Online optimization method for ethylene cracking furnace based on pattern search algorithm

Abstract

The present application relates to the technical field of petrochemical ethylene production, and is an online optimization method for ethylene cracking furnace based on pattern search algorithm. The data of industrial ethylene cracking furnace system is collected, a cracking furnace mechanism model is established to predict products, the mass fraction of each product simulation is obtained, the sum of square differences between the volume fraction of simulated products and actual products is minimized as the objective function, the outlet temperature deviation Delta COT is obtained, a fuel gas prediction model and an ultra-high pressure steam prediction model are established, the objective function is selected, the pattern search optimization algorithm is used, the COT and the gas hydrocarbon ratio corresponding to the maximum objective function are obtained, the yield of each product after optimization is predicted, and the yield is written back to the cracking furnace controller in a certain step. The present application combines the characteristics of high prediction accuracy of mechanism model, uses the pattern search optimization algorithm, accelerates the optimization convergence speed, improves the yield of target products, and better meets the needs of real-time and economic benefits in the field.

Description

Technical Field

[0001] This invention relates to the field of petrochemical ethylene production technology, and is an online optimization method for ethylene cracking furnaces based on a pattern search algorithm. Background Technology

[0002] The ethylene industry is the leading sector in the petrochemical industry, and ethylene production is a key indicator of a country's petrochemical development level. The cracking furnace is the central component of an ethylene plant; its control significantly impacts the overall product quality and output of the plant, and also affects the stable operation of downstream production units (such as polyethylene, polypropylene, and ethylene glycol production units).

[0003] Cracking depth is a crucial indicator of the extent of reaction in a cracking furnace. Key factors influencing cracking depth include furnace outlet temperature (COT), the vapor-to-hydrogen ratio, outlet pressure, and feedstock composition. Among these, COT has the greatest impact on ethylene and propylene yields, as well as the ethylene-to-propylene yield ratio. Currently, most ethylene cracking units in China use COT to characterize cracking depth, and the COT setpoint is often selected based on specified values ​​or empirical values ​​provided by the cracking furnace patent holder under the designed feedstock conditions. However, domestic ethylene producers now generally face complex feedstock sources and significant fluctuations in oil properties. This means that a fixed COT production operation mode cannot promptly reflect changes in cracking depth and product yield caused by variations in feedstock composition and furnace operating conditions. Therefore, it is necessary to determine the optimal cracking depth in real-time based on the current operating status of the cracking furnace.

[0004] Establishing accurate product yield models for cracking furnaces can improve the accuracy of product yield predictions after furnace optimization. Currently, there are three main types of cracking product yield prediction models: empirical models, mechanistic models, and semi-empirical / semi-mechanistic models combining both. While empirical models converge quickly, significant fluctuations in market oil prices and the influence of refining units on feedstock composition lead to substantial changes in feedstock composition. When new operating conditions arise, the cracking furnace model requires re-collection of data and retraining, consuming considerable human and material resources and resulting in poor model timeliness. Mechanistic models and semi-empirical / semi-mechanistic models offer higher accuracy in product yield prediction. However, in actual production, changes in feedstock properties and process operating conditions cause real-time changes in the operating status of the cracking furnace. Therefore, the online mechanistic models of each cracking furnace change accordingly with various operating conditions. This necessitates real-time calibration of the online cracking furnace models to enable adaptive adjustments based on the actual production process characteristics of the industrial unit, ensuring model accuracy. Summary of the Invention

[0005] This invention provides an online optimization method for ethylene cracking furnaces based on a pattern search algorithm, which overcomes the shortcomings of the existing technologies and can effectively solve the problems of complex raw material sources, large fluctuations in oil properties, and low product yield in ethylene production.

[0006] The technical solution of this invention is achieved through the following measures: an online optimization method for ethylene cracking furnaces based on a pattern search algorithm, which is carried out according to the following steps: First, data from the industrial ethylene cracking furnace system is collected, including cracking furnace operating conditions, oil property data, and furnace structure information, and a cracking furnace mechanism model is established for product prediction; Second, using the cracking furnace mechanism model, based on the current cracking furnace operating conditions, the simulated mass fraction of each product is obtained; Third, the simulated mass fraction of the product is converted into a volume fraction, and then the golden section method is used, with the objective function being minimizing the sum of squares of the differences between the simulated product volume fraction and the actual product volume fraction, so that the simulated product yield is close to the actual product yield, and the deviation of the outlet temperature is obtained. ΔCOT; Fourth step, based on the current furnace type, cracking feedstock, and on-site operating conditions, establish prediction models for fuel gas and ultra-high pressure steam; Fifth step, select the objective function based on actual on-site requirements; Sixth step, using the determined objective function and based on the current cracking furnace operating conditions, use a pattern search optimization algorithm, taking the COT and gas-hydrocarbon ratio as optimization variables, and call the corrected mechanism model from the third step to obtain the COT and gas-hydrocarbon ratio values ​​corresponding to the maximum objective function; Seventh step, based on the calculated optimized COT and gas-hydrocarbon ratio values, call the corrected mechanism model to predict the optimized yields of each product and the propylene-ethyl ratio, and then write the propylene-ethyl ratio and optimized gas-hydrocarbon ratio values ​​back to the cracking furnace controller with a certain step size.

[0007] The following are further optimizations and / or improvements to the above-mentioned technical solution:

[0008] In the first step above, the cracking furnace operating conditions include feed load, dilution steam ratio, feed temperature, feed pressure, outlet temperature and outlet pressure, and the oil property data includes liquid phase feed property data and gas phase feed property data.

[0009] The above liquid-phase feedstock attribute data includes the mass percentage and distillation range data of n-alkanes, isoalkanes, alkenes, cycloalkanes, and aromatics, while the gas-phase feedstock attribute data includes the volume fraction of methane, ethylene, and propylene.

[0010] The above furnace structure includes the dimensional information of the effective length, wall thickness, and inner diameter of the pyrolysis furnace and radiant furnace tubes.

[0011] In the second step above, the cracking furnace mechanism model is either a commercial ethylene cracking furnace simulation software or a model established based on the basic principles of physical and chemical reaction processes.

[0012] In the third step mentioned above, the products include hydrogen, methane, ethylene, ethane, propylene, and propane.

[0013] In the fourth step above, the on-site operating conditions include the cracking feedstock, COT, gas-hydrocarbon ratio, fuel gas composition, feed flow rate, and high-temperature flue gas preheating boiler feedwater (BFW) flow rate.

[0014] In the fourth step above, the fuel gas prediction model is based on the global thermal balance relationship of the pyrolysis furnace, that is, the relationship between the total heat released by fuel gas combustion and the total heat absorbed by the furnace tubes, the heat carried away by the flue gas and the heat loss of the pyrolysis furnace, to predict fuel gas consumption.

[0015] In the fourth step above, the prediction model for ultra-high pressure steam is based on the developed TLE and pyrolysis furnace convection section model. In addition to solving the three basic equations of mass, energy and momentum conservation, the convection section model also involves solving the convective heat transfer equation. The TLE convection section model is used to predict the core heat transfer formula by inputting saturated water flow rate, gasification rate and thermosiphon parameters.

[0016] In the fifth step above, the objective functions include the ethylene yield maximization function, the propylene yield maximization function, the diene yield maximization function, the triene yield maximization function, the single-pass high-by-interest yield maximization function, the overall high-by-interest yield maximization function, and the benefit maximization function.

[0017] The superior effects of this invention are as follows:

[0018] First, this invention eliminates the drawbacks of the traditional method of selecting and configuring models based on the experience of skilled workers and current working conditions.

[0019] Second, this invention combines the high accuracy of product prediction in mechanistic models with a correction method that uses the golden section method to fit the volume yield of light components. This method is also applicable to the correction of different cracking furnace models and has wide applicability.

[0020] Third, this invention establishes a prediction model for fuel gas and SS steam, as well as a product yield distribution obtained based on the correction of the cracking furnace mechanism model, providing reasonable predictions for the production of ethylene cracking units in petrochemical enterprises.

[0021] Fourth, the present invention determines the objective function by sending an optimization selection signal through DCS, which can meet the needs of enterprises to select different optimization indicators according to different actual production requirements, thus having good flexibility.

[0022] Fifth, based on the determined objective function and the corrected mechanism model, this invention utilizes a pattern search optimization algorithm to accelerate the optimization convergence speed, improve the yield of the target product, and better meet the needs of real-time performance and economic benefits in the field. Attached Figure Description

[0023] Appendix Figure 1 This is a process flow diagram of the pyrolysis furnace in Embodiment 11 of the present invention.

[0024] Appendix Figure 2 This is the global thermal balance diagram of the furnace in Embodiment 11 of the present invention.

[0025] Appendix Figure 3 This is a model diagram of the simulated product yield of the ethylene plant in Embodiment 11 of the present invention.

[0026] Appendix Figure 4 This is the optimized execution flowchart in Embodiment 11 of the present invention. Detailed Implementation

[0027] The present invention is not limited to the following embodiments, and the specific implementation can be determined according to the technical solution of the present invention and the actual situation.

[0028] The present invention will be further described below with reference to embodiments:

[0029] Example 1: The online optimization method for ethylene cracking furnaces based on pattern search algorithms is performed as follows: First, data from the industrial ethylene cracking furnace system is collected, including cracking furnace operating conditions, oil property data, and furnace structure information, to establish a cracking furnace mechanism model for product prediction; Second, using the cracking furnace mechanism model, based on the current cracking furnace operating conditions, the simulated mass fraction of each product is obtained; Third, the simulated mass fraction of the product is converted into a volume fraction, and then the golden section method is used, with the objective function being minimizing the sum of squares of the differences between the simulated and actual product volume fractions, so that the simulated product yield is close to the actual product yield, and the deviation of the outlet temperature ΔCOT is obtained; Fourth, The process involves several steps: First, based on the current furnace type, pyrolysis feedstock, and on-site operating conditions, establish prediction models for fuel gas and ultra-high pressure steam. Second, select an objective function based on actual on-site requirements. Third, using the determined objective function and the current pyrolysis furnace operating conditions, employ a pattern search optimization algorithm, taking the COT to hydrocarbon ratio as optimization variables, and calling the corrected mechanistic model from the third step to obtain the COT and hydrocarbon ratio values ​​corresponding to the maximum objective function. Fourth, based on the calculated optimized COT to hydrocarbon ratio values, call the corrected mechanistic model to predict the optimized yields of each product and the propylene-ethylhexane ratio, and then write the propylene-ethylhexane ratio and the optimized hydrocarbon ratio back to the pyrolysis furnace controller with a certain step size.

[0030] Example 2: As an optimization of the above example, in the first step, the cracking furnace operating conditions include feed load, dilution steam ratio, feed temperature, feed pressure, outlet temperature and outlet pressure, and the oil property data includes liquid phase feed property data and gas phase feed property data.

[0031] Example 3: As an optimization of the above examples, the liquid phase feedstock property data includes the mass percentage content and distillation range data of n-alkanes, isoalkanes, alkenes, cycloalkanes, and aromatics, and the gas phase feedstock property data includes the volume fraction of methane, ethylene, and propylene.

[0032] Example 4: As an optimization of the above examples, the furnace structure includes the dimensional information of the effective length, wall thickness, and inner diameter of the pyrolysis furnace radiant furnace tube.

[0033] Example 5: As an optimization of the above examples, in the second step, the cracking furnace mechanism model is a model established using commercial ethylene cracking furnace simulation software or based on the basic principles of physical and chemical reaction processes.

[0034] Example 6: As an optimization of the above examples, in the third step, the products include hydrogen, methane, ethylene, ethane, propylene, and propane.

[0035] Example 7: As an optimization of the above example, in the fourth step, the on-site operating conditions include cracked feedstock, COT, gas-hydrocarbon ratio, fuel gas composition, feed flow rate, and high-temperature flue gas preheating boiler feedwater (BFW) flow rate.

[0036] Example 8: As an optimization of the above example, in the fourth step, the fuel gas prediction model is based on the global heat balance relationship of the pyrolysis furnace, that is, the relationship between the total heat released by fuel gas combustion and the total heat absorbed by the furnace tube, the heat carried away by the flue gas and the heat loss of the pyrolysis furnace, to predict fuel gas consumption.

[0037] Example 9: As an optimization of the above example, in the fourth step, the prediction model for ultra-high pressure steam is predicted by the developed TLE and pyrolysis furnace convection section model. In addition to solving the three basic equations of mass, energy and momentum conservation, the convection section model also involves solving the convective heat transfer equation. The prediction of the TLE convection section model is achieved by solving the core heat transfer formula under the conditions of inputting saturated water flow rate, gasification rate and thermosiphon parameters.

[0038] Example 10: As an optimization of the above example, in the fifth step, the objective function includes the ethylene yield maximization function, the propylene yield maximization function, the diene yield maximization function, the triene yield maximization function, the single-pass high-by-product yield maximization function, the overall high-by-product yield maximization function, and the benefit maximization function.

[0039] To make the technical solution of the present invention clearer and easier to understand, the present invention will be further described in detail below with reference to the accompanying drawings and Embodiment 11.

[0040] Example 11: Optimization of cracking in an industrial ethylene cracking furnace using naphtha as feedstock.

[0041] Industrial pyrolysis furnace systems such as Figure 1 As shown, Figure 1 In the diagram, BA101 is the cracking furnace, which includes an upper convection section and a lower radiation section; F101, F102, F103, F104, and F105 are the liquid phase feed, dilution steam, bottom fuel gas, sidewall fuel gas, and cracking gas, respectively; F1C101, F1C102, and F1C103 are the feed flow meter, steam flow meter, and fuel gas flow meter, respectively; AI101 is the raw material density meter; T1105 and A1105 are the cracking furnace outlet cracking gas temperature indicator and cracking gas online analyzer, respectively; T1C105 is the COT controller; and A1C105 is the cracking depth controller. Figure 1 The system comprises a cascade control system consisting of a cracking depth control system and a COT control system. The measured values ​​of the COT control system originate from the thermocouple TI105 at the outlet of the cracking furnace tube. The set value is calculated from the output of the cracking depth controller AIC105. The data from the online cracking gas composition analyzer AI105 on site is used as a reference and is used as the measured value of the cracking depth controller AIC105.

[0042] Models for simulating product yield in ethylene plants, such as Figure 3 As shown, Figure 3 The operating conditions of the cracking furnace include feed load, dilution steam ratio, feed temperature, feed pressure, outlet temperature, and outlet pressure. Oil product properties include feed density, mass percentage of n-alkanes, isoalkanes, olefins, cycloalkanes, and aromatics, information on different levels of distillation layers, and percentages of methane, ethylene, and propylene in the cracked gas. The furnace structure includes the effective length, wall thickness, and inner diameter of the radiant furnace tubes. The mechanistic model uses commercially available ethylene cracking furnace simulation software or is based on fundamental principles of chemical engineering and chemical reaction engineering, with the cracking reaction mechanism being a free radical reaction mechanism. Simulated product predictions include hydrogen, methane, ethylene, ethane, propylene, propane, 1,3-butadiene, and benzene.

[0043] Predictive models for fuel gas and ultra-high pressure steam (SS) were established. The required on-site operating conditions mainly include pyrolysis feedstock, COT (coal-to-water ratio), steam-to-hydrocarbon ratio, fuel gas composition, feed flow rate, and high-temperature flue gas preheating boiler feedwater (BFW) flow rate. The fuel gas prediction model is primarily based on the global heat balance relationship within the pyrolysis furnace, such as... Figure 3 As shown, the relationship between the total heat released by fuel gas combustion and the total heat absorbed by the furnace tubes, the heat carried away by the flue gas, and the heat loss of the pyrolysis furnace is used to predict fuel gas consumption.

[0044] The prediction of ultra-high pressure steam SS quantity is mainly carried out through the development of TLE and pyrolysis furnace convection section models. In addition to solving the three basic equations of mass, energy and momentum conservation, the convection section model also involves solving the convective heat transfer equation. The TLE model mainly solves the core heat transfer formula by inputting conditions such as saturated water flow rate, gasification rate and thermosiphon parameters.

[0045] The following methods can be used to optimize the cracking depth and gas-to-hydrocarbon ratio, with specific solutions as follows: Figure 4 As shown, the following steps are first taken: First, the operating conditions of the current cracking furnace are collected, including outlet temperature, current feed load, outlet temperature, outlet pressure, and inlet temperature. The naphtha feedstock properties are also collected, including n-alkanes, isoalkanes, cycloalkanes, aromatics, olefins, and some distillation layer information. The furnace tube structure information of the currently operating cracking furnace is also collected. A cracking furnace mechanism model is then established to predict the products, obtaining the yields of key components such as hydrogen, methane, ethylene, ethane, propylene, and propane under the current operating conditions. Then, based on the actual component yields obtained from online instruments, the golden section method is used to minimize the deviation between the simulated and actual component yields, thus determining the COT (Combined Oxidation Time) deviation. Based on the on-site operating conditions, maximizing efficiency is selected as the objective function. Then, a pattern search optimization method is used to set the optimization range for outlet temperature and vapor-to-hydrocarbon ratio. The mechanism model is then invoked to obtain the optimized propylene-to-ethylene ratio and vapor-to-hydrocarbon ratio. The optimized values ​​of the propylene-to-ethylene ratio and vapor-to-hydrocarbon ratio are limited, and a write-back step size is set. Using the current cracking furnace controller value as the initial value, the optimized value is gradually approximated and finally written to the cracking furnace controller. When the next sampling time arrives, the operating conditions and oil properties of the current time period are selected, the model is recalibrated, optimized, and written back.

[0046] 1. Modeling and prediction of fuel gas forecasting models

[0047] A global thermal balance analysis of the pyrolysis furnace chamber is performed, as shown in equation (1).

[0048] Q r =Q ab +Q flue +Q loss (1)

[0049] In equation (1), Q R Q is the total heat released by the combustion of fuel gas. ab Q represents the total heat absorption of all furnace tubes. flue Q represents the enthalpy change from the imported fuel gas and air to the high-temperature flue gas at the outlet, that is, the heat carried away by the flue gas. loss This refers to the heat loss from the pyrolysis furnace.

[0050] Q r Q ab Q flue They can also be expressed as the following formulas:

[0051]

[0052]

[0053]

[0054] Cracking furnaces typically maintain a certain excess air ratio to ensure complete combustion of the fuel gas, therefore the total heat release Q r This can be simply expressed as the lower calorific value ΔH of each component of the fuel gas. i Quality fraction x i The total heat absorption Q of all furnace tubes is the product of the fuel gas flow rate F and the total heat absorption Q of all furnace tubes. ab By measuring the heat flux q flux The heat Q carried away by the flue gas is obtained by integrating along the axial direction of the furnace tube. flue It is the composition of the flue gas in the furnace and the temperature T across the section. flue The function of the furnace heat loss Q loss Typically, the heat release Q r 1%.

[0055] Based on the above analysis, fuel gas modeling and prediction specifically include the following steps:

[0056] 1) Determine the on-site operating conditions (including cracking feedstock, COT, gas-to-hydrogen ratio, fuel gas composition, flow rate, etc.), and calculate the total heat release Q. r and heat loss Q loss .

[0057] 2) Perform coupled simulation to calculate the total heat absorption Q of the furnace tubes. ab By combining formulas (1) and (4), the cross-section temperature T of the flue gas can be solved. flue To address the uncertainties of on-site operating conditions, heat transfer correction parameters are defined and adjusted. This causes the flue gas to cross section temperature T flue The calculated values ​​of the model match the measured values, thus completing the modeling process.

[0058] 3) When optimizing the application, the new furnace tube operating conditions are used as input to perform furnace tube simulation calculations to obtain the total heat absorption Q. ab Based on the heat transfer correction parameter α obtained from the modeling, and combined with formulas (1), (2) and (4), the fuel gas flow rate F under the new operating conditions is predicted.

[0059] Because the pyrolysis characteristics of different pyrolysis feedstocks vary significantly, to ensure the effectiveness of the fuel gas prediction model, similar to the pyrolysis product yield model, this invention establishes a separate fuel gas prediction model for each feedstock and furnace type. Furthermore, the heat transfer correction parameter d is periodically updated to match the continuous changes in fuel gas composition, temperature, and other operating conditions during pyrolysis furnace operation.

[0060] 2. Modeling and Prediction of Ultra-High Pressure Steam SS Prediction Model

[0061] 1) In addition to solving the three basic equations of mass, energy, and momentum conservation, the convection section model also involves solving the convective heat transfer equation. The flow of fluid in the boiler feedwater section, high-pressure steam superheating section, and dilution steam superheating section is mainly a single-phase (gas phase or liquid phase) flow heat transfer process. The calculation of the convective heat transfer coefficient without phase change in a horizontal circular pipe is shown in equation (5):

[0062]

[0063] 2) In the raw material preheating section, the raw material is heated and evaporated, thus a two-phase flow heat transfer process exists. In the initial stage of evaporation, the pipe wall is completely surrounded by liquid and is in a fully wetted state. At this time, the heat transfer mechanism at the pipe wall is mainly liquid-phase flow boiling heat transfer. As the evaporation process proceeds, part of the pipe wall gradually dries, and the pipe wall comes into contact with the gaseous fluid. At this time, not only liquid-phase flow boiling heat transfer but also gas-phase flow heat transfer exists at the pipe wall. The two-phase flow heat transfer coefficient is calculated as shown in equation (6):

[0064]

[0065] h cg and h cl These are the relative heat transfer coefficients of the gas phase and the boiling heat transfer coefficients of the liquid phase, respectively. θ dry It is the drying angle, determined by the current two-phase flow pattern: when the pipe wall is completely wetted, θ dry =0, the process inside the tube is a liquid phase flow, boiling, and heat transfer process; when θ dry When π = 2π, the heat transfer coefficient of two-phase flow is the average heat transfer coefficient of the pipe wall circumference.

[0066] 3) The heat transfer process of flue gas in the convection section includes two parts: convection and radiation. The calculation method for the former is shown in formula (7):

[0067] Q cov =h cf A(T g -T w (7)

[0068]

[0069] Where h cf Let f be the heat transfer coefficient of the flue gas, A be the total heat transfer area, and f be the heat transfer coefficient of the flue gas. g and T w These are the flue gas temperature and the pipe wall temperature, respectively.

[0070] Since the convective heat transfer of flue gas in the tube bundle is not only related to the properties of the flue gas, its flow rate, and the structure of the furnace tubes, but also affected by the arrangement of the tube bundle, formula (8) introduces C H and This describes the influence of structural factors such as tube spacing, tube diameter, and arrangement on heat transfer.

[0071] 3) At the bottom of the convection section, the temperature of the flue gas reaches 800 to 900°C, so the radiative heat of the flue gas cannot be ignored. Since the tubes in the tube bundle are densely arranged, we can consider the tube bundle as a shell surrounding the flue gas. Therefore, the radiative heat transfer from the flue gas to the tube bundle can be approximated as the radiative heat transfer between the gas and the shell. The radiative heat transfer formula is shown in equation (9):

[0072]

[0073] Where, ε c and ε g These represent the emissivity of the furnace tubes and the flue gas, respectively, α. g This is the absorption coefficient of the flue gas. Since there is also strong thermal coupling between the tube bundle and the convection chamber in the convection section, coupled simulation is needed to obtain accurate modeling results. The specific steps are as follows:

[0074] a) Initialize the initial value of the flue gas temperature distribution outside the furnace tubes of each functional segment. Based on the feed temperature and flow rate of raw materials, boiler feedwater, high-pressure steam and other streams, perform flow evaporation simulation in the tube bundle and calculate the heat flux of each layer of furnace tubes in each functional segment.

[0075] b) Based on the heat flux, flue gas inlet temperature and flow rate obtained above, calculate the flue gas temperature distribution outside the furnace tubes of each functional section.

[0076] c) Repeat steps (a) and (b) until the difference between the flue gas temperature distribution results obtained from two adjacent iterations is less than 1℃, the calculation converges, and the final simulation result is output.

[0077] 5) For TLE (Transmission-Oriented Leakage), the high-temperature pyrolysis gas mixture containing pyrolysis products and water vapor exiting the radiant section of the pyrolysis furnace is rapidly cooled to a certain temperature through its tube bundle to reduce secondary reactions in the high-temperature pyrolysis gas and thus ensure the yield of the target product. Boiler feedwater, preheated in the convection section of the pyrolysis furnace, is transferred to the steam drum to replenish the heat exchange fluid. Through the siphon principle, saturated water in the steam drum is sent to the shell side of the waste heat boiler via the downcomer. It undergoes a strong heat exchange process through the waste heat boiler tube wall and the high-temperature pyrolysis gas mixture within the tube side, transforming into saturated steam. This saturated steam is then transported to the steam drum via the upcomer and then to the outside of the steam drum via the saturated steam delivery pipe. In this process, on the one hand, the rapid cooling of the waste heat boiler rapidly cools the pyrolysis gas mixture, reducing secondary reactions and ensuring the yield of the target product; on the other hand, the strong heat exchange process in the waste heat boiler converts high-temperature thermal energy into high-temperature saturated steam, achieving high-grade thermal energy recovery from the high-temperature pyrolysis gas. The mechanism model incorporates a TLE model, which mainly includes three types of input methods: saturated water flow rate, vaporization rate, and thermosiphon parameters. Its core heat transfer formula is shown in equation (10):

[0078]

[0079] The right side of the equation represents the heat released during the cooling and heat exchange process of the cracked gas in the TLE, where F is the cracked gas flow rate, f g，i and T g，o Here, ΔH and ΔH represent the temperatures of the cracked gas at the TLE inlet and outlet, respectively. On the left side of the equation, the heat removed by the saturated water cooling medium in the shell side through evaporation is equal to its enthalpy of vaporization ΔH at the saturation temperature. vap With gas output F s The product of . Therefore, the saturated steam production entering the convection section can be calculated according to equation (10).

[0080] 6) Determine the on-site operating conditions (including the type of pyrolysis material, feed flow rate, COT, gas-hydrocarbon ratio, BFW flow rate, etc.), and calculate the composition distribution of pyrolysis gas products through the pyrolysis furnace tubes.

[0081] 7) Solve the convection section model and TLE model shown in steps 1-5 simultaneously to calculate the steam drum gas generation, phosphorus-free feedwater flow rate, and total SS gas generation. Then, correct the model parameters based on the actual measured values ​​(steam drum gas generation, SS temperature before and after superheating, etc.).

[0082] 8) During optimized application, the new furnace tube operating conditions are used as input to perform furnace tube simulation calculations to obtain the composition distribution of cracked gas products. The convection section and TLE (Transmission Effluent Length) are then coupled and solved to predict the SS (Superconducting Solids) flow rate under the new operating conditions. Addressing the issue that the characteristics of the cracking furnace vary significantly with the cracking feedstock, this invention establishes separate SS prediction models for each feedstock and furnace type. Furthermore, the parameters of the convection section model and TLE model are updated in real-time based on the actual operating conditions, thereby ensuring the real-time effectiveness of the SS prediction models.

[0083] 3. The golden section method correction mechanism model includes the following steps.

[0084] 1) Using the established mechanism model, the current operating conditions are used as input variables to simulate the mass yield of six key components, namely hydrogen, methane, ethylene, ethane, propylene, and propane. The mass yield is then converted into volume yield. The volume yield of these six components is collected using online analytical instruments. The yields of the key components are shown in Table 1.

[0085] 2) Based on the simulated and actual quantities of key components in Table 1, the objective function for mechanistic model correction is established as shown in equation (11):

[0086]

[0087] Note: a i (i = 1, 2, 3, 4, 5, 6) represent the weighting coefficients.

[0088] 3) Determine the optimization variables, input variables, and output variables. Optimization variable: outlet temperature (COT); Input variables: inlet temperature, outlet temperature, cracking furnace feed load, outlet pressure, gas-hydrogen ratio, inlet pressure, raw material industrial index attribute values, furnace tube structure information; Output variables: mass yield of components such as hydrogen, methane, ethylene, ethane, propylene, propane, 1,3-butadiene, and benzene.

[0089] 4) Golden Section Correction Steps

[0090] a) Assume the optimization range of the optimization variable is [a, b], and set the precision eps > 0.

[0091] b) Take lo = a + 0.382(ba) and hi = a + 0.618(ba), call the mechanism model, and calculate the corresponding objective function values ​​obj(lo) and obj(hi) when the outlet temperature is lo and hi.

[0092] c) If obj(lo) > obj(hi), then go to (d); otherwise go to (e).

[0093] d) If hi-lo < eps, then stop the calculation, output the COT simulation value of the model correction = hi, and obj(hi), thus obtaining the model deviation ΔCOT = simulated COT value - current actual COT value. Otherwise, let a = lo, lo = hi, hi = a + 0.618(ba) and go to (c).

[0094] e) If hi-lo < eps, then stop the calculation, output the COT simulation value of the model correction = lo, and obj(lo), thus obtaining the model deviation ΔCOT = simulated COT value - current actual COT value. Otherwise, let b = hi, hi = lo, lo = a + 0.382(ba) and turn to (c).

[0095] 4. Optimization of cleavage depth includes the following steps:

[0096] 1) The objective function is determined based on the optimization index selection signal sent by the DCS. The objective functions include maximizing ethylene yield, maximizing propylene yield, maximizing diene yield, maximizing triene yield, maximizing single-pass high-by-product yield, maximizing overall high-by-product yield, and maximizing efficiency. Efficiency maximization considers not only the yield and corresponding price factors of key cracking products (product yield and price information are shown in Table 2), but also fuel gas consumption and SS steam production during the optimization period. Assume the current cracking furnace feed rate, DS rate, fuel gas consumption, and SS ultra-high pressure steam production are F0, F... FG F DS F SS The raw material prices are P0 and P, respectively. FG P DS P SS The product quality, yield, and price information are shown in Table 2. The mathematical expression of the objective function is shown in Equation (12):

[0097]

[0098] 2) Determine the optimization range of the actual COT and gas-hydrogen ratio on site.

[0099] 3) Collect current cracking furnace operating data and naphtha analysis data as input conditions for the mechanism model.

[0100] 2)4) The objective function is optimized using a pattern search optimization algorithm. The main idea of ​​the pattern search algorithm is to start from a base point and alternately perform two types of searches: axial search and pattern search. The axial search proceeds sequentially along the n coordinate axes to determine the new base point and the direction that is conducive to the decrease of the function value. The pattern search proceeds along the line connecting two adjacent base points, attempting to make the function value decrease faster. The specific steps are as follows.

[0101] a) Determine the objective function and optimization variables based on the selection signal sent on the DCS, as shown in Equation (13).

[0102]

[0103] b) Given an initial point x 1 ∈R n Initial step size δ, acceleration factor a≥1, reduction rate β∈(0,1), precision ε>0, let y 1 =x 1 k = 1, j = 1.

[0104] c) Axial search

[0105] If f(y) j +δe j )>f(y j If y = 0, then let y = 0. j+1 =y j +δe j (d)

[0106] If f(y) j -δe j )>f(y j If y = 0, then let y = 0. j+1 =y j -δe j (d)

[0107] Otherwise, let y j+1 =y j .

[0108] Note:e j This indicates the direction of the j-th coordinate axis.

[0109] d) If j≤n, then let j=j+1 and turn (c). If f(y) n+1 )>f(x k Turn (e) or turn (f)

[0110] e) Let x k+1 =y n+1 y 1 =x k+1 +a(x k+1 -x k Let k = k + 1, j = 1. (C)

[0111] f) If δ≤ε, then stop the calculation to obtain the advantage x. k Otherwise, let δ = β*δ, and let y 1 =x k x k+1 =x k Let k = k + 1, j = 1, then turn (c).

[0112] 3) Substitute the optimized values ​​of COT and hydrocarbon ratio calculated by the pattern search algorithm into the mechanism model to calculate the optimized propylene-ethyl ratio.

[0113] The optimized propylene-to-ethylene ratio per opt , Gas-hydrogen ratio dor opt Amplitude limiting is applied, assuming the on-site technicians set the cracking furnace depth controller range to [per] low ,per high The range of the vapor-hydrocarbon ratio controller is [dor]. low ,dor high The propylene-to-ethyl ratio and hydrocarbon vapor ratio, obtained after limiting the amplitude, are shown in Formulas 14 and 15:

[0114]

[0115]

[0116] 4) Considering the tracking performance of the cracking furnace controller, it is necessary to start from the controller's setpoint. Assume the current cracking depth and the gas-to-hydrogen ratio controller values ​​are per... now dor now Set the step size (step is generally between 0.001 and 0.005) to gradually approximate the value after the amplitude limiting process. Assume that the value sent to the controller in this optimization step is per. new dor now As shown in equations 16 and 17:

[0117]

[0118]

[0119] If the optimization cycle is reached and a new optimization cycle is required, the model needs to be recalibrated in the next optimization cycle due to changes in the cracking furnace controller. The calibration steps are performed according to step (4) of the golden section method calibration mechanism model. After the model is calibrated, steps 1 to 7 of the cracking depth optimization are performed. The optimization cycle is set according to the real-time requirements of the site, and a recommended time is 30 to 60 minutes.

[0120] In summary, this invention combines the high accuracy of mechanistic model product prediction with the use of pattern search optimization algorithm to accelerate optimization convergence speed and improve target product yield, thus better meeting the needs of real-time performance and economic benefits in the field.

[0121] The above technical features constitute the embodiments of the present invention, which have strong adaptability and implementation effect. Unnecessary technical features can be added or removed according to actual needs to meet the needs of different situations.

[0122] Table 1

[0123]

[0124] Table 2

[0125]

Claims

1. An online optimization method for ethylene cracking furnaces based on a pattern search algorithm, characterized in that... The following steps are to be taken: First, collect data from the industrial ethylene cracking furnace system, including cracking furnace operating conditions, oil property data, and furnace structure information, and establish a cracking furnace mechanism model for product prediction. The second step is to use the pyrolysis furnace mechanism model to obtain the simulated mass fraction of each product based on the current pyrolysis furnace operating conditions. The third step is to convert the simulated mass fraction of the product into a volume fraction, and then use the golden section method to minimize the sum of squares of the difference between the simulated product volume fraction and the actual product volume fraction as the objective function, so that the simulated product yield is close to the actual product yield, and obtain the deviation of the outlet temperature ΔCOT. The fourth step is to establish a prediction model for fuel gas and a prediction model for ultra-high pressure steam based on the current furnace type, pyrolysis feedstock, and on-site operating conditions. Fifth, select the objective function based on the actual needs on site; Sixth, using the determined objective function and the current cracking furnace operating conditions, use a pattern search optimization algorithm, taking the COT and gas-hydrocarbon ratio as optimization variables, and call the corrected mechanism model from the third step to obtain the COT and gas-hydrocarbon ratio values ​​corresponding to the maximum objective function; Seventh, based on the calculated optimized COT and gas-hydrocarbon ratio values, call the corrected mechanism model to predict the optimized yields of each product and the propylene-ethyl ratio, and then write the propylene-ethyl ratio and the optimized gas-hydrocarbon ratio back to the cracking furnace controller with a certain step size.

2. The online optimization method for ethylene cracking furnace based on pattern search algorithm according to claim 1, characterized in that... In the first step, the cracking furnace operating conditions include feed load, dilution steam ratio, feed temperature, feed pressure, outlet temperature and outlet pressure, and the oil property data includes liquid phase feed property data and gas phase feed property data.

3. The online optimization method for ethylene cracking furnace based on pattern search algorithm according to claim 2, characterized in that... The liquid phase feedstock attribute data includes the mass percentage and distillation range data of n-alkanes, isoalkanes, alkenes, cycloalkanes, and aromatics, while the gas phase feedstock attribute data includes the volume fraction of methane, ethylene, and propylene.

4. The online optimization method for ethylene cracking furnace based on pattern search algorithm according to claim 1, 2, or 3, characterized in that... The furnace structure includes the effective length, wall thickness, and inner diameter of the pyrolysis furnace and radiant furnace tubes.

5. The online optimization method for ethylene cracking furnace based on pattern search algorithm according to claim 4, characterized in that... In the second step, the cracking furnace mechanism model is established using commercial ethylene cracking furnace simulation software or a model based on the basic principles of physical and chemical reaction processes.

6. The online optimization method for ethylene cracking furnace based on pattern search algorithm according to claim 1, 2, 3, or 5, characterized in that... In the third step, the products include hydrogen, methane, ethylene, ethane, propylene, and propane.

7. The online optimization method for ethylene cracking furnace based on pattern search algorithm according to claim 6, characterized in that... In the fourth step, the on-site operating conditions include the cracking feedstock, COT, gas-hydrocarbon ratio, fuel gas composition, feed flow rate, and high-temperature flue gas preheating boiler feedwater (BFW) flow rate.

8. The online optimization method for an ethylene cracking furnace based on a pattern search algorithm according to claim 1, 2, 3, 5, or 7, characterized in that... In the fourth step, the fuel gas prediction model is based on the global heat balance relationship of the pyrolysis furnace, namely the relationship between the total heat released by fuel gas combustion and the total heat absorbed by the furnace tubes, the heat carried away by the flue gas and the heat loss of the pyrolysis furnace, to predict fuel gas consumption.

9. The online optimization method for ethylene cracking furnace based on pattern search algorithm according to claim 8, characterized in that... In the fourth step, the prediction model for ultra-high pressure steam is based on the developed TLE and pyrolysis furnace convection section model. In addition to solving the three basic equations of mass, energy and momentum conservation, the convection section model also involves solving the convective heat transfer equation. The TLE convection section model is used to solve the core heat transfer formula by inputting saturated water flow rate, gasification rate and thermosiphon parameters.

10. The online optimization method for an ethylene cracking furnace based on a pattern search algorithm according to claim 1, 2, 3, 5, 7, or 9, characterized in that... In the fifth step, the objective functions include the ethylene yield maximization function, the propylene yield maximization function, the diene yield maximization function, the triene yield maximization function, the single-pass high-by-interest yield maximization function, the overall high-by-interest yield maximization function, and the benefit maximization function.

Images