Predicting Coking and Fouling Progression to Improve Production Scheduling in Chemical Production Plants

Abstract

To predict the future evolution of the health state of equipment and / or processing units of a chemical production plant such as a steam cracker, a computer-implemented method is provided that builds a data-driven model of future key performance indicators based on current key performance indicators, current process conditions, and process conditions within a prediction time horizon.

Description

Technical Field

[0001] The present invention relates to a computer-implemented method and apparatus for predicting the progression of degradation in equipment of a chemical production plant. Background of the Invention

[0003] Chemical production plants can suffer from coking or other fouling processes. An example of such a chemical production plant is a steam cracker. The steam cracker uses steam to "crack" crude oil, products such as naphtha or liquefied petroleum gas (LPG) at a temperature of about 850 °C. This produces olefins such as ethylene, propylene and C4 products as well as aromatic compounds such as benzene, toluene, xylene, etc. They are basic chemical products based on fossil fuels. During the operation of a steam cracker, some equipment or processing units may suffer from the accumulation of coke residues. For example, during operation, the inner wall of the cracking coil in the furnace of the steam cracker suffers from the formation of a coke layer. The same problem also occurs in the transfer line heat exchanger located directly downstream of the furnace. Since this reduces the performance of both types of equipment, the coke must be removed regularly by burnout or mechanically. If the coke is not removed at the appropriate time, it may cause unplanned production losses due to reduced efficiency or asset failure. Similar coking, or more generally, fouling processes also reduce the efficiency of process equipment in other chemical plants.

[0004] Another example of such a chemical production plant is a dehydrogenation reactor, as described in the following publication: R. Kelling, G. Kolios, C. Tellaeche, U. Wegerle, V. M. Zahn, A. Seidel-Morgenstern: "Development of a control concept for catalyst regeneration by coke combustion", Chemical Engineering Science, Vol. 83, 2012, pp. 138-148. Other examples can be found in Jens R. Rostrup-Nielsen: "Industrial relevance of coking", Catalysis Today, Vol. 37, No. 3, 1997, pp. 225-232.

[0005] As another example, an aniline catalyst for a fluidized bed reactor also suffers from the formation of coke during operation. This can lead to a reduction in aniline production, insufficient heat dissipation, a decrease in bed height and ultimately to the collapse of fluidization. Therefore, the coke must be removed periodically by oxidation, which results in the end of the furnace run period. Summary of the Invention

[0007] There may be a need to provide a system to provide a reasonable prediction of the evolution of the expected health state of the equipment and / or processing units of a chemical production plant within a prediction horizon.

[0008] The object of the present invention is achieved by the subject matter of the independent claims, wherein further embodiments are incorporated in the dependent claims. It should be noted that the aspects described below of the present invention also apply to computer-implemented methods and devices.

[0009] A first aspect of the present invention provides a computer-implemented method for predicting the progression of degradation in equipment of a chemical production plant, comprising:

[0010] a) obtaining (110) future values of at least one operating parameter of the equipment,

[0011] wherein the at least one operating parameter has an impact on the degradation of the equipment, and

[0012] wherein the at least one operating parameter is known and / or controllable within the prediction horizon such that the future values of the at least one operating parameter can be determined within the prediction horizon;

[0013] b) using (120) a prediction model to estimate future values of at least one key performance indicator within the prediction horizon based on an input data set comprising the future values of the at least one operating parameter,

[0014] wherein the prediction model is parameterized or trained based on a sample set of historical data comprising at least one process variable and the at least one operating parameter, wherein the at least one process variable is used to determine the at least one key performance indicator; and

[0015] c) predicting (130) the progression of degradation of the equipment within the prediction horizon based on the future values of the at least one key performance indicator.

[0016] In other words, a data-driven prediction model is used to predict the degradation of equipment of a chemical production plant such as a steam cracker. In particular, under the assumption of a given scenario (i.e., one or more given operating parameters within the prediction horizon, such as plant load and cracking temperature), the future evolution of the degradation key performance indicators is predicted. Based on this information, necessary actions can be implemented to prevent unplanned production losses due to the degradation or failure of process equipment. For example, the scheduling and adjustment of the downtime between different furnaces of a steam cracker can be improved, for example, by avoiding parallel downtimes of two or more furnaces. The data typically used for the prediction model in this context is created by sensors in the plant close to the production process.

[0017] Using the future values of the one or more operating parameters, which can be used to simulate "what-if" scenarios, such as changes in process conditions, such as reduced feed load, feed composition, and reactor temperature within the prediction horizon. It should be noted that the proposed prediction model does not infer the future operating state from the past and / or current operating states, but requires the user to input future operating parameters in order to account for future changing operating conditions of the plant, such as changing requirements from connected plants and / or optimization of production performance. The use of future operating parameter values can account for future changes in plant operation. The key performance indicators are a function of the load on the system. By using the values of the future operating parameters, future load on the system can be included, for example, for prediction. Allowing the values of the future operating parameters to vary based on the planning in the plant can provide additional degrees of freedom, which can improve the quality of the prediction model and can make the prediction more robust.

[0018] For example, the prediction model can be used as a production planning tool for aniline production. Operations in an aniline plant are typically adjusted during the furnace period to account for changing requirements from connected plants and / or to optimize catalyst and production performance. Therefore, it may be beneficial to explicitly set these operating parameters and account for changing future operating parameters in the model.

[0019] Hereinafter, the value of the at least one key performance indicator obtained during the current operation of the device may also be referred to as the current value of the at least one key performance indicator. The value of the at least one key performance indicator obtained during the past operation of the device may also be referred to as the past value of the at least one key performance indicator. Similarly, the value of the at least one operating parameter obtained during the current operation of the device may be referred to as the current value of the at least one operating parameter. The value of the at least one operating parameter obtained during the past operation of the device may be referred to as the past value of the at least one operating parameter. This also applies to other parameters described hereinafter.

[0020] Two exemplary instances will be described in detail hereinafter. A brief overview of these two exemplary instances will be described hereinafter.

[0021] In the first instance, during operation, the inner wall of the cracking coil in the furnace of a steam cracker is subject to the formation of a coke layer. The same problem also occurs in the transfer line heat exchanger (TLE) located directly downstream of the furnace. Since this reduces the performance of both equipment types, the coke must be removed regularly, either by burning out or mechanically.

[0022] To schedule this maintenance program, it is highly advantageous to know at least 1 - 2 weeks in advance when a certain critical coil wall temperature (CWT) or critical TLE outlet temperature is reached. In this way, for example, by avoiding parallel downtimes of two or more furnaces, the scheduling and adjustment of downtimes between different furnaces of the steam cracker can be improved.

[0023] The key performance indicator (KPI) for coking in the cracking coils can be the coil wall temperature (CWT); for the TLEs, it can be their outlet temperature.

[0024] For the two KPIs, a prediction model can be developed based on historical production data. It can predict the coking process in the cracking coils and TLEs within the upcoming 4 weeks with a quantified confidence interval. Additionally, using the future values of the at least one operating parameter, it can be used to simulate "what-if" scenarios, i.e., changes in process conditions such as reduced feed load, feed type, or cracking temperature.

[0025] Using the prediction model of the steam cracker, in the quantification of the remaining useful life (RUL) of the furnace / TLEs before the next maintenance event, the risk of several furnaces being shut down in parallel can be reduced. Additionally, since alternative production scenarios can be simulated, such as reduced feed load, feed type, or cracking temperature in the near future, it can allow the process administrator to proactively delay cleaning procedures, e.g., synchronizing them with other maintenance tasks.

[0026] In a second example, the aniline catalyst used in the fluidized bed reactor also suffers from coke formation during operation. This leads to catalyst deactivation and thus a reduction in aniline conversion, which in turn results in shorter furnace cycles, more frequent catalyst regeneration, and thus a loss in production. The second effect is a reduction in bed height, in some cases accompanied by a reduction in fluidization quality and thus a reduction in heat transfer. Therefore, the coke must be removed periodically by ending the oxidative run period.

[0027] Also in this case, predicting the reactor performance and the end of the furnace cycle 1 - 2 weeks in advance can be highly beneficial. This is useful for the production planning of other related plants and also for optimizing performance and catalyst life.

[0028] Here, the KPIs are coking (through the pressure drop across the bed), bed height, the heat transfer coefficient between the bed and the heat exchanger tubes, and aniline conversion.

[0029] For all KPIs, the historical data of one reactor can be used to develop a prediction model. The same type of prediction model can also be developed for more reactors worldwide. As in the case of the steam cracker, the model can predict the coking progress of the aniline catalyst and the evolution of bed height, heat transfer coefficient, and aniline conversion with a quantified confidence interval until the end of the said furnace cycle. It can also simulate "what-if" situations, such as changes in process conditions such as reduced feed load, feed composition, and reactor temperature.

[0030] The prediction model of the aniline reactor can have at least one of the following benefits:

[0031] a) Predictions using pressure drop, bed height, heat transfer coefficient, and aniline conversion can extend catalyst life. This can result in lower catalyst consumption.

[0032] b) The possibility of simulating alternative production scenarios such as feed load, feed composition, and reactor temperature allows for the optimization of aniline production and better production planning as well as coordination with connected units.

[0033] Through the predictive model of the aniline fluidized bed reactor, process management can monitor the estimated progress of coking in the near future and the evolution of bed height, heat transfer coefficient, and aniline conversion. All KPIs are used to predict the furnace campaign length, thus allowing the scheduling of catalyst regeneration and thus allowing for better planning. Additionally, catalyst life can be extended by minimizing coke accumulation. This can be achieved by optimizing process conditions, where different future process conditions can be simulated.

[0034] Despite the wide variety of affected equipment and chemical production units, and the completely different physical or chemical degradation processes underlying them, the degradation of these equipment can share some of the following characteristics:

[0035] 1. The equipment of the chemical production unit under consideration can have one or more key performance indicators that can quantify the progress of degradation.

[0036] 2. On a time scale longer than the typical production time scale, e.g., for the batch time of a discontinuous process or the typical time between setpoint changes for a continuous process, the key performance indicators more or less monotonically drift to higher or lower values, indicating the occurrence of irreversible degradation phenomena. However, on shorter time scales, the key performance indicators may exhibit fluctuations that are not driven by the degradation process itself, but rather by changing process conditions or background variables such as ambient temperature.

[0037] 3. The key performance indicators roughly return to their benchmarks after maintenance events such as the cleaning of fouled heat exchangers, the replacement or regeneration of deactivated catalysts, etc.

[0038] 4. Degradation is driven by creep, inevitable wear, and / or tear of the process equipment.

[0039] In other words, the evolution of the key performance indicators is largely determined by process conditions rather than by uncontrolled external factors, which allows for predicting the evolution of the degradation key performance indicators within a specific time range given the process conditions planned within that time range.

[0040] The key performance indicator can be selected from parameters including the following: parameters included in a set of measured process data, such as the outlet temperature of a transfer line heat exchanger as described below, and / or a derived parameter that represents a function of one or more parameters included in a set of measured process data. For example, although catalyst activity is not directly measured among the process variables, it itself indicates a decrease in the yield and / or conversion rate of the process.

[0041] As used herein, the term "current" refers to the most recent measurement, since measurements of some equipment may not be taken in real time.

[0042] As used herein, the term "future" refers to a point in time within a prediction horizon. As will be explained below, the useful prediction horizon for equipment degradation is typically between a few hours and several months. The prediction horizon used is determined by two factors. First, the prediction must be accurate enough to be used as a basis for decision-making. To achieve accuracy, input data for future production plans must be available, which is the case when there are only a limited number of days or weeks in the future. In addition, the prediction model itself may lack accuracy due to the underlying prediction model structure or poorly defined parameters, which is a result of the noisy and limited nature of the historical data set used for model identification. Second, the prediction horizon must be long enough to address relevant operational issues, such as taking maintenance actions and making planning decisions.

[0043] For example, predict the progress of degradation in at least one of the following equipment: a heat exchanger suffering from fouling or other scaling processes due to the formation and / or polymerization of a coke layer; a pipe in which the fouling or other scaling process due to the formation and / or polymerization of a coke layer hinders mass flow; a fixed-bed reactor suffering from fouling or other scaling processes due to the formation, polymerization, and / or deposition of solid materials from upstream unit operations; a fluidized-bed reactor suffering from fouling or other scaling processes due to the formation, polymerization, and / or deposition of solid materials from upstream unit operations; and a filter with reduced efficiency due to polymerization and / or deposition of solid materials from upstream unit operations.

[0044] Generally, the above and below methods and apparatuses can be applied to any equipment component that suffers from coking or other fouling processes. For example, due to the formation or polymerization of a coke layer, or due to microbial or inorganic deposits, a heat exchanger may suffer from coking or other fouling processes. For a detailed discussion of the coking / fouling phenomenon in heat exchangers, refer to the following publications: Cai H, Krzywicki A, Oballa MC, Coke formation in Steam crackers for ethylene production; Chemical Engineering and Processing: Process Intensification 2002, 41(3): 199-214 and Müller-Steinhagen H., Heat exchanger fouling: Mitigation and cleaning techniques. IChemE, 2000. As another example, due to the formation and polymerization of a coke layer, the mass flow through a pipeline may be hindered by coking or other fouling processes. As another example, the performance of a fixed bed reactor and a fluidized bed reactor may suffer from coking or other fouling processes due to the formation, polymerization, and / or deposition of solid materials from upstream unit operations. As another example, the efficiency of a filter may be reduced due to polymerization and / or deposition of solid materials from upstream unit operations.

[0045] For example, a chemical production plant includes at least one of the following: a dehydrogenation reactor, a steam cracker, or a fluidized bed reactor.

[0046] Generally, the above and below methods and apparatuses can be used for any chemical production plant having equipment components that may suffer from coking or other fouling processes. An example of such a chemical production plant is a steam cracker. Another example of such a chemical production plant is a dehydrogenation reactor or a fluidized bed reactor.

[0047] In one example, step b) further includes determining past values of the at least one key performance indicator based on the at least one process variable measured during past operation of the equipment within a predefined period before the current operation. Step c) further includes obtaining past values of the at least one operating parameter of the equipment during past operation. In step d), the input data set further includes the past values of the at least one key performance indicator and the past values of the at least one operating parameter.

[0048] The past value can also be referred to as the lag value. Thus, by using lag variables to introduce feedback over time, the prediction model is more robust. In contrast, a model without lag variables represents a system that responds specifically to current events. The predefined time window for the lag value / lag window can be selected by the model developer, for example, according to the type of the device. For example, the predefined time period can be 5%, 10%, or 15% of the typical time period between two maintenance actions of the device.

[0049] According to an embodiment of the present invention, the computer-implemented method further comprises obtaining values of at least one operating parameter of the device obtained during current and / or past operation of the device. The input data set comprises the values of the at least one operating parameter obtained during current and / or past operation of the device.

[0050] In other words, under the assumption of a given scenario, i.e., one or more given operating parameters, such as the device load and the cracking temperature within the prediction horizon, the future evolution of the degradation key performance indicator can be predicted based on current and / or past operating parameters. Therefore, the prediction of the key performance indicator can be more accurate.

[0051] According to an embodiment of the present invention, the computer-implemented method further comprises obtaining at least one process variable measured during current and / or past operation of the device, and determining values of at least one key performance indicator based on the at least one process variable obtained during current and / or past operation of the device. The input data set further comprises the values of the at least one key performance indicator obtained during current and / or past operation of the device.

[0052] In other words, the future evolution of the degradation key performance indicator is predicted based on current and / or past key performance indicators reflecting the current and / or past health state of the device. Therefore, the prediction of the future evolution of the degradation key performance indicator is more accurate.

[0053] According to an embodiment of the present invention, the computer-implemented method further comprises repeating steps a) to c) within another prediction horizon.

[0054] This prediction horizon can be referred to as the first prediction horizon, and this another prediction horizon can also be referred to as the second prediction horizon.

[0055] Using an iterative method with lag variables can generally produce better predictions in the near future (in the case of aniline, 1-2 weeks). This is due to the fact that the model uses continuously updated initial values. In the more distant future (in the case of aniline, >4 weeks), a simpler model without lag variables tends to have better accuracy. In the even more distant future, the poor performance comes from the accumulation of prediction errors due to the iterative nature of the model.

[0056] According to an embodiment of the present invention, the other prediction time domain partially overlaps with this prediction time domain. Alternatively, the other prediction time domain is separated from this prediction time domain.

[0057] In other words, the second prediction time domain, i.e., the other prediction time domain, can start after the first prediction time domain. Alternatively, the two prediction time domains may partially overlap.

[0058] According to an embodiment of the present invention, the value of the at least one key performance indicator is determined based on at least one transformed process variable of a function representing the at least one process variable.

[0059] For example, the transformation can be a simple summation of feed rates in two or more pipes, where the two or more pipes all flow in the same transfer line exchanger (TLE). Another approach is to use an energy / mass balance or a non-linear transformation on the raw data to increase the information content of the data.

[0060] For example, if one or more values of the at least one operating parameter in the input data set and / or sample set violate a predefined set of operating ranges, the one or more values of the at least one operating parameter are eliminated.

[0061] In other words, when one or more values of the at least one operating parameter violate a predefined set of operating ranges, the prediction model is corrected by filtering or eliminating the one or more values of the at least one operating parameter in the input data set and / or sample set. In this way, during the training phase and / or during the phase of using the prediction model to estimate future values of the at least one key performance indicator, unreasonable observations are filtered out. Some examples of unreasonable observations are provided in Section A (TLE) of this disclosure.

[0062] According to an embodiment of the present invention, the prediction model includes a multiple linear regression model, optionally with regularization.

[0063] Regularization can be used to introduce additional information in order to solve ill-posed problems or prevent overfitting.

[0064] According to an embodiment of the present invention, the historical data includes historical values of the at least one process variable and the at least one operating parameter for at least 10 cleaning cycles, preferably at least 30 cleaning cycles.

[0065] As used herein, the term "cycle" can refer to the time period between two consecutive cleaning procedures, which can be 2 - 3 months for a steam cracker. For aniline, the term "cycle" can also be referred to as a furnace period.

[0066] According to an embodiment of the present invention, the device includes at least one of a steam cracking furnace, a heat exchanger for a pipeline for transporting a steam cracker, and an aniline catalyst.

[0067] The steam cracking furnace and the heat exchanger for the pipeline can also be referred to as key components because their health status has a greater impact on the overall efficiency of the steam cracker. In particular, during operation, the inner wall of the cracking coil in the furnace of the steam cracker is subject to the formation of a coke layer. The same problem also occurs in the heat exchanger for the pipeline directly downstream of the furnace. Since this reduces the performance of both types of equipment, the coke must be removed regularly by burnout or mechanically. The aniline catalyst for the fluidized bed reactor is subject to the formation of coke during operation. This may lead to a reduction in aniline production, insufficient heat dissipation, a decrease in bed height, and ultimately a breakdown of fluidization. Therefore, the coke must be removed periodically by oxidation at the end of the furnace period.

[0068] In one example, the at least one key performance indicator of coking in the steam cracking furnace includes the tube metal temperature of the cracking coil in the steam cracking furnace. The at least one key performance indicator of coking in the heat exchanger for the pipeline includes the outlet temperature of the heat exchanger for the pipeline.

[0069] In one example, the at least one process variable for determining the at least one key performance indicator of the steam cracker furnace includes the tube metal temperature of the cracking coil in the steam cracker furnace. The at least one process variable for determining the at least one key performance indicator of the heat exchanger for the pipeline includes the outlet temperature of the heat exchanger for the pipeline.

[0070] In one example, the current value of the at least one operating parameter includes the current naphtha feed load and / or the current cracking temperature.

[0071] In one example, the future value of the at least one operating parameter includes at least one of the following: the future naphtha feed load within the prediction time domain, the future cracking temperature within the prediction time domain, and the feed load accumulated within the prediction time domain and weighted by the weight fraction of each component of the different naphtha components that will pass through the heat exchanger for the pipeline within the prediction time domain. The naphtha components include at least one of normal paraffins, isoparaffins, naphthenes, aromatic compounds, and olefins.

[0072] In one example, the current value and the future value of the at least one operating parameter of the transmission pipeline heat exchanger further include at least one of: a current liquefied petroleum gas feed load, a future liquefied petroleum gas feed load, and a feed load accumulated over a forecast horizon and weighted by a weight fraction of each of a certain amount of different liquefied petroleum gas components passing through the transmission pipeline heat exchanger over the forecast horizon. The liquefied petroleum gas component includes at least one of propane, n-butane, isobutane, isobutylene, 1-butane, trans-2-butene, cis-2-butene, and pentane.

[0073] Another aspect of the present invention provides an apparatus for predicting degradation progress in equipment of a chemical production plant, comprising:

[0074] a) an input unit (210) configured to receive:

[0075] - a future value of at least one operating parameter of said device,

[0076] wherein the at least one operating parameter has an effect on degradation of the device, and

[0077] wherein the at least one operating parameter is known and / or controllable within the prediction time domain; b) a processing unit (220) configured to:

[0078] - using a forecasting model to estimate a future value of the at least one key performance indicator within a forecast horizon based on an input data set comprising future values of the at least one operating parameter,

[0079] wherein the predictive model is parameterized or trained based on a sample set comprising historical data of the at least one process variable and the at least one operating parameter, wherein the at least one process variable is used to determine the at least one key performance indicator; and

[0080] - predicting degradation progression of the device within a prediction horizon based on future values of the at least one key performance indicator; and

[0081] c) An output unit (230) configured to output the predicted degradation progress in the device.

[0082] According to an embodiment of the present invention, the input unit is configured to obtain a value of at least one operating parameter of the device during current and / or past operation of the device. The input data set comprises the values of the at least one operating parameter obtained during current and / or past operation of the device.

[0083] According to an embodiment of the present invention, the input unit is configured to obtain at least one process variable measured during the current and / or past operation of the device. The processing unit is configured to determine the value of at least one key performance indicator based on the at least one process variable obtained during the current and / or past operation of the device. The input data set further includes the value of the at least one key performance indicator obtained during the current and / or past operation of the device.

[0084] According to an embodiment of the present invention, the processing unit is configured to repeatedly perform the estimation within another prediction horizon.

[0085] According to an embodiment of the present invention, the other prediction horizon partially overlaps with the prediction horizon. Alternatively, the other prediction horizon is separated from the prediction horizon.

[0086] According to an embodiment of the present invention, the processing unit is configured to eliminate one or more values of the at least one operating parameter if the one or more values of the at least one operating parameter in the input data set and / or the sample set violate a predefined set of operating ranges.

[0087] According to an embodiment of the present invention, the prediction model includes a multiple linear regression model, optionally with regularization.

[0088] Another aspect of the present invention provides a computer program unit for guiding a device, which, when executed by a processing unit, is adapted to execute the method.

[0089] Another aspect of the present invention provides a computer-readable medium storing a program unit. Brief Introduction of the Drawings

[0090] These and other aspects of the present invention will become apparent and be further elucidated from the embodiments described by way of example hereinafter and with reference to the drawings, in which:

[0091] Figure 1 A flowchart of a computer-implemented method for predicting the degradation progress in a device of a chemical production plant is shown.

[0092] Figure 2A The measured outlet temperature of TLE A and the corresponding 14-day prediction are shown.

[0093] Figure 2B The prediction errors of models with different prediction horizons for TLE A and TLE B are shown.

[0094] Figure 3 The 14-day prediction result of TLE B is shown.

[0095] Figure 4A and4B shows the principle of the prediction model.

[0096] Figure 5A and 5B shows an example of an extended prediction horizon.

[0097] Figure 6 schematically shows a device for predicting the degradation progress in a device for a chemical production plant.

[0098] It should be noted that the drawings are purely schematic and not drawn to scale. In the drawings, elements corresponding to elements already described may have the same reference numerals. Examples, embodiments or optional features, whether indicated as non-limiting or not, should not be construed as limiting the invention claimed.

[0099] Detailed description of the embodiments

[0100] Method for predicting degradation progress

[0101] Figure 1 shows a flowchart illustrating a computer-implemented method 100 for predicting the degradation progress in a device for a chemical production plant such as a steam cracker.

[0102] In step 110, i.e., step a), future values of at least one operating parameter of the device are obtained. The at least one operating parameter has an impact on the degradation of the device. The at least one operating parameter is known and / or controllable within the prediction horizon such that the future values of the at least one operating parameter can be determined within the prediction horizon.

[0103] Examples of operating parameters are naphtha feed load and cracking temperature. The at least one operating parameter has an impact on the degradation progress of the device. In other words, only the operating parameters relevant to determining the degradation of the device are selected. The at least one operating parameter is known and / or controllable within the prediction horizon such that the future values of the at least one operating parameter can be planned or anticipated within the prediction horizon.

[0104] A useful prediction horizon for device degradation is typically between several hours and several months. The prediction horizon used is determined by two factors. First, the prediction must be accurate enough to be used as a basis for decision-making. To achieve accuracy, input data for future production plans must be available, which is only available for a limited prediction horizon.

[0105] In addition, the prediction model itself may lack accuracy due to the underlying prediction model structure or due to poorly defined model parameters, which may be the result of noise and limited nature of the historical data set used for model identification. Second, the prediction horizon must be long enough to address relevant operating issues, such as taking maintenance actions and making planning decisions.

[0106] Optionally, during the current and / or past operation of the device, at least one process variable is measured, for example, by one or more sensors. Examples of process variables may include, but are not limited to, temperature, pressure, flow rate, level, and composition. For the device, appropriate sensors can be selected that provide information about the health status of the device under consideration. The sensors can be selected based on experience and process understanding.

[0107] The device can be one of the critical components because the health status of the critical components has a stronger impact on the maintenance activities of chemical production plants such as steam crackers or dehydrogenation reactors. The source of such information regarding the selection of critical components may be bad actor analysis or general experience of operation. Examples of devices include, but are not limited to, heat exchangers that suffer from coking or other fouling processes due to the formation and / or polymerization of coke layers; pipes in which the coking or other fouling processes due to the formation and / or polymerization of coke layers impede mass flow; fixed bed reactors that suffer from coking or other fouling processes due to the formation, polymerization, and / or deposition of solid materials from upstream unit operations; fluidized bed reactors that suffer from coking or other fouling processes due to the formation, polymerization, and / or deposition of solid materials from upstream unit operations; and filters that experience reduced efficiency due to polymerization and / or deposition of solid materials from upstream unit operations.

[0108] Optionally, a current value of at least one key performance indicator is determined based on the at least one process variable obtained during the current operation of the device. Optionally, a past value of the at least one key performance indicator can be determined based on the at least one process variable measured by one or more sensors during the past operation of the device within a predefined period before the current operation. In other words, in addition to the current value of the at least one key performance indicator, the past value of the at least one key performance indicator, i.e., the lag value, is also determined. The predefined period before the current operation can be set by the model developer. For example, the predefined period can be 10% of the period between two maintenance actions of the device.

[0109] Key performance indicators may include one or more measured process variables, which represent raw measurement values. Optionally, the current value and / or past value of the at least one key performance indicator is determined based on at least one transformed process variable that is a function representing the at least one process variable. In other words, the raw measurement values are mathematically combined into new variables, such as temperature compensated for pressure, or mass flow calculated from volume flow measurement values. The new variables, i.e., the transformed process variables, may be created to include a version of the measurement that is most familiar to the process operator or to improve the correlation structure of the data for a prediction model. The key performance indicators may be defined by a user (e.g., a process operator) or by a statistical model, such as an anomaly score of the distance in a multivariate space of the measured relevant process variables to the "healthy" state (i.e., the state without degradation) of the device, such as the Hotelling T 2 -score or DModX distance.

[0110] Optionally, the current value of at least one operating parameter of the device during the current operation is obtained. Optionally, the past value, i.e., the lag value, of the at least one operating parameter of the device during a past operation may be obtained.

[0111] In step 120, i.e., step b), a prediction model is used to estimate the future value of the at least one key performance indicator within a prediction time horizon. The input to the prediction model includes an input data set that includes the future values of the at least one operating parameter.

[0112] Optionally, the input data set includes the values of the at least one operating parameter obtained during the current and / or past operation of the device.

[0113] Optionally, the input data set further includes the values of the at least one key performance indicator obtained during the current and / or past operation of the device, as it provides information about the health state of the device during the current and / or past operation. Including the current and / or past values of the at least one key performance indicator as input can improve the prediction accuracy.

[0114] The prediction model is parameterized or trained based on a sample set of historical data that includes the at least one process variable and the at least one operating parameter. Since the size of the sample set affects the performance of the prediction model, the historical data preferably includes historical values of the at least one process variable and the at least one operating parameter for at least 10 cleaning cycles, preferably at least 30 cleaning cycles. For example, the prediction model may use 80% of the historical cleaning cycles to calibrate the model and 20% of the historical cleaning cycles to verify the goodness of fit or prediction accuracy of the model. It is also important to periodically recalibrate the model to address process variations not captured by the model.

[0115] Optionally, when one or more values of the at least one operating parameter violate a predefined set of operating ranges, the prediction model is corrected by filtering or eliminating the one or more values of the at least one operating parameter in the input data set and / or the sample set. In this way, unreasonable observations are filtered out during the training phase and / or during the phase of using the prediction model to estimate future values of the at least one key performance indicator.

[0116] An example of a prediction model is a multiple linear regression (MLR) model that models the relationship between two or more explanatory variables and a response variable by fitting a linear equation to observed data, where the explanatory variables are optionally the current values of the at least one key performance indicator, the current values of the at least one operating parameter, and the future values of the at least one operating parameter, and the response variable is the future value of the at least one key performance indicator.

[0117] In step 130, i.e., step c), the degradation progression of the device within the prediction time horizon is predicted based on the future value of the at least one key performance indicator.

[0118] It should be understood that the above operations can be performed in any suitable order, such as sequentially, simultaneously, or a combination thereof, and in cases where a particular order is necessary, depending on, for example, input / output relationships.

[0119] Examples of prediction models

[0120] To show that the prediction model is also applicable to predicting the health state of the device in the previous days and weeks, three device examples are provided, including a steam cracking furnace, a transfer line exchanger (also known as a TLE) of a steam cracker, and an aniline catalyst used in a fluidized bed reactor.

[0121] A. TLE

[0122] Since the TLE is prone to coking, its cooling capacity decreases over time, which may cause the gas temperature at the outlet to increase. For a newly cleaned TLE, this temperature is approximately 380°C. It rises to 470 - 480°C within 1 - 3 months, which is the threshold for device cleaning (either by burnout or mechanically). In terms of the foregoing section, the outlet temperature is a key performance indicator for monitoring the degradation progression. To schedule the cleaning task, it is clearly very beneficial to know at least 1 - 2 weeks in advance when the critical outlet temperature will be reached.

[0123] Another benefit of predicting TLE coking is that alternative scenarios can be simulated, such as reducing the feed load or cracking temperature in the near future, thereby allowing the process administrator to proactively delay the cleaning procedure, for example, synchronizing the cleaning procedure with other maintenance tasks. For these reasons, a reliable prediction of the outlet temperature is a great benefit.

[0124] A1.TLE A

[0125] The prediction target for TLE A is the outlet temperature, i.e., the future value of the key performance indicator. For clarity, this means that we estimate the outlet temperature value at a fixed time shift in the future (the prediction horizon) through the prediction model.

[0126] The input quantities for the model can be selected by first selecting all quantities known or believed to affect coking, for example by an expert in the field. Then, some of these quantity features are combined, an appropriate mathematical form for the prediction model is selected, and the model is calibrated against historical data. It is also preferable to avoid the model memorizing irrelevant features from the historical data, thereby failing to produce accurate predictions for new data (the "overfitting problem").

[0127] As a result, the following features are selected as input to the model:

[0128] a. At the current time t 现在 Quantity measured (on the day the forecast was made):

[0129] - Current naphtha feed load [t / h], averaged over all cracking coils connected to TLE A;

[0130] - current cracking temperature (at 90% of the coil length) in [°C]; and

[0131] - Current TLE outlet temperature;

[0132] b. Quantities that can be reliably predicted within the forecast horizon, because they are (at least in principle) known in the future or even controlled by:

[0133] - The day the forecast is made 未来 The naphtha feed load, in [t / h];

[0134] -t 未来 Cracking temperature at [°C];

[0135] - the amount of different naphtha components passing through the TLE during the forecast horizon, in [t]. For clarity, this is defined as the feed load accumulated over the forecast horizon and weighted by the weight fraction of the components in the feed, and for the i-th component by m i express:

[0136]

[0137] Instead of considering all 29 measured components, we include weight fractions aggregated into the following component categories:

[0138] n-paraffins;

[0139] · Isoalkanes;

[0140] · Naphthenes;

[0141] · Aromatic compounds; and

[0142] · Olefins.

[0143] As an additional step in feature engineering, the input quantities of all the above models can be extended to higher polynomial orders, e.g., to a second-order factorization model for all inputs. The next step is to use multiple linear regression to determine the relationship between these input quantities and the target key performance indicator (future outlet temperature) based on historical data. For this purpose, historical values of all quantities for the last 48 cleaning cycles (∼10 years) are collected using a sampling rate of 1 hour. Several criteria are applied to filter out the "bad" observations of the system, i.e., the data points not used for regression:

[0144] - Observations of TLE being decoked at time t 现在 or t 未来 ;

[0145] - Observations with "abnormal" low or very high feed loads at time t 现在 or t 未来 ;

[0146] - Observations with "abnormal" cracking temperature values at time t 现在 or t 未来 ;

[0147] - Observations where one of the input variables or the output is not measured.

[0148] Next, the dataset is split into a validation set (8 cleaning cycles) and a training set (40 cycles). Only the training set is used for multiple linear regression.

[0149] A robustification method of the least squares fitting procedure of MLR is used to mitigate the harmful effects of outliers on the model accuracy.

[0150] The prediction error of the model (for 1-σ confidence) is quantified by the root mean square deviation (RMSE) between the predicted model estimates and their values within the prediction time domain. The RMSE of the validation set and the training set are of similar magnitude, indicating that the model does not suffer from overfitting. Figure 2AThe measured outlet temperature of an exemplary history section and the corresponding 14-day prediction are shown. In particular, the true value 10 of the outlet temperature is compared with the corresponding 14-day predicted values of the training data 12 between October 2010 and November 2010 and between February 2011 and April 2011. The true value 10 of the outlet temperature is compared with the corresponding 14-day predicted values of the validation data 14 between October 2010 and February 2011. For clarity, if the feed load and feed composition have been accurately predicted within the prediction time domain, curves 12 and 14 are the predicted values for a given day, which are calculated for the 14 days prior to that time.

[0151] The agreement between the predicted values and the actual values is very good. The instantaneous changes (caused by sudden changes in the feed load) and the overall increasing trend caused by coking are captured. Figure 2B The prediction errors of the model for different prediction time domains are shown. As expected, the uncertainty increases with the longer the future prediction time. However, the model produces acceptable predictions (error below 10 °C) for a time domain up to one month.

[0152] A2.TLE B

[0153] Very similar to TLE A, a prediction model for the outlet temperature of TLE B was established. However, contrary to furnace A where TLE A is located, furnace B where TLE B is located is not dedicated to processing naphtha, but can also be fed with LPG or a mixture of the two ("co-cracking"). This results in a greater variation in the cycle length (i.e., the time interval between two cleaning procedures). Therefore, the plant personnel have a much weaker "gut feeling" about coking in furnace B than in furnace A, which increases the benefit of a reliable prediction model.

[0154] In addition, the possibility of simulating the impact of switching feedstocks (e.g., from naphtha to liquefied petroleum gas) on the degradation progress is an additional benefit of the TLE B prediction model.

[0155] As the input to the TLE B prediction model, we start with the same quantities as for TLE A and add the following:

[0156] - The current LPG feed load, in [t / h];

[0157] - The LPG feed load [t / h] at time t 未来 ; and

[0158] - The quantities of different LPG components, which include:

[0159] · Propane;

[0160] · n-Butane;

[0161] · Isobutane;

[0162] · Isobutene:

[0163] · 1-Butene;

[0164] · trans-2-Butene;

[0165] · cis-2-Butene; and

[0166] · Pentane (total).

[0167] Except for the larger range of effective cracking temperature values, the criteria for filtering out unreasonable observations are similar to TLEA. All other steps of the model building process are as described for TLEA.

[0168] The results of the 14-day prediction are shown in Figure 3 Specifically, the true value 10 of the outlet temperature is compared with the corresponding 14-day predicted values of the training data 12 between July 2008 and September 2008 and between December 2008 and February 2009. The true value 10 of the outlet temperature is compared with the corresponding 14-day predicted values of the validation data 14 between September 2008 and December 2008. The prediction model describes the historical data of the training and validation data up to an acceptable error of ∼8 °C. The prediction error of TLE B is slightly greater than that of TLE A for all time domains, which reflects the expectations of the plant personnel since the co-cracking operation makes coke formation more difficult to predict.

[0169] Therefore, the developed prediction model for the outlet temperature - the outlet temperature is the central key performance indicator of degradation - well describes the available 10-year historical data. For predictions up to two weeks, it allows predictions with an accuracy of + / -7 °C (at the 1-σ confidence level), which is small compared to the temperature window of the normal variation of the outlet temperature. In addition, the model can simulate what impact the feed load, cracking temperature, and feed composition in the upcoming days will have on the progression of the outlet temperature, enabling the operation to schedule the next cleaning procedure for a convenient date.

[0170] B. Steam Cracker

[0171] For steam crackers, a method similar to that for TLEs is used. Key performance indicators for coke formation in a steam cracker include the tube metal temperature of the cracking coils in the steam cracker. To determine the key performance indicators, the tube metal temperature of the cracking coils in the steam cracker is the process variable to be measured. To predict future values of the key performance indicators, i.e., the tube metal temperature of the cracking coils, a prediction model is developed based on historical production data, such as for the past 10 years, using the method described in Section A above. The prediction model can predict the coke formation progress of the cracking coils for 4 or more weeks in the future with a quantified confidence interval. In addition, it is used to simulate "what-if" scenarios, i.e., changes in process conditions, i.e., changes in operating parameters, such as reduced feed load, feed type, or cracking temperature.

[0172] The operating parameters, effective ranges, derived features, regression methods, etc. are the same as those described above for TLEs.

[0173] C. Aniline catalyst

[0174] In the case of aniline catalysts, the accumulation of coke also causes degradation of the catalyst's performance and, consequently, a reduction in the production of aniline. The accumulation of coke and the resulting increase in the mass of the catalyst particles cause an increase in the pressure drop across the bed in the fluidized bed reactor. Changes in the mass and size of the catalyst particles further cause changes in the bed height and the heat transfer coefficient (also abbreviated as the k-value) between the catalyst particles and the heat exchanger inside the reactor. In addition, due to the deactivation of the catalyst, coking also reduces the aniline conversion rate. In terms of the previous section, the pressure drop across the bed is the key performance indicator for monitoring the degradation process. However, the heat transfer coefficient, bed height, and aniline conversion rate are the key performance indicators for determining the end of the furnace period and the start of catalyst regeneration. The end of the furnace period is typically defined by a decrease in the bed height below the level of the heat exchanger, a reduction in the heat transfer coefficient, and / or an aniline conversion rate below a certain level. Each of these situations ultimately leads to the end of the furnace period / cycle. The end of the furnace period is typically reached between 2 - 5 weeks. To schedule catalyst regeneration, it is clearly very advantageous to know the evolution of the key performance parameters for at least the next 1 - 2 weeks and thereby know the approximate end of the furnace period / cycle.

[0175] As in the previous examples, another benefit of prediction is the possibility of simulating alternative scenarios, such as increased feed load, feed composition, or reactor temperature, which allows the process administrator to adjust catalyst regeneration and thus shut down connected devices.

[0176] To achieve better prediction results, certain preprocessing steps are applied to the model's target and input parameters. The model's target parameters are key performance indicators, specifically the pressure drop across the bed, bed height, heat transfer coefficient, and the aniline conversion rate individually. For all target parameters, smoothing algorithms ranging from simple moving average to double exponential smoothing algorithms are applied. The model's input parameters are selected similarly to the TLE example. First, for example, all quantities known or thought to affect coking, heat transfer coefficient, bed height, and conversion rate are selected by experts. Additionally, irrelevant inputs are ignored as this only reduces the model accuracy. Furthermore, for better model accuracy, outliers of the target as well as input parameters are removed. Then the appropriate mathematical form of the prediction model is selected. This includes determining the number of past and future values considered in one iteration of the model. Additionally, a regularization algorithm and parameters are selected to avoid over-precise prediction of the training data and inadequate generalization of the model, also known as overfitting.

[0177] The following features are selected as possible inputs to the model:

[0178] a. Operating parameters measured in the past and present and to be operationally defined in the future:

[0179] · Flow rate of MNB to the reactor,

[0180] · Reactor temperature,

[0181] · Flow rate of recycle gas,

[0182] · Hydrogen concentration in the off-gas;

[0183] b. Process variables calculated in the past and present, and process variables obtainable from future operating parameters:

[0184] · Age of the catalyst, in terms of the number of tons of MNB produced per ton of catalyst since the first use of the catalyst,

[0185] · Length of the furnace run, in terms of the number of tons of MNB produced per ton of catalyst during the furnace run,

[0186] · Average coking rate of the current or previous furnace run,

[0187] c. Initial target variables:

[0188] · Initial bed height,

[0189] · Initial heat transfer coefficient,

[0190] · Initial pressure drop.

[0191] Depending on the operation of the reactor and thus the dependence of the prediction of the target variables on the corresponding inputs, different selections of input parameters can be chosen for each reactor. In addition, the past and current values of the target variables (pressure drop across the bed, heat transfer coefficient, and bed height) are also used as inputs to the model.

[0192] For the training, hyperparameter tuning, and validation of the model, the method of nested cross-validation is used. Depending on the age of the reactor, different numbers of furnace campaigns can be used to train the pending model from several years to over 10 years. Here, five folds are used for both the inner and outer folds of the nested cross-validation. The prediction error of the model (for 1-σ confidence) is quantified by the root mean square deviation (RMSE) between the predicted model estimates and their values in the prediction time horizon. Here, the average of the RMSE on the test set over external resampling is used because they give an estimate of the generalization error of the model. Here, the error also increases with the increase in the prediction time horizon. The small difference between the training and test sets indicates that the model does not suffer from overfitting.

[0193] In the long term, we assume that the evolution of the error over time is similar to that of a random walk, while in the short term, the error is dominated by the inherent difference between the model and the real data. Therefore, we propose the following model:

[0194]

[0195] where the constant offset "a" represents the short-term error, while the sqrt(t) dependence represents the long-term evolution of the random walk.

[0196] Principle of the proposed method

[0197] Now refer to Figure 4A and 4B which illustrate the principle of the proposed method. The proposed prediction model predicts the key performance indicators based on future operating parameters.

[0198] Line 2000 reflects the start of the first iteration of the prediction, which can be the current time. The past direction is indicated by arrow 2500, while the future direction is indicated by arrow 2600. Figure 4A The values depicted in are the key performance indicators, including the past values shown as hollow circles 2100 and solid circles 2200. The solid circles also reflect the past values used in the prediction model. In a non-limiting example, these key performance indicators can be one or more of the key performance indicators described in the above examples, such as the pressure drop across the bed (coke content), heat transfer coefficient, bed height, and conversion rate of aniline catalyst, as well as the outlet temperature of the TLE.

[0199] The time span for selecting past values incorporated into the model is depicted as 2300. The prediction horizon is depicted as 2400. The predicted KPI values are shown as 2700. In some instances, there may be only one prediction iteration. In some cases, it may be advantageous to extend the prediction horizon. This can be accomplished by repeating the prediction during or at the end of the first prediction horizon.

[0200] Figure 4B The corresponding operating values and process variables are shown. In Figure 4B the operating values and process variables 3000, 3100, 3200 are plotted along the time axis. Past operating values and process variables not considered in the prediction model are depicted as 3000. Past operating values and process variables considered in the prediction are depicted as 3100, and future operating values and process variables considered in the prediction are depicted as 3200 and are within the prediction horizon 2400. Examples of these operating values and process variables are described in the "Examples of Prediction Models" section above.

[0201] In some instances, only future operating parameters can be used for prediction.

[0202] In some instances, only future process variables can be used as inputs to the prediction model.

[0203] The next prediction cycle starts at 4000, which marks the [[ID=1 June 16]] Figure 4A end of the first prediction horizon 2400 in

[0204] In some instances, there may be only one prediction iteration.

[0205] In some instances, it may be advantageous to extend the prediction horizon. This can be accomplished by repeating the prediction at the end of the first prediction horizon. Figure 5A and 5B shows an example. Alternatively (not shown), this can be accomplished by repeating during the first prediction horizon, i.e., the first and second iterations can overlap each other.

[0206] Using an iterative method with lagged variables can generally produce better predictions in the near future (in the case of aniline, 1 - 2 weeks). This is due to the fact that the model uses continuously updated initial values. In the more distant future (in the case of aniline, > 4 weeks), a simpler model without lagged variables may tend to have better accuracy. In the even more distant future, the poor performance comes from the accumulation of prediction errors due to the iterative nature of the model.

[0207] Go to Figure 5A which shows the second iteration of the prediction of the key parameter metrics. This prediction cycle starts at 4000, which marks the Figure 4AThe end of the first prediction time horizon 2400 in []. The dashed line 4000 marks the same time in FIGS. 4 and 6. Here, the time span for selecting past values incorporated into the model is depicted as 4300. Here, the prediction time horizon is depicted as 4400. The predicted KPI value for this iteration is depicted as the empty cross 4700. The hash triangles represent the future operating values and process variables for the second iteration. The full prediction now covers the time span from the dotted line 2000 to the dashed line 6000. Additionally, the iteration can be similar to the moving window method. Here, the window moves to the end of the first prediction time span.

[0208] The prediction time horizon can be adjusted according to the needs of the problem. For example, the prediction time horizon can be a few days or weeks in the future. For example, for scheduling the regeneration of a catalyst, it is clearly very advantageous to know the evolution of the key performance parameters over at least the next 1 - 2 weeks and thus know the approximate end of the furnace period / cycle. For example, the end of the furnace period / cycle is typically reached between 2 - 5 weeks.

[0209] The time window 2300, i.e., the time span for selecting past values incorporated into the model, can also be adjusted according to the needs of the problem. For example, the time window can be 10%, 20%, 30%, 40%, or 50% of the time period between two maintenance actions of the device.

[0210] The prediction model can be a multiple linear regression model. In some instances, the prediction model can be a multiple linear regression model with regularization. Regularization is a method of introducing additional information to solve ill - posed problems or prevent overfitting. One way of regularization is to add a constraint to the loss function:

[0211] Regularized loss = Loss function + Constraint

[0212] There are various different forms of constraints that can be used for regularization. Examples include but are not limited to, Ridge regression, Lasso, and elastic net.

[0213] Device for predicting the degradation progress

[0214] Figure 6 Schematically shows a device 200 for predicting the degradation progress in a device such as a steam cracker in a chemical production plant. The device 200 includes an input unit 210, a processing unit 220, and an output unit 230.

[0215] [[ID=**25**]]The input unit 210 is configured to receive future values of at least one operating parameter of the device. The at least one operating parameter has an impact on the degradation of the device, and the at least one operating parameter is known and / or controllable within the prediction time horizon such that the future values of the at least one operating parameter can be determined within the prediction time horizon.

[0216] Optionally, the input unit 210 is further configured to receive the value of the at least one operating parameter obtained during the current or past operation of the device.

[0217] Optionally, the input unit 210 is configured to receive at least one process variable measured during the current and / or past operation of the device.

[0218] Thus, in one example, the input unit 210 may be implemented as an Ethernet interface, a USB(TM) interface, a wireless interface such as WiFi(TM) or Bluetooth(TM), or any suitable data transfer interface enabling data transfer between the input peripheral and the processing unit 220.

[0219] The processing unit 220 is further configured to use a prediction model, such as a multiple linear regression model, to estimate future values of the at least one key performance indicator within a prediction horizon based on an input data set. The input data set includes the value of the at least one operating parameter obtained during the current and / or past operation of the device and future values of the at least one operating parameter. The prediction model is parameterized or trained based on a sample set of historical data including the at least one process variable and the at least one operating parameter. The processing unit 220 is further configured to predict the degradation progress of the device within the prediction horizon based on the future values of the at least one key performance indicator.

[0220] Optionally, the input unit 210 is configured to obtain the value of at least one operating parameter of the device during the current and / or past operation of the device. The input data set includes the value of the at least one operating parameter obtained during the current and / or past operation of the device.

[0221] Optionally, the processing unit 220 is configured to determine the value of the at least one key performance indicator based on the at least one process variable. The input data set further includes the value of the at least one key performance indicator obtained during the current and / or past operation of the device.

[0222] Optionally, the processing unit is configured to repeatedly perform the estimation within another prediction horizon. The other prediction horizon (also referred to as the second prediction horizon) partially overlaps with this prediction horizon (also referred to as the first prediction horizon). Alternatively, the two prediction horizons may be separated from each other. This is explained above and in particular with respect to Figure 5A and 5B in the embodiments shown.

[0223] Accordingly, the processing unit 220 may execute computer program instructions to perform various processes and methods. The processing unit 220 may refer to an application specific integrated circuit (ASIC), an electronic circuit, a processor (shared, dedicated, or group), and / or a memory (shared, dedicated, or group), combinational logic circuitry, and / or other suitable components that execute one or more software or firmware programs, or a part thereof, or include such components. Additionally, the processing unit 220 may be connected to a volatile or non-volatile memory, a display interface, a communication interface, etc., as is known to those skilled in the art.

[0224] The output unit 230 is configured to output the predicted degradation progress in the device.

[0225] Accordingly, in one example, the output unit 230 may be implemented as an Ethernet interface, a USB(TM) interface, a wireless interface such as WiFi(TM) or Bluetooth(TM), or any equivalent data transfer interface that enables data transfer between the output peripheral device and the processing unit 230.

[0226] This exemplary embodiment of the present invention covers computer programs that use the present invention from the start and computer programs that turn existing programs into programs using the present invention through updates.

[0227] Furthermore, the computer program unit may be capable of providing all the necessary steps to complete the program of the exemplary implementation of the method as described above.

[0228] According to another exemplary embodiment of the present invention, a computer-readable medium, such as a CD-ROM, is provided, wherein the computer-readable medium has a computer program unit stored thereon, and the computer program unit is described in the foregoing sections.

[0229] The computer program may be stored and / or 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 may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems.

[0230] However, the computer program may also be presented via a network similar to the World Wide Web and downloaded into the working memory of a data processor. According to another exemplary embodiment of the present invention, a medium for making a computer program unit available for download is provided, and the computer program unit is configured to execute the method according to one of the previously described embodiments of the present invention.

[0231] It should be noted that the embodiments of the present invention are described with reference to different topics. In particular, some embodiments are described with reference to method-type claims, while other embodiments are described with reference to apparatus-type claims. However, those skilled in the art will understand from the context description that, unless otherwise indicated, any combination between features related to different topics is also considered to be disclosed together with this application, in addition to any combination of features belonging to a single type of topic. However, all features can be combined to provide a synergistic effect that exceeds the simple sum of the features.

[0232] Although the present invention has been illustrated and described in detail in the drawings and the foregoing description, such illustration and description are to be considered illustrative or exemplary and not restrictive. The present invention is not limited to the disclosed embodiments. Other variations of the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention, by studying the drawings, the disclosure and the dependent claims.

[0233] In the claims, the wording "comprising" does not exclude other elements or steps, and the indefinite article "a" or "an" does not exclude a plurality. A single processor or other unit 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 indicate that a combination of these measures cannot be used advantageously. Any reference signs in the claims should not be construed as limiting the scope.

Claims

1. A computer-implemented method (100) for predicting the degradation progress in a device of a chemical production plant, comprising: a) obtaining (110) future values of at least one operating parameter of the device, wherein the at least one operating parameter has an impact on the degradation of the device, and wherein the at least one operating parameter is known and / or controllable within a prediction time domain such that the future values of the at least one operating parameter can be determined within the prediction time domain; b) using (120) a prediction model to estimate future values of at least one key performance indicator within the prediction time domain based on an input data set including the future values of the at least one operating parameter, wherein the at least one key performance indicator is a measure of the degradation progress in the device of the chemical production plant; wherein the prediction model is parameterized or trained based on a sample set of historical data including at least one process variable and the at least one operating parameter, wherein the at least one process variable is used to determine the at least one key performance indicator; and c) predicting (130) the degradation progress of the device within the prediction time domain based on the future values of the at least one key performance indicator.

2. The computer-implemented method according to claim 1, further comprising: - obtaining values of at least one operating parameter of the device during the current and / or past operation of the device; wherein the input data set includes the values of the at least one operating parameter obtained during the current and / or past operation of the device.

3. The computer-implemented method according to claim 1 or 2, further comprising: - obtaining at least one process variable measured during the current and / or past operation of the device; - determining values of at least one key performance indicator based on the at least one process variable obtained during the current and / or past operation of the device; wherein the input data set further includes the values of the at least one key performance indicator obtained during the current and / or past operation of the device.

4. The computer-implemented method according to claim 1 or 2, further comprising: - repeating steps a) to c) within another prediction time domain.

5. The computer-implemented method according to claim 4, wherein the another prediction time domain partially overlaps with the prediction time domain; or wherein the another prediction time domain is separated from the prediction time domain.

6. The computer-implemented method according to claim 1 or 2, wherein the prediction model includes an optionally regularized multiple linear regression model.

7. The computer-implemented method according to claim 1 or 2, wherein the device includes at least one of the following: - a steam cracking furnace; - a transfer line heat exchanger of a steam cracker; and - an aniline catalyst.

8. A device (200) for predicting the degradation progress in a device of a chemical production plant, comprising: a) an input unit (210) configured to receive future values of at least one operating parameter of the device, wherein the at least one operating parameter has an impact on the degradation of the device, and wherein the at least one operating parameter is known and / or controllable within a prediction time domain; b) a processing unit (220) configured to: - Using a prediction model to estimate future values of at least one key performance indicator within a prediction horizon based on an input data set that includes future values of the at least one operating parameter, wherein the at least one key performance indicator is a measure of the degradation progress in the equipment of the chemical production plant; wherein the prediction model is parameterized or trained based on a sample set of historical data that includes at least one process variable and the at least one operating parameter, and wherein the at least one process variable is used to determine the at least one key performance indicator; and - Predicting the degradation progress of the equipment within the prediction horizon based on the future values of the at least one key performance indicator; and c) An output unit (230) configured to output the predicted degradation progress in the equipment.

9. The apparatus according to claim 8, wherein the input unit is configured to obtain values of at least one operating parameter of the equipment during current and / or past operations of the equipment; and wherein the input data set includes the values of the at least one operating parameter obtained during current and / or past operations of the equipment.

10. The apparatus according to claim 8 or 9, wherein the input unit is configured to obtain at least one process variable measured during current and / or past operations of the equipment; wherein the processing unit is configured to determine values of at least one key performance indicator based on the at least one process variable obtained during current and / or past operations of the equipment; and wherein the input data set further includes the values of the at least one key performance indicator obtained during current and / or past operations of the equipment.

11. The apparatus according to any one of claims 8-10, wherein the processing unit is configured to repeatedly perform the estimation within another prediction horizon.

12. The apparatus according to claim 11, wherein the another prediction horizon partially overlaps with the prediction horizon; or wherein the another prediction horizon is separated from the prediction horizon.

13. The apparatus according to any one of claims 9-12, wherein the prediction model includes a multiple linear regression model optionally with regularization.

14. A non-transitory computer-readable medium having a sequence of instructions stored thereon, which when executed by a processor cause the processor to perform the method according to any one of claims 1-7.

Images