination of exact delivery times and delivery quantities takes place during the scheduling process. Furthermore, an exact quantity and schedule structure of the products to be delivered to customers is required.

The accuracy of the models used at the scheduling level is greatly improved compared to the planning level, so that a sufficiently precise specification of the management variables for the process control level can be guaranteed. For example, non-linear aspects of the sub-systems under consideration and blending are increasingly taken into account here. Nevertheless, these are usually not rigorous models. Simplified models such as empirical models are mostly used.

In contrast to the higher-level planning level, resolution is also sought when representing the time, which enables the generation of time profiles for the process control level that are sufficiently precise for a smooth operational process. Here, which, in contrast to the higher-level planning level, only agglomerates quantity data to a very limited extent. The required, sensible resolution generally depends on the dynamic behavior of the systems under consideration. To solve the optimization problems that occur at the scheduling level, there are largely only software tools based on simulations that support the personnel concerned in generating schedules based on manual specifications.

The process control level is used to generate suitable control parameters for the control systems of the plants in the refinery. The objective here is to track the plant operating point as accurately as possible based on the command values specified by the scheduling level, while at the same time ensuring safe plant operation. Other aspects, such as energy-saving system operation, can be taken into account within the scope of the remaining degrees of freedom, but are only of minor importance. The time periods considered at this level are in the range of seconds to minutes. In addition to the temporal profiles of management variables specified by the scheduling level, different measurement variables from the ongoing system operation are required and used at this level. Simplified, robust plant models are used on the process control level, which generally guarantee the control and regulation of the plant in real time. A large number of methods and established commercial systems exist, for example for advanced process control, in the control and regulation of process engineering plants.

Due to the complexity and complexity of the scheduling problems encountered in a refinery, several people are typically responsible for scheduling. The areas of responsibility are often functionally structured. The following breakdown is known, for example: Scheduling of the ships unloading or loading at a pier,

- Scheduling the use of different types of crude oil and the respective crude oil tanks as well as the processing of the atmospheric residual and the associated tanks,

- Scheduling the liquid gas production line and the relevant product tanks and scheduling of the gasoline production line and the relevant component and product tanks and - Scheduling the kerosene production line and the relevant component and product tanks.

Depending on the structure and organizational structure within a refinery, tasks can be divided up differently

The scheduling of the individual areas is generally not automated, but according to heuristics gained through daily experience. In many cases, the scheduler is supported by a software tool based on spreadsheets.

With this, for example, tank levels are determined for specific times of a specific planning period depending on the operating parameters specified by the scheduler for the systems under consideration. The plant and mixture models used in this tool are largely self-created and can be extremely inaccurate. With the help of this tool, the scheduler now tries to create a feasible schedule that, for example, does not show any fuel tank violations. This activity can be very time-consuming, especially in situations in which a large part of the existing planning freedom is already being used.

In most cases, this type of schedule creation lacks any criteria for comparing the quality of a feasible schedule that has been created with that of another. Quality criteria like that when executing the schedule Profits achieved or the flexibility of the schedule are recorded intuitively, but can only rarely be quantified and included in the decision-making process. Furthermore, regardless of the type of division of tasks, problems arise in coordinating the different areas of scheduling, since these influence each other. If the schedule for the pier occupancy is changed in the example given above, for example because a tanker loaded with crude oil is late, this also changes the specifications for the scheduling of the crude oil and thus also the schedules for gas, petrol and kerosene production , Often, the changed information is only communicated to the responsible persons too late. The data required for scheduling is generally not kept centrally. Due to the lack of or different evaluation criteria of the respective schedulers for the quality of the schedules you have created, a decision about improving the quality of a schedule at the expense of the quality of another schedule is usually not possible.

As already shown above, the schedule to be used for the entire refinery for a period of time results from several iteration steps of plans of the individual schedulers which have been developed almost independently of one another. The resulting overall schedule is usually discussed as part of a daily or shift-by-layer discussion with the managers responsible for the system. Problems that may become visible here can cause the schedule to be changed again before it is finally implemented.

It is disadvantageous for the current situation that decisions are made in the local area of responsibility, whereby the respective decision-makers know only insufficiently the situation of upstream or downstream aggregates. Decision-makers currently use my Excel sheets for your analysis. The necessary global decision is made in a meeting.

Known first simulation-based implementations, such as the SIPP-II solution from Chemtech or the Schedule ++ from OR Soft, work with simple models.

On the basis of this prior art, the object of the invention is to improve the data determination in the scheduling level, so that improved control data are available to the control and regulating units of the refinery's systems.

This object is achieved by a method with the features specified in claim 1. Advantageous refinements and developments of the invention are specified in the dependent claims. Claims 13 to 21 relate to a tool for controlling the operation of the systems provided in a refinery or a petrochemical industrial complex.

The advantages of the invention are, in particular, that the decomposition of an overall problem enables stable, comprehensible partial solutions and also creates the possibility of including more complex, high-quality models. For example, octane numbers and knock resistance can be taken into account when modeling product properties along the value chain. There is an optimization of sizes and not just a simulation.

The holistic view carried out according to the present invention leads in the direction of a global optimum for an entire system. The results obtained are reproducible and documentable, the quality of the results is improved. Further advantageous properties of the invention result from its exemplary explanation with reference to the figures. Show it

1 shows diagrams for explaining the principle of horizontal and vertical decomposition within refineries, FIG. 2 shows diagrams for explaining a deco position concept for numerical optimization, FIG. 3 shows a diagram which illustrates a generalization of the horizontal decomposition and a coordination of two individual systems, FIG 4 shows a diagram which illustrates a decomposition into four individual systems and a coordination of these four individual systems,

5 shows a flowchart to illustrate the optimization method taking place at the scheduling level, FIG. 6 shows a diagram of an application example and FIGS. 7-10 show diagrams to illustrate determined and estimated target function values.

The present invention relates to a method for controlling the operation of the systems provided in a refinery or petrochemical industrial complex. According to this procedure, rough specifications for the required raw materials and product quantities to be produced are determined in one planning level. The rough specifications determined in the planning level are converted in a scheduling level into data that correspond to the temporal courses of management figures. The data determined in the scheduling level are fed as control variables to the control and regulating units provided in the plants of the refinery. These generate control variables for the respective system, taking into account the management variables mentioned. In addition to the command variables, this determination of the control variables also includes which are obtained from the ongoing plant operation.

At the planning level, refinery-wide planning takes place for periods of one to six months. At this level, methods of linear programming (LP) are mainly used. The strictly linear programming is expanded by recursive methods which, for example, allow the representation of the mixture of streams. Nonlinear behavior of systems is only mapped to a limited extent within the so-called LP models by piecewise linear modeling. Within the planning level, the types and quantities of the crude oils to be purchased and the quantities of end products to be produced are essentially determined. The aim is to maximize the profit for the entire refinery in the planning period under consideration. Decision variables at this level of decomposition for the planning period under consideration are the quantities of different crude oils to be bought, the distribution of material flows within the refinery, the plant operating methods and the distribution of the blending components among the end products. Additional conditions for the optimization problems in the planning level are essentially upper and lower limits for the quantities of purchases and sales, upper and lower limits for throughput through sub-systems and product specifications. The one on the

Specifications generated at the planning level, such as the amount of crude oil purchased, are taken into account when scheduling the fuel line, among other things.

The subject of the present invention is, in particular, a systematic approach for optimizing the reference variables determined at the scheduling level. The starting point for the need for such optimization is generally short-term events or circumstances that make it necessary to change the way the systems are operated. An example of such an event is the short-term availability of particularly cheap crude oil on the spot market. In In this case - as far as the refinery's storage and processing capacities allow - the production volume can be increased in the short term in order to benefit from the currently cheap crude oil. Another example of such an event is the short-term failure of an aggregate, which is associated with a temporary reduction in the tank capacity. In this case, the production may have to be sensibly reduced in order to cope with the short-term event. As part of this adaptation of the operation of the systems to short-term events or conditions, technological, logistical and economic aspects must be taken into account.

A basic idea of the invention is to carry out the scheduling in the sense of a holistic view, taking into account the largest possible balance space, for example the entire refinery. The invention allows the inclusion of high-quality process models, in particular the modeling of product properties along the value chain. The modeling can be non-linear. There is a mathematical optimization of variables.

This is essentially accomplished by first decomposing an equipment system into at least two individual systems, each of these systems having at least one variable suitable for numerical optimization. Coordination is then carried out in such a way that the variables of one of the individual systems are first optimized and then factors determined during this optimization are used in the optimization of the variables in the other individual system, and vice versa. If such coordination has taken place in the horizontal direction, then a link to further decision levels is made in the sense of an integration concept or a concept for vertical coordination. In the context of horizontal or spatial decomposition, a superordinate optimization problem is divided according to functional criteria. For example, the entire refinery is divided into plant strings, which in turn are divided into individual plants. This is illustrated in FIG. 1, which shows a diagram on the left to explain the principle of horizontal decomposition within a refinery. Reference number 1 denotes the entire refinery, reference number 2 the individual production lines of the refinery and reference number 3 the process units or plants of the refinery. The arrow 4 symbolizes the increasing accuracy of the models used as the complexity of the optimization problems decreases.

In the context of vertical or temporal decomposition, a higher-level optimization problem is divided according to temporal aspects. In this way, a planning period can be divided into several shorter periods, which are individually optimized. This is shown on the right side of FIG. 1. The planning level is illustrated with the reference number 6, the scheduling level with the reference number 7 and the process control level with the reference number 8, each of which is plotted over the time axis. The arrow 5 symbolizes the increasing complexity of the optimization problems, the longer the period under consideration.

FIG. 2 shows diagrams of a developed decomposition concept, which can be used for the numerical optimization of the production planning and control of the gasoline line of a refinery. The corresponding decomposition levels are created by a combined application of horizontal and vertical decomposition. In the decomposition structure for numerical optimization, the scheduling level present in the decomposition structure according to FIG. 1 is divided into two levels, namely a long-term scheduling level 7a and a short-term scheduling level 7b. Otherwise, the structures according to FIG. 1 and FIG. 2 match.

A general form of the horizontal deco position concept according to the invention is presented below.

A prerequisite for coordination according to the developed concept is the generation of Lagrangian factors or comparable information when solving the respective optimization problem. Optimization algorithms such as random search methods or algorithms based on heuristics are unsuitable for the coordination strategy presented. An example of information comparable to the Lagrangian factors are the shadow prices generated in the context of the simplex process.

Another prerequisite for coordination is the existence of market prices for the flows that cross the system boundaries. In addition, valid start values for the optimization variables must be available at the start of the optimization. These valid start values can be based on existing empirical values, can be obtained in the context of a simulation or can also be determined in the context of an optimization process.

The systems or equipment systems for the optimization problems considered in the process engineering, continuous production can generally be divided horizontally into two smaller systems. The systems that were created can also be divided up to a reasonable extent from a process engineering point of view into two further subsystems. In this way, an entire equipment system can be broken down into individual systems with sizes suitable for numerical optimization. The division of such a system into two individual systems with any number of system boundaries and beyond Currents flowing between the subsystems are shown in FIG.

Reference number 9 denotes the overall system and reference numbers 10 and 11 the two subsystems into which the overall system is divided. The arrows shown in FIG. 3 symbolize currents that flow into the system, leave it or flow between the subsystems. Each of these flows can be described by a flow vector, the components of which describe the mass or volume flow and the properties of the corresponding flow. Typical properties are, for example, the research octane number RON, the motor octane number MOZ and the density.

The procedure is as follows:

First, an optimization problem for the second system 11 is solved for a valid operating point of the overall system, in which the current vectors for the currents fed back into the system 10 remain fixed to the values predetermined by the starting point of the optimization. In this first solution to the optimization problem, Lagrangian factors are generated, from which weighting factors for the currents flowing from the system 10 into the system 11 at the starting operating point can be determined. The weighting factors determined in this way are used in the target function fi of the optimization problem now to be solved for the system 10. The objective function to be maximized is as follows:

Here ω denotes a vector of the weighting factors, S a current vector, P a process variable vector, B the operating costs and E the revenue. The indices i denote all currents flowing from the system 10 into the system 11. Additional conditions of the optimization problem to be solved here are the quantities and properties of the flows entering the system 10, upper and lower quantity restrictions for intermediate products, specifications and prices of intermediate products, plant models for the system 10 and restrictions for step sizes.

The current vectors resulting from the solution of this optimization problem are now specified in the auxiliary conditions for the optimization problem of the system 11 that is now to be solved. For this purpose, from the previous solution of the optimization problem for the system 10, the weighting factors from the system 11 into the system 10 flowing currents determined and used in the objective function f ₂ for the optimization problem of the system 11. The objective function to be maximized is as follows:

i

The indices i denote all currents that flow from the system 11 into the system 10. The constraints on this optimization problem are the quantities and properties of the flows entering system 11, upper and lower quantity restrictions for intermediate products, specifications and prices of intermediate products, plant models for system 11 and restrictions for step sizes.

In the following it is checked whether the termination criterion for the optimization is fulfilled. To converge the between To ensure currents flowing in the systems, it is checked whether the change in all current vector components of these currents between two iterations passes below certain values. If none of the corresponding current vector components changes by a significant value between two runs, the optimization is terminated and the results are output. If the currents have not yet converged, the new optimal current vectors determined for the system 11 are used in the secondary conditions for the system 10 and a new optimization is carried out.

The current vector can contain information about the mass flow, about compositions, about temperature and pressure, as well as about physical properties such as density and octane number.

The procedure described above is explained in more detail below on the basis of the flow diagram shown in FIG.

The reference number 19 denotes the start of the optimization process of the overall system 9. According to step 20, the optimization problem for the system 11 with fixed current vectors is solved. Then a determination of new weighting factors for the system 10 takes place in step 21. Taking these new weighting factors into account, a solution of the optimization problem for the system 10 takes place in step 22. Then, according to step 23, new current vectors are specified for the system 11. In the subsequent step 24, new weighting factors for the system 11 are determined. These new weighting factors are used in step 25 to solve the optimization problem for the system 11. In step 26, a query is made as to whether the abort criterion is met. If the termination criterion is not met, then new current vectors are specified for the system 10 in step 29 and the system returns to Step 21. On the other hand, if the termination criterion is met, the results are output in step 27. Step 28 denotes the end of the optimization of the overall system 9.

FIG. 4 shows a diagram which illustrates a decomposition of an overall system 12 into four individual systems 13, 14, 15, 16 and a coordination of these four individual systems. In the case of this system, in the same way as was described in connection with FIGS. 3 and 5, the systems 14 and 16 are optimized in order to form a new system 17. This is followed by an optimization with regard to the two systems 15 and 17, forming a new system 18. Finally, there is an optimization with respect to the systems 13 and 18. In the course of the above-described coordination of the individual systems, each iteration step involves a new optimization of the respectively formed one new systems required.

As is clear from the example shown, the effort for coordination increases with the number of systems to be coordinated. In the context of a horizontal decomposition, it must therefore be considered whether subdivision into a large number of small systems justifies the coordination effort that arises.

In principle, the procedure outlined above also allows the optimization of the gasoline line to be integrated into an optimization of the entire refinery. For example, the LP model, which mostly exists in refineries, could be used for the equipment system that is not part of the gasoline chain. The required interface information for determining the change in target function of the individual systems to be optimized depending on the change in secondary conditions can be obtained from the shadow prices of the LP model or from the Lagrangian factors of the optimization problem for the gasoline train. One way of coordinating the various vertical planning levels is to specify results from the optimization of the higher level as secondary conditions for the optimization problem of the lower level.

As is also possible within the framework of the developed decomposition strategy, a number of optimal blending recipes can be determined, for example, at the higher level of long-term scheduling. These recipes are used within the secondary conditions of the subordinate level of scheduling in order to optimize a certain objective function at this level by determining an optimal recipe order.

The objective function of the subordinate level can also be used to take into account the intentions pursued on the upper level on the lower decomposition level. In this way, deviations from the specifications created at the higher planning level can be minimized on the lower planning level. In the decomposition concept presented, for example, at the subordinate planning level of scheduling, the optimal timing of recipes can minimize the deviations from fuel levels at the end of the period of the long-term scheduling optimization problem.

In the following, a suitable concept for coordinating the sub-problems arising from decomposition in the context of production planning and production control of the gasoline train is discussed in more detail.

For the system reformer blending system of a refinery, a coordination concept for the optimization problems that arise from a decomposition of the optimization problem for the long-term scheduling level is initially shown for a period of time. The overall system 30 considered here, which is shown in FIG. 6, consists of the subsystems reformer 32 and blender 24. A series of educt streams, such as the reformer feedstro 31 and components 36 for gasoline blending, enter the balance space of this overall system 30. Furthermore, a number of streams leave the balance area. In the case of the blending system 24, these are products 35 manufactured according to specific specifications and in the case of the reformer system 32, intermediate products 37 such as hydrogen. Within the system under consideration, the reformate stream 33 leaving the reformer 32 is used as a component for the blending 24.

Each individual current within this overall system is described by a so-called current vector S _j , the

Components Sij contain the mass or volume flow and properties of the corresponding flow. The index i denotes a specific current, the index j denotes a specific component of the current vector in question. Typical properties are, for example, the research octane number RON, the motor octane number MOZ or the density p. The following applies:

S, = f (p _t )

Here, the vector P _t describes the plant operating mode set in the subsystem t. This vector contains all process variables to be optimized in the subsystem, such as pressures, temperatures or, in the case of blending, volume flows of individual components that are blinded for each end product.

As part of an optimization at the long-term scheduling level, the reformer's driving style for the entire system and the optimal distribution of the resulting blending components among the different end products must be determined. The profit achieved in the entire system is made up of the profits achieved in both systems. For the coordination of the two systems, the question arises how a change in the optimization variables of one subsystem affects the target function of the other subsystem. In the presence of such quantitative information, an iterative procedure can be developed which, starting from a valid operating point of the overall system, changes the operating points of the individual systems in such a way that an improved operating state for the overall system is ultimately achieved.

As will be shown below, the analysis of the Lagrangian factors at the optimum of a subsystem 2 provides information about the partial change in the target function f of this subsystem as a function of the changes in the current vector components Sι, j of the current flowing from system 1 to system 2 You win. The vector of these weighting factors is referred to as ^ below:

If the vector CÖ _{J is known} for a current Si that flows from system 1 into system 2, the change in the target function of this system depending on changes in the current generated in system 1 can be carried out in a certain range around the respective optimum of system 2 and thus predict changes in system 1 process variables. The calculation of the changed current vectors as a function of the changed process variables takes place here via the system model. The optimization problems that arise at the long-term scheduling level can be described as:

min f (x)

taking into account the secondary conditions g in the form:

g (x) = 0

Typical constraints of the multiblend optimization problem are, for example, volume balances and mixing rules to meet product specifications. Such a mixing rule reflects the functional relationship between the optimization variables of the multi-blending problem X, that is to say the quantities of component streams blended into a specific product, the properties of the component streams contained in the stream vectors S ₍ and a property of the product produced). These secondary conditions are generally formulated as follows , where the constant e _k corresponds to a value specified by specifications:

For optimization problems with constraints, the Lagrangian function L is defined with the vector λ of the Lagrangian factors λk belonging to the constraints as follows:

L (x) = f (x) + λ-g (x)

The corresponding Lagrangian factors λ _{k are} generally determined using numerical optimization methods.

In the following, the calculation of the change ω in the optimal value of a target function f as a function of the change in a current vector component S is first used as an example. gone, which only occurs in a single constraint g of the optimization problem.

Let f ^{* be} the value of the target function at the optimum X, which in turn depends on the current vector component S considered: f = f (χ- (S »

The following also applies to the secondary condition g ^{* considered} :

g ^* = g (x ^* (S)) = 0

as well as the Kuhn-Tucker condition:

Vf ^* + λ ^* (S) -Vg ^* = 0

According to the general chain rule:

dg ^* (s, x '(s)) ₌ ag ^* (s, x ^* (s)) ög ^* (s, x ^' (s)) dx ög ^* (s, x ^* (s)) dS as; dS ^"' δχ ^* _n ^' dS

₌ * L _{+ V} g ^{* τ} . ^ = 0 öS dS

For the change of the target function depending on the current vector component S, the following applies with the chain rule:

ω = - ^d - * f * = _n g _M ra _f d _{) f} f * -—dx ^* - = Vf * τ dx ^* dS dS dS

It follows:

d, *. * / _Λ \ „* τ dx ω = —f = -λ (S) -Vg • - dS dS

For the change of the target function depending on the property S follows:

Since a current vector component cannot only appear in one constraint of an optimization problem, the above approach is extended to all n constraints of the optimization problem under consideration. For this follows:

ω ₌ dQi λ ^* ι (S) - + dg ₂ λ ^* ₂ (S) + + λ _k (S) dg _k dS ^, 'dS ^{^} ' dS ^Λ dS

With the help of this equation, the weighting factor ω with respect to a current vector component occurring in these auxiliary conditions can be calculated using the Lagrangian factors from auxiliary conditions of an optimization problem and the partial derivatives of these auxiliary conditions according to the current vector components S.

The immediate validity of the determined weighting factors for predicting the change in objective function when the associated constraint changes is only given for infinitesimally small changes in the respective constraint. This is based on the following:

- The estimation of the change of a nonlinear objective function by a constant term corresponds to a linear

target function and can therefore only be valid in the immediate vicinity of the specific optimum. - An improvement of the optimum in a certain direction may be restricted by other constraints. In this case too, the weighting factor must be determined again.

In the following, a typical multiblend optimization problem is used to check the extent to which the weighting factors obtained can be used to predict the change in objective function depending on the change in secondary conditions.

The optimization problem considered here was created on the basis of a real quantity and schedule structure for products and starting materials. The blending models used were also calibrated according to real operating data.

The example optimization problem was first solved in the course of the following investigation. The optimum f ^* ait found in this way is identified in FIGS. 7-10 by the vertical axis.

At the optimum found, the weighting factors C0i, j with respect to selected flow properties and the volume flow of the reformate were determined using the method described above.

Using the weighting factors obtained in this way, new optimal values of the objective function f ^* _nS u, estimated as a function of the change in a component of a current vector ΔSi, j, were estimated:

f new, estimated ~ ^ old ⁺ "T ^ T- 'ΔSy - f _old + GO ^• ΔSj S ^αb i, j

These values obtained in this way were compared with the values f ^* _n eu, which result from a real change in the constraint by ΔS _± , j and then again solving the optimization problem with the changed constraint while keeping all other constraints constant. If the optimization problem could no longer be solved by changing a constraint, no corresponding entry was made for f ^* _ne u in the figures.

The vapor pressure, RON, MON and the volume flow of the reformate were selected as components of a flow vector to be examined. Nonlinear blending models were used to determine the current properties of the manufactured products.

In order to ensure comparability of the individual optimization results despite different orders of magnitude of the current vector components under consideration and Lagrangian factors belonging to the constraints, the range of current vectors under consideration for the investigation was limited to values at which the predicted change in the target function did not exceed ten percent of the initial value.

FIG. 7 shows the estimated and ascertained values of the target function around the optimum f ^* _a ιt when the vapor pressure changes. The estimated and determined values for the target function agree over a wide range of vapor pressure.

FIG. 8 shows the change in the objective function as a function of a change in the research octane number.

Here, the predicted changes in a narrow area around the initial value agree well with the true changes, but the change is underestimated in the area of higher octane numbers. If the octane number of the reformate is greatly reduced to approximately 94 RON, no solution to the optimization problem can be achieved. Also for the dependency of the optimal target function value on the number of engine octets shown in FIG. 9, a good agreement between the predicted and true value applies to the majority of the values. However, a problem of non-linear optimization becomes clear with this series of measurements:

One of the values f ^* _n eu is significantly above the corresponding value of f ^* _ne u, estimated- The points f ^* new to the right of this pair of values should have at least the same objective function value, since by a further increase in the number of motor octets for which no upper limit is specified by specifications, no deterioration of the optimal objective function value should arise. Obviously, only a local optimum was achieved at these points due to the specification of unsuitable starting values.

Similar to FIG. 9, the difficulties in nonlinear programming become clear again in FIG. While a good forecast accuracy is achieved in the area of increased volume flows, the optimized values in the area of a volume flow that is lower than in the base case are exclusively above the values obtained by estimation. Furthermore, a better target function value is achieved for these points than in the area of higher volume flows. Since the upper tank level limit for reformate was not limiting for the optimization problem under consideration, this behavior is again due to a local optimum that was found by the optimization algorithm on the basis of the specified start values.

In summary, it can be said that in all the cases considered, the tendency of changes in the constraints to affect the objective function was correctly predicted. In many cases a very good agreement is reached for the respective, albeit local, optimum. The developed method is therefore basically suitable to work in a tight

Area around which the respective optima predict the change in the objective function. Based on this result, the development of a step-by-step coordination strategy for the optimization problems under consideration.

As has been shown, the weighting factors can be used to predict changes in the optimal target function value depending on the changes in current vectors. This information can be used to specifically change the driving style of the systems upstream of the blending in such a way that the properties and quantities of blending component flows in the overall system that are changed in this way achieve a higher profit.

In the first step, values are specified for the current vectors of the reformate. These current vectors should correspond to a possible driving style of the reformer, for example the current driving style at the time of planning.

In the second step, the respective multi-glare optimization problem is now solved with the specified current vector for the reformate current. In addition to the determination of optimal values for the optimization variables and the value of the target function at the optimum, this optimization also calculates the Lagrangian factors for all constraints of the optimization problem.

With the help of these Lagrangian factors related to the individual constraints, the weighting factors regarding the individual components of the current vector of the reformate are now determined.

Now the separate reformer optimization problem is solved. Here, the vector with the operating parameters of the reformer is to be changed in such a way that the greatest possible increase in the profit of the overall system consisting of reformer and blender is achieved compared to the current state. The following issues are considered: - Changes in the profit made from blending. The estimation is carried out using the weighting factors determined with regard to a component of the reformate current vector;

- change in the operating costs of the reformer caused by a change in the way of operating;

- Changes in revenues that result from the change in sellable intermediate products manufactured in the reformer.

The target function of the reformer optimization problem is obtained by summarizing the weighting factors belonging to the current vector.

^{max =} ® Ref 'l §Ref (FRef, new j ^" ~ §Ref

J

~ [^ Ref γ ^" Re., New j ~ ^ ef fRef.alt j J

⁺ l ^ Ref V ^" Ref, new / ~ ^ Ref, V Ref, old J

Additional conditions of the optimization problem to be solved here are the quantities and properties of the flows entering the system, upper and lower quantity restrictions for intermediate products, specifications and prices of intermediate products, the reformer model and limits for step sizes.

The reformer model establishes the connection between the operating parameter vectors and the current vector of the reformate. Furthermore, the model also determines the electricity vectors of the intermediate products and the operating costs.

The secondary conditions for limiting step sizes between the old and new set of operating parameters of individual iteration steps are of particular importance. As already mentioned, the validity of the prediction of objective function changes is limited to a narrow range around the respective optimum. In order not to leave this area, the changes in the new operating state compared to the old state should be limited in order to avoid convergence problems in the iterative solution of the two optimization problems. Possible strategies for determining the step sizes are:

- a restriction of the absolute change of components of the process variable vectors from new to old state:

Here, the value Si specifies an upper limit for the change in an operating parameter P _ne u, ι of the reformer compared to the previous iteration step.

- a restriction of the absolute change of current vector components from new to old state:

The value ει, j specifies an upper limit for the change in a component of a current vector compared to the previous iteration step

a limitation of the predicted change in the objective function by changing individual current vector components:

The parameter ε is a user-definable barrier for the upper limit of the change in target function generated by the change in the current vector component. The first two procedures assume that there is sufficient prediction accuracy of the weighting factors in a certain, absolute range of current vector or process variable vector components around the optimum determined. In the third case, on the other hand, it is assumed that the accuracy of the estimate using weighting factors depends on the change in the objective function.

Furthermore, in contrast to the first two procedures, the weighting factors belonging to the current vector components are taken into account in the last procedure shown.

In some cases, weighting factors for current vector components assume very high values. These values are usually only valid in an extremely narrow range around the respective optimum. By restricting the step size with constraints, the permissible step size for current vector components with high weighting factors is more restricted than the step size for other current vector components.

The quality of the individual strategies and the breadth of the steps used must still be checked by means of optimization calculations. A combination of individual strategies may also prove to be effective.

In the following it is checked whether the termination criterion for the optimization has been met. Possible termination criteria are, for example: exceeding a maximum number of iteration steps,

- exceeding a maximum number of iteration steps with no or only a certain marginal improvement or - falling below a minimal change in current vectors between two iteration steps. If the termination criterion is not met, the improved current vectors resulting from the solution of the reformer optimization problem are specified as new current vectors for the multi-glare optimization problem.

Then the multiblend optimization problem is solved again and then the reformer optimization problem until the optimization criterion is met.

Claims

claims

1. A method of controlling the operation of the facilities provided in a refinery or a petrochemical industrial complex, wherein

- rough specifications for the required raw materials and product quantities to be produced are determined in one planning level,

- The rough specifications determined in the planning level are converted in a scheduling level into data that correspond to the temporal courses of management figures,

- The data determined in the scheduling level are supplied as control variables to the control and regulating units provided in the systems and - The control and regulating units of the systems in each case generate the control variables for the respective system, taking into account the control parameters, so that

- To generate the data generated in the scheduling level, an equipment system is decomposed into at least two individual systems, each with at least one variable suitable for numerical optimization,

- The variables of one of the individual systems are optimized and

- Factors determined during this optimization are used for the purpose of coordination when optimizing the variables in the other individual system.

2. The method according to claim 1, characterized in that a target function is specified with respect to each individual system and the target function is changed as part of the optimization process by the factors determined in the respective other individual system.

3. The method of claim 2, d a d u r c h g e k e n n z e i c h n e t that the target function is changed depending on the amounts and properties of flowing material flows.

4. The method according to any one of the preceding claims, d a d u r c h g e k e n n z e i c h n e t that the optimization of a variable takes place using an iteration process.

5. The method as claimed in claim 4, so that a check is carried out as part of the iteration process as to whether the variable determined in each case fulfills an abort criterion.

6. The method as claimed in one of the preceding claims, in that starting values for the variables to be optimized are specified at the beginning of an optimization process.

7. The method according to any one of the preceding claims, that the decomposition is a horizontal decomposition and the optimization of the variables of the individual systems is a horizontal coordination.

8. The method according to any one of the preceding claims, characterized in that the scheduling level is divided into two sub-levels, one of which is a long-term scheduling level and the other a short-term scheduling level.

9. The method according to any one of the preceding claims, that a vertical coordination is carried out in which a level is linked to at least one further level.

10. The method according to claim 9, characterized in that the vertical coordination by specifying results from the optimization of a respective superordinate level as ancillary conditions for the optimization problem of the subordinate level or by using the objective function of a subordinate level in order to achieve the to take account of the objective pursued at the upper level at the lower level.

11. The method according to any one of the preceding claims, d a d u r c h g e k e n n z e i c h n e t that high-quality process models are used in the optimization process, which correspond to a modeling of product properties.

12. The method according to claim 11, so that a non-linear modeling is carried out.

13. Tool for controlling the mode of operation of the plants provided in a refinery or a petrochemical industrial complex, with an equipment system being decomposed into at least two individual systems, each with at least one variable suitable for numerical optimization, in order to generate the data generated in the scheduling level Which tool has means for optimizing the variables of one of the individual systems and for using factors determined in this optimization for the purpose of coordinating when optimizing the variables in the other individual system.

14. Tool according to claim 13, and that it has means that are provided with respect to each individual system for specifying a target function and for changing the target function in the course of the optimization process by means of the factors determined by the respective other individual system.

15. Tool according to claim 14, d a d u r c h g e k e n n z e i c h n e t that it has means for changing the target function depending on the quantities and properties of flowing material flows.

16. Tool according to one of claims 13 - 15, d a d u r c h g e k e n n z e i c h n e t that it has means for optimizing a variable using an iteration process.

17. Tool according to one of claims 13 - 16, d a d u r c h g e k e n n e e c h n e t that it has means for performing a horizontal decomposition and a horizontal coordination.

18. Tool according to one of claims 13 - 17, that it also means that it has means for performing vertical coordination.

19. Tool according to claim 18, characterized in that it has means for vertical coordination by specifying results from the optimization of a respective higher level as a secondary condition for the optimization problem of the lower level or by using the target function of a respective lower level in order to to pursue the objective pursued at the upper level at the lower level.

20. Tool according to one of claims 13 - 19, d a d u r c h g e k e n n z e i c h n e t that it has data which correspond to a modeling of product properties.

21. Tool according to claim 20, d a d u r c h g e k e n n z e i c h n e t that the models are non-linear.