Method and system for operating an electrochemical reactor

Abstract

A method for operating an electrochemical reactor is disclosed. The reactor includes an anode and a cathode as electrodes. Further, the reactor is configured to receive a starting material feed, to electrochemically react the starting material in the presence of the catalyst coated electrode to produce a product, and to output the product. The reactor is operated based on a set of operating parameters including at least a voltage across an electrode of the reactor and a dosing scheme for feeding a catalyst solution and / or a catalyst precursor solution to a starting material. The method includes receiving a reactor model describing a behavior of the reactor including a relationship between at least a voltage across an electrode and a dosing regime in which a catalyst solution and / or a catalyst precursor solution is fed to a starting material, an optimized feeding scheme is determined by optimizing a feeding scheme in the feeding of a catalyst solution and / or a catalyst precursor into a starting material using a reactor model to minimize the voltage, and the optimized feeding scheme for use in operating the reactor is output. In addition, the invention discloses a corresponding system as well as a computer program product and a computer readable storage medium.

Description

[0001] The present disclosure relates to a method for operating an electrochemical reactor, such as a hydrochloric acid electrolysis cell, a corresponding system, a computer program product and a computer readable storage medium.

[0002] In the production of isocyanates by phosgenation of the corresponding amines, hydrogen chloride is produced as a by-product. The recovery of chlorine by electrolysis of hydrochloric acid formed from hydrogen chloride offers the opportunity to close the chlorine cycle. This is very desirable, especially from the perspective of sustainability.

[0003] When electrolyzing aqueous hydrochloric acid in diaphragm cells, it is known to reduce the cell voltage by adding metal salts to the electrolyte. During the electrolysis, these metals are deposited on the cathode of the electrolyzer. The resulting layer reduces the hydrogen overvoltage. For example, US 3 336 209 describes a method for reducing the hydrogen overvoltage in the electrolysis of aqueous hydrochloric acid to form hydrogen and chlorine in a diaphragm electrolyzer by adding a metal salt to the electrolyte, wherein the metal salt is a palladium salt, added in an amount providing 0.2 to 2 grams of palladium per square meter of active electrode surface.

[0004] Experience has shown that in order to maintain high performance, the metal deposited on the cathode needs to be replenished during the operation of the electrolyzer. The basic options for replenishment are continuous and discontinuous addition of metal salts. Continuous dosing of metal salt solution into the starting material feed may not be the optimal operating mode, especially if the desired production output of the electrolyzer varies over time. In the discontinuous case, addition at regular predetermined intervals may also not achieve the best effect of the cathode coating, because saturation of the cathode does not lead to a further reduction in the hydrogen overvoltage, while the cost of the metal salt is still incurred.

[0005] Therefore, there is still a need for a data-driven model for timely addition of the metal salt solution to the electrolyte of the electrolyzer.

[0006] EP 3 696 619 A1 discloses a system and method for determining operating conditions of a chemical production plant including at least one catalytic reactor. Operating data and a catalyst age indicator are received via a communication interface. At least one target operating parameter of the operating conditions of a preset production run or a current production run is determined. The at least one target operating parameter of the operating conditions can be used to monitor and / or control the chemical production plant.

[0007] CA3150624A1 relates to a method for monitoring the operating characteristics of an industrial gas equipment complex including a plurality of industrial gas equipment. The method involves assigning a machine learning model to each industrial gas equipment, training the corresponding machine learning model based on historical time-dependent operating characteristic data, and executing the trained machine learning model to predict the operating characteristics for a predetermined future time period. In addition, the predicted operating characteristic data can be compared with the measured operating characteristic data to identify deviations in the performance of the industrial gas equipment and determine predicted future resources, future failures and / or predicted future maintenance.

[0008] CN113780672A describes an artificial intelligence-based method for optimizing electrolyzer parameters. The method focuses only on obtaining historical production data generated during the electrolyzer production process, obtaining process parameters that affect current efficiency and daily alkali output, and using principal component analysis to obtain the weight ranking of the impact of process parameters on KPI indicators (such as current efficiency, daily alkali output, and electricity consumption per ton of alkali). A mathematical model is established based on KPI indicators, weight ranking, and control parameters.

[0009] The object of the present disclosure is to provide a method and system that supports optimization of a dosing schedule of a catalyst solution and / or a catalyst precursor solution for operation of an electrochemical reactor.

[0010] This object is achieved by a method according to claim 1 and a system according to claim 13. A computer program product is the subject of claim 14 and a computer-readable storage medium is the subject of claim 15. Preferred embodiments are the subject of the dependent claims. They are freely combinable unless the context clearly indicates otherwise.

[0011] Therefore, the present disclosure provides a method for operating an electrochemical reactor. The reactor may be one of one or more reactors in several reactors, which are electrically connected to each other and / or can receive their starting material feeds in parallel. The reactor includes an anode and a cathode as electrodes. The reactor is configured to cause the starting material to undergo an electrochemical reaction to generate a product in the presence of a catalyst-coated electrode, and the reactor is configured to receive the starting material feed and output the product. The reactor is operated based on a set of operating parameters, which at least include the voltage on the electrodes of the reactor and the feeding scheme of the catalyst solution and / or catalyst precursor solution into the starting material feed. The set of operating parameters may also include the current flowing to the reactor electrodes.

[0012] Preferably, the anode and cathode are separated from each other by a separator, in particular a membrane or a diaphragm cloth, which is mainly known in the art.

[0013] According to the present invention, the method comprises:

[0014] receiving a reactor model, the reactor model describing the behavior of the reactor including at least the relationship between the voltage on the electrodes and the dosing schedule of the catalyst solution and / or catalyst precursor solution into the starting material feed,

[0015] determining an optimized dosage by optimizing the dosage of catalyst solution and / or catalyst precursor to the starting material feed using a reactor model to minimize voltage, and

[0016] The optimized feed schedule is output for use in running the reactor.

[0017] According to the present invention, the step of receiving the reactor model comprises:

[0018] Receive reactor metadata, where the reactor metadata includes the following:

[0019] the age of the components in the reactor, and at least one of the type of components in the reactor, the number of components in the reactor, and the geometry of the reactor, or

[0020] at least one of the type of components in the reactor, the number of components in the reactor, and the geometric configuration of the reactor,

[0021] receiving sensor data representing the current state of the reactor, and

[0022] The reactor model is generated by training a machine learning model based on the reactor metadata and the sensor data, wherein the machine learning model models at least the voltage on the electrode.

[0023] In the context of the present disclosure, the term "electrochemical reactor" shall refer to a reactor in which electrical energy is generated by a chemical reaction or is used to drive a chemical reaction (electrolysis). An example of an electrochemical reactor is a filter-press reactor with parallel plate electrodes. Other but less common examples are stirred tanks with rotating cylindrical electrodes. The catalyzed electrochemical reactions and their scale do not limit the scope of the present disclosure. For example, the present disclosure may be applicable to the industrial electrolysis of water, and to the generation of electrical energy by portable fuel cells used in homes and cars.

[0024] The present disclosure uses a reactor model that describes at least part of the reactor behavior. This behavior includes the relationship between the voltage on at least the electrode and the dosing scheme of the catalyst solution and / or catalyst precursor solution. The reactor model can be generated based on the reactor in operation, such as by training the model with historical data. However, the reactor model can also be pre-calculated and used in the optimization step. When there is no sensor data and / or other detailed information about the reactor in operation (such as at the beginning of the process), the latter method is particularly useful. When the domain knowledge about the process allows the same model to be used over a long period of time until the model needs to be updated to the changing conditions, this method may also be useful. In this case, an initial, possibly simpler model can be constructed without training with historical data, such as formulating a linear equation based on domain knowledge, which represents the dependence of voltage on different factors (such as temperature, electrolyte concentration and current). It should be noted that linear models do not necessarily produce simple optimization problems, i.e., the boundaries of the domain. Optimization needs to take into account the current price of the catalyst precursor solution, which usually means that optimization needs to strike a balance between catalyst cost and power saving. In both cases, i.e., using a pre-computed model or a generated model, the available reactor model may be updated during operation, for example based on the duration of the operation, measurements, etc. It should be noted that the present disclosure is not limited to the use of a single reactor model. The present disclosure may also use two or more reactor models to model the behavior of the reactor. For example, this may be helpful if the reactor is modeled in different operating states, or if several reactor models relate to different aspects of the reactor operation.

[0025] In one embodiment, the reactor model may also include a dependency on the current flowing to the electrode. In this case, the modeling behavior may include at least the voltage on the electrode, the current flowing to the electrode, and the relationship between the charging scheme of the catalyst solution and / or the catalyst precursor solution. If the current is applied, for example, by a constant current power supply, the dependency on the current can be replaced by a constant. It should be noted that "flowing to the electrode" should also include the current flowing out of the electrode, because the flow direction can be reflected by a signed number or an unsigned number.

[0026] In the present disclosure, optimization may refer to any mathematical function or method that aims to find the minimum or maximum of an objective function, preferably taking into account given parameter bounds. The minimum / maximum may be a global minimum / maximum. Typical optimization methods include basin hopping algorithms, differential evolution algorithms, dual annealing algorithms, simple homology global optimization algorithms, and trivial brute force approaches, to name just a few that may be useful.

[0027] In this case, the objective function can be represented by a reactor model. In one embodiment, the optimization algorithm can manually generate the input parameter group of the model and trigger the model to predict each group of input parameters. The generation of the input parameter group can occur randomly, for example, with a guide / directional iteration step or a combination thereof. Preferably, the algorithm is limited to the input parameters generated only within the reasonable boundaries of the operating conditions. The prediction generated by the model can be saved together with the corresponding parameter group for predicting. For example, an optimized target can be the cost optimization of the catalyst addition frequency to a reactor. The optimization can be carried out by generating different catalyst addition modes and frequencies for a fixed desired operating condition group. Then, the optimization function can trigger the prediction of the voltage of each group of catalyst addition frequency, calculate the cost of applying the catalyst under a given frequency and compare it with the electricity cost, which is obtained by multiplying the current with the voltage and the electricity price. The optimization can also find out the minimum charge amount of the catalyst solution, the minimum wear of the reactor or the maximum gain of the product, just to name a few non-limiting examples.

[0028] The optimization step (which can also be considered as an optimization submethod) can receive one or more optimization targets and one or more operating parameters with one or more boundaries for optimization. For example, a possible optimization target selection can be provided to an operator and received by the optimization submethod.

[0029] For example, a single electrolyzer, a subset of electrolyzers, or an entire cell chamber can be optimized. In some cases, differences in reactor design may require building a separate model for each reactor design. Each reactor type can then be optimized separately.

[0030] In the case of one or more optimization objectives, one or more optimal solutions can be determined, which may or may not include a global optimal solution. One or more optimal solutions can be provided to a monitoring system and can be displayed on a display device. In this case, an operator of a production facility including (one or more) operating reactors can choose between the provided solutions and different scenarios for future operation, or can directly trigger process changes through an electronic process control device. In another exemplary scenario, (one or more) optimal solutions are submitted as signals to the electronic process control device and can directly affect the operation of the equipment. In another exemplary scenario, these electronic signals may require the approval of authorized personnel before changing the operation of the equipment.

[0031] In one embodiment, the reactor is one of a plurality of reactors, wherein the plurality of reactors are electrically connected to each other and / or receive their respective starting material feeds in parallel, and determining the optimized feeding scheme includes optimizing the feeding scheme of each of the plurality of reactors and determining the optimized feeding scheme by combining the optimized feeding schemes of each of the plurality of reactors.

[0032] In this case, there may be multiple models and / or distributed optimization for a single reactor or a subset of reactors. If all reactors only require a single catalyst feed, then integrating the optimized charging scheme may be useful. In a preferred embodiment, integrating different optimization results can take the form of averaging the optimal charging frequency for different reactors. Other strategies may be to use catalyst only with minimum frequency or maximum frequency, which depends on, for example, possible cost restrictions.

[0033] According to the present invention, the step of receiving the reactor model comprises:

[0034] Receive reactor metadata, where the reactor metadata includes the following:

[0035] the age of components in the reactor and at least one of the type of components in the reactor, the number of components in the reactor and the geometric configuration of the reactor, or at least one of the type of components in the reactor, the number of components in the reactor and the geometric configuration of the reactor,

[0036] receiving sensor data representing the current state of the reactor, and

[0037] A reactor model is generated by training a machine learning model based on reactor metadata and sensor data, wherein the machine learning model models at least a voltage on an electrode.

[0038] Therefore, this method can be regarded as a sequence of model building submethod and optimization submethod.It should be understood that the order of submethods is not crucial.The model building submethod can be performed before the optimization submethod is performed.In this case, the optimization will use the reactor model generated by the model building submethod so that the optimization can be based on the latest reactor model.However, the optimization submethod can also be performed before the execution of the model building submethod.In this case, the optimization submethod can use the reactor model received earlier, and the model building submethod can use the optimized charging scheme, and can be regarded as updating the reactor model for the next optimization submethod.

[0039] In this model building sub-method, two types of data are received: metadata and sensor data. Metadata refers to data specific to the reactor configuration. Different types of reactor configurations are known from the prior art, such as Fluorinated Ionomers (W. Grot, 2011, 2nd edition, Elsevier, Chapter 5.1, Electrolysis). At a high level, the most prominent geometric configuration of the reactor may be an Uhde tank (including independently sealed single tanks), or other types of monopolar electrolyzers and bipolar electrolyzers (which are sealed by forces applied to the ends of the tank stack), or other geometric configurations with different inter-electrode gap sizes (e.g., finite gap, zero gap, hybrid) and different liquid and gas inlet and outlet designs. For a filter press bipolar electrolyser, the metadata preferably includes the following information: information on the number of cells in the electrolyser, component design information, such as electrode design (e.g., electrode thickness, louvered anode, perforated "C" profile anode, etc.), component type information (e.g., diaphragm type or frame type), component material information (e.g., electrode material or electrode coating material, or age information of these individual components. In general, the term metadata may emphasize that this type of data provides information about the status of the electrochemical reactor / all electrochemical reactors and related components in the process. In general, metadata is not sensitive to the way the equipment is operated. There may be exceptions, such as when components in a reactor are known to age faster when they are operated at maximum load or are frequently started and stopped. This may be taken into account in the metadata.

[0040] Metadata can include the physical location of each reactor (e.g., represented by an identification number) and the operating time period of each reactor. Based on this information, a model can be constructed that models performance as a function of the reactor age, the number, type and age of components in the reactor, and the geometric configuration of the reactor. In a preferred embodiment, the metadata group is stored in the form of a flat table, wherein each row can contain all metadata available for an electrolyzer within a given time range. In this case, the time range can be specified by the date when a new (or repaired) electrolyzer is started at a given location and the date when the operation of a given electrolyzer is terminated (e.g., due to maintenance) at a given physical location. For a given row in the data group, all metadata (including location, operation start date, and end date) can be stored in a separate column.

[0041] The inventors have found that the material and type of the diaphragm, the material and type of the frame, the age of the frame, diaphragm and electrodes have a large impact on the variance of the predicted voltage. The normal mean absolute error in voltage is about 20 mV per single cell (normalized voltage over a one month period). 50 mV is typically the accuracy that a skilled technician operating such equipment will achieve. The inventors have observed that ignoring the mentioned reactor metadata results in a jump in the mean absolute error from about 20 mV to about 150-250 mV.

[0042] The sensor data comprises (one or more) measured values ​​of (one or more) sensors measuring one or more physical values ​​in (one or more) reactors or (one or more) components connected to (one or more) reactors. For example, this sensor / these sensors may include a temperature sensor, a current sensor, a voltage sensor when the electrolyte flows into the electrolyzer, or a sensor that records the timing and / or amount of catalyst (precursor) added to the starting material feed. The sensor data may also include information that can reconstruct the measurement time and / or the order of the sensor data. In one embodiment, the sensor data is time-stamped.

[0043] Sensor data for at least one reactor and at least two separate operating time points may be received. Considering the combination of sensor data and metadata, there may also be a set of metadata corresponding to at least one reactor, wherein the metadata for this reactor covers at least two separate operating time points. In a preferred embodiment, the sensor data set is stored in the form of a flat table, wherein a row contains all available sensor data in a separate column. The time stamp when the sensor data was collected and the physical location of the sensor may be stored in a separate column.

[0044] In some application scenarios, sensor data is acquired asynchronously. In one embodiment, the data of each sensor can be interpolated to support calculation of interpolation for all sensors at a common sampling frequency (e.g., in whole minutes, whole hours, etc.).

[0045] When generating a reactor model, the received metadata and the received sensor data can be used to train the machine learning model. This step can be regarded as the parameterization of the model to (better) fit the modeled reactor. It should be noted that training a machine learning model may also include training two or more machine learning models. In the present disclosure, a machine learning model may refer to any mathematical function that can describe the behavior of a reactor and map one or more input variables to one or more target variables. The mathematical function can be iteratively adjusted by a learning algorithm in order to optimize the fit of the function to historical data (training data). The historical data can be derived from the operation of an electrochemical reactor similar to the reactor to be modeled. Typical machine learning algorithms can include multiple linear regression, support vector machines, decision tree-based methods (such as random forest algorithms), neural networks, and the like.

[0046] In one embodiment, generating a reactor model includes:

[0047] Generate a combined dataset by combining reactor metadata and sensor data,

[0048] A subset of the combined dataset is generated, the subset including at least the voltages on the electrodes, and a machine learning model is trained using the subset.

[0049] When generating a combined data set, the metadata and sensor data of the corresponding reactor can be combined using unique identifiers present in both the metadata and the sensor data. Other methods of combining metadata and sensor data can be used, such as by storing the data sets under specific file names and then combining the data based on certain rules that the files belong to the same.

[0050] Generating a combined data set may include matching the physical location where the sensor data was collected, as well as the time stamp of the sensor data, with the corresponding parameter groups from the metadata. In a preferred embodiment, the time stamp ("Sensor"-Time Stamp) and the physical location of the electrolyzer ("Sensor"-Location) are used to "look up" the corresponding row in the metadata table, where the "Sensor"-Time Stamp is within the range specified by the run start date and end date. Subsequently, the two rows are merged. In a preferred embodiment, three additional values ​​are calculated from the "Sensor-Date" and added to separate columns of the merged row: the age of the electrolyzer, the age of the frame, and the age of the graphite electrode. Finally, the merged row is stored in a list together with all other merged rows. The list of merged rows is then converted into an array-like data structure for further processing.

[0051] There may be practical reasons for dividing the data into a subset containing the voltage and a subset containing the remaining data, and may prevent the model from being able to predict the voltage using the voltage as an input variable. In this case, the term model may refer to a data-driven machine learning model or some other form of data-driven mathematical model that is parameterized so as to be able to reproduce the voltage of the electrochemical cell or stack of electrochemical cells from the remaining data set.

[0052] Subsets can be generated by dividing the combined data set into two or more parts. One or more parts can be used for the selection and training of a suitable machine learning algorithm, so it can be called a training data set. (One or more) other parts can be used as a test data set to verify the ability of the machine learning model to generalize to new data. It is worth noting that in order to obtain a realistic estimate of future performance, the time order should be maintained throughout the process of processing the entire data set until it is divided into a training set and a test set. Due to the stability of sensor readings on an industrial scale, the shuffling of data points will result in data points in the test set that are very similar to some data points in the training data set. This may lead to undetectable "overfitting", which will manifest as very similar prediction errors in both the training data set and the test data set.

[0053] In a preferred embodiment, the split is performed according to the general scheme of cross validation of time series. The size of a single test set can be selected to match the desired level of prediction. For example, if a production expert needs to make predictions every week, each test set can be selected to cover one week. In fact, the algorithm then uses the first week of the data set as a training set and the second week as a test set. After training with the data from the first week, the algorithm will continue to use the first two weeks as training sets and the third week as a test set. Continuing in this mode, the algorithm will eventually reach all points of its training except the last week. In order to obtain performance estimates, one possibility is to average the errors made for all test sets. Alternatively, in some cases, it is possible to choose to use any previous model for the optimization task.

[0054] In one embodiment, the sensor data indicates at least one of the following: current consumption of the reactor, voltage on the electrodes of the reactor, inflow of starting material into the reactor, outflow of product from the reactor, concentration of starting material in the solvent introduced into the reactor, and dosing schedule of catalyst solution and / or catalyst precursor solution into the starting material feed. The sensor data can provide detailed information about the status and / or behavior of (one or more) reactors.

[0055] In one embodiment, two machine learning models are constructed to capture the different behaviors of iridium coated electrolyzers and uncoated electrolyzers. In a preferred embodiment, both models are constructed using a random forest algorithm.

[0056] According to a preferred embodiment, the sensor data columns that store information about the timing and / or amount of catalyst feeding are further processed and / or aggregated. In this way, a finer-grained adsorption / desorption behavior of the catalyst can be captured by a machine learning model. In one embodiment, the amount of catalyst fed can be summed up per day during a limited time interval before the corresponding time mark. Additional columns can be created to store the aggregated feeding events, thereby providing information about how much catalyst was fed in different time intervals before the corresponding data point.

[0057] In another embodiment, if a machine learning model capable of capturing time-dependent dynamic behavior is used, the aforementioned aggregation of time-dependent catalyst dosing events may not be necessary. Examples of implementing such machine learning models include, but are not limited to, recurrent neural networks and convolutional neural networks.

[0058] In one embodiment, outputting the optimized charging scheme comprises communicating the optimized charging scheme to the operator and communicating the optimized charging scheme to at least one of the process control systems of the reactor. If some operation constraints must be met and / or if a qualified selection is required to be performed to the optimized charging scheme, then communicating the optimized charging scheme to the operator may be beneficial. Communicating the optimized charging scheme to the process control system may support the use of the optimized charging scheme quickly and automatically when operating (one or more) reactors.

[0059] In one embodiment, the method further comprises feeding the catalyst solution and / or catalyst precursor solution into the starting material feed according to an optimized feeding scheme. In this way, determining the optimized feeding scheme can have a direct impact on the process in the (one or more) reactors.

[0060] In one embodiment, the reactor is one of a plurality of reactors, the plurality of reactors being electrically connected to each other, wherein the starting material feed is a common starting material feed for at least a portion of the reactors, and wherein the reactors receive the starting material feed in parallel. In this way, the reactor can be part of an apparatus and can operate more efficiently. In addition, the use of a plurality of reactors can increase the output of the system.

[0061] In one embodiment, the electrochemical reaction includes reduction of hydrogen cations to hydrogen gas and / or reduction of oxygen to hydroxide ions.

[0062] In one embodiment, the electrochemical reaction includes oxidation of chloride anions to chlorine gas and / or oxidation of oxygen anions to oxygen gas.

[0063] In one embodiment, the starting material comprises aqueous hydrochloric acid and / or aqueous sodium chloride.

[0064] In one embodiment, the catalyst comprises nickel, palladium, platinum, cobalt, rhodium, iridium, copper, antimony, silver, molybdenum or a combination of at least two of the foregoing metals, particularly platinum and iridium, and the selection of a specific catalyst material may be based on cost considerations.

[0065] In one embodiment, determine to optimize charging scheme and / or receiving reactor model and carry out repeatedly, preferably regularly.In this way, model and / or charging scheme can adapt to the changing behavior of (one or more) reactors, for example, due to aging or change of starting material feeding.The frequency of repetition may depend on the dynamics of operation.It may be carried out per hour, per day, by a few days, by week or by month, just to name a few.In one embodiment, determine to optimize charging scheme and / or receiving reactor model and carry out once per hour or shorter.In another embodiment, determine to optimize charging scheme and / or receiving reactor model and carry out once per month or shorter.

[0066] The present disclosure also encompasses a system for operating an electrochemical reactor,

[0067] The reactor comprises an anode and a cathode as electrodes, and is configured to receive a starting material feed, electrochemically react the starting material in the presence of a catalyst-coated electrode to generate a product, and output the product.

[0068] The system is configured to perform the method according to the present disclosure and comprises:

[0069] a reactor model processing entity configured to receive and / or update and / or generate a reactor model, the reactor model describing the behavior of the reactor including at least the relationship between the voltage on the electrodes and the dosing scheme of the catalyst solution into the starting material feed,

[0070] a feed optimizer configured to optimize a feed schedule of a catalyst solution into a starting material feed using a reactor model to minimize voltage, and

[0071] The output interface is configured to output a feeding plan for use in running the reactor.

[0072] According to one embodiment, the anode and cathode are separated from each other in the reactor by a separator, preferably a membrane or a diaphragm cloth.

[0073] The present disclosure also covers a computer program product comprising instructions, which, when a computer executes the program, cause the computer to perform the method according to the present disclosure; and a computer-readable storage medium comprising instructions, which, when a computer executes the instructions, cause the computer to perform the method according to the present disclosure.

[0074] The invention will be further described with reference to the following drawings, without wishing to limit the disclosure to the described embodiments.

[0075] Figure 1 A flow chart of a sub-method according to the present disclosure is shown.

[0076] Figure 2 A flow chart of another sub-method according to the present disclosure is shown.

[0077] Figure 3 A data processing device system according to the present disclosure is shown.

[0078] Figure 1 and Figure 2 The embodiments shown in assume that the steps of the method of operating one or several or more electrochemical reactors can be grouped into a model building sub-method and an optimization sub-method. The model building sub-method comprises steps for generating a model of the electrochemical reactor or several / multiple electrochemical reactors.

[0079] Figure 1 A flow chart of a model building sub-method 100 is shown. This model building sub-method 100 is executed at a predetermined frequency X, for example once an hour, once a day, once a week, or once every four weeks.

[0080] Ideally, the method begins at a starting point 110. Next, at step 120, data is received from a data source 150 for sensor data and a data source 160 for system descriptors. The reception of sensor data and system descriptor data may occur simultaneously or sequentially, as appropriate.

[0081] The data source 160 may be a database or a plurality of databases.The system descriptor includes reactor metadata indicating the number, type and age of components in the reactor, the geometry of the reactor and identifiers.

[0082] The model built in the model building sub-method 100 is available in a box 140 labeled “Saved Model Instance”.

[0083] Figure 2 An important part of this sub-method is the optimization sub-method 200, which is executed at a predetermined frequency X, which can be related to Figure 1The frequency X in is defined similarly. The optimization sub-method 200 starts at step 210 and continues at step 300. Step 300 is performed for each machine (ie for each reactor).

[0084] At step 300, a saved model instance is received (block 250). The saved model can be accessed by Figure 1 The model building sub-method shown in (and provided in box 140), or can be generated based on conventional domain knowledge of the (one or more) reactors being operated. In addition, in step 300, sensor data is received from data source 260, cost parameters for optimization are received from data source 270, and system descriptors are received from data source 280. In step 300, the model loaded from box 250 can be a fitted model instance. In addition to the received data, step 300 can also use data generated during execution of previous time periods. The optimization goal of executing step 300 is to optimize (i.e., minimize) the voltage measured at the electrodes of each reactor examined. In Figure 2 In the embodiment of, three different optimization scenarios are determined: scenario A, scenario B and scenario C. Each scenario can reflect different operation and / or optimization strategies. The corresponding optimization results are output in steps 310, 320 and 330. When each reactor is optimized, the optimization results of different scenarios (A, B, C, ...) are combined in step 220. In addition, the result of this combination is output as an optimized charging scheme for use when the reactor is running. It can be output to an operator or an automated process control system. In step 230, the recommended optimized operating conditions are implemented, and the reactor is operated with the optimized charging scheme. The method ends in step 240.

[0085] Figure 3 A data processing device system 1000 configured to perform a method according to the present disclosure is shown. In some embodiments, the system 1000 may include one or more servers 1010. The (one or more) servers 1010 may be configured to communicate with one or more client computing platforms 1070 according to a client / server architecture and / or other architectures. The (one or more) client computing platforms 1070 may be configured to communicate with other client computing platforms through the (one or more) servers 1010 and / or according to a peer-to-peer architecture and / or other architectures. Users may access the system 1000 through the (one or more) client computing platforms 1070.

[0086] The server(s) 1010 may be configured via machine readable instructions 1040. The machine readable instructions 1040 may include one or more instruction modules. The instruction modules may include computer program modules. The instruction modules may include one or more of a model building sub-method module 1050, an optimization sub-method module 1060, and / or other instruction modules.

[0087] The model building sub-method module 1050 may be configured to perform the model building step of the method according to the present disclosure. The optimization sub-method module 1060 may be configured to perform the optimization step of the method according to the present disclosure.

[0088] In some embodiments, server(s) 1010, client computing platform(s) 1070, and / or external resources 1080 (e.g., databases) may be operably linked via one or more electronic communication links. For example, such electronic communication links may be established at least in part via a network (e.g., the Internet and / or other networks). It is to be understood that this is not intended to be limiting, and the scope of the present disclosure includes embodiments in which server(s) 1010, client computing platform(s) 1070, and / or external resources 1080 may be operably linked via some other communication medium.

[0089] A given client computing platform 1070 may include one or more processors configured to execute computer program modules. The computer program modules may be configured to enable an expert or user associated with a given client computing platform 1070 to interact with the system 1000 and / or external resources 1080, and / or to provide other functionality attributed herein to the client computing platform(s) 1070. By way of non-limiting example, a given client computing platform 1070 may include one or more of a desktop computer, a laptop computer, a handheld computer, a tablet computing platform, a netbook, a smartphone, and / or other computing platforms.

[0090] External resources 1080 may include information sources external to system 1000, external entities participating in system 1000, and / or other resources. In some implementations, some or all of the functionality attributed herein to external resources 1080 may be provided by resources included in system 1000.

[0091] Server(s) 1010 may include electronic storage 1020, one or more processors 1030, and / or other components. Server(s) 1010 may include communications links or ports to enable information to be exchanged with a network and / or other computing platforms. Figure 3 The description of the server(s) 1010 in the description is not intended to be limiting. The server(s) 1010 may include a number of hardware, software, and / or firmware components that operate together to provide the functionality attributed herein to the server(s) 1010. For example, the server(s) 1010 may be implemented by a computing platform cloud operating with the server(s) 1010.

[0092] Electronic storage 1020 may include non-transitory storage media that electronically stores information. The electronic storage media of electronic storage 1020 may include one or both of system memory that is provided integrally with (i.e., substantially non-removable) server(s) 1010 and / or removable storage that is removably connected to (i.e., via, for example, a port (e.g., a USB port, a FireWire port, etc.) or a drive (e.g., a disk drive, etc.). Electronic storage 1020 may include one or more of optically readable storage media (e.g., optical disks, etc.), magnetically readable storage media (e.g., magnetic tape, a magnetic hard drive, a floppy disk drive, etc.), charge-based storage media (e.g., EEPROM, RAM, etc.), solid-state storage media (e.g., a flash drive, etc.), and / or other electronically readable storage media. Electronic storage 1020 may include one or more virtual storage resources (e.g., cloud storage, a virtual private network, and / or other virtual storage resources). Electronic storage 1020 may store software algorithms, information determined by processor(s) 1030, information received from server(s) 1010, information received from client computing platform(s) 1070, and / or other information that enables server(s) 1010 to function as described herein.

[0093] Processor(s) 1030 may be configured to provide information processing capabilities in server(s) 1010. Thus, processor(s) 1030 may include one or more of a digital processor, an analog processor, a digital circuit designed to process information, an analog circuit designed to process information, a state machine, and / or other mechanisms for electronically processing information. Figure 3 1030 is shown as a single entity, but this is for illustrative purposes only. In some embodiments, processor(s) 1030 may include multiple processing units. These processing units may be physically located within the same device, or processor(s) 1030 may represent processing functions of multiple devices operating in coordination. Processor(s) 1030 may be configured to execute module 1050, module 1060, and / or other modules via software; hardware; firmware; some combination of software, hardware, and / or firmware; and / or other mechanisms for configuring processing capabilities on processor(s) 1030. As used herein, the term "module" may refer to any component or group of components that perform the functions attributed to a module. This may include one or more physical processors, processor-readable instructions, circuits, hardware, storage media, or any other component during execution of processor-readable instructions.

[0094] It should be understood that although Figure 3 Modules 1050 and 1060 are shown as being implemented in a single processing unit, but in embodiments where processor(s) 1030 include multiple processing units, one or more of modules 1050 and / or 1060 may be implemented remotely from other modules. The descriptions of the functionality provided by the different modules 1050 and 1060 described herein are for illustrative purposes and are not intended to be limiting, as any module may provide more or less functionality than described. For example, one or more of modules 1050 and / or 1060 may be eliminated, and some or all of their functionality may be provided by other modules in modules 1050 and / or 1060. As another example, processor(s) 1030 may be configured to execute one or more additional modules that may perform some or all of the functionality attributed to one of modules 1050 and / or 1060 below.

Claims

1. A method for operating an electrochemical reactor, The reactor comprises an anode and a cathode as electrodes, and is configured to receive a starting material feed, electrochemically react the starting material in the presence of a catalyst-coated electrode to generate a product, and output the product. wherein the reactor is operated based on a set of operating parameters, the operating parameters comprising at least the voltage on the electrodes of the reactor and the feeding scheme of the catalyst solution and / or catalyst precursor solution into the starting material feed, The method comprises: a receiving reactor model, which describes the behavior of the reactor, including at least the relationship between the voltage on the electrodes and the dosing schedule of the catalyst solution and / or catalyst precursor solution into the starting material feed, wherein the receiving reactor model comprises: Reactor metadata is received, wherein the reactor metadata includes the following: the age of components in the reactor, and at least one of the type of components in the reactor, the number of components in the reactor, and the geometry of the reactor, or at least one of the type of components in the reactor, the number of components in the reactor, and the geometric configuration of the reactor, receiving sensor data representing a current state of the reactor, and generating a reactor model by training a machine learning model based on the reactor metadata and the sensor data, wherein the machine learning model models at least a voltage on an electrode, - determining an optimized dosing schedule by optimizing the dosing schedule of the catalyst solution and / or catalyst precursor into the starting material feed using a reactor model to minimize the voltage, and - Outputting an optimized dosing schedule for use in operating the reactor.

2. The method according to claim 1, wherein the reactor is one of a plurality of reactors, the plurality of reactors being electrically connected to each other and / or receiving their respective starting material feeds in parallel, and wherein determining an optimized feeding scheme comprises optimizing the feeding scheme of each of the plurality of reactors and determining the optimized feeding scheme by combining the optimized feeding schemes of each of the plurality of reactors.

3. The method according to claim 1 or 2, wherein generating a reactor model comprises: Generate a combined dataset by combining reactor metadata and sensor data, generating a subset of the combined data set, the subset comprising at least the voltages on the electrodes, and A machine learning model is trained using the subset.

4. According to the method according to any of the preceding claims, the sensor data indicates at least one of the following: the current consumption of the reactor, the voltage on the electrodes of the reactor, the inflow of starting material into the reactor, the outflow of product from the reactor, the concentration of starting material in the solvent introduced into the reactor, and the dosing scheme of catalyst solution and / or catalyst precursor solution into the starting material feed.

5. The method according to any one of the preceding claims, wherein outputting the optimized dosing plan comprises at least one of: communicating the optimized dosing plan to an operator, and communicating the optimized dosing plan to a process control system of the reactor.

6. The method according to any one of the preceding claims, further comprising dosing a catalyst solution and / or a catalyst precursor solution into the starting material feed according to an optimized dosing schedule.

7. The method according to any one of the preceding claims, wherein the reactor is one of a plurality of reactors, the plurality of reactors being electrically connected to each other, wherein the starting material feed is a common starting material feed to at least a portion of the reactors, and wherein the reactors receive the starting material feed in parallel.

8. A method according to any one of the preceding claims, wherein the electrochemical reaction comprises the reduction of hydrogen cations to hydrogen gas and / or the reduction of oxygen to hydroxide ions.

9. A method according to any one of the preceding claims, wherein the electrochemical reaction comprises oxidation of chloride anions to chlorine gas and / or oxidation of oxygen anions to oxygen gas.

10. A process according to any one of the preceding claims, wherein the starting material comprises aqueous hydrochloric acid and / or aqueous sodium chloride solution.

11. The process of any one of the preceding claims, wherein the catalyst comprises nickel, palladium, platinum, cobalt, rhodium, iridium, copper, antimony, silver, molybdenum, or a combination of at least two of the foregoing metals.

12. The method according to any of the preceding claims, wherein the determination of the optimized feeding scheme and / or the receiving reactor model is performed repeatedly, preferably regularly.

13. A system for operating an electrochemical reactor, The reactor comprises an anode and a cathode as electrodes, and is configured to receive a starting material feed, cause the starting material to undergo an electrochemical reaction in the presence of a catalyst-coated electrode or a catalyst-deposited electrode to generate a product, and output the product. The system is configured to perform the method according to any one of claims 1 to 12 and comprises: a reactor model processing entity configured to receive and / or update and / or generate a reactor model, the reactor model describing the behavior of the reactor including at least the relationship between the voltage on the electrodes and the dosing scheme of the catalyst solution and / or catalyst precursor solution into the starting material feed, a feed optimizer configured to optimize a feed schedule of a catalyst solution into a starting material feed using a reactor model to minimize voltage, and An output interface is configured to output a feeding plan for use in running the reactor.

14. Computer program product comprising instructions which, when a computer executes said program, cause the computer to perform the method according to any one of claims 1 to 12.

15. A computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to perform the method according to any one of claims 1 to 12.

Images