System for determining individual product quantity for plurality of chemical reactors

Abstract

The present invention relates to a system (100) for determining individual product quantities of a plurality of chemical reactors (11, 12,..., 1K) that contribute to a combined product quantity. The system comprises a measured value providing unit (101) which provides each reactor with a measured individual reactant amount, and an artificial intelligence providing unit (102) which provides each reactor with artificial intelligence (21, 22,..., 2K), the artificial intelligence is trained to provide, as an output, individual product quantities of the reactors when individual reactant quantities of the reactors are received as an input, the individual product quantities combining into a combined product quantity associated with the individual reactant quantities received as an input. The system further comprises a single product quantity determination unit (103) that determines a single product quantity for the reactors based on the measured single reactant quantities and trained artificial intelligence. The system enables improved chemical production efficiency when multiple reactors facilitate combined product quantities.

Description

Technical Field

[0001] The present invention relates to a system, method and computer program for determining the individual product amounts of a plurality of chemical reactors that contribute to a combined product amount. Background Art

[0002] In a chemical production process, it is often necessary to carry out reactions simultaneously in more than one chemical reactor. However, usually only a single product output can be provided, that is, a single product amount composed of the product amounts produced by individual reactors. For example, the conduits from individual reactors can converge into a common supply conduit, where the product can only be obtained via the common supply conduit. In this case, if the individual product amounts of multiple reactors cannot be measured, it may be difficult to control these individual reactors. However, sub-optimal control of the reactors may lead to inefficiencies in the production process. Summary of the Invention

[0003] The object of the present invention is to be able to improve the efficiency of a chemical production process in which a plurality of chemical reactors contribute to a combined product amount.

[0004] In a first aspect of the present invention, there is provided a system for determining the individual product amounts of a plurality of chemical reactors that contribute to a combined product amount, wherein the system comprises:

[0005] - a measurement value providing unit configured to provide a measured individual reactant amount for each of the plurality of chemical reactors,

[0006] - an artificial intelligence providing unit configured to provide a trained artificial intelligence for each of the chemical reactors, wherein the provided trained artificial intelligence is trained to provide the individual product amounts of the chemical reactors as outputs when receiving the individual reactant amounts of the chemical reactors as inputs, and these individual product amounts combine into a combined product amount associated with the individual reactant amounts received as inputs, and

[0007] - an individual product amount determining unit configured to determine the individual product amounts of the plurality of chemical reactors based on the measured individual reactant amounts and the trained artificial intelligences.

[0008] Thus, multiple artificial intelligences are used to individually model the chemical reactors, wherein the artificial intelligence is trained such that the individual product amounts provided as outputs combine into a combined product amount associated with the individual reactant amounts received as inputs. It has been found that this allows for the accurate determination of the individual product amounts of multiple reactors based on the corresponding individual reactant amounts of the multiple reactors, and the determined individual product amounts can be used for the optimized control of the reactors, thereby achieving an increase in the efficiency of the corresponding production process.

[0009] The determined single product amount can replace the corresponding measured value. If only combined product amounts are accessible, such measurement of single product amounts may not be possible. And when such measurement is possible, these measurements may no longer be needed when using the provided system.

[0010] Even if all reactors are identical and the supplied combined product amount is known, it can be challenging to control a chemical production process in multiple chemical reactors that contribute to a combined product amount without being able to measure the single product amounts of the reactors, because the amounts of reactants supplied to individual reactors and other process parameters selected to control the reactions running in individual reactors may differ from each other. For practical reasons, the amounts of reactants supplied to individual reactors and other process parameters of multiple reactors can be intentionally selected differently. However, even when the same control is desired, this may only be achievable with limited accuracy, where even relatively small differences in single reactant amounts and other process parameters can have a relatively large impact on the single product amounts produced by individual reactors. The accurate determination of the single product amounts of multiple chemical reactors that contribute to a combined product amount provides richer information based on which the production process can be controlled.

[0011] The single reactant amount can refer to the amount of one reactant among multiple reactants or the corresponding amounts of more than one reactant, where the multiple reactants are chemically distinguishable from each other. Measuring only the amount of one of these reactants in each of the multiple chemical reactors may be sufficient. However, especially for complex chemical reactions, the amounts of more than one reactant (especially all reactants) in each chemical reactor can also be measured.

[0012] The combined product amount can refer to the amount of one product among multiple products, where the multiple products are chemically distinguishable from each other. The combined product amount mentioned herein can particularly be the product whose production is the main purpose of the chemical reactor. Another product among the multiple products can be produced only as a side reaction. However, these other products may still be useful, such as for other reactions in other reactors.

[0013] The multiple chemical reactors can be acetylene reactors for producing acetylene. In particular, acetylene or a raw form that can be subsequently processed into actual acetylene can be produced from oxygen and natural gas as reactants in multiple chemical reactors. In addition to acetylene, syngas can also be produced. In this particular case, for example, the measured single reactant amounts can refer to the individual amounts of natural gas and oxygen supplied to each of the multiple reactors, and the combined product amount can refer to the amount of acetylene produced jointly by the multiple chemical reactors.

[0014] The term "artificial intelligence" is understood in this text to include any type of machine learning model. Artificial intelligence that has been found to be particularly useful for this purpose is artificial neural networks and regression trees, in particular gradient-boosted trees such as the XGBoost model. However, it should be understood that these are just specific examples of the types of artificial intelligence that can be used.

[0015] Artificial intelligence can be specifically trained such that when receiving the amount of a single reactant of a chemical reactor as input, these artificial intelligences provide the amount of a single product of the chemical reactor as output, and the amounts of these single products add up to the combined product amount. Thus, "combined" can specifically refer to the sum. The amounts of the individual products and the combined product can be expressed, for example, in volume or weight, particularly in volume flow or weight flow (i.e., the volume or weight supplied or conveyed per unit time, respectively).

[0016] The association between the amount of a single reactant received by the artificial intelligence as input and the combination of the amounts of single products provided by the artificial intelligence as output can specifically correspond to the assignment made for training the artificial intelligence. For example, the combined product amount can be associated with the amount of a single reactant received by the artificial intelligence as input because it has been used as the combined training output of the artificial intelligence that receives the amount of a single reactant as input during the combined training of multiple artificial intelligences. Thus, the training data for the combined training of the artificial intelligence can include, for example, a) the measured amount of a single reactant as training input data, and b) the measured combined product amount as combined training output data. The artificial intelligence can be trained such that the amounts of single products provided as output when receiving the measured amount of a single reactant used as training input data add up to the measured combined product amount used as combined training output data.

[0017] The data "pairs" of a) the measured amount of a single reactant and b) the measured combined product amount can also be obtained at the moment when the production process is to be controlled (i.e., the state of the production process is to be inspected and / or changed). Thus, the association between the amount of a single reactant received by the artificial intelligence as input and the combination of the amounts of single products provided by the artificial intelligence as output can also correspond to the assignment between a) the amounts of single reactants measured for multiple reactors at the moment when the production process is to be controlled and b) the combined product amount that can be expected and / or measured at that moment. In other words, the trained artificial intelligence can be regarded as a model of the "true" data that can be measured during production and characterizes the ongoing production process.

[0018] It should be understood that perfect training of artificial intelligence is generally impossible in practice. Therefore, the output provided by the trained artificial intelligence may only roughly combine or sum to obtain the combined product amount, that is, the combined product amount associated with the individual reactant amounts received by the artificial intelligence as input. This applies not only to "real" data but also to training data, one reason being that overtraining should be avoided.

[0019] The provided trained artificial intelligence can be trained to provide the individual product amounts of these chemical reactors as output when additionally receiving input values derived from these individual reactant amounts. Although the additional input values can be derived only from the individual reactant amounts, that is, in the absence of other information, such as in terms of a function that depends only on the individual reactant amounts, for example, it has been found that using such additional input values can more accurately determine the individual product amounts. For example, the artificial intelligence can receive the individual reactant amounts of the corresponding chemical reactor and the ratios between the individual reactant amounts as input. For example, in the case of producing acetylene from natural gas and oxygen, it may be preferable to use artificial intelligence that not only receives the amounts of natural gas and oxygen as input but also additionally receives the ratios between the corresponding amounts of oxygen and natural gas in multiple chemical reactors. It is somewhat unexpected that such additional input can improve the performance of the system because these ratios do not carry more information than the separately measured amounts of oxygen and natural gas themselves, as they can be calculated by dividing one of these measured individual amounts by the other.

[0020] To determine the control parameters of the chemical reactions in multiple chemical reactors that contribute to the combined product amount, a system can be used that includes:

[0021] - A system for determining the individual product amounts of multiple chemical reactors as described above, and

[0022] - A control parameter determination unit configured to determine the control parameters of these chemical reactions in the multiple chemical reactors based on the determined individual product amounts.

[0023] Since the individual product amounts can be accurately determined using multiple separate artificial intelligences, the control parameters can be determined that allow for optimized control of the chemical reactors, thereby enabling more efficient production.

[0024] In particular, the control parameter determination unit may be configured to determine the amount of a single reactant to be fed to a plurality of chemical reactors and / or other process parameters of a plurality of chemical reactions based on the determined amount of a single product. Thus, the determined control parameters may particularly correspond to the amount of a single reactant to be fed to a plurality of chemical reactors and / or other process parameters of a plurality of chemical reactions. It should be understood that the plurality of reactions may be substantially chemically identical, where the "plurality" is only due to the reactions being carried out in different reactors, resulting in relatively small differences.

[0025] A human operator may control the plurality of chemical reactors based on the determined control parameters. Alternatively, a control system for controlling the plurality of chemical reactors may be provided, where the control system may be configured to control the plurality of chemical reactors based on the determined control parameters. Controlling the reactors based on the determined control parameters may refer to adjusting the actually observed control parameters to the determined control parameters. For example, the flow rate of one or more reactants to the reactors may be increased or decreased. The determined control parameters may particularly refer to target control parameters.

[0026] The control parameter determination unit may be configured to further determine the control parameters based on these measured amounts of a single reactant and / or the measured combined product amount. In this way, the current state of the corresponding reactor can be taken into account. This may make the determination of the control parameters more efficient, as the state of the reactor may limit the control parameters achievable within a desired time window, such that other control parameters do not need to be considered as candidates.

[0027] Additionally or alternatively, the control parameter determination unit may be configured to determine the control parameters based on a trained artificial intelligence. For example, a trained artificial intelligence may be used to determine the amount of a single product for a candidate amount of a single reactant. Then, those candidate amounts of a single reactant that produce the most favorable amount of a single product by using the trained artificial intelligence may be selected as the actual amount of a single reactant to be fed to the plurality of chemical reactors. As will be understood, the term "most favorable" may refer to any measure for evaluating the performance of the plurality of chemical reactors. For example, the "most favorable" amount of a single product may not necessarily be the highest, but may be those amounts of a single product that satisfy a predefined relationship with respect to the candidate amounts of a single reactant.

[0028] The control parameter determination unit may be configured to determine the control parameters of these chemical reactions such that, without reducing the combined product amount, the combined amount of at least one reactant among the reactants of the plurality of chemical reactors is minimized as much as possible. In this way, resource-saving chemical production can be achieved. Additionally, the cost for supplying at least one reactant and thus the total production cost can also be minimized as much as possible.

[0029] In particular, the control unit may be configured to control the chemical reaction such that, without changing the amount of the combined product, the combined amount of at least one of the reactants in the plurality of chemical reactors is minimized as much as possible. Thus, in such an embodiment, the amount of the combined product is neither decreased nor increased. In other words, the amount of the combined product is maintained.

[0030] In fact, the control unit may also be configured not to minimize the combined amount of a specific reactant, but to minimize the combined production cost. Since the costs for supplying different reactants and for controlling the chemical reaction according to specific process parameters (which may include specific energy consumption) may not remain constant over time, minimizing the combined production cost does not necessarily correspond to minimizing the amount of a specific reactant.

[0031] It has been found that the dependence of the combined product amount on the amounts of a plurality of individual reactants and / or other process parameters is generally relatively complex and typically includes several local minima when understood as a function of the space of the possible reactant amounts and / or other process parameters of the plurality of chemical reactors. Therefore, preferably, the corresponding minimization is performed globally (i.e., in the space of the possible reactant amounts and / or other process parameters of the plurality of chemical reactors). In particular, evolutionary algorithms such as differential evolution or genetic algorithms can be used for the corresponding minimization.

[0032] In an embodiment, the control parameter determination unit may be configured to determine these control parameters by minimizing a quantity that can be expressed by the following function:

[0033]

[0034] Under the following constraints

[0035] y(t)·b≥y, (1b)

[0036] And (1c)

[0037]

[0038] Where contains the individual amounts of P reactants in K reactors at time t, contains the cost of the P reactants, b ∈ {0,1} K indicates whether the K reactors are active, y(t) refers to the individual product amounts of the K reactors determined by these trained artificial intelligences for these individual reactant amounts X(t), y refers to the desired minimum combined product amount, s1 and s2 are predefined constants, contains the measured individual reactant amounts, and Include predefined limits for these individual reactant amounts. It should be understood that k is an integer row index ranging from 1 to K, and p is an integer column index ranging from 1 to P. It has been found that equations (1a) and (1d) form a suitable starting point for differential evolution ("DE"). By using these equations, it can also be specifically taken into account that in practice, there may be technical limitations on how quickly the flow rate of the reactants supplied to an individual chemical reactor can be changed, where these limitations can be reflected in a certain fraction of the current (measured) flow rate, up to which the flow rate can be decreased or increased in a given control cycle, as can be represented by s1 and s2 respectively.

[0039] Alternatively, for example, a genetic algorithm ("GA") can be used to optimize the control parameters. Then, the control parameter determination unit can be specifically configured to determine these control parameters by minimizing the quantity that can be expressed by the following function:

[0040]

[0041] Under the following constraints

[0042] And (2b)

[0043]

[0044] where, again, Include the individual amounts of P reactants in K reactors at time t, Include the cost of these P reactants, b ∈ {0,1} K Indicate whether these K reactors are active, y(t) refers to the individual product amounts of these K reactors determined by these trained artificial intelligences for these individual reactant amounts X(t), y refers to the desired minimum combined product amount, s1 and s2 are predefined constants, Include the measured individual reactant amounts, and Include the predefined limits for these individual reactant amounts. Additionally, Is a weight indicating the degree of influence of the constraint conditions known from equation (1b) now entering the function to be minimized as a penalty term. Again, it should be understood that k is an integer row index ranging from 1 to K, and p is an integer column index ranging from 1 to P.

[0045] Both differential evolution and genetic algorithms can be carried out in their different variants. Since some of them can allow for a faster and / or more accurate determination of the optimal control parameters, an "optimization of the optimization program" can be performed, that is, optimizing a series of methods considered for optimizing the control parameters.

[0046] Regarding considering differential evolution to find the optimal control parameters, the search for its optimal variant can be limited to the variants indicated in Table 1(a) below, where N pop refers to the size of each population considered, D refers to the number of control parameters considered, CR refers to the crossover probability, F refers to the differential weight, and "strategy" refers to the evolutionary strategy used, for example, in SciPy v1.9.2. Regarding considering genetic algorithms to find the optimal control parameters, the search for its optimal variant can be limited to the variants indicated in Table 1(b) below, where N pop similarly refers to the size of each population considered, CR similarly refers to the crossover probability, MR refers to the mutation probability, and "crossover" refers to the type of crossover considered.

[0047]

[0048] Table 1a, Table 1b: Parameter selection for a) differential evolution (DE, left) and b) genetic algorithm (GA, right) considered for control parameter optimization in the embodiment.

[0049] In order to optimize the control parameters in a meaningful way, the control parameter determination unit relies on the artificial intelligence that has been trained previously.

[0050] Preferably, the trained artificial intelligence can be obtained through a training method, that is, it can be obtained through a training method that includes:

[0051] - Providing the measured single reactant amount of each chemical reactor in the plurality of chemical reactors and the measured combined product amount of the plurality of chemical reactors associated with these measured single reactant amounts,

[0052] - Providing the estimated single product amount of each chemical reactor in the plurality of chemical reactors,

[0053] - Providing the artificial intelligence to be trained for each of the plurality of chemical reactors,

[0054] - Initially training the provided artificial intelligence so that the initially trained artificial intelligence provides the corresponding estimated single product amount as output when receiving these measured single reactant amounts as input,

[0055] - Determining the adjusted single product amount of each chemical reactor in the plurality of chemical reactors according to a method that can be expressed by the following formula

[0056]

[0057] where, refers to the individual product amounts provided as output by these preliminarily trained artificial intelligences when receiving these measured individual reactant amounts as input, where K is the number of chemical reactors, y refers to the combined product amount of this measurement, refers to these adjusted individual product amounts, and where S has the following form

[0058]

[0059] where I K is the identity matrix of dimension K, and each entry in the first row of S is equal to 1, and

[0060] - Supplementary training of these preliminarily trained artificial intelligences such that the supplementary trained artificial intelligence provides the corresponding adjusted individual product amount y′ as output when receiving these measured individual reactant amounts as input.

[0061] The above definitions of preliminary training and supplementary training should not be misinterpreted such that the output provided by the correspondingly trained artificial intelligence needs to perfectly match the corresponding target output. Instead, the target output (i.e., the estimated individual product amounts during preliminary training and the adjusted individual product amount y′ during supplementary training) is used as the training output data for the training input data, which is the measured individual reactant amounts, where for each of the preliminary training and supplementary training, i.e., when considered separately, known training procedures can be followed.

[0062] An artificial intelligence can be understood as a model that associates the individual reactant amounts and possibly other input amounts (which can be collectively referred to as x1, x2, …, x P ) with the individual product amounts where the individual product amounts provided as output by the artificial intelligence change during the training process, i.e., from the value y produced by the preliminary training to more precise values, which are assumed to more accurately represent the actual unmeasurable individual product amounts.

[0063] The quantity y′ can be understood as the adjusted combined product amount. However, it is preferably only regarded as a hypothetical combined product amount because the combined product amount has been measured compared to the individual product amounts, where these measured values are preferably reliable.

[0064] Given the form of S and the shape of P, the above equation (3) for the adjusted individual product amount can be expressed in detail as

[0065]

[0066] and

[0067]

[0068] such that due to the form of S, it also follows

[0069]

[0070] This means that the adjusted individual product amounts add up to the "hypothetically" adjusted combined product amount. Thus, the adjusted amounts y′ and y′ behave as expected from their "true" counterparts (i.e., from the measured combined product amount and the corresponding unmeasurable individual product amounts). At the same time, this will generally not apply to the corresponding vectors (y, y) T , because the individual product amounts y produced as the output of the preliminary training generally cannot be expected to have added up to the actually measured combined product amount y. If the combination of the individual product amounts is expected to be different from the combined product amount obtained by addition, S can be adjusted accordingly, especially its first row.

[0071] Estimated individual product amounts for multiple chemical reactors can be provided based on the measured combined product amount and / or the measured individual reactant amounts. In particular, they can be provided so as to combine (i.e., specifically add up) to the measured combined product amount. For example, if it is measured that multiple chemical reactors all receive the same individual reactant amount, the individual product amounts can also be estimated to be the same, i.e., the measured combined product amount divided by the number of chemical reactors. This particular estimated value can also be referred to as the average value. More generally, the decomposition of the measured combined product amount according to the measured individual reactant amounts can be used to estimate the individual product amounts. That is, if it is measured that multiple chemical reactors receive different individual reactant amounts, the individual product amounts can be estimated to be related to the measured combined product amount in the same way as the measured individual reactant amounts are related to the combination of the measured individual reactant amounts. A specific reactant among the reactants can be selected as the basis for such decomposition. In other words, the estimated individual product amounts can be such that they are related to the measured combined product amount in the same way as the individual reactant amounts measured for a specific reactant among the reactants are related to the combination of the individual reactant amounts measured for that specific reactant among the reactants. These combinations can specifically refer to sums.

[0072] Therefore, when the estimated individual product amounts are represented as and the individual product amounts obtained as the output from the supplementary training are represented as y″, the sequence can be regarded as a sequence of continuously improved approximations of the unmeasurable actual individual product amounts. It should be noted that even if the estimated individual product amounts are chosen so as to add up to the measured combined product amount y, i.e., Nor can it be assumed that the single product quantity y obtained as output from the preliminary training is also the case. Thus, even so, it is necessary to assume in general that Because preliminary training will generally have the effect that the output provided by the artificial intelligence deviates from the training output, so as to find a "compromise" between them, which is common for machine learning. Of course, it cannot be excluded that for some single reactant quantities that have already been used as training inputs, the preliminarily trained artificial intelligence provides an output that matches the corresponding training output, that is, the corresponding estimated single product quantity And thus added to obtain the measured combined product quantity y associated with the single reactant quantity.

[0073] These measured single reactant quantities and these measured combined product quantities provided in the training method can be measured over time, so that multiple single reactant quantities and associated combined product quantities measured at several time points are provided, where the steps of providing estimated single product quantities, preliminarily training these artificial intelligences, determining adjusted single product quantities, and supplementary training these artificial intelligences are performed for these multiple single reactant quantities and associated combined product quantities measured at these several time points.

[0074] If the chemical reactions of interest, that is, each chemical reaction in multiple chemical reactors, occur fast enough, and if the time required for the single products from multiple reactors to be transported and transported along a combined route to a location where the combined product quantity can be measured is short enough, then any time delay can be ignored to a sufficient approximation, that is, the combined product quantity measured at a given time point can be assigned to the single reactant quantity measured at that time point. Otherwise, the assignment between the measured combined product quantity and the measured single reactant quantity can be applied based on a predetermined time delay.

[0075] In the case where the measured values for training are obtained over time, some quantities used to describe the training method acquire time dependence. Although for practical reasons, measurements can be made only at specific time points, so that the time dependence can also be indicated by an additional index, for the sake of presentation, the time dependence can still be represented as if it were continuous, in order to distinguish it from the index indicating the chemical reactor / artificial intelligence and the input quantity received by the artificial intelligence. Then, for example, the measured single reactant quantity and the combined product quantity will be representable as x 1,2 = x 1,2 (t) and y = y(t), where t indicates the corresponding measurement moment. Thus, the preliminary training will produce an output quantity y = y(t), where, as defined in equation (3) above, the determination of the estimated single product quantity and the determination of the adjusted single product quantity can be performed identically for all measurement moments t, resulting in quantities etc.

[0076] When the training input data, i.e., a) the measured amount of a single reactant x 1,2 and possibly other input amounts x 3,…,P , and b) the training output data, i.e., the corresponding amounts etc. are preferably related to the same time points such that the pairs of training input data and training output data can be conveniently indexed by the same t, it should be understood that the training input data and training output data can also originate from different time points. In other words, the techniques disclosed herein can be generalized to prediction applications, i.e., applications where, based on the measured amount of a single reactant at a first time point, the amount of a single product can be predicted for a second time point located after the first time point.

[0077] These adjusted amounts of the single product y′ can be determined using the following formula:

[0078] P = (S T W -1 S) -1 S T W -1 , (8)

[0079] where W has any of the following forms:

[0080] a) W ∝ I K+1 ,

[0081] b)

[0082] where refers to the error or deviation of y (i.e., the amount of the single product provided as an output by the preliminarily trained artificial intelligence) relative to the estimated amount of the single product determined for the measured amount of the single reactant and the combined product amount indicated by t = 1…T tot . Option b) can also be written as As an alternative to option b), can be used which can also be written as Another option is to choose

[0083] c) W ∝ diag(S1 K ),

[0084] where 1 K refers to the K-dimensional vector 1 with all entries equal to 1 K =(1,…,1) T .

[0085] It has been found that any of these choices for P can achieve good accuracy for the adjusted amounts of the single product.

[0086] Preferably, in this training method, the adjusted individual product amounts are repeatedly determined and the artificial intelligence is supplemented and trained, wherein, in each repetition:

[0087] - The individual product amounts provided as outputs by the previously trained artificial intelligence when receiving these measured individual reactant amounts as inputs are assumed to be the individual product amounts to be adjusted, so as to determine further adjusted individual product amounts based thereon, and

[0088] - The artificial intelligence is supplemented and trained such that the supplemented and trained artificial intelligence provides these further adjusted individual product amounts as outputs when receiving these measured individual reactant amounts as inputs.

[0089] As described above, the quantity y′ is preferably regarded as the hypothetical combined product quantity. Accordingly, it can be discarded between repetitions, which means that the value thereof determined in one repetition during the process of determining the adjusted individual product amounts is not used in the next repetition, i.e., for determining any other adjusted individual product amounts. Instead, the corresponding measured combined product quantity can be used again. Therefore, when repeatedly supplementing and training, equation (3) can be used again, wherein y is replaced by the individual product amounts determined by the artificial intelligence generated by the previous supplementary training, but not y′.

[0090] Furthermore, in order to adjust the individual product amounts between supplementary trainings, an equation corresponding to equation (8) can be used again, particularly in conjunction with any one of options a) to c) of W above. If option b) is used, then the quantities e t are then redefined to refer to i) the error or deviation of the individual product amounts provided as outputs by the corresponding previously supplemented and trained artificial intelligence when receiving the measured individual reactant amounts as inputs relative to ii) the corresponding individual product amounts used as the training outputs of the corresponding previous supplementary training. Since the estimated individual product amounts are used as the training outputs during the preliminary training, it can also be said that for each adjustment, the training outputs used in the corresponding previous training can be used. On the other hand, since in this embodiment, the training outputs for supplementary training are the adjusted training results of the corresponding previous training, it can be said that in the adjustment prepared for the corresponding next supplementary training, the quantities e t refer to the error or deviation between the current training result and the adjusted training result of the corresponding previous training.

[0091] The preliminary training, supplementary training, and any repetition of supplementary training can each be carried out in various ways. For example, known training procedures can be followed, which may depend on the type of artificial intelligence used, wherein, in particular, a loss function whose type is basically unrestricted can be used.

[0092] While as described above, it may be preferable to train multiple artificial intelligences not in a single training but in a number of training epochs, it has been found that similar or even more preferred embodiments can be achieved using only a single training as long as a suitable loss function is chosen for the single training.

[0093] According to one of the embodiments that can be achieved without the need to train in a number of training epochs as described above, the (single) training of the artificial intelligence is performed using the following loss function:

[0094]

[0095] where, refers to the estimated single product amount of the K reactors at a given time t, y(t) refers to the single product amount at time t as determined by the K artificial intelligences at a given stage of the (single) training, and α is a predefined training parameter.

[0096] It should be understood that in principle, the loss function given by equation (9) can also be used in different epochs of the step-by-step training procedure outlined above, i.e., for initial training, supplementary training, and any repetition of supplementary training. Although for initial training, equation (9) can be used identically, for one or more supplementary trainings, in equation (9) can be replaced by a corresponding adjusted version of the single product amount as determined by the previously trained corresponding artificial intelligence, i.e., by the corresponding training output amount for the multiple artificial intelligences in the corresponding training epoch. That is, for the original supplementary training, in equation (9) can then be replaced by the adjusted single product amount y ′ (t) as determined according to equation (3), and if the supplementary training is repeated as described above, then as used in equation (9) can be reset for each repetition to the corresponding further adjusted single product amount, i.e., for example, reset to SPy″(t) for the first repetition in terms of the terms introduced further above.

[0097] As can be understood from equation (9), the loss function is intended to minimize, using its first term, the deviation between a) the training output (which can particularly correspond to the initially estimated single product amount in the case of a single training or in the case of the initial training epoch in the step-by-step training process outlined above) and b) the actual output y of the trained artificial intelligence, and to minimize, using its second term, the deviation between a) the training output and b) the corresponding combination (particularly the sum) of the actual output (i.e., a) entries of and b) entries of y). The latter is obvious by noting that as long as Refers to the estimated individual product amounts, as in the case of a single training, and if the estimated individual product amounts are chosen such that they add up to the measured combined product amount y, then the second term simplifies to

[0098] The following is another possible loss function that can be used specifically in the method of a single training but can also generalize well to multi-stage training, as also indicated above for the loss function from equation (9):

[0099]

[0100] This alternative loss function, which is the L1 analogue of the previous one, considers that the larger deviation between and y(t) is less important, which can help to avoid overfitting of the artificial intelligence to

[0101] In the case where a trained artificial intelligence is available, for any newly measured set of individual reactant amounts for multiple chemical reactors, the corresponding individual product amounts can be determined even if they may not be measurable. As already indicated above, this can allow for improved control of multiple chemical reactors. This is because more refined optimization of the control parameters can be carried out. For example, as already outlined above, specific evolutionary algorithms can be applied, such as differential evolution algorithms or genetic algorithms.

[0102] It should be noted that optimization can also be performed on a series of artificial intelligences under consideration in order to find the artificial intelligence to be actually implemented. When only considering a given type of artificial intelligence, this process can be referred to as hyperparameter optimization. When considering more than one type of artificial intelligence, this optimization can be extended to different types of artificial intelligence. Typically, the optimization performed on artificial intelligence includes: pre-selecting one or more types of artificial intelligence with different hyperparameters, training these artificial intelligences, and then comparing these trained artificial intelligences in terms of performance according to a predefined performance metric. Thus, for example, in order to find the artificial intelligence to be used for optimizing control parameters, artificial neural networks and XGBoost models with different hyperparameters can be pre-selected, trained, and then compared, where then the one with the best performance among them can be selected for actual use. The term "best performance" can refer, for example, to the degree to which the corresponding artificial intelligence can reproduce the measured combined product amount from the corresponding measured reactant amounts (i.e., the measured validation data).

[0103] In the case of using an artificial neural network or an XGBoost model as the artificial intelligence, the hyperparameters can be limited to the values given in Table 2a and Table 2b below, respectively.

[0104] ​

[0105] Table 2a, Table 2b: Hyperparameter selection of artificial intelligence considered in embodiments with a) artificial neural network (ANN, left) and b) XGBoost model (XGB, right).

[0106] In addition, the determination of the training and optimization control parameters of the artificial intelligence can be repeated over time, for example, at predefined intervals. In this way, changes in different chemical reactors and / or their environments can be taken into account. Then, the control parameters of the reactor can be set (re) accordingly. The periodic intervals for training the artificial intelligence, optimizing the control parameters, and (re) setting the control parameters to the corresponding "new" optimal control parameters can be referred to as control cycles. A typical control cycle can have a duration of, for example, one hour.

[0107] On the other hand, the present invention relates to a training system for training multiple artificial intelligences to be used for determining the individual product amounts of multiple chemical reactors that contribute to a combined product amount, wherein the training system includes:

[0108] - A training data providing unit configured to provide training data, where the training data includes a) training input data corresponding to the individual reactant amounts received by the multiple chemical reactors, and b) training output data corresponding to the combined product amount associated with these individual reactant amounts provided as input data,

[0109] - An artificial intelligence providing unit configured to provide an artificial intelligence to be trained for each of these chemical reactors, and

[0110] - A training unit configured to train these artificial intelligences in a combined training using the training data such that these trained artificial intelligences provide the individual product amounts of these chemical reactors as outputs when receiving the individual reactant amounts of these chemical reactors as inputs, and these individual product amounts are combined into a combined product amount associated with these individual reactant amounts received as inputs. The training performed by the training unit can be of any of the above types.

[0111] The present invention also relates to a method for determining the individual product amounts of multiple chemical reactors that contribute to a combined product amount, wherein the method includes:

[0112] - Providing the measured individual reactant amounts for each of the multiple chemical reactors,

[0113] - Provide a trained artificial intelligence for each of these chemical reactors, wherein the provided trained artificial intelligence is trained to provide a single product quantity of these chemical reactors as an output when receiving a single reactant quantity of these chemical reactors as an input, and these single product quantities are combined into a combined product quantity associated with these single reactant quantities received as an input, and

[0114] - Determine the single product quantity of the plurality of chemical reactors based on these measured single reactant quantities and these trained artificial intelligences. As further described above, the method can be performed by the corresponding system in any of its embodiments.

[0115] Another aspect of the present invention relates to a training method for training a plurality of artificial intelligences to be used for determining the single product quantity of a plurality of chemical reactors that contribute to a combined product quantity, wherein the training method includes:

[0116] - Provide training data, wherein the training data includes a) training input data corresponding to the single reactant quantities received by the plurality of chemical reactors, and b) training output data corresponding to the combined product quantity associated with these single reactant quantities provided as input data,

[0117] - Provide an artificial intelligence to be trained for each of these chemical reactors, and

[0118] - Train these artificial intelligences in a combined training such that the trained artificial intelligences provide the single product quantity of these chemical reactors as an output when receiving the single reactant quantities of these chemical reactors as an input, and these single product quantities are combined into a combined product quantity associated with these single reactant quantities received as an input. The method can be performed by the above training system in any of its embodiments.

[0119] On the one hand, the present invention also relates to a computer program for determining the single product quantity of a plurality of chemical reactors that contribute to a combined product quantity, wherein the program includes program code means for causing the above system for determining the single product quantity to execute the corresponding method for determining the single product quantity.

[0120] In addition, on the one hand, the present invention relates to a computer program for training a plurality of artificial intelligences to be used for determining the single product quantity of a plurality of chemical reactors that contribute to a combined product quantity, wherein the program includes program code means for causing the above training system to execute the above training method.

[0121] It should be understood that the above aspects, specifically the system as described in claim 1, the method as described in claim 13, the computer program as described in claim 14, as well as the training system, training method, and corresponding computer program have similar and / or identical preferred embodiments, particularly as defined in the dependent claims.

[0122] It should be understood that the preferred embodiments of the present invention may also be any combination of the dependent claims or the above embodiments and the corresponding independent claims.

[0123] These and other aspects of the present invention will become apparent and be elucidated with reference to the embodiments described below. BRIEF DESCRIPTION OF THE DRAWINGS

[0124] Figure 1 An installation for acetylene production is schematically and exemplarily shown,

[0125] Figure 2 A system for determining the amount of a single product is schematically and exemplarily shown,

[0126] Figure 3 Multiple artificial intelligences are schematically and exemplarily shown,

[0127] Figure 4a 、 Figure 4b An increase in production efficiency achievable in one embodiment is exemplarily shown,

[0128] Figure 5a 、 Figure 5b An increase in production efficiency achievable in another embodiment is exemplarily shown, and

[0129] Figure 6 A method for determining the amount of a single product is schematically and exemplarily shown. DETAILED DESCRIPTION

[0130] Figure 1Schematically and exemplarily shows a chemical production facility 10 operating a production process. The facility 10 includes ten chemical reactors 11, 12, ..., 1K, and oxygen (O2) and natural gas (NG) are fed to each chemical reactor as reactants. A plurality of chemical reactors 11, 12, ..., 1K operate chemically corresponding reactions, thereby producing chemically corresponding products. In this case, acetylene (AC) in its original form and synthesis gas (SG) are produced from the reactants O2 and NG. Due to differences in the amounts of reactants fed to different reactors and the behavior of the reactors, for example, each of the reactors 11, 12, ..., 1K produces a single product amount. The outputs of the individual reactors (i.e., the single product amounts) are combined via ducts and directed into a fractionation tower 12'. In the illustrated embodiment, three groups of reactors are formed, and a separate fractionation tower 12' is provided for each of the three groups. From the fractionation tower 12', the product (i.e., the product amounts that have been partially combined) is further directed to a compressor 13', and from the compressor 13' to a dedicated device 14' for separating the product into its chemical components (i.e., the original AC and SG). Since the chemical reactions occurring in the reactors 11, 12, ..., 1K involve gas cracking, the separation process performed by the device 14' can be referred to as cracked gas separation. Although a compressor 13' is still provided separately for each of the three reactor groups, after passing through the compressor 13', the products are combined such that the total product amount produced by the ten reactors 11, 12, ..., 1K enters the device 14'. Although not shown in Figure 1 , after the original AC and SG are separated from each other in the device 14', the original AC is compressed and processed into the final form of AC in an acid scrubber, while the SG is directed to a lean gas scrubber. Since the main purpose of the facility 10 is to produce acetylene, the amount of acetylene output by the device 14' will be referred to as "the" combined product amount, and the amount of synthesis gas output by the device 14' can be regarded as a by-product for this purpose. Since both the acetylene and the synthesis gas produced are typically used in other chemical production processes, it should be understood that the embodiments described herein can also be used when alternatively considering the amount of synthesis gas produced as the combined product amount whose production is to be optimized.

[0131] Figure 2System 100 is schematically and exemplarily shown for determining the individual product amounts of a plurality of chemical reactors contributing to a combined product amount. The system includes a measurement value providing unit 101 configured to provide the measured individual reactant amounts for each of the plurality of chemical reactors. Further, system 100 includes an artificial intelligence providing unit 102 configured to provide a trained artificial intelligence for each of the chemical reactors, wherein the provided trained artificial intelligence is trained to provide the individual product amounts of the chemical reactors as output when receiving the individual reactant amounts of the chemical reactors as input, and the individual product amounts are combined into a combined product amount associated with the individual reactant amounts received as input. System 100 further includes an individual product amount determining unit 103 configured to determine the individual product amounts of the plurality of chemical reactors based on the measured individual reactant amounts and the trained artificial intelligences.

[0132] System 100 can be used to model Figure 1 production facility 10 shown in. In this way, information about the reactions running in the individual chemical reactors 11, 12,..., 1K can be obtained, even when the individual product amounts produced by the respective reactors 11, 12,..., 1K (i.e., the individual product streams led from the individual reactors 11, 12,..., 1K to the fractionation column 12') cannot be measured. Generally, in a production process as Figure 1 shown, the individual product amounts produced by the individual chemical reactors 11, 12,..., 1K cannot be measured, in contrast to which the individual reactant amounts supplied to the individual reactors 11, 12,..., 1K (i.e., the amounts of oxygen and natural gas supplied to the reactors 11, 12,..., 1K in the Figure 1 example) and the combined product amount (i.e., in particular the total amount of acetylene in the Figure 1 example) can be measured. Knowing the individual product amounts enables optimal control of the plurality of reactors 11, 12,..., 1K and thereby increases the efficiency of the entire production process. For example, for a given combined product amount, the natural resources consumed in production can be reduced. Thus, in the above example of acetylene production, the amount of natural gas required can be reduced. This can not only reduce production costs and thereby gain a competitive advantage, but also increase economic and political independence.

[0133] Figure 3Schematically and by way of example, the structure of artificial intelligence that can be provided by an artificial intelligence providing unit is shown, i.e., it can be used to model individual reactors 11, 12, ..., 1K. In the case shown, these artificial intelligences are identical in structure and are in the form of artificial neural networks 21, 22, ..., 2K. Since a separate artificial intelligence is provided for each of the chemical reactors 11, 12, ..., 1K, there are K artificial intelligences, which can be considered to form a larger combined artificial intelligence 20. In Figure 1 the example shown, K = 10. The artificial neural networks 21, 22, ..., 2K being "identical in structure" can in particular mean that they share the same set of hyperparameters, while their training can of course result in different internal parameters of the artificial neural networks, thereby resulting in individual models of the corresponding chemical reactors 11, 12, ..., 1K.

[0134] In Figure 3 the embodiment of, the selected artificial neural network includes three layers, i.e., a single hidden layer. Via the input layer, a single reactant quantity x 1,2 is received, and via the output layer, a single product quantity is provided Furthermore, the single product quantities provided by the K artificial neural networks are added together, thereby forming a combined product quantity as a combined output. This combined output can be regarded as a prediction or estimation of the combined product quantity by the combined artificial neural network 20, and the facility 10 will produce this combined product quantity in the case of feeding the given reactant quantities x 1,2 received as inputs by the artificial intelligences 21, 22, ..., 2K to its reactors 11, 12, ..., 1K.

[0135] In the embodiment shown, in addition to the single reactant quantity, the artificial neural networks 21, 22, ..., 2K also receive one or more other input values x 1,2 obtained from the single reactant quantity x 3,..,P . Thus, the number (P) of inputs received by each of the K artificial neural networks can be higher than the number of reactants involved in the chemical reaction. For example, in the context of acetylene production, it has been found that in addition to the amounts of oxygen (O2) and natural gas (NG) provided to an individual reactor, using their ratio (e.g., the amount of O2 received divided by the amount of NG received) simultaneously as a control parameter for controlling acetylene production is beneficial.

[0136] Figure 3An embodiment of modeling reactors 11, 12, ..., 1K using an artificial neural network is shown. In other embodiments, other types of artificial neural networks can be used. In particular, gradient boosting trees, especially those from the XGBoost library, can be used alternatively. When using an alternative artificial intelligence to model individual reactors 11, 12, ..., 1K, the internal structure of the artificial intelligence will be different, but they can work with the same input and output data as described above with respect to Figure 3 In addition, it may still be preferred that multiple artificial intelligences are used to implement a larger combined artificial intelligence, and the output of the combined artificial intelligence is formed by combining (in particular, adding) the outputs provided by individual artificial intelligences.

[0137] To train the artificial intelligence, a training system is used, which includes a training data providing unit configured to provide training data, where the training data includes a) training input data corresponding to the amount of a single reactant received by a plurality of chemical reactors, and b) training output data corresponding to the amount of combined product associated with these amounts of single reactants provided as input data. The training system further includes an artificial intelligence providing unit and a training unit. The artificial intelligence providing unit is configured to provide an artificial intelligence to be trained for each of these chemical reactors, and the training unit is configured to train these artificial intelligences in a combined training using the training data such that these trained artificial intelligences provide the amounts of single products of these chemical reactors as outputs when receiving the amounts of single reactants of these chemical reactors as inputs, and these amounts of single products are combined into the amount of combined product associated with these amounts of single reactants received as inputs.

[0138] The training data can be during ongoing production (such as during Figure 1When the facility 10 shown in is in operation, it is collected via measurement. Thus, from such measurements, for example, a) the measured amounts of oxygen and natural gas supplied to the K reactors 11, 12, ..., 1K at a given point in time and the ratios between these amounts, and b) the amount of acetylene recovered from the device 14' at the same point in time can be established. Among them, these pairs collected over time can be used as training data. In a similar manner, validation data can be obtained, and the validation data can be used together with the training data to train the corresponding artificial intelligence. The collected training (and validation) data can be regarded as combined training (and validation) data. Although the combined training (and validation) data can be used for a single combined training of multiple artificial intelligences, it is also preferably to divide the combined training into several stages. In each of these stages, multiple artificial intelligences are trained using single training data. The corresponding single training data can be obtained from the combined training data and / or from the training results of the corresponding previous training stage. In particular, the training unit of the training system can be configured to implement the following training method.

[0139] In the first step of the training method, the measured single reactant amount of each chemical reactor among multiple chemical reactors and the measured combined product amount of the multiple chemical reactors associated with these measured single reactant amounts are provided. In other words, the combined training data is provided.

[0140] In the second step of the training method, the estimated single product amount of each chemical reactor among the multiple chemical reactors is provided. These estimated single product amounts can correspond to the average value of the combined product amounts provided in the first step of the training method, where the average value can be weighted according to the amount of one of the reactants supplied to the corresponding reactors 11, 12, ..., 1K. For example, the average value can be weighted according to the amount of natural gas supplied to the single reactors 11, 12, ..., 1K. This weighting can produce a more accurate estimate because it can be expected that the more natural gas is supplied to the reactor, the more acetylene the reactor contributes to the total amount of acetylene produced.

[0141] In the third step of the training method, an artificial intelligence to be trained is provided for each of the multiple chemical reactors 11, 12, ..., 1K. For example, an artificial neural network 21, 22, ..., 2K as shown in can be provided. Figure 3 shown.

[0142] In the fourth step of the training method, the artificial intelligence is preliminarily trained so that the preliminarily trained artificial intelligence provides the corresponding estimated single product amount as output when receiving the measured single reactant amount as input. In addition to this selection of training data, the preliminary training of a single artificial intelligence can be carried out in a manner known for the corresponding type of artificial intelligence. For example, a known loss function can be used.

[0143] In the fifth step of the training method, the adjusted individual product amounts for each of the plurality of chemical reactors 11, 12, ..., 1K are determined according to the method expressible by the above equations (3) and (4). In particular, in order to determine the matrix P used in equation (3), equation (8) can be used in conjunction with any of the choices a) to c) for the selection of the matrix W used therein.

[0144] Then, in the sixth step of the training method, the preliminarily trained artificial intelligence is supplemented with training such that the supplemented artificial intelligence provides the corresponding adjusted individual product amounts as outputs when receiving the measured individual reactant amounts as inputs. Similarly, apart from the selection of the training data, the supplementary training of the individual artificial intelligence can also be carried out in a manner known for the corresponding type of artificial intelligence. For example, known loss functions can also be used for the supplementary training.

[0145] To collect the training (and validation) data, the measured individual reactant amounts and the measured combined product amounts provided in the first step of the training method can be measured over time such that a plurality of individual reactant amounts and the associated combined product amounts measured at several time points are provided. Then, the steps of providing the estimated individual product amounts (second step), preliminarily training these artificial intelligences (fourth step), determining the adjusted individual product amounts (fifth step), and supplementing the training of these artificial intelligences (sixth step) can be carried out for the plurality of individual reactant amounts and the associated combined product amounts measured at several time points.

[0146] Preferably, the fifth and sixth steps of the training method are repeated, wherein in each repetition, the individual product amounts provided as outputs by the previously trained artificial intelligence when receiving these measured individual reactant amounts as inputs are assumed to be the individual product amounts to be adjusted in order to determine the further adjusted individual product amounts based thereon, and the artificial intelligences are supplemented with training such that the supplemented artificial intelligences provide these further adjusted individual final product amounts as outputs when receiving the corresponding individual reactant amounts as inputs. To avoid overtraining of the artificial intelligence, a stopping criterion can be applied, wherein if the stopping criterion is met, the repeated supplementary training is terminated. For example, the stopping criterion can be selected such that it is met as long as at least one of the following conditions is satisfied: a) a predetermined number of repetitions has been experienced, b) the performance of the artificial intelligence has not improved after a predetermined number of repetitions, wherein the performance can be measured according to the mean absolute error of the sum of the individual outputs of the artificial intelligence relative to the measured combined product amount, and the mean absolute error is evaluated on the training data set.

[0147] In an alternative training method, the combined training is based on the same training data but is not divided into several stages. Instead, a known training protocol can be followed, where one of the functions given in the above equations (9), (10) is used as the loss function for training.

[0148] Irrespective of the training protocol followed, the hyperparameters of the artificial intelligence used separately can be optimized. In principle, the optimization of the hyperparameters can be performed such that the artificial intelligence is trained with different hyperparameter selections, where the different trained artificial intelligences are then compared in terms of performance. However, it may be more efficient to make a final selection of the hyperparameters before entering the actual training. Thus, for example, the hyperparameter optimization can be performed based on an estimated single product amount assumed to be the training output data. In particular, in the case of the multi-stage training involving preliminary training and one or more supplementary trainings indicated above, the hyperparameter optimization can be performed only on the preliminarily trained artificial intelligence.

[0149] Once the artificial intelligence is trained for a certain selection of hyperparameters, they are able to accurately model the reactors 11, 12,..., 1K.

[0150] Then, in order to determine the control parameters of the chemical reactions in the plurality of chemical reactors 11, 12,..., 1K, a system can be provided that includes: a system 100 for determining the single product amount of the plurality of chemical reactors; and a control parameter determination unit configured to determine the control parameters of the chemical reactions in the plurality of chemical reactors 11, 12,..., 1K based on the determined single product amount. The control parameter determination unit is preferably configured to further determine the control parameters based on the measured single reactant amount and / or the measured combined product amount, more particularly such that the combined amount of at least one of the reactants in the reactants of the plurality of chemical reactors is minimized as much as possible without reducing the combined product amount. For example, in the case of acetylene production, for a given desired total amount of acetylene produced, the amount of natural gas used in the production can be minimized as much as possible. Particularly preferred schemes for finding the optimal control parameters include differential evolution and genetic algorithms as discussed above with respect to equations (1a) to (1d) and (2a) to (2c) respectively.

[0151] Figure 4a and Figure 4b Schematically and exemplarily shows the improvement in efficiency achievable for the actual production facility 10 according to the above embodiment. Figure 4ais a scatter plot, where each point represents the state of a production facility at a given moment in the past. On the horizontal axis, the combined amount of acetylene produced is indicated in kilograms per hour, and on the vertical axis, the money spent on the reactants consumed is indicated in euros (EUR) per hour. As will be understood, for a fixed position in the horizontal direction, the higher the production efficiency, the lower the position of the corresponding point in the scatter plot in the vertical direction. On the other hand, for a fixed position in the vertical direction, the higher the production efficiency, the further to the right the position of the corresponding point in the scatter plot in the horizontal direction. From Figure 4a it can be seen that the production efficiency of the production facility under consideration varies within the considered time interval because of the relative distribution of the points in the scatter plot. Figure 4a The two smaller point clouds highlighted therein correspond to the states of the production facility under consideration on the same day, where the upper point cloud of the two point clouds consists of points representing the production state without optimization of the control parameters, and the lower point cloud of the two point clouds represents the production state achievable through the above-mentioned optimization of the control parameters. Thus, it can be observed that by implementing the optimization of the control parameters as described above, a significant increase in production efficiency can be achieved on that day. Figure 4b This is further demonstrated by the graph showing the money spent on the reactants consumed on the corresponding day over the course of a day, also in euros per hour. The upper plotted line of the two plotted lines corresponds to the unoptimized production state, while the lower plotted line of the two lines corresponds to the optimized production state. The approximate width of the gap between these two lines exceeds 200 EUR per hour throughout the day, from which it is obvious that there is significant cost-saving potential. It should be understood that this cost-saving potential is closely related to the potential for saving resources (especially natural gas in the case of acetylene production).

[0152] Figure 5a and Figure 5b are Figure 4a and Figure 4b differ only in the database, i.e., in the production facility 10 for which data has been collected. Similarly for this different production facility, the potential for significantly increasing production efficiency by adopting the optimization of the control parameters as described above becomes obvious.

[0153] Figure 6Method 200 is schematically and exemplarily shown for determining the individual product amounts of a plurality of chemical reactors that contribute to a combined product amount. The method includes: a step 201 of providing a measured individual reactant amount for each of the plurality of chemical reactors; and a step 202 of providing a trained artificial intelligence for each of these chemical reactors, wherein the provided trained artificial intelligence is trained to provide, when receiving the individual reactant amounts of these chemical reactors as input, the individual product amounts of these chemical reactors as output, and these individual product amounts are combined into a combined product amount associated with the individual reactant amounts received as input. Further, method 200 includes: a step 203 of determining the individual product amounts of the plurality of chemical reactors based on the measured individual reactant amounts and the trained artificial intelligence. The method can be performed, for example, by system 100.

[0154] From the study of the drawings, the present disclosure, and the appended claims, those skilled in the art can understand and realize other variations of the disclosed embodiments when practicing the claimed invention.

[0155] 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.

[0156] A single unit or device may implement the functions of several items recited in the claims. The fact that certain measures are recited in mutually different dependent claims does not mean that a combination of these measures cannot be used advantageously.

[0157] A program such as providing reactant amounts or other data, providing artificial intelligence, determining individual product amounts or control parameters, any training of the artificial intelligence, etc., which is performed by one or several units or devices, can be performed by any other number of units or devices. These processes can be implemented as program code means of a computer program and / or dedicated hardware.

[0158] A computer program product can be stored / distributed on a suitable medium, such as an optical storage medium or a solid-state medium provided together with or as part of other hardware, but can also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems.

[0159] Any reference signs in the claims should not be construed as limiting the scope.

[0160] The present invention relates to a system for determining the individual product amounts of a plurality of chemical reactors that contribute to a combined product amount. The system includes a measurement value providing unit and an artificial intelligence providing unit. The measurement value providing unit provides the measured individual reactant amounts for each reactor, and the artificial intelligence providing unit provides artificial intelligence for each reactor. These artificial intelligences are trained to provide the individual product amounts of these reactors as outputs when receiving the individual reactant amounts of these reactors as inputs, and these individual product amounts are combined into a combined product amount associated with the individual reactant amounts received as inputs. The system further includes an individual product amount determining unit that determines the individual product amounts of these reactors based on these measured individual reactant amounts and the trained artificial intelligence. The system enables an improvement in chemical production efficiency when a plurality of reactors contribute to a combined product amount.

Claims

1. A system (100) for determining the individual product amounts of a plurality of chemical reactors (11, 12, ..., 1K) that contribute to the amount of a combined product, wherein, The system includes: - A measurement value providing unit (101) configured to provide a measured amount of a single reactant for each of the plurality of chemical reactors, - An artificial intelligence providing unit (102) configured to provide a trained artificial intelligence (21, 22,..., 2K) for each of the chemical reactors (11, 12,..., 1K), wherein the provided trained artificial intelligences (21, 22,..., 2K) are trained to provide an amount of a single product of the chemical reactors (11, 12,..., 1K) as an output when receiving the amount of a single reactant of the chemical reactors (11, 12,..., 1K) as an input, and the amounts of these single products are combined into an amount of a combined product associated with the amounts of these single reactants received as an input, and - A single product amount determining unit (103) configured to determine the amounts of single products of the plurality of chemical reactors (11, 12,..., 1K) based on the measured amounts of the single reactants and the trained artificial intelligences (21, 22,..., 2K).

2. The system according to claim 1, wherein The plurality of chemical reactors (11, 12,..., 1K) are acetylene reactors for producing acetylene.

3. The system according to claim 1 or 2, wherein The provided trained artificial intelligences (21, 22,..., 2K) are trained to provide the amounts of single products of the chemical reactors (11, 12,..., 1K) as an output when additionally receiving an input value obtained from the amounts of the single reactants.

4. A system for determining control parameters of chemical reactions in a plurality of chemical reactors (11, 12,..., 1K) that contribute to the amount of a combined product, wherein, The system includes: - A system (100) for determining the amounts of single products of a plurality of chemical reactors (11, 12,..., 1K) according to any one of claims 1 to 3, - A control parameter determining unit configured to determine control parameters of the chemical reactions in the plurality of chemical reactors (11, 12,..., 1K) based on the determined amounts of single products.

5. The system according to claim 4, wherein, The control parameter determining unit is configured to further determine the control parameters based on the measured amounts of the single reactants and / or the measured amount of the combined product.

6. The system according to claim 4 or 5, wherein, The control parameter determining unit is configured to determine the control parameters of the chemical reactions such that, without reducing the amount of the combined product, the combined amount of at least one of the reactants in the reactants of the plurality of chemical reactors (11, 12,..., 1K) is reduced as much as possible.

7. The system according to any one of the preceding claims, wherein, The trained artificial intelligences (21, 22,..., 2K) can be obtained by a training method, the training method including: - Providing the measured amount of a single reactant of each chemical reactor in the plurality of chemical reactors (11, 12,..., 1K) and the measured amount of the combined product of the plurality of chemical reactors (11, 12,..., 1K) associated with the measured amounts of the single reactants, - Providing an estimated amount of a single product of each chemical reactor in the plurality of chemical reactors (11, 12,..., 1K), - Provide an artificial intelligence (21, 22, ..., 2K) to be trained for each of the plurality of chemical reactors (11, 12, ..., 1K), - Initially train the provided artificial intelligence (21, 22, ..., 2K) such that the initially trained artificial intelligence (21, 22, ..., 2K) provides a corresponding estimated single product amount as output when receiving the measured single reactant amount as input, - Determine the adjusted single product amount for each chemical reactor in the plurality of chemical reactors (11, 12, ..., 1K) according to a method that can be expressed by the following formula wherein, refers to the individual product amounts provided as outputs by these preliminarily trained artificial intelligences (21, 22, ..., 2K) when receiving these measured individual reactant amounts as inputs, where K is the number of chemical reactors (11, 12, ..., 1K), and y refers to the combined product amount of this measurement, refers to these adjusted individual product amounts, and wherein S has the following form where I K is the identity matrix of dimension K, and each entry in the first row of S is equal to 1, and - Supplementary train the initially trained artificial intelligence (21, 22, ..., 2K) such that the supplementary trained artificial intelligence (21, 22, ..., 2K) provides a corresponding adjusted single product amount as output when receiving the measured single reactant amount as input.

8. The system according to claim 7, wherein, Measure the single reactant amounts and the combined product amounts provided in this training method over time, such that a plurality of single reactant amounts and associated combined product amounts measured at several time points are provided, wherein the steps of providing the estimated single product amounts, initially training the artificial intelligence (21, 22, ..., 2K), determining the adjusted single product amounts, and supplementary training the artificial intelligence (21, 22, ..., 2K) are performed for the plurality of single reactant amounts and associated combined product amounts measured at the several time points.

9. The system according to claim 8, wherein, These adjusted single product amounts are determined using the following formula: P = (S T W -1 S) -1 S T W -1 , wherein, W h has any one of the following forms: a) W ∝ I K+1 , b) wherein, refers to the error of these estimated single product amounts determined for the single reactant amounts and combined product amounts measured with respect to y for t = 1...T tot and, c)W h ∝ diag(S1 K ), Among them, 1 K refers to the K-dimensional vector 1 with all entries being 1 K =(1,…,1) T .

10. The system according to claim 8 or 9, wherein Repeat determining the adjusted single product amounts and supplementary training the artificial intelligence (21, 22, ..., 2K) in this training method, wherein in each repetition: - The single product amounts provided as output by the previously trained artificial intelligence (12) when receiving the measured single reactant amount as input are assumed to be the single product amounts to be adjusted in order to determine further adjusted single product amounts based thereon, and - The artificial intelligence (21, 22, ..., 2K) is supplementary trained such that the supplementary trained artificial intelligence (21, 22, ..., 2K) provides a further adjusted single final product amount as output when receiving the corresponding single reactant amount as input.

11. The system according to claim 4 alone or in combination with any one of claims 5 to 10, wherein the control parameter determination unit is configured to determine the control parameters by minimizing a quantity that can be expressed by the following function: Under the following constraints y(t)·b≥y, and Among them, The individual amounts of P reactants in K reactors at time t, including the cost of the P reactants, b ∈ {0, 1} K indicating whether the K reactors (11, 12, ..., 1K) are active, y(t) refers to the individual product amounts of the K reactors determined by these trained artificial intelligences (21, 22, ..., 2K) for these individual reactant amounts X(t), y refers to the desired minimum combined product amount, and s1 and s2 are predefined constants, including the measured individual reactant amounts, and including the predefined limits of these individual reactant amounts, or wherein the control parameter determination unit is configured to determine these control parameters by minimizing an amount expressible by the following function: Under the following constraints And Among them, similarly, contains the individual amounts of P reactants in K reactors at time t, contains the cost of the P reactants, b ∈ {0, 1} K indicates whether the K reactors are active, y(t) refers to the individual product amounts of the K reactors determined by these trained artificial intelligences for these individual reactant amounts X(t), y refers to the desired minimum combined product amount, s1 and s2 are predefined constants, contains the measured individual reactant amounts, and contains the predefined limits of these individual reactant amounts, and is a predefined weight.

12. The system according to any one of the preceding claims, wherein, The training of the artificial intelligence (21, 22, ..., 2K) is performed using the following loss function: Or is performed using the following loss function: wherein, refers to the estimated single product amount of the K reactors (11, 12,..., 1K) at a given time t, y(t) refers to the single product amount at time t as determined by K artificial intelligences (21, 22,..., 2K), and α is a predefined training parameter.

13. A method (200) for determining the individual product amounts of a plurality of chemical reactors (11, 12,..., 1K) contributing to the combined product amount, wherein, The method includes: - Provide (201) the measured single reactant amounts for each of the plurality of chemical reactors, - Provide (202) trained artificial intelligence for each of these chemical reactors, wherein the provided trained artificial intelligence is trained to provide, as output, a single product amount for these chemical reactors when receiving a single reactant amount for these chemical reactors as input, and these single product amounts are combined into a combined product amount associated with these single reactant amounts received as input, and - Determine (203) the single product amounts for the plurality of chemical reactors based on these measured single reactant amounts and these trained artificial intelligence.

14. A computer program for determining the individual product amounts of a plurality of chemical reactors contributing to the amount of a combined product, wherein, The program includes program code means for causing the system according to claim 1 to perform the method according to claim 13.

Images