Assistance in controlling a chemical reactor

Abstract

An assistance system (100) for assisting in controlling a chemical reaction run in a chemical reactor (10) for producing a plurality of products (20) is presented. The system comprises a) a product demands providing unit (101) configured to provide one or more future demands (40a, 40b, 50) for each of the products, b) a reaction model providing unit (102) configured to provide a reaction model, the reaction model indicating, for each of a plurality of combinations of values of a plurality of reaction control parameters (30a, 30b), a produced amount of each of the products, and c) a reaction parameter determining unit (103) configured to determine future reaction control parameter values based on the one or more future demands and the reaction model. This allows for a more efficient control of chemical reactions run in chemical reactors.

Description

[0001] Assistance in controlling a chemical reactor

[0002] FIELD OF THE INVENTION

[0003] The invention relates to an assistance system and an assistance method for assisting in controlling a chemical reaction in a chemical reactor for producing a plurality of products, and to a corresponding computer program. Furthermore, the invention relates to a control method for controlling a chemical reaction run in a chemical reactor for producing a plurality of products. The invention also relates to a method for the manufacture of a plurality of chemical products using the assistance system, the assistance method or the method for controlling a chemical reaction according to the invention.

[0004] BACKGROUND OF THE INVENTION In the chemical industry, chemical reactions are run in large chemical reactors in order to satisfy industrial demands for chemical products. The control of such chemical reactions requires careful planning, since the supply of reactants has to be arranged for and reaction conditions in the chemical reactors cannot be changed arbitrarily fast. Therefore, assistance for an efficient control of chemical reactions run in chemical reactors is needed. SUMMARY OF THE INVENTION

[0005] It is an object of the invention to provide assistance for an efficient control of chemical reactions run in chemical reactors. According to a first aspect, an assistance system for assisting in controlling a chemical reaction run in a chemical reactor for producing a plurality of products is provided, wherein the assistance system comprises a) a product demands providing unit configured to provide one or more future demands for each of the plurality of products, and b) a reaction model providing unit configured to provide a reaction model, the reaction model indicating, for each of a plurality of combinations of values of a plurality of reaction control parameters, a produced amount of each of the plurality of products. Furthermore, the assistance system comprises c) a reaction parameter determining unit configured to determine future reaction control parameter values based on the one or more future demands and the reaction model.

[0006] For a sufficiently efficient control of a chemical reaction resulting in a single product, just a future demand for this single product would need to be provided in combination with a corresponding reaction model. It has been found, however, that for controlling a chemical reactor used to produce a plurality of products, a more efficient control can be achieved based on one or more future demands for each of the plurality of products. In particular, it has been realised that many industrially relevant chemical reactions result in a plurality of products being produced simultaneously, wherein a respective reaction control can in that case be rendered more efficient if not only a future demand for one of the products is used, such as only a future demand for a dominant one among the products. Likewise, if the chemical reactor is used in a batch production process in which different chemical reactions are successively run in the chemical reactor, wherein each of the chemical reactions results in only a single product, an overall more efficient control of the chemical reactor can be achieved if not only a future demand for the product resulting from a single one of the chemical reactions is used as a basis.

[0007] The reaction model may be a machine learning model which has been trained using training data indicating, for each of a plurality of past combinations of values of the plurality of reaction control parameters, a past produced amount of each of the plurality products.

[0008] Hence, the model can reflect real, actually observed relations between reaction control parameters and produced product amounts. The reaction running in the chemical reactor can in this way be modelled accurately. In particular, the accuracy of the model can be controlled relatively reliably depending on a training effort taken for the model. The training effort may include the provision of the training database. Particular machine-learning models that may be used as reaction model include decision trees and neuronal networks. The reaction model may particularly be a continuous model. For instance, it may establish a relation between essentially a continuum of combinations of values of the plurality of reaction control parameters and corresponding produced amounts of the plurality of products in terms of a continuous production function. In this way, very accurate models can be built, which allow for a significantly more efficient control of chemical reaction than if, for instance, only a few discrete control points were known.

[0009] The reaction control parameters may be associated with one or more of: a type of catalyst used for the chemical reaction, an age of a catalyst used for the chemical reaction, a pressure associated with the chemical reactor, a temperature associated with the chemical reactor, an amount of one or more reactants fed into the chemical reactor and / or a ratio between two or more amounts thereof. These reaction control parameters have been found to have a significant correlation with the produced product amounts.

[0010] A pressure associated with a chemical reactor can refer, for instance, to a pressure inside the chemical reactor, a pressure before the chemical reactor and / or a pressure after the chemical reactor. Similarly, a temperature associated with a chemical reactor can refer, for instance, to a temperature inside the chemical reactor, a temperature before the chemical reactor and / or a temperature after the chemical reactor. In this regard, the terms “before” and “after” may refer to a respective measurement location along a material flow, such as a gas or fluid flow towards, into and out of the chemical reactor. A pressure and / or temperature before and / or after a chemical reactor may correspond, for instance, to a pressure and / or temperature, respectively, in a heat exchanger or a distillation tower. The heat exchanger or distillation tower may be arranged, in the direction of the material flow, before and / or after the chemical reactor.

[0011] Possible reaction control parameters that may be used also include reaction control parameters associated with an amount of one or more of the chemical products being produced that are fed into the chemical reactor again, or a ratio between two or more amounts thereof. In other words, parameters associated with one or more product recycle streams may be used as, or for determining, reaction control parameters. The one or more products being recycled may be understood as further reactants.

[0012] In general, it may be desirable to find a compromise regarding the complexity of the reaction model and its accuracy. Using more reaction control parameters may allow for a more accurate model, but may increase its complexity. On the other hand, less reaction control parameters may render the model simple, but may lead to an insufficiently accurate model. The above indicators have been found to be most significant and can hence render making the compromise easier. Also, they have been found to be sufficiently available for a wide range of reactions and / or reactors, which is particularly relevant for training reaction models of the machine-learning type.

[0013] The plurality of products can include several compounds which are structural analogues of each other. The individual produced amounts of compounds being structural analogues of each other that are produced in a same reaction can be particularly difficult to control. Hence, assistance in doing so is particularly needed in this case.

[0014] The reaction parameter determining unit can be configured to a) use the reaction model to determine, for each of a plurality of candidate future combinations of values of the plurality of reaction control parameters, a candidate future produced amount of each of the plurality of products, and to b) determine the future reaction control parameter values based on the one or more future demands and the candidate future produced amounts of the products.

[0015] Hence, when trying to infer, from given future demands for the products, consequences on how the reactor should be controlled, the reaction model may be used as a bridge between a) reaction control parameters and b) produced amounts and product demands. Since there will typically be several ways of satisfying a given set of future product demands, i.e. , several possible combinations of reaction control parameter values, an optimization may be carried out using a predefined optimization target. The optimization may be carried out, for instance, such that a deviation between the candidate future produced amounts and the future demands decreases until an optimization target associated with the reaction control parameters and optionally also the candidate produced amounts is met.

[0016] The one or more future demands can include a first future demand and a second future demand for each of the products, wherein the first future demand is a demand which is required to be satisfied and the second future demand is a demand whose satisfaction is optional.

[0017] The distinction between the first future demand and the second future demand has been realized to be relevant in the chemical industry. For instance, a first future demand could be a demand whose satisfaction is required by legal contract. However, future product demands may also need to be satisfied unconditionally if they are needed for maintaining an internal production process at a chemical plant. Non-maintenance of an internal production process may raise security concerns or increase an energy consumption at the plant, since reactors may need to be shut-down temporarily and subsequently be started again, which will typically not be energy-efficient. In the above indicated optimization, satisfaction of the first future demand may be considered as a necessary optimization condition, whereas satisfaction of the second future demand may be considered as further, but non-necessary optimization condition.

[0018] In particular, the assistance system may further comprise a price providing unit configured to provide, for each of a plurality of candidate degrees of satisfaction of the second future demand for each of the products, a price achievable by selling the respective product at an amount that satisfies the second future demand to the respective degree, wherein the reaction parameter determining unit may be configured to determine the future reaction control parameter values based further on the achievable prices.

[0019] A price achievable for selling a product can serve as a reliable and, in particular, normally well-available indicator of an actual need for the product. Distinguishing between the demand for a product and an actual need for the product can allow for a more sustainable production. Moreover, taking into account achievable prices for different degrees of satisfaction of the respective second demands can allow for a more cost-efficient production.

[0020] The assistance system can further comprise a cost providing unit configured to provide, for each of a plurality of candidate future combinations of values of the plurality of reaction control parameters and / or for each of a plurality of candidate future produced amounts of each of the plurality of products, a candidate cost which would be caused by running the chemical reactor accordingly, wherein the reaction parameter determining unit may be configured to determine the future reaction control parameter values based further on the candidate costs. Also in this way a cost-efficient operation of a chemical reactor can be realized. Particularly combining the achievable prices and the costs can allow for a high costefficiency.

[0021] The assistance system can also be suitable for controlling storage levels of storages for storing the products produced in the chemical reactor. Additionally or alternatively, the assistance system can further comprise a storage level indicator providing unit configured to provide a storage level indicator for each of the products, wherein the storage level indicators are indicative of a current and / or a future storage level of the respective product. The reaction parameter determining unit may be configured to determine the future reaction control parameter values based further on the storage level indicators. The possibility to store produced product amounts allows to temporarily decouple the production from the satisfaction of the demands to some degree. An overproduction can temporarily be tolerable if the excess produced product amounts can be stored and used for satisfying corresponding demands at a later point in time. At that later point in time, then, and underproduction can be tolerable, since the product demands may be satisfied using stored product amounts. However, storage capacities will generally be limited. Therefore, a more efficient reactor control can be achieved by taking into account storage level indicators. In fact, it has been found that this particularly increases the flexibility in the reactor control, since storage capacities can be deliberately used. It may be preferable to not always temporarily align the production with the demands.

[0022] The reaction parameter determining unit may be configured to determine the future reaction control parameter values maximizing a contribution margin associated with satisfying the second future demands for each of the products under consideration of one or more of: predefined target storage levels, required minimum storage levels, maximum storage capacities for the plurality of storages.

[0023] Hence, the above indicated optimization process can be carried out such that the contribution margin is maximized. This requirement can be used as an optimization condition, such as in terms of an optimization target. The contribution margin may particularly refer to a contribution margin 1 .

[0024] The reaction parameter determining unit may be configured to determine the future reaction control parameter values by use of an evolutionary algorithm, particularly a genetic algorithm. Such algorithms have been found to lead to good optimization results. In particular, the optimization process can be carried out relatively efficiently, while at the same time enough optimization conditions can be considered.

[0025] The plurality of products may be produced in a single continuous chemical reaction run in the chemical reactor. Hence, the plurality of products may be produced essentially simultaneously. In consequence, any choice of reaction control parameter values will have an essentially immediate effect on the produced amounts for all of the products. Controlling the reactor efficiently can therefore be rather complex and necessitates assistance.

[0026] On the other hand, the chemical reactor can be part of a batch production plant and the plurality of products may be produced sequentially in respective different chemical reactions run in the chemical reactor. In this case, reaction control parameter values need to be found for each of the chemical reactions. Also this can be a relatively complex task and therefore necessitates assistance. The sequential production of the products will typically not be perfect, meaning that the respective chemical reactions might typically be chosen so as to produce a respective one of the plurality of products to an amount that is significantly higher than for the other products. However, some amount of all products may still be produced in each of the reactions. Therefore, in principle, a different combination of optimal reaction control parameter values might need to be found for each of the chemical reactions.

[0027] According to a related aspect, an assistance method for assisting in controlling a chemical reaction run in a chemical reactor for producing a plurality of products is presented, wherein the assistance method includes a) providing one or more future demands for each of the plurality of products, and b) providing a reaction model, the reaction model indicating, for each of a plurality of combinations of values of a plurality of reaction control parameters, a produced amount of each of the plurality products. Furthermore, the method includes c) determining future reaction control parameter values based on the one or more future demands and the reaction model.

[0028] In a further aspect, the present disclosure relates to a control method for controlling a chemical reaction run in a chemical reactor for producing a plurality of products, wherein the control method includes controlling the chemical reactor according to reaction control parameter values determined as defined above.

[0029] In an aspect, the present disclosure also includes one or more products produced in the chemical reactor according to the above control method.

[0030] A further aspect of the present disclosure relates to a computer program for assisting in controlling a chemical reaction run in a chemical reactor for producing a plurality of products, wherein the program comprises program code means for causing the assistance system to carry out the assistance method.

[0031] In an aspect related to the control method, the present disclosure also includes a use of the assistance system and / or the computer program for controlling a chemical reaction run in a chemical reactor for producing a plurality of products. In particular, the reaction control parameters determined according to the assistance system and / or via the corresponding computer program may be used for controlling the chemical reactor. The present disclosure also includes one or more products produced in this way. Still another aspect of the invention relates to a method for the manufacture of a plurality of chemical products in a chemical reactor by conversion of one or more educts, comprising the steps of (i) feeding the one or more educts to the chemical reactor, (ii) converting the one or more educts in the chemical reactor using reaction control parameter values determined as defined in claim 12 or using the assistance system according to any of claims 1 to 11 . A preferred embodiment of the present invention relates to the production of ethyleneamines obtained by the conversion of monoethanolamine (MEOA) with ammonia (NH3) in the presence of hydrogen and an amination catalyst. The reaction of monoethylene glycol (MEG) with ammonia can be affected in the liquid phase or the gas phase. Gas phase reactions are disclosed, for example, in CN 102190588 A and CN 102233272 A.

[0032] The reaction of MEOA and ammonia is described, for example, in US 2,861 ,995 A, DE 1 172 268 B and US 3,112,318 A. An overview of the various process variants of the reaction of MEOA with ammonia can be found, for example, in the PERP Report No. 138 “Alkyl- Amines”, SRI International, 03 / 1981 (especially pages 81-99, 117). The conversion of MEOA with NH3 usually yields a plurality of chemical products, such as ethyleneamines and ethanolamines. Examples for ethyleneamines obtained by the conversion of MEOA with NH3 include ethylene diamine (EDA), diethylene triamine (DETA), triethylene tetramine (TETA), tetraethylene pentaamine (TEPA), piperazine (PIP) and aminoethyl piperazine (AEP) and higher ethyleneamines. Preferred alkanolamines are aminoethylethanolamine (AEEA) and hydroxyethyl piperazine (HEP). As specified below, the assistance system, the assistance method and / or the control method of the present invention can be used to control the manufacture of the ethyleneamines and ethanolamines to optimally meet the demands. The chemical products manufactured according to the method of the present invention can be further converted in one or more steps to obtain a further chemical product or chemical material. In further preferred embodiment, monoethylene glycol (MEG) can be converted with NH3 in the presence of hydrogen and an amination catalyst. Preferably the conversion of MEG with ammonia is conducted in the liquid phase according to US 4,111 ,840 A, US 3,137,730 A, DE 1 172 268 B, WO 2007 / 093514 A1 , WO 2007 / 093552 A1 , WO 2018 / 224316 A1 , WO 2018 / 224315 A1 , WO 2018 / 224321 A1 and WO 2020 / 17085 A1.

[0033] Accordingly, still another aspect of the present invention relates to the manufacture of a chemical product or material wherein at least one of the plurality of chemical products is further converted in one or more steps to obtain a further chemical product or chemical material. Ethyleneamines and ethanolamines can be used as a starting material for polyamides, herbicides or herbicide intermediates, pharmaceuticals or pharmaceutical intermediates, resins or coating, chelating agents, adhesives, personal care products, lubricants, textile resins, paper auxiliaries, fuel additives, corrosion inhibitors, surfactants, and many other products.

[0034] The manufacturing methods described above are not exclusive and merely illustrative of the large number of manufacturing processes producing a plurality of chemical products which can be produced according to the present invention. Instead of producing a plurality of products jointly as co-products in a single reactor or a series of reactors, the plurality of products can also be produced sequentially in a single reactor or series of reactors in so- called multi-product plants.

[0035] It shall be understood that the aspects described above, specifically the system of claim 1 and the method of claims 12 and 13 have similar and / or identical preferred embodiments, in particular as defined in the dependent claims.

[0036] It shall be understood that a preferred embodiment of the invention can also be any combination of the dependent claims or above embodiments with the respective independent claim.

[0037] These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter.

[0038] BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Fig. 1 shows schematically and exemplarily an assistance system for assisting in controlling a chemical reaction run in a chemical reactor for producing a plurality of products,

[0040] Fig. 2 shows schematically and exemplarily the chemical reactor, products produced in the chemical reaction run therein, corresponding educts and further reaction control parameters,

[0041] Fig. 3 shows, split in three parts 3A to 3C, exemplarily the structure of a decision tree usable as reaction model,

[0042] Fig. 4 shows exemplarily an evolution of a loss function in the course of training a reaction model, Fig. 5 shows exemplarily product amounts estimated using the trained reaction model,

[0043] Fig. 6 shows exemplarily future reaction control parameter values determined based on one or more future demands and a reaction model,

[0044] Fig. 7 shows schematically and exemplarily a distribution of different kinds of future product demands and corresponding achievable prices,

[0045] Fig. 8 shows schematically and exemplarily an assistance method for assisting in controlling a chemical reaction run in a chemical reactor for producing a plurality of products, and

[0046] Fig. 9 shows schematically and exemplarily a display output corresponding to three different sets of future reaction control parameter values determined for a same reaction model and same product demands.

[0047] DETAILED DESCRIPTION OF EMBODIMENTS

[0048] Fig. 1 shows schematically and exemplarily an assistance system 100 for assisting in controlling a chemical reaction run in a chemical reactor 10 for producing a plurality of products 20 as shown schematically and exemplarily in Fig. 2. The assistance system 100 comprises a product demands providing unit 101 configured to provide one or more future demands 40a, 40b, 50 for each of the plurality of products 20 as shown schematically and exemplarily in Fig. 7. Furthermore, the assistance system 100 comprises a reaction model providing unit 102 configured to provide a reaction model, the reaction model indicating, for each of a plurality of combinations of values of a plurality of reaction control parameters 30a, 30b as shown schematically and exemplarily in Fig. 2, a produced amount of each of the plurality of products 20. Moreover, the assistance system 100 comprises a reaction parameter determining unit 103 configured to determine future reaction control parameter values based on the one or more future demands 40a, 40b, 50 and the reaction model. Besides, in the illustrated embodiment, the assistance system 100 comprises a price providing unit 104, a cost providing unit 105 and a storage level indicator providing unit 106, which will be explained in more detail further below.

[0049] As indicated in Fig. 2, for the sake of illustration, it is exemplarily assumed herein that the products 20 produced in the chemical reactor 10 refer to ethyleneamines (EEAs, or EAs), which may mainly comprise ethylenediamine (EDA), but besides that also diethylenetriamine (DETA), triethylenetriamine (TETA), piperazine (PIP), aminoethylpiperazine (AEPIP), aminoethylethanolamine (AEEA), hydroxyethylpeperazine (HEP) and possibly further homologues, derivates and / or conversion products. Hence, a plurality of compounds are being produced that are structural analogues to each other. The educts used for the chemical reaction running in the chemical reactor 10 are exemplarily assumed to refer to ammonia (NH3), monoethanolamine (MEOA) and hydrogen (H2). To be precise, EEAs may therefore be produced in this exemplary case by conversion of MEOA with NH3 in the presence of H2 yielding various ethyleneamines, such as ethylene diamine (EDA), diethylene triamine (DETA), triethylene tetramine (TETA), tetraethylene pentaamine (TEPA), piperazine (PIP), aminoethyl piperazine (AEP) and higher ethyleneamines, and alkanolamines, such as aminoethylethanolamine (AEEA) and hydroxyethyl piperazine (HEP). Besides the educt amounts supplied into the chemical reactor 10, the chemical reaction is in this case also characterized by a pressure (P) and a temperature (T) in the chemical reactor 10. Furthermore, an amination catalyst (C) may be used in the reaction, which will also influence the chemical reaction 10. Parameters associated with the educts, particularly their supplied amounts or a ratio thereof, are viewed in this example as forming a first set of reaction control parameters 30b. Furthermore, parameters associated with the pressure, the temperature and the catalyst, particularly a type, an amount and / or age of the catalyst in the reactor 10 or an amount of catalyst supplied to the reactor 10, are viewed in this example as forming a second set of reaction control parameters 30a.

[0050] In the illustrated case, the parameters 30b include parameters not being associated with the original educts, but with one or more of the products 20 that are being fed back into the reactor 10. Feedbacks of products like EDA, AEEA and MEOA, which may be realized by one or more product recycle streams 31 as indicated in Fig. 2, may correspond to a reuse of one or more of the products as educts for the chemical reaction running in the chemical reactor 10. Crude or purified products can be fed again into the chemical reactor 10 together with raw materials like NH3, MEOA and H2. This can be beneficial for a temperature control of an overall process of which the chemical reaction running in the chemical reactor 10 is part of, due to the relation between the amount of starting material and the exothermic degree of the reaction. Moreover, via product recycle streams, further changes in the selectivity of a multi-product reactions can be effected, such as when products can react back or if multiple consecutive reactions are run in the chemical reactor 10. For instance, EDA can be converted to DETA upon addition of another equivalent of MEOA. Similarly, AEEA can be converted to DETA by reaction with ammonia. Thus, products being recycled can be viewed as intermediates in the production of other products. As a consequence, the overall product ratios may be influenced. The reaction model provided by the reaction model providing unit can particularly be a machine learning model which has been trained using training data indicating, for each of a plurality of past combinations of values of the plurality of reaction control parameters 30a, 30b, a past produced amount of each of the plurality of products 20. Good results have been achieved when using neural networks and decision trees, which will also be regarded as regression trees, as machine learning model. Machine learning models like these, which are suitable for providing several outputs instead of only a single output, may be preferred, since in this way a single model may be used to estimate the produced amounts for all of the products 20, which allows the model to take into account dependencies between the products 20. However, in principle, the plurality of produced product amounts could also be modeled by outputs of, for instance, an individual model for each of the products 20.

[0051] Figs. 3A to 3C show schematically and exemplarily how a simplified version of a decision tree that could be used as reaction model can be set up from a training data set. In this simplified example, only MEOA and NH3 are considered as educts, and the respective feeds, i.e. , the amounts of MEOA and NH3, respectively, supplied into the chemical reactor, are considered as reaction control parameters. Moreover, six products corresponding to different EEAs or EEA mixtures are considered. Each data point in the training data set, i.e., each sample, comprises a) a past value for each of the reaction control parameters considered, wherein these values could also be referred to as the training input data part of the sample, and b) corresponding past produced amounts for each of the products considered, wherein these produced amounts could also be referred to as the training output data part of the sample.

[0052] In particular, Figs. 3A to 3C show how the training data set splits into subsets depending on a segmentation into sub-ranges of the reaction control parameter values. While the first box on the top or origin of the tree shown in Fig. 3A still corresponds to the whole training data set, the boxes below, i.e. further into the tree structure, each correspond to a different subset thereof. As indicated in the boxes, the whole training data set as well as its subsets are characterized in terms of values corresponding to an average produced amount for each of the products across the respective (sub-)set and a corresponding squared error, which may be viewed as a measure for a combined average deviation of the produced product amounts from their averages. The values for the educt feeds and the produced product amounts indicated in the boxes of Figs. 3A to 3C are given in units of tons per day.

[0053] The segmentation of the training dataset into subsets depends on the criteria indicated in the top row of each of the boxes illustrated from which arrows emerge to a respective subsequent level of the tree, which may also be regarded as branching points. As indicated, the boxes on the second level of the tree correspond to the two training data subsets consisting of a) samples with an amount of MEOA fed into the reactor of below or equal to 172.817 tons per day and b) samples with an amount of MEOA fed into the reactor of above 172.817 tons per day, respectively. Furthermore, the boxes on the third level of the illustrated tree arise from the two subsets a) and b) by segmenting the subset a) further into c) samples with an amount of MEOA fed into the reactor of below or equal to 171.922 tons per day and d) samples with an amount of NH3 fed into the reactor of above 171 .922 tons per day (the two boxes on the left of the third level, shown as upper boxes in Fig. 3B), and by segmenting the subset b) further into c’) samples with an amount of NH3 fed into the reactor of below or equal to 270.192 tons per day and d’) samples with an amount of NH3 fed into the reactor of above 270.192 tons per day (the two boxes on the right of the third level, shown as upper boxes in Fig. 3C), respectively, thereby intermediately arriving at four subsets c), d), c’), d’) of the whole training data set. A further division of the subsets c), d) and c’) based on an MEOA feed threshold value of 170.701 tons per day for subset c), a conversion threshold value of 2.054 for subset d) and an NH3 feed threshold value of 269.452 tons per day leads to a final number of seven subsets of the whole training data set, with could be regarded as forming the leaves of the tree. The conversion threshold may be a threshold of a conversion indicator, a conversion indicator being generally indicative of a respective amount of an educt like MEOA or NH3 being converted into one or more products.

[0054] Thus, for instance, the training data set as a whole consists of 537 samples, wherein, among these 537 samples, the average produced amount of the first product is approximately 126.4 tons per day, the average produced amount of second product is approximately 43.9 tons per day, etc. On the other hand, for instance, the smallest training data subset (leftmost leaf of the illustrated tree, shown in the bottom left corner in Fig. 3B), corresponding to an amount of MEOA fed into the reactor of below or equal to 170.701 tons per day, comprises 3 samples, among which the average produced amount of the first product is approximately 121 .8 tons per day, the average produced amount of the second product is approximately 40.6 tons per day, etc.

[0055] The criteria in the top row of each of the boxes illustrated from which arrows emerge to a respective subsequent level of the tree correspond to the decision rules to be learned by the tree during training. Since they split the training data set into subsets, they and / or the corresponding boxes could also be regarded as branching points of the tree, as already indicated above. Assuming fixed decision rules, such as after training has been finished, a given combination of reaction control parameter values, i.e. the input data of any single further sample, can be fed into the tree, first branching point or node, whereupon the tree determines, by virtue of the learned decision rules, into which sub-range the input falls and hence which output data can be associated with the input in terms of produced product amounts. The bottom boxes in Figs. 3B and 3C, which may be viewed as the leaves of the decision tree, insofar correspond to the predictions that can be generated by the tree. For instance, assuming the decision rules indicated in the illustrated tree, for an MEOA feed below 170.701 tons per day, an approximate produced amount 121.8 tons per day would be predicted for the first product.

[0056] It should be emphasized again that the simplified decision tree illustrated by Figs. 3A to 3C only considers supplied amounts of the two educts MEOA and NH3 as reaction control parameters, whereas more generally the reaction model used might also consider reaction control parameters associated with a molar ratio between the supplied educts, feedbacks of one or more of the products or a kind of compressor selected for the reaction, for instance.

[0057] Furthermore, it should be emphasized that the accuracy of a trained machine learning model such as a decision tree or also a neural network will at least partly be dependent on the training data set, i.e., a quality thereof. It may therefore preferred that the training data set used for training the machine learning model, i.e., the past reaction control parameter values and the corresponding past produced amounts for the plurality of products used for training, are selected based on a quality assessment thereof. The quality assessment may involve, for instance, an assessment of whether the training data stems from a normal reaction state and / or a reactor state. In particular, data from times where a reactor was turned on or turned off may be excluded from the training data. Such times may be identified as times where the collected data was not steady to a predefined degree.

[0058] Fig. 4 shows exemplarily how a prediction accuracy of a machine learning-type reaction model such as a decision tree or neural network evolves in the course of its training. As shown, the training can proceed in several iterations, wherein in each of the iterations, a deviation between the produced product amounts predicted by the model and the produced product amounts indicated by the training data can be captured in terms of a loss function, and wherein this loss function may be used to adapt the model, preferably such that the loss function decreases to below a predefined limit. In particular, a class of models of a same type may be considered, wherein the models in the class differ in hyperparameters. For instance, decision trees may be understood as forming a class of models differing from each other in terms of hyperparameters such as a number of branching points. Likewise, neural networks may be understood as forming a class of models differing from each other in terms of hyperparameters such as a number of layers. The model training may not only include an optimization of a specific member of a class of models, such as by adapting the sub-ranges of reaction control parameters indicated by the decision rules of a decision tree, but also an optimization of the hyperparameters. In the graph of Fig. 4, it can be seen how a value of the loss function, which in this case is a maximum absolute percentage error (MAPPE) loss function, decreases for each of six products relatively rapidly within less than the first 100 training iterations and then remains approximately constant for the rest of the training.

[0059] Fig. 5 shows exemplary predictions of a reaction model that has already been trained, i.e., whose training has been finished. For each of six products being produced in the modeled reactor, referred to as EEA 1 to EEA 5, a respective produced amount as determined by the model is indicated in dependence on an amount of one of the educts - in this case MEOA - being supplied into the reactor, wherein other reaction control parameters are held constant. On the vertical axis, the respective produced amount can be read off in terms of a normalized value, wherein the normalization has been carried out with respect to a maximum produced amount of the respective product. On the horizontal axis, the supplied MEOA amount can be read off in terms of a normalized value as well, wherein the normalization has been carried out with respect to a maximum amount of supplied MEOA. It will be appreciated from the shown graph that already the relation between a single reaction control parameter and the plurality of produced product amounts can be highly complex. For instance, it cannot even be concluded that all produced product amounts will generally increase with increasing amounts of MEOA supplied. In contrast, in the illustrated case, the produced amount of EEA 2 decreases with the supplied MEOA amount for high values of the latter. Furthermore, the produced amount of EEA 6 first decreases up to a supplied MEOA amount of approximately 50% of the maximum and thereafter increases again to about its initial value. Such complicated relations are practically impossible to properly consider for a human aiming for an efficient control of a chemical reactor, even for an expert.

[0060] Once a reaction model has been established, such as by training a machine learning model as indicated above, and hence a relation between reaction control parameter values and corresponding produced product amounts that can be expected for the respective reaction control parameter values is provided, i.e., once this relation can be assumed to be known, this knowledge can be used to adapt the reaction control parameter values according to which the respective chemical reactor is operated to a given product demands. This adaptation of the control parameters values is, however, preferably subject to a further optimization process. In order to optimize the control parameter adaptation, the reaction parameter determining unit 103 is, in the embodiment described in detail herein, configured to use the reaction model to determine, for each of a plurality of candidate future combinations of values of the plurality of reaction control parameters 30a, 30b, a candidate future produced amount of each of the plurality of products 20. Hence, a plurality of possible choices for a future operation of the chemical reactor is considered, and for each of these choices an expected outcome in terms of produced product amounts is determined. The future reaction control parameter values are then determined based on the one or more future demands 40a, 40b and the candidate future produced amounts of the plurality of products 20.

[0061] Fig. 6 shows exemplary optimization results for two different months, wherein the optimization results are tabulated on a per week basis. The considered reaction control parameters shown in Fig. 6 include in this case the amount of MEOA fed into the chemical reactor and the molar ratio between the amount of NH3 and the amount of MEOA fed into the chemical reactor. Furthermore, the respective produced amounts of two products - in this exemplary case EDA and AEEA - being recycled, i.e., fed again into the chemical reactor 10, are used as process control parameters. Besides, the tables shown in Fig. 6 include respective conversion rates, which refer in this case to the molecular share of MEOA being converted. For instance, a conversion rate of 50% would mean that half of the MEOA fed into the reactor 10 has reacted to any of the possible products, without being specific to which of the products are being formed.

[0062] Fig. 7 shows schematically and exemplarily characteristics of a combined future demand decomposing into three parts 40a, 40b, 50. A first future demand is formed from a first part 40a and a second part 40b, and a second future demand is formed from a third part 50. The graph shown in Fig. 7 relates to a future demand for a single one of the products 20, while corresponding graphs could be presented for each of the other products 20. The first future demand 40a, 40b and the second future demand 50 differ from each other in that the first future demand 40a, 40b is required to be satisfied and the second future demand 50 is a demand whose satisfaction is optional. For instance, the demand 40a could refer to an amount of a respective product that is needed internally, i.e. for further processes at a production plant at which also the reactor to be controlled is located. Furthermore, the demand 40b could refer to an amount of the product for which external supply contracts exist. It may be desired that both internal product demands and demands required by contract with external parties are treated such that they are required to be satisfied unconditionally. On the other hand, the demand 50 could refer to a potential demand from further buyers of the product, to which the product may be sold via the spot market, i.e., on a relatively short- term basis. In that case, since there may exist different spot markets, the demand 50 may be decomposed further into respective market-specific demands.

[0063] While the horizontal axis of the graph shown in Fig. 7 indicates a size or extent of the respective product demands 40a, 40b, 50 for a given future time interval, which could be given in terms of tons, for instance, the vertical axis of the graph indicates a price 60 achievable by satisfying the respective demand 40a, 40b, 50 for a given degree of satisfaction. A graph like the one shown in Fig. 7 could be provided for different degrees of satisfaction of the demands 40a, 40b, 50. However, the first future demands 40a, 40b are preferably being fully satisfied. Therefore, in particular, the price providing unit 104 may be configured to provide, for each of a plurality of candidate degrees of satisfaction of only the second future demand 50 for each of the products 20, a price 60 achievable by selling the respective product at an amount that satisfies the second demand 50 to the respective degree. The reaction parameter determining unit 103 is then preferably configured to determine the future reaction control parameter values based further on the achievable prices 60.

[0064] Furthermore, the cost providing unit 105 may be configured to provide, for each of a plurality of candidate future combinations of values of the plurality of reaction control parameters 30a, 30b and / or for each of a plurality of candidate future produced amounts of each of the plurality of products 20, a candidate cost which would be caused by running the respective chemical reactor accordingly. The reaction parameter determining unit 103 would then preferably be configured to determine the future reaction control parameter values based further on the candidate costs.

[0065] Moreover, the storage level indicator providing unit 106 may be configured to provide a storage level indicator for each of the products 20, wherein the storage level indicators are indicative of a current and / or a future storage level of the respective product. The reaction parameter determining unit 103 may then be configured to determine the future reaction control parameter values based further on the storage level indicators.

[0066] Hence, for instance, the future reaction control parameter values may be determined by balancing prices achievable by satisfying the second future demands 50 against the costs for doing so, wherein furthermore storage levels for the respective products 20 may be considered in order to determine, for instance, whether it is possible and may also be more cost-efficient to satisfy the second future demands 50 from the storages. If the future reaction control parameter values are determined in consideration of storage level indicators, it may be preferred that the assistance system 100 is also suitable for controlling storage levels of storages for storing the products 20 produced in the chemical reactor 10. For instance, apart from the future reaction control parameter values, also storage control parameters based on which the storage levels of the storages can be controlled may be determined by the assistance of the system 100.

[0067] In particular, the reaction parameter determining unit 103 may be configured to determine the future reaction control parameter values maximizing a contribution margin associated with satisfying the second future demands 50 for each of the products 20 under consideration of one or more of: predefined target storage levels, required minimum storage levels, maximum storage capacities for the plurality of storages. Hence, for instance, the future reaction control parameter values may be determined such that the contribution margin associated with satisfying the second future demands 50 for each of the products 20 is maximized under the constraint that one or more predefined target storage levels are maintained up to a predefined tolerance, or under the constraint that one or more storage levels should not fall below a predefined minimum or cannot possibly rise above a maximum storage capacity.

[0068] In principle, the reaction parameter determining unit 103 may use any of a wide range of possible optimization algorithms to determine the future reaction control parameter values using the reaction model and in dependence on one or more of the future demands, the achievable prices, the costs and the storage levels. For instance, a genetic algorithm may be used. Conditions and / or targets like a maximized contribution margin or storage levels may be considered in such algorithms in terms of constraints to be satisfied by the solutions, i.e. the future reaction control parameters values to be determined.

[0069] Notably, the type of process in which the plurality of products 20 are produced is not particularly limited. For instance, the plurality of products 20 may be produced in a single continuous chemical reaction run in the chemical reactor 10. On the other hand, it is also possible that the chemical reactor 10 is part of a batch production plant and the plurality of products 20 are produced sequentially a respective different chemical reactions run in the chemical reactor 10. These two cases could also be described as a simultaneous production of the plurality of products 20 on the one hand, and a subsequent production of the plurality of products 20 on the other hand. While, in the former case, the determined future reaction control parameters may relate to control parameters for the respective single reaction, in the latter case, the determined future reaction control parameters may correspond to several sets of control parameters of which each set relates to a different one of several subsequent reactions.

[0070] Fig. 8 shows schematically and exemplarily an assistance method 200 for assisting in controlling a chemical reaction run in a chemical reactor 10 for producing a plurality of products 20. The assistance method 200 includes a step 201 of providing one or more future demands 40a, 40b, 50 each of the plurality of products 20, and a step of providing 202 a reaction model, wherein the reaction model indicates, for each of a plurality of combinations of values of a plurality of reaction control parameters 30a, 30b, a produced amount of each of the plurality of products 20. Furthermore, the method 200 includes a step 203 of determining future reaction control parameter values based on the one or more future demands 40a, 40b, 50 and the reaction model.

[0071] In this particular embodiment of the method 200, in accordance with the above described system 100, the future reaction control parameter values are determined further based on prices, costs and storage levels. Accordingly, the method 200 further includes a step 204 of providing, for each of a plurality of candidate degrees of satisfaction of a second future demand 50 for each of the products 20, wherein the second future demand 50 is a demand whose satisfaction is optional, a price 60 achievable by selling the respective product at an amount that satisfies the second future demand 50 to the respective degree. Furthermore, in a step 205 of the method 200, for each of a plurality of candidate future combinations of values of the plurality of reaction control parameters 30a, 30b and / or for each of a plurality of candidate future produced amounts of each of the plurality of products 20, a candidate cost is provided which would be caused by running the chemical reactor 10 accordingly. Moreover, the method 200 includes a step 206 of providing a storage level indicator for each of the products 20, wherein the storage level indicators are indicative of a current and / or a future storage level of the respective product. In step 203, the future reaction control parameters are then determined based further on the provided prices, costs and storage level indicators from steps 204, 205 and 206.

[0072] In a control method for controlling a chemical reaction run in a chemical reactor 10 for producing a plurality of products in the 20, the chemical reactor 10 could then be controlled using reaction control parameter values determined in accordance with the method 200. An exemplarily control method of this kind could be schematically illustrated by Fig. 8 as well, wherein as additional step a step of controlling the chemical reactor using the determined future reaction control parameters could be included. Fig. 9 shows schematically and exemplarily a display output generated by a computer program implementing the method 200. The display output could be part of a user interface of the computer program, for instance.

[0073] In particular, Fig. 9 illustrates three different optimization results, corresponding to different optimization parameters that might be chosen by a user. For instance, as shown in Fig. 5, the user may be able to choose between a) an optimization focused on a high contribution margin (“CM1-Focus”), b) an optimization focused on high storage levels (“Inventory-Focus”) and c) an optimization balanced between the two extremes a) and b) (“Balanced”).

[0074] In an exemplary embodiment, for determining the future reaction control parameter values based on the one or more future demands 40a, 40b, 50 and the reaction model, a nondominant front of candidate solutions in the form of candidate future combinations of values of the plurality of reaction control parameters 30a, 30b is generated using the non-dominant sorting genetic algorithm NSGA-II. Out of these candidate solutions, that candidate solution is selected which corresponds to the 50% percentile of the values of an inventory target function indicating target storage levels for the plurality of products 20.

[0075] Hence, from a plurality of candidate solutions, which may correspond to a pareto front of solutions found using NSGA-II, each solution may be evaluated using a target function, wherein the target function may, for instance, indicate target storage levels for the plurality of products 20. As a result, a plurality of target function values corresponding to respective candidate solutions are determined, wherein the candidate solution corresponding to the median among the plurality of determined target function values is selected.

[0076] Based on this solution, an optimization of the inventory target function and a contribution margin target function indicating a target contribution margin is carried out using a genetic algorithm with further auxiliary constraints. For optimizing the contribution margin, it is imposed as auxiliary constraint that the storage levels for the plurality of products number 20 should never be below a minimum inventory target function value following from the nondominant front of candidate solutions arising from the NSGA-II optimization. An analogous auxiliary constraint may be imposed regarding the value of the contribution margin during inventory optimization, i.e., when optimizing for the storage levels. The results following for contribution margin optimization and inventory optimization, respectively, are shown on the left and on the right in Fig. 9. In the middle, a "balanced" solution is shown, which in this case corresponds to the 50% percentile. However, also any other result of the solutions following from the NSGA-II optimization could be considered as a "balanced" solution. As seen in Fig. 9, the future reaction control parameters may be determined for the respective next six months, for instance, such that also corresponding produced product amounts can be determined for the respective next six months using the reaction model in order to facilitate planning based thereon. In particular, as was indicated in Fig. 6, a separate set of reaction control parameters may be determined for each week in the six months. Moreover, such a determination may be repeated every week, such that a new production plan spanning the respective next six months in the form of weekly intervals may be generated on a weekly basis. It will of course be understood that these exemplary time windows are just practically convenient choices, and could in principle be chosen in any other way.

[0077] Moreover, while the above detailed disclosure mainly related to the production of EEAs, in principle any other chemical production process could be controlled similarly. Besides, as already indicated, other reaction models than the ones mentioned above may be used, and also other optimization algorithms. In particular, the disclosure is not limited to decision trees, neural networks and genetic optimization algorithms.

[0078] Other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention, from a study of the drawings, the disclosure, and the appended claims.

[0079] In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite article "a" or "an" does not exclude a plurality.

[0080] A single unit or device may fulfill the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.

[0081] Procedures like the providing of future demands for the products, the providing of the reaction model, the providing the achievable prices, of the candidate costs and of the storage level indicators, the determining of the future reaction control parameter values, etc., performed by one or several units or devices, can be performed by any other number of units or devices. These procedures can be implemented as program code means of a computer program and / or as dedicated hardware.

[0082] A computer program product may be stored / distributed on a suitable medium, such as an optical storage medium or a solid-state medium, supplied together with or as part of other hardware, but may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems.

[0083] Any reference signs in the claims should not be construed as limiting the scope.

[0084] An assistance system for assisting in controlling a chemical reaction run in a chemical re- actor for producing a plurality of products is presented. The system comprises a) a product demands providing unit configured to provide one or more future demands for each of the products, b) a reaction model providing unit configured to provide a reaction model, the reaction model indicating, for each of a plurality of combinations of values of a plurality of reaction control parameters, a produced amount of each of the products, and c) a reaction parameter determining unit configured to determine future reaction control parameter values based on the one or more future demands and the reaction model. This allows for a more efficient control of chemical reactions run in chemical reactors.

Claims

Claims:

1. An assistance system (100) for assisting in controlling a chemical reaction run in a chemical reactor (10) for producing a plurality of products (20), wherein the assistance system (100) comprises: a product demands providing unit (101) configured to provide one or more future demands (40a, 40b, 50) for each of the plurality of products (20), a reaction model providing unit (102) configured to provide a reaction model, the reaction model indicating, for each of a plurality of combinations of values of a plurality of reaction control parameters (30a, 30b), a produced amount of each of the plurality of products (20), a reaction parameter determining unit (103) configured to determine future reaction control parameter values based on the one or more future demands (40a, 40b, 50) and the reaction model.

2. The assistance system (100) as defined in claim 1 , wherein the reaction model is a machine learning model which has been trained using training data indicating, for each of a plurality of past combinations of values of the plurality of reaction control parameters (30a, 30b), a past produced amount of each of the plurality products (20).

3. The assistance system (100) as defined in any of the preceding claims, wherein the reaction control parameters (30a, 30b) are associated with one or more of: a type of catalyst used for the chemical reaction, an age of a catalyst used for the chemical reaction, a pressure in the chemical reactor (10), a temperature in the chemical reactor (10), an amount of one or more reactants fed into the chemical reactor (10) and / or a ratio between two or more amounts thereof.

4. The assistance system (100) as defined in any of the preceding claims, wherein the plurality of products (20) include several compounds which are structural analogues of each other.

5. The assistance system (100) as defined in any of the preceding claims, wherein the reaction parameter determining unit (103) is configured to a) use the reaction modelto determine, for each of a plurality of candidate future combinations of values of the plurality of reaction control parameters (30a, 30b), a candidate future produced amount of each of the plurality of products (20), and to b) determine the future reaction control parameter values based on the one or more future demands (40a, 40b) and the candidate future produced amounts of the products (20).

6. The assistance system (100) as defined in any of the preceding claims, wherein the one or more future demands (40a, 40b, 50) include a first future demand (40a, 40b) and a second future demand (50) for each of the products (20), wherein the first future demand (40a, 40b) is a demand which is required to be satisfied and the second future demand (50) is a demand whose satisfaction is optional.

7. The assistance system (100) as defined in claim 6, further comprising: a price providing unit (104) configured to provide, for each of a plurality of candidate degrees of satisfaction of the second future demand (50) for each of the products (20), a price (60) achievable by selling the respective product at an amount that satisfies the second future demand (50) to the respective degree, wherein the reaction parameter determining unit (103) is configured to determine the future reaction control parameter values based further on the achievable prices (60).

8. The assistance system (100) as defined in any of the preceding claims, further comprising: a cost providing unit (105) configured to provide, for each of a plurality of candidate future combinations of values of the plurality of reaction control parameters (30a, 30b) and / or for each of a plurality of candidate future produced amounts of each of the plurality of products (20), a candidate cost which would be caused by running the chemical reactor (10) accordingly, wherein the reaction parameter determining unit (103) is configured to determine the future reaction control parameter values based further on the candidate costs.

9. The assistance system (100) as defined in any of the preceding claims, further being suitable for controlling storage levels of storages for storing the products (20) produced in the chemical reactor (10), wherein the assistance system (100) further comprises: a storage level indicator providing unit (106) configured to provide a storage level indicator for each of the products (20), wherein the storage level indicators are indicative of a current and / or a future storage level of the respective product, wherein the reaction parameter determining unit (103) is configured to determine the future reaction control parameter values based further on the storage level indicators.

10. The assistance system (100) as defined in claims 6 to 9, wherein the reaction parameter determining unit (103) is configured to determine the future reaction control parameter values maximizing a contribution margin associated with satisfying the second future demands (50) for each of the products (20) under consideration of one or more of: predefined target storage levels, required minimum storage levels, maximum storage capacities for the plurality of storages.11 . The assistance system (100) as defined in any of claims 1 to 10, wherein the chemical reactor (10) is part of a batch production plant and the plurality of products (20) are produced sequentially in respective different chemical reactions run in the chemical reactor (10).

12. An assistance method (200) for assisting in controlling a chemical reaction run in a chemical reactor (10) for producing a plurality of products (20), wherein the assistance method (200) includes: providing (201) one or more future demands (40a, 40b, 50) for each of the plurality of products (20), providing (202) a reaction model, the reaction model indicating, for each of a plurality of combinations of values of a plurality of reaction control parameters (30a, 30b), a produced amount of each of the plurality of products (20),determining (203) future reaction control parameter values based on the one or more future demands (40a, 40b, 50) and the reaction model.

13. A method for the manufacture of a plurality of chemical products in a chemical reactor (10) by conversion of one or more educts, comprising the steps of (i) feeding the one or more educts to the chemical reactor (10), (ii) converting the one or more educts in the chemical reactor (10) using reaction control parameter values (30a, 30b) determined according to claim 12 or using the assistance system according to any of claims 1 to 11 .

14. A method according to claim 13, where the plurality of products (20) are one or more ethyleneamines and / or one or more ethanolamines and the one or more educts are ammonia and an alcohol selected from the group consisting of ethanolamine and ethylene glycol.

15. A method according to claim 13 or 14, wherein at least one of the plurality of products produced in the method according to claim 13 or 14 is further converted in one or more steps to obtain a further chemical product or chemical material.

Images