Method, device, storage medium and processor for optimization of a cracking furnace operation

Abstract

This invention provides a method, apparatus, storage medium, and processor for optimizing the operation of a pyrolysis furnace, belonging to the field of chemical automatic control technology. The optimization method for the pyrolysis furnace operation includes: continuously acquiring the pyrolysis depth of the pyrolysis furnace and forming a first curve reflecting the change in the pyrolysis furnace's operating effect within a preset time period; continuously acquiring the operating parameters and reaction condition parameters of the pyrolysis furnace, and inputting the operating parameters and reaction condition parameters into first and second prediction models of deep learning, respectively, to obtain first and second predicted values ​​of the pyrolysis depth of the pyrolysis furnace at time T+T0 after the current time T, and forming second and third curves reflecting the change in the pyrolysis furnace's operating effect within the preset time period; when any one of the similarities between any two of the first, second, and third curves exceeds a preset similarity value, correcting the pyrolysis furnace operation to make the changing trends of the first, second, and third curves consistent. This method can quickly and effectively optimize the real-time operation of a pyrolysis furnace.

Description

Technical Field

[0001] This invention relates to the field of chemical automatic control technology, and specifically to an optimization method, apparatus, storage medium, and processor for the operation of a pyrolysis furnace. Background Technology

[0002] Ethylene and propylene are the most basic raw materials in petrochemicals and the foundation for the production of various important organic chemical products. There are many methods for their production, but tubular furnace steam cracking technology is the most mature. Therefore, the production scale, output and technology of tubular cracking furnaces indicate the development level of a country's petrochemical industry.

[0003] A tubular pyrolysis furnace consists of a convection section and a radiant section. The pyrolysis feedstock and dilution steam are mixed in a specific ratio and first heated in the convection section tubes. The feedstock is vaporized and superheated to the initial pyrolysis temperature (i.e., the transverse temperature) before entering the radiant section tubes for the pyrolysis reaction. Fuel gas nozzles are arranged at the bottom and / or sidewalls of the pyrolysis furnace, through which fuel gas enters the furnace chamber. The combustion of the fuel gas provides the heat required for the pyrolysis reaction, causing the feedstock to undergo a pyrolysis reaction to produce pyrolysis products such as ethylene, propylene, and butadiene, which are then sent to downstream processes for cooling and separation.

[0004] In the pyrolysis production process, the operating conditions of the pyrolysis furnace significantly impact its yield. In actual production, there are two main ways to adjust the operating conditions of the pyrolysis furnace: one is based on reference values ​​provided by the pyrolysis furnace manufacturer and human experience. However, in actual production, the source of the pyrolysis oil is usually not fixed, and different oils are switched as needed during production. Therefore, current human experience cannot optimize and adjust the pyrolysis furnace to its optimal operating condition. The other method is to use an online component analyzer to measure the relative content of key components in the pyrolysis gas and adjust the production operating conditions accordingly. However, online analyzers require a certain amount of time to analyze a sample, exhibiting significant lag and randomness, which is detrimental to the stable operation of the pyrolysis furnace. Furthermore, if the online analyzer malfunctions, this method will fail, potentially leading to an emergency shutdown of the pyrolysis furnace and severe economic losses. Therefore, to achieve rapid optimization and adjustment of the pyrolysis furnace operating conditions, a fast and effective real-time operation optimization method for pyrolysis furnaces is urgently needed. Summary of the Invention

[0005] The purpose of this invention is to provide an optimization method for pyrolysis furnace operation, which can quickly and effectively optimize the real-time operation of the pyrolysis furnace.

[0006] To achieve the above objectives, embodiments of the present invention provide an optimization method for the operation of a pyrolysis furnace, the method comprising:

[0007] The pyrolysis depth of the pyrolysis furnace is continuously acquired, and a first curve reflecting the changes in the pyrolysis furnace's operating performance over a preset time period is generated.

[0008] The operating parameters of the cracking furnace are continuously acquired and input into the first prediction model of deep learning to obtain the first predicted value of the cracking depth of the cracking furnace at time T+T0 after the current time T, and form a second curve reflecting the change of the cracking furnace operation effect within a preset time period.

[0009] The reaction condition parameters of the pyrolysis furnace are continuously acquired and input into a second prediction model based on deep learning to obtain a second predicted value of the pyrolysis depth at time T+T0 after the current time T, and to form a third curve reflecting the changes in the pyrolysis furnace's operating effect over a preset time period; and

[0010] The similarity between each pair of the first curve, the second curve, and the third curve is calculated. If any one of them exceeds the preset similarity standard value, the operation of the pyrolysis furnace is corrected so that the changing trends of the first curve, the second curve, and the third curve are consistent.

[0011] Preferably, the cracking depth is determined based on the cracking feedstock and cracking reaction mechanism, wherein the cracking depth includes the ethylene content in the cracked gas or the propylene-ethylene content ratio in the cracked gas;

[0012] Continuously obtain the pyrolysis depth of the pyrolysis furnace, including:

[0013] The content of each target component in the pyrolysis gas at the current moment is measured by gas chromatography; and

[0014] The cleavage depth is calculated based on the content of each target component.

[0015] Preferably, the operating parameters are the operation and control conditions of the pyrolysis furnace determined according to the type of furnace tubes and the model of the pyrolysis furnace. The operating parameters include one or more of the following: total steam feed ratio, total feed amount, COT, target outlet temperature, feed rate of each furnace tube group, feedwater flow rate, bottom fuel gas flow rate, sidewall fuel flow rate, steam drum liquid level, radiant section furnace pressure, and feed cross section pressure.

[0016] Furthermore, the first prediction model is a trained long short-term memory network model. The training dataset used includes a set of training samples, which are the historical operation data of the current cracking furnace in cracking the current raw material. The training samples include sample values ​​of operating parameters and corresponding sample values ​​of cracking depth.

[0017] The training process of the first prediction model includes:

[0018] Obtain the initialized Long Short-Term Memory (LSTM) network model. The input of the LSM network model is the sample values ​​of the operating parameters, and the output is the predicted value of the rift depth.

[0019] Input the sample values ​​of the operating parameters into the long short-term memory network model, and calculate the deviation between the predicted value of the rupture depth and the sample value of the rupture depth through the error loss function;

[0020] Training ends when the deviation is less than a first preset threshold or when the training reaches a first preset number of iterations; otherwise, the long short-term memory network model is corrected, and the next round of training continues.

[0021] The model with the smallest deviation is selected as the final first prediction model.

[0022] Preferably, the reaction condition parameters are the feed ratio, pyrolysis gas and intermediate product content, and monitored values ​​of temperature and pressure conditions of the pyrolysis reaction determined according to the reaction mechanism of the pyrolysis feedstock. The reaction condition parameters include one or more of the following: pyrolysis gas pressure, bottom fuel gas pressure, temperature deviation and COT deviation of each furnace tube group, feed cross section temperature of each furnace tube group, pyrolysis gas outlet temperature of the waste heat boiler, furnace temperature, outlet temperature of each group of feed furnaces, and outlet temperature of the first stage of feed mixing and preheating of each group.

[0023] Furthermore, the second prediction model is a trained long short-term memory network model. The training dataset used includes a set of training samples, which are historical reaction condition data of the current cracking furnace cracking the current raw material. The training samples include sample values ​​of reaction condition parameters and corresponding cracking depth sample values.

[0024] The training process for the second prediction model includes:

[0025] Obtain the initialized Long Short-Term Memory (LSTM) network model. The input of the LSM network model is the sample values ​​of the reaction condition parameters, and the output is the predicted value of the cleavage depth.

[0026] The sample values ​​of reaction condition parameters are input into the long short-term memory network model, and the deviation between the predicted value of pyrolysis depth and the sample value of pyrolysis depth is calculated by the error loss function.

[0027] Training ends when the deviation is less than the second preset threshold or when the training reaches the second preset number of iterations; otherwise, the long short-term memory network model is corrected, and the next round of training continues.

[0028] The model with the smallest deviation is selected as the final second prediction model.

[0029] Preferably, the training dataset is obtained by collecting historical operating data of the pyrolysis furnace, and then classifying and filtering the data.

[0030] Historical operating data includes: properties of cracking feedstock oil, type of cracking furnace tubes, relevant parameters monitored by DCS, and corresponding ethylene and propylene yields;

[0031] The classification of historical operating data includes: classifying cracking attributes according to the properties of cracking feedstock, based on one or more of the following performance indicators: relative density, distillation range, group composition, viscosity, hydrogen content, and average molecular weight of the oil.

[0032] The filtering of historical operating data includes deleting operating data that meets the following conditions: the error of any one of the relevant parameters, ethylene yield, and propylene yield exceeds the corresponding preset range, or there are abnormal operating conditions.

[0033] Preferably, the preset time period is n1 minutes, T0 is n2 minutes, and both n1 and n2 are between 0 and 300.

[0034] Preferably, both n and n2 are between 1 and 30.

[0035] Optionally, the pyrolysis furnace can be of type 1, 1-1, 2-1, 4-1, 4-1-1-1, 2-1-1-1, 1-1-1-1, or 8-4-2-1.

[0036] Optionally, the cracking feedstock for the cracking furnace can be ethane, propane, LPG, naphtha, diesel, aviation kerosene, or hydrotreated tail oil.

[0037] On the other hand, the present invention provides an optimization apparatus for the operation of a pyrolysis furnace, the apparatus comprising:

[0038] The operation monitoring module continuously acquires the pyrolysis depth of the pyrolysis furnace and generates a first curve reflecting the changes in the pyrolysis furnace's operating performance within a preset time period;

[0039] The first prediction module continuously acquires the operating parameters of the cracking furnace and inputs the operating parameters into the first prediction model of deep learning to obtain the first predicted value of the cracking depth of the cracking furnace at time T+T0 after the current time T, and forms a second curve reflecting the change of the cracking furnace operation effect within a preset time period.

[0040] The second prediction module continuously acquires the reaction condition parameters of the cracking furnace and inputs the reaction condition parameters into the second prediction model of deep learning to obtain the second predicted value of the cracking depth of the cracking furnace at time T+T0 after the current time T, and forms a third curve reflecting the change of the cracking furnace operation effect within a preset time period.

[0041] The correction module calculates the similarity between each pair of the first curve, the second curve, and the third curve. When any one of them exceeds the preset similarity standard value, the operation of the pyrolysis furnace is corrected to make the changing trends of the first curve, the second curve, and the third curve consistent.

[0042] Preferably, the operating parameters are the operation and control conditions of the pyrolysis furnace determined according to the type of furnace tubes and the model of the pyrolysis furnace. The operating parameters include one or more of the following: total steam feed ratio, total feed amount, COT, target outlet temperature, feed rate of each furnace tube group, feedwater flow rate, bottom fuel gas flow rate, sidewall fuel flow rate, steam drum liquid level, radiant section furnace pressure, and feed cross section pressure.

[0043] The first prediction model is a trained long short-term memory network model. The training dataset used includes a set of training samples, which are the historical operation data of the current cracking furnace in cracking the current raw material, including sample values ​​of operating parameters and corresponding sample values ​​of cracking depth.

[0044] The training process of the first prediction model includes:

[0045] Obtain the initialized Long Short-Term Memory (LSTM) network model. The input of the LSM network model is the sample values ​​of the operating parameters, and the output is the predicted value of the rift depth.

[0046] Input the sample values ​​of the operating parameters into the long short-term memory network model, and calculate the deviation between the predicted value of the rupture depth and the sample value of the rupture depth through the error loss function;

[0047] Training ends when the deviation is less than a first preset threshold or when the training reaches a first preset number of iterations; otherwise, the long short-term memory network model is corrected, and the next round of training continues.

[0048] The model with the smallest deviation is selected as the final first prediction model.

[0049] Preferably, the reaction condition parameters are the feed ratio, pyrolysis gas and intermediate product content, and temperature and pressure conditions of the pyrolysis reaction determined according to the reaction mechanism of the pyrolysis feedstock. The reaction condition parameters include one or more of the following: pyrolysis gas pressure, bottom fuel gas pressure, temperature deviation and COT deviation of each furnace tube group, feed cross section temperature of each furnace tube group, pyrolysis gas outlet temperature of the waste heat boiler, furnace temperature, outlet temperature of each group of feed furnaces, and outlet temperature of each group of feed mixing and preheating section.

[0050] The second prediction model is a trained long short-term memory network model. The training dataset used includes a set of training samples, which are historical reaction condition data of the current cracking furnace cracking the current raw material. The training samples include sample values ​​of reaction condition parameters and corresponding cracking depth sample values.

[0051] The training process for the second prediction model includes:

[0052] Obtain the initialized Long Short-Term Memory (LSTM) network model. The input of the LSM network model is the sample values ​​of the reaction condition parameters, and the output is the predicted value of the cleavage depth.

[0053] The sample values ​​of reaction condition parameters are input into the long short-term memory network model, and the deviation between the predicted value of pyrolysis depth and the sample value of pyrolysis depth is calculated by the error loss function.

[0054] Training ends when the deviation is less than the second preset threshold or when the training reaches the second preset number of iterations; otherwise, the long short-term memory network model is corrected, and the next round of training continues.

[0055] The model with the smallest deviation is selected as the final second prediction model.

[0056] Furthermore, the training dataset was obtained by collecting historical operating data from the pyrolysis furnace, and then classifying and filtering the data.

[0057] Historical operating data includes: properties of cracking feedstock oil, type of cracking furnace tubes, relevant parameters monitored by DCS, and corresponding ethylene and propylene yields;

[0058] The classification of historical operating data includes: classifying cracking attributes according to the properties of cracking feedstock, based on one or more of the following performance indicators: relative density, distillation range, group composition, viscosity, hydrogen content, and average molecular weight of the oil.

[0059] The filtering of historical operating data includes deleting operating data that meets the following conditions: the error of any one of the relevant parameters, ethylene yield, and propylene yield exceeds the corresponding preset range, or there are abnormal operating conditions.

[0060] On the other hand, the present invention provides a machine-readable storage medium storing instructions for causing a machine to perform an optimized method for operating a pyrolysis furnace according to the present application.

[0061] On the other hand, the present invention provides a processor for running a program, wherein the program is run to perform an optimized method for pyrolysis furnace operation according to the present application.

[0062] The above technical solution obtains a first curve representing the trend of the pyrolysis furnace's operating performance within a preset time period. Then, using a first prediction model based on deep learning and the pyrolysis furnace's operating parameters, a second curve is generated, representing the trend of the first predicted value of the pyrolysis furnace's performance within the same time period. This second curve represents the trend of the first predicted value of the pyrolysis furnace's performance within the same time period. Finally, using a second prediction model based on deep learning and the pyrolysis furnace's reaction condition parameters, a third curve is generated, representing the trend of the second predicted value of the pyrolysis furnace's performance within the same time period. The trends of the three curves are compared within the same time period, and the operation of the pyrolysis furnace is corrected when the trends do not match, enabling real-time optimization of the pyrolysis furnace's operation.

[0063] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0064] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0065] Figure 1 This is a flowchart of an embodiment of the optimized operation method of the pyrolysis furnace according to this application;

[0066] Figure 2 yes Figure 1 A schematic diagram of the training process of the first prediction model in the embodiment; and

[0067] Figure 3 This is a structural diagram of an embodiment of the optimization device for pyrolysis furnace operation according to this application. Detailed Implementation

[0068] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0069] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0070] Additionally, it should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0071] This invention provides a method for optimizing the operation of a pyrolysis furnace, which can quickly and effectively optimize the real-time operation of the pyrolysis furnace. The flow of one embodiment of the pyrolysis furnace operation optimization method of this application is as follows: Figure 1 As shown, it includes:

[0072] Step 1: Continuously acquire the pyrolysis depth of the pyrolysis furnace and generate a first curve reflecting the changes in the pyrolysis furnace's operating performance within a preset time period;

[0073] Step 2: Continuously acquire the operating parameters of the pyrolysis furnace and input the operating parameters into the first prediction model of deep learning to obtain the first predicted value of the pyrolysis depth of the pyrolysis furnace at time T+T0 after the current time T, and form a second curve reflecting the change of the pyrolysis furnace operation effect within a preset time period.

[0074] Step 3: Continuously acquire the reaction condition parameters of the pyrolysis furnace and input these parameters into the second prediction model of deep learning to obtain the second predicted value of the pyrolysis depth at time T+T0 after the current time T, and form a third curve reflecting the changes in the pyrolysis furnace operating effect within a preset time period; and

[0075] Step 4: Calculate the similarity between each pair of the first curve, the second curve, and the third curve. If any one of them exceeds the preset similarity value, adjust the operation of the pyrolysis furnace to make the changing trends of the first curve, the second curve, and the third curve consistent.

[0076] In actual production, pyrolysis furnace operators can refer to the vector composed of various operating variables of the pyrolysis furnace and its corresponding pyrolysis depth, the first and second predicted values ​​of the pyrolysis depth, and select the optimal operating conditions of the pyrolysis furnace according to the actual production needs on site, and adjust and optimize the operating status of the pyrolysis furnace.

[0077] In this embodiment, the cracking furnace can be of type 1, 1-1, 2-1, 4-1, 4-1-1-1, 2-1-1-1, 1-1-1-1, or 8-4-2-1. The cracking feedstock of the cracking furnace is ethane, propane, LPG, naphtha, diesel, aviation kerosene, or hydrotreated tail oil, etc.

[0078] It should be noted that steps 1-3 above refer to the continuous monitoring of relevant parameters of the pyrolysis furnace. In practice, these parameters can be monitored periodically, at regular intervals, or simultaneously or alternately as needed, and are not necessarily limited to the order of steps 1-3. The similarity between any two of the first, second, and third curves can be determined by Euclidean distance, Manhattan distance, Chebyshevsky distance, Minkowski distance, standardized Euclidean distance, Mahalanobis distance, cosine similarity, etc.

[0079] It should also be noted that the pyrolysis furnace is typically in continuous operation, and its operating parameters and reaction condition parameters are continuously acquired according to this application. Therefore, the first and second predicted values ​​obtained based on the operating parameters and reaction condition parameters can form a change curve corresponding to the pyrolysis furnace's operating time. In this embodiment, the first and second curves showing the change in the pyrolysis furnace's operating effect within a preset time period are partial curves extracted from the aforementioned change curves according to actual needs. This time period can be set between 0 and 5 hours, preferably the most recent 1-30 minutes, to determine whether the trends of the first and second predicted values ​​within this period are consistent. Furthermore, the value of T0 can be determined reasonably during the training process of the first and second prediction models; therefore, the value can be the same as or different from the length of the preset time period.

[0080] In step 1, the cracking depth can be determined based on the cracking feedstock and the cracking reaction mechanism. For example, if the cracking feedstock is naphtha, the cracking depth can be the propylene-ethylene content ratio in the cracked gas, i.e., the cracking reaction depth. In other words, in this application, the cracking depth is a parameter reflecting the operating effect of the cracking furnace. In actual operation, it can also be the ethylene or propylene yield, and the ethylene and propylene content in the cracked gas at the current moment can be measured using an online analyzer such as a gas chromatograph to obtain the propylene-ethylene content ratio in the cracked gas. Depending on the cracking feedstock, the cracking depth can also be the feedstock conversion rate, methane yield, methane / propylene yield ratio, hydrogen content of the cracked liquid product, outlet temperature, kinetic depth, or the yield of C3 and lighter components, etc.

[0081] Taking naphtha as the cracking feedstock as an example, the composition of its cracking products is constrained by various factors, such as feedstock quality, residence time, cracking furnace tube configuration, hydrocarbon partial pressure, and reaction temperature. In steps 2 and 3, the operating parameters and reaction condition parameters of the cracking furnace can be selected from the DCS system's detection parameters based on experience, reflecting the cracking furnace's operating conditions; or all parameters corresponding to the cracking furnace's operation and reaction conditions can be selected from the DCS system's detection parameters; or important parameters can be pre-determined based on artificial intelligence algorithms.

[0082] In step 2, the first prediction model is a trained long short-term memory network model based on time series analysis. The training samples are the historical operating data of the current cracking furnace in cracking the current raw material. That is, the first prediction model is a model that reflects the correspondence between cracking furnace operation and cracking depth. By analyzing the historical operating data of the cracking furnace, a prediction model for cracking depth is established. In step 3, the second prediction model is a trained long short-term memory network model. The training samples are the historical reaction condition data of the current cracking furnace in cracking the current raw material. That is, the second prediction model is a prediction model based on reaction mechanism. By analyzing the historical reaction condition data of the cracking furnace, a prediction model for cracking depth is established, reflecting the correspondence between reaction conditions and cracking depth in the cracking furnace.

[0083] It should also be noted that this application can not only use the cracking depth of the cracking furnace as the object of monitoring and prediction, but also, depending on different cracking production purposes, any one of the following can be used as the object of monitoring and prediction: the maximum value of ethylene, propylene, ethylene and propylene in the cracking products, the maximum yield of triene, or the maximum yield of cracking product value. That is, the "cracking depth" in this embodiment can be replaced with any one of the above monitoring objects so that the cracking reaction proceeds in the desired direction.

[0084] Compared with the prior art, the technical advantages of this invention include:

[0085] (1) The long short-term memory algorithm based on time series analysis is used to analyze the historical operating data of the cracking furnace and establish a prediction model for cracking depth. Compared with the traditional time series analysis method, this method has obvious advantages in dealing with long historical data sequences. While ensuring that the information of the nearest time nodes is fully utilized, it can also ensure that the information of nodes with far similarity is not forgotten, thereby improving the utilization rate of historical data and the accuracy of the prediction model.

[0086] (2) A second prediction model is established based on the pyrolysis reaction mechanism, which can be mutually verified with the prediction results of the first prediction model. At the same time, it reduces the dependence on online analysis instruments and improves the real-time performance of the pyrolysis furnace operation optimization method.

[0087] (3) Establish a multi-objective real-time operation optimization model for cracking furnaces, providing a set of multiple control schemes for on-site cracking furnace operators, which can be flexibly selected according to on-site production needs, effectively improving the overall production efficiency and economic benefits of the ethylene plant.

[0088] In some implementations, the training dataset used by the first prediction model is a set of training samples. These training samples are historical operational data of the current cracking furnace cracking the current feedstock. The training samples include sample values ​​of operating parameters and corresponding cracking depths. The training samples are obtained from offline operating data of the cracking furnace over the past 5-10 years, collected in a time-series manner. The collected data includes the properties of the cracking feedstock oil, the type of cracking furnace tubes, all relevant operating parameters monitored on the DCS, and the corresponding diene yields. This data is used to establish a cracking furnace cracking depth database. Then, based on experience, important parameters reflecting the operating conditions of the cracking furnace are selected from the detection parameters of the DCS system. The operating parameters are the cracking furnace operation control conditions determined according to the type of cracking furnace tubes and the cracking furnace model. These operating parameters include one or more of the following: total steam feed ratio, total feed volume, COT (Cost of Oxide), target outlet temperature, feed rate of each furnace tube group, feedwater flow rate, bottom fuel gas flow rate, sidewall fuel flow rate, steam drum level, radiant furnace pressure, and feed cross-section pressure.

[0089] In some implementations, the composition of naphtha cracking products is considered to be influenced by various factors, such as the quality of the cracking feedstock, residence time, cracking furnace tube configuration, hydrocarbon partial pressure, and reaction temperature. Operating parameters include the cracking feedstock flow rate, cracking feedstock temperature and pressure, dilution steam feed flow rate, dilution steam feed temperature, cracking furnace cross section temperature, cracking furnace cross section pressure, cracking furnace tube outlet temperature and pressure, waste heat boiler outlet temperature, cracking furnace furnace temperature, and burner fuel flow rate.

[0090] In some implementations, different furnace tube configurations are used for cracking furnaces, such as single-pass furnace tubes, 1-1 type, 2-1 type and 4-1 type two-pass furnace tubes, and 1-1-1-1 type and 2-1-1-1 type four-pass furnace tubes. For example, ethylene cracking furnaces using naphtha as cracking feedstock often use 1-1 type two-pass furnace tubes or 2-1 type two-pass furnace tubes. Therefore, when acquiring historical operating condition data of the cracking furnace, in addition to collecting cracking furnace operating parameter data, it is also necessary to include the corresponding cracking furnace tube configuration parameters in the cracking furnace cracking depth database.

[0091] Before training the first prediction model, the training dataset or the pyrolysis furnace pyrolysis depth database must be classified and filtered.

[0092] In some implementations, naphtha is classified according to its cracking properties, and naphtha cracked products are divided into L categories with similar cracking performance. The classification is based on its main performance indicators, namely the relative density, distillation range, group composition (PONA value), viscosity, hydrogen content, and average molecular weight of the product.

[0093] Based on the classification, the relevant data in each oil product classification are corrected, and the collected data is cleaned to delete cracking furnace operating condition data and diene yield data containing obvious errors. Data cleaning refers to screening the extracted historical operating condition data of the cracking furnace according to the range of relevant operating condition parameters and diene yield range during the operation of the naphtha cracking to ethylene cracking furnace, and removing abnormal operating condition values ​​with obvious errors.

[0094] Furthermore, the corrected data is integrated, with 50%-95% of the total data used as the training set and the remaining 50%-5% used as the test set.

[0095] refer to Figure 2 As shown, during the training process of the first prediction model, the input is the sample value of the operation parameters of the training sample n, where the training sample n is the nth training sample selected from a set of training samples in the above training dataset, and the sample value of the operation parameters should include the operation conditions of the pyrolysis furnace at time t for all selected operation parameters; the output is the predicted value of the pyrolysis depth, and the prediction deviation of the current first prediction model is determined by comparing the predicted value of the pyrolysis depth with the sample value of the pyrolysis depth in the training sample n through the error loss function.

[0096] It should be noted that the input operating parameter sample value is the cracking furnace operating condition at time t. Correspondingly, the cracking depth sample value used for comparison with the predicted value should be the actual cracking depth at time t+n (n≥1), and the output is the predicted cracking depth at time t+n (n≥1). Here, the value of n can be any time before the end of the current feedstock cracking production, such as 1 min, 5 min, 10 min, 13 min, 15 min, 24 min, 35 min, 42 min, 48 min, 54 min, 60 min, or even longer periods such as 2 hours, 3 hours, or 5 hours.

[0097] The training process of the first prediction model is as follows: First, an initialized Long Short-Term Memory (LSTM) network model is obtained. The input of the LSM network model is the sample value of the operating parameters, and the output is the predicted value of the rift depth. Then, multiple training sessions are conducted according to a preset number of training sessions and a preset deviation threshold. Each training session includes inputting the sample value of the operating parameters into the LSM network model, calculating the deviation between the predicted value of the rift depth and the sample value of the rift depth using an error loss function, and ending the training when the deviation is less than the first preset threshold or when the training reaches the first preset number of sessions. Otherwise, the LSM network model is corrected, and the next round of training continues. Finally, the model with the smallest deviation is selected as the final first prediction model.

[0098] In some implementations, the root mean square error loss is used to calculate the deviation between the two, and the models saved after multiple training sessions are selected, with the one with the smallest root mean square error being the final prediction model.

[0099] The second prediction model is a trained long short-term memory network model. The training dataset can be a set of training samples selected from the cracking furnace cracking depth database established above. The training samples are the historical reaction condition data of the current cracking furnace cracking the current raw material. The training samples include sample values ​​of reaction condition parameters and corresponding cracking depth sample values.

[0100] It should be noted that the input operating parameter sample value is the cracking furnace operating condition at time t. Correspondingly, the cracking depth sample value used for comparison with the predicted value should be the actual cracking depth at time t+n (n≥1), and the output is the predicted cracking depth at time t+n (n≥1). Here, n can be any time before the end of the current feedstock cracking production, such as 1 min, 5 min, 10 min, 13 min, 15 min, 24 min, 35 min, 42 min, 48 min, 54 min, 60 min, or even longer periods like 2 hours, 3 hours, or 5 hours. Preferably, n is 1-30 minutes.

[0101] In some embodiments, the reaction condition parameters are the feed ratio of the pyrolysis reaction, the content of pyrolysis gas and intermediate products, and the monitored values ​​of temperature and pressure conditions of the pyrolysis reaction, determined according to the reaction mechanism of the pyrolysis feedstock. The reaction condition parameters include one or more of the following: pyrolysis gas pressure, bottom fuel gas pressure, temperature deviation and COT deviation of each furnace tube group, feed cross section temperature of each furnace tube group, pyrolysis gas outlet temperature of the waste heat boiler, furnace temperature, outlet temperature of each group of feed furnaces, and outlet temperature of the first stage of feed mixing and preheating of each group.

[0102] The training process of the second prediction model is as follows: First, an initialized Long Short-Term Memory (LSTM) network model is obtained. The input of the LSM network model is the sample value of the reaction condition parameter, and the output is the predicted value of the rift depth. Then, multiple training sessions are conducted according to a preset number of training sessions and a preset deviation threshold. Each training session includes inputting the sample value of the reaction condition parameter into the LSM network model, calculating the deviation between the predicted value of the rift depth and the sample value of the rift depth using an error loss function, and ending the training when the deviation is less than the second preset threshold or when the training reaches the second preset number of sessions. Otherwise, the LSM network model is corrected, and the next round of training continues. Finally, the model with the smallest deviation is selected as the final second prediction model.

[0103] In some implementations, a virtual component model is established based on the main performance parameters of the pyrolysis feedstock. A second prediction model is then established on this basis as a pyrolysis reaction mechanism model (hereinafter referred to as the pyrolysis reaction mechanism model). The pyrolysis furnace operating conditions at time t are used as input variables to calculate the predicted value of the pyrolysis depth corresponding to the operating conditions. This predicted value is compared with the pyrolysis depth operating value obtained by the online pyrolysis gas analyzer to ensure that the relative error between the two is less than a given threshold of 2%. Otherwise, the pyrolysis reaction mechanism model is corrected, and finally, an optimal pyrolysis reaction mechanism model is established.

[0104] In some implementations, naphtha is used as the cracking feedstock. Since naphtha is a mixture of different hydrocarbons, its specific hydrocarbon composition is difficult to obtain through measurement. In actual industrial operations, its cracking performance is mainly characterized based on its key performance indicators. However, when establishing a specific cracking reaction mechanism model, it is necessary to construct its detailed hydrocarbon composition. For example, for naphtha as a special cracking feedstock, a virtual component model corresponding to its key performance indicators is established, mainly including C5-C12 alkanes and their isomers. Based on this, a cracking reaction mechanism model corresponding to this hydrocarbon composition is established. During the cracking process, the calculation results of the mechanism model are compared in real time with the measurement data from the online cracking gas analyzer, and the operation of the cracking furnace is optimized accordingly.

[0105] In some implementations, the goal is to maximize ethylene and propylene yields. The maximum and minimum allowable values ​​of each operating variable of the cracking furnace are given as constraints to establish a multi-objective real-time operation optimization model for the cracking furnace. Finally, the optimized setpoints of each operating variable of the cracking furnace are obtained, forming the optimal solution geometry of each control variable and its corresponding ethylene and propylene yields. This is provided to the on-site operators so that they can select the corresponding optimal operating conditions to optimize and adjust the operation of the cracking furnace in real time according to the current actual production needs.

[0106] Current actual production needs mainly include the following situations:

[0107] Based on the current market prices of ethylene and propylene, determine the specific values ​​of the output variables to maximize profits;

[0108] • Based on the different requirements of downstream processes of the cracking furnace for ethylene and propylene, select the one with the highest demand and determine the specific value of the output variable;

[0109] • Based on the range of the ratio of ethylene to propylene yield on site, we aim to obtain the highest overall yield of ethylene and propylene, and thus determine the specific value of the output variable;

[0110] When the production process does not have specific requirements for the yield of other products, and only the ethylene yield is of concern,

[0111] Choose the option with the highest ethylene yield and determine the specific values ​​for the output variables.

[0112] This invention provides an optimized device for the operation of a pyrolysis furnace, the structure of which is as follows: Figure 3 As shown, the system includes an operation monitoring module, a first prediction module, a second prediction module, and a correction module. The operation monitoring module continuously acquires the pyrolysis depth of the pyrolysis furnace and generates a first curve reflecting the changes in the furnace's operating performance over a preset time period. The first prediction module continuously acquires the operating parameters of the pyrolysis furnace and inputs these parameters into a deep learning first prediction model to obtain a first predicted value of the pyrolysis depth at time T+T0 after the current time T, generating a second curve reflecting the changes in the furnace's operating performance over the preset time period. The second prediction module continuously acquires the reaction condition parameters of the pyrolysis furnace and inputs these parameters into a deep learning second prediction model to obtain a second predicted value of the pyrolysis depth at time T+T0 after the current time T, generating a third curve reflecting the changes in the furnace's operating performance over the preset time period. The correction module calculates the similarity between each pair of the first, second, and third curves. If any one of these curves exceeds a preset similarity standard value, the pyrolysis furnace operation is corrected to ensure that the changing trends of the first, second, and third curves are consistent.

[0113] In this embodiment, the operation monitoring module is used to monitor changes in the operation effect of the cracking furnace, specifically the changes in the monitored value of the cracking depth (or the calculated value obtained from the monitored value). The cracking depth can be the ethylene content in the cracked gas or the propylene-ethylene ratio in the cracked gas. The first prediction module and the second prediction module obtain the first and second predicted values ​​of the cracking depth based on the monitored operating parameters and reaction condition parameters of the cracking furnace, respectively. When the changing trends of the monitored value of the cracking depth and the first and second predicted values ​​are inconsistent, it is considered that the current operation of the cracking furnace needs to be optimized. The vector composed of the various operating variables of the cracking furnace and its corresponding cracking depth, as well as the first and second predicted values ​​of the cracking depth, are provided to the cracking furnace operator so that he / she can select the optimal operating conditions of the cracking furnace according to the actual production needs on site, and adjust and optimize the operating status of the cracking furnace.

[0114] In this embodiment, the training methods of the first prediction model and the second prediction model, as well as the training dataset used, can be referred to the implementation method of the optimization method for pyrolysis furnace operation, and will not be repeated here.

[0115] This invention also provides an optimization device for pyrolysis furnace operation, including a processor and a memory. The aforementioned operation monitoring module, first prediction module, second prediction module, and correction module are all stored as program units in the memory, and the processor executes the aforementioned program units stored in the memory to achieve the corresponding functions.

[0116] The processor contains a kernel, which retrieves the corresponding program unit from memory. One or more kernels can be configured; adjusting kernel parameters determines if there are any deviations in the pyrolysis furnace and makes corresponding adjustments.

[0117] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0118] This invention provides a storage medium storing a program that, when executed by a processor, implements the optimization method for pyrolysis furnace operation as described in this application.

[0119] This invention provides a processor for running a program, wherein the program executes the optimization method for pyrolysis furnace operation of this application.

[0120] This invention provides an apparatus including a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the optimized method for operating a pyrolysis furnace as described in this application. The apparatus described herein may be a server, PC, PAD, mobile phone, etc.

[0121] This application also provides a computer program product that, when executed on a data processing device, is adapted to perform the program of the optimized method steps for initializing the operation of the pyrolysis furnace of this application.

[0122] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0123] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0124] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0125] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0126] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0127] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0128] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0129] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0130] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. An optimization method for pyrolysis furnace operation, comprising: The pyrolysis depth of the pyrolysis furnace is continuously acquired, and a first curve reflecting the changes in the pyrolysis furnace's operating performance over a preset time period is generated. The operating parameters of the pyrolysis furnace are continuously acquired and input into the first prediction model of deep learning to obtain the first predicted value of the pyrolysis depth of the pyrolysis furnace at time T+T0 after the current time T, and a second curve reflecting the change of the pyrolysis furnace operation effect within the preset time period is formed. The reaction condition parameters of the pyrolysis furnace are continuously acquired and input into the second prediction model of deep learning to obtain the second predicted value of the pyrolysis depth of the pyrolysis furnace at time T+T0 after the current time T, and a third curve reflecting the change of the pyrolysis furnace operation effect within the preset time period is formed. and The similarity between each pair of the first curve, the second curve, and the third curve is calculated. If any one of them exceeds a preset similarity value, the operation of the pyrolysis furnace is corrected to make the changing trends of the first curve, the second curve, and the third curve consistent. The operating parameters are the pyrolysis furnace operation control conditions determined according to the pyrolysis furnace tube type and pyrolysis furnace model. These operating parameters include one or more of the following: total steam feed ratio, total feed volume, COT (Cost-to-Oil) temperature, target outlet temperature, feed rate for each furnace tube group, feedwater flow rate, bottom fuel gas flow rate, sidewall fuel flow rate, steam drum liquid level, radiant furnace pressure, and feed cross section pressure. The reaction condition parameters are the feed ratio, pyrolysis gas and intermediate product content, and temperature and pressure conditions of the pyrolysis reaction determined according to the reaction mechanism of the pyrolysis feedstock. The reaction condition parameters include one or more of the following: pyrolysis gas pressure, bottom fuel gas pressure, temperature deviation and COT deviation of each furnace tube group, feed cross section temperature of each furnace tube group, pyrolysis gas outlet temperature of the waste heat boiler, furnace temperature, outlet temperature of each group of feed furnaces, and outlet temperature of each group of feed mixing and preheating section.

2. The method for optimizing the operation of a pyrolysis furnace according to claim 1, characterized in that, The cracking depth is determined based on the cracking feedstock and cracking reaction mechanism, wherein the cracking depth includes any one of the following: feedstock conversion rate, methane yield, propylene / ethylene yield ratio, methane / propylene yield ratio, hydrogen content of the cracked liquid product, outlet temperature, kinetic depth, or yield of C3 and lighter components. The continuous acquisition of the pyrolysis depth of the pyrolysis furnace includes: The content of each target component in the pyrolysis gas at the current moment is measured by gas chromatography; and The cleavage depth is calculated based on the content of each target component.

3. The method for optimizing the operation of a pyrolysis furnace according to claim 1, characterized in that, The first prediction model is a trained long short-term memory network model. The training dataset used includes a set of training samples, which are the historical operation data of the current cracking furnace cracking the current raw material. The training samples include sample values ​​of operation parameters and corresponding sample values ​​of cracking depth. The training process of the first prediction model includes: Obtain an initialized long short-term memory network model, wherein the input of the long short-term memory network model is the sample value of the operating parameters, and the output is the predicted value of the rupture depth; The sample values ​​of the operating parameters are input into the long short-term memory network model, and the deviation between the predicted rift depth and the sample values ​​of the rift depth is calculated using the error loss function. Training ends when the deviation is less than a first preset threshold or the training reaches a first preset number of iterations; otherwise, the long short-term memory network model is corrected, and the next round of training continues. The model with the smallest deviation is selected as the final first prediction model.

4. The optimized operation method of the pyrolysis furnace according to claim 1, characterized in that, The second prediction model is a trained long short-term memory network model. The training dataset used includes a set of training samples, which are historical reaction condition data of the current cracking furnace cracking the current raw material. The training samples include sample values ​​of reaction condition parameters and corresponding cracking depth sample values. The training process of the second prediction model includes: Obtain an initialized long short-term memory network model, wherein the input of the long short-term memory network model is the sample value of the reaction condition parameter, and the output is the predicted value of the cleavage depth; The sample values ​​of the reaction condition parameters are input into the long short-term memory network model, and the deviation between the predicted rupture depth and the sample rupture depth is calculated using the error loss function. Training ends when the deviation is less than a second preset threshold or when the training reaches a second preset number of iterations; otherwise, the long short-term memory network model is corrected, and the next round of training continues. The model with the smallest deviation is selected as the final second prediction model.

5. The optimized method for operating a pyrolysis furnace according to claim 3 or 4, characterized in that, The training dataset is obtained by collecting historical operating data of the pyrolysis furnace, and then classifying and filtering the data. The historical operating data includes: the properties of the cracking feedstock oil, the type of cracking furnace tubes, relevant parameters monitored by DCS, and the corresponding ethylene and propylene yields; The classification of the historical operating data includes: classifying the cracking attributes according to the properties of the cracking feedstock, based on one or more of the following performance indicators: relative density, distillation range, group composition, viscosity, hydrogen content, and average molecular weight of the oil. The filtering of the historical operating data includes deleting operating data that meets the following conditions: the error of any one of the relevant parameters, the ethylene yield, and the propylene yield exceeds the corresponding preset range, or there are abnormal operating conditions.

6. The method for optimizing the operation of a pyrolysis furnace according to claim 1, characterized in that, The preset time period is n1 minutes, and T0 is n2 minutes, where n1 and n2 are both between 0 and 300.

7. The method for optimizing the operation of a pyrolysis furnace according to claim 6, characterized in that, Both n and n2 are between 1 and 30.

8. The method for optimizing the operation of a pyrolysis furnace according to claim 1, characterized in that, The pyrolysis furnace is of type 1, 1-1, 2-1, 4-1, 4-1-1-1, 2-1-1-1, 1-1-1-1, or 8-4-2-1.

9. The method for optimizing the operation of a pyrolysis furnace according to claim 1, characterized in that, The cracking feedstock of the cracking furnace is ethane, propane, LPG, naphtha, diesel, aviation kerosene, or hydrotreated tail oil.

10. An optimization apparatus for pyrolysis furnace operation, comprising: The operation monitoring module continuously acquires the pyrolysis depth of the pyrolysis furnace and generates a first curve reflecting the changes in the pyrolysis furnace's operating performance within a preset time period; The first prediction module continuously acquires the operating parameters of the pyrolysis furnace and inputs the operating parameters into the first prediction model of deep learning to obtain the first predicted value of the pyrolysis depth of the pyrolysis furnace at time T+T0 after the current time T, and forms a second curve reflecting the change of the pyrolysis furnace operation effect within the preset time period. The second prediction module continuously acquires the reaction condition parameters of the pyrolysis furnace and inputs the reaction condition parameters into the second prediction model of deep learning to obtain the second predicted value of the pyrolysis depth of the pyrolysis furnace at time T+T0 after the current time T, and forms a third curve reflecting the change of the pyrolysis furnace operation effect within the preset time period. The correction module calculates the similarity between each pair of the first curve, the second curve, and the third curve. If any one of them exceeds a preset similarity standard value, the pyrolysis furnace operation is corrected to ensure that the changing trends of the first curve, the second curve, and the third curve are consistent. The operating parameters are the pyrolysis furnace operation control conditions determined according to the pyrolysis furnace tube type and pyrolysis furnace model. These operating parameters include one or more of the following: total steam feed ratio, total feed volume, COT (Cost-to-Oil) temperature, target outlet temperature, feed rate for each furnace tube group, feedwater flow rate, bottom fuel gas flow rate, sidewall fuel flow rate, steam drum liquid level, radiant furnace pressure, and feed cross section pressure. The reaction condition parameters are the feed ratio, pyrolysis gas and intermediate product content, and temperature and pressure conditions of the pyrolysis reaction determined according to the reaction mechanism of the pyrolysis feedstock. The reaction condition parameters include one or more of the following: pyrolysis gas pressure, bottom fuel gas pressure, temperature deviation and COT deviation of each furnace tube group, feed cross section temperature of each furnace tube group, pyrolysis gas outlet temperature of the waste heat boiler, furnace temperature, outlet temperature of each group of feed furnaces, and outlet temperature of each group of feed mixing and preheating section.

11. The optimized apparatus for pyrolysis furnace operation according to claim 10, characterized in that, The first prediction model is a trained long short-term memory network model. The training dataset used includes a set of training samples, which are the historical operation data of the current cracking furnace in cracking the current raw material, including sample values ​​of operation parameters and corresponding sample values ​​of cracking depth. The training process of the first prediction model includes: Obtain an initialized long short-term memory network model, wherein the input of the long short-term memory network model is the sample value of the operating parameters, and the output is the predicted value of the rupture depth; The sample values ​​of the operating parameters are input into the long short-term memory network model, and the deviation between the predicted rift depth and the sample values ​​of the rift depth is calculated using the error loss function. Training ends when the deviation is less than a first preset threshold or the training reaches a first preset number of iterations; otherwise, the long short-term memory network model is corrected, and the next round of training continues. The model with the smallest deviation is selected as the final first prediction model.

12. The optimized apparatus for pyrolysis furnace operation according to claim 10, characterized in that, The second prediction model is a trained long short-term memory network model. The training dataset used includes a set of training samples, which are historical reaction condition data of the current cracking furnace cracking the current raw material. The training samples include sample values ​​of reaction condition parameters and corresponding cracking depth sample values. The training process of the second prediction model includes: Obtain an initialized long short-term memory network model, wherein the input of the long short-term memory network model is the sample value of the reaction condition parameter, and the output is the predicted value of the cleavage depth; The sample values ​​of the reaction condition parameters are input into the long short-term memory network model, and the deviation between the predicted rupture depth and the sample rupture depth is calculated using the error loss function. Training ends when the deviation is less than a second preset threshold or when the training reaches a second preset number of iterations; otherwise, the long short-term memory network model is corrected, and the next round of training continues. The model with the smallest deviation is selected as the final second prediction model.

13. The optimized apparatus for pyrolysis furnace operation according to claim 11 or 12, characterized in that, The training dataset is obtained by collecting historical operating data of the pyrolysis furnace, and then classifying and filtering the data. The historical operating data includes: the properties of the cracking feedstock oil, the type of cracking furnace tubes, relevant parameters monitored by DCS, and the corresponding ethylene and propylene yields; The classification of the historical operating data includes: classifying the cracking attributes according to the properties of the cracking feedstock, based on one or more of the following performance indicators: relative density, distillation range, group composition, viscosity, hydrogen content, and average molecular weight of the oil. The filtering of the historical operating data includes deleting operating data that meets the following conditions: the error of any one of the relevant parameters, the ethylene yield, and the propylene yield exceeds the corresponding preset range, or there are abnormal operating conditions.

14. A machine-readable storage medium storing instructions for causing a machine to perform: an optimized method for operating a pyrolysis furnace as described in any one of claims 1-9.

15. A processor, characterized in that, Used to run a program, wherein the program is run to perform: an optimized method for operating a pyrolysis furnace as described in any one of claims 1-9.

Images