Construction method and equipment of catalytic cracking prediction model, medium and product

Abstract

The invention relates to the technical field of industrial control, in particular to a construction method of a catalytic cracking prediction model, equipment, a medium and a product. The method comprises the following steps: acquiring device data of a catalytic cracking riser, and determining kinetic parameters of a mechanism model according to a preset multi-strategy improved algorithm; and according to a preset hyper-parameter optimization algorithm, hyper-parameters of the initial data model are optimized, and a data driving model is constructed based on the optimized hyper-parameters. And according to a preset weight adaptive strategy, performing fusion processing on the mechanism model and the data driving model to generate a hybrid model, thereby predicting the product yield of the catalytic cracking process through the hybrid model, and improving the prediction efficiency and accuracy of the catalytic cracking reaction process.

Description

Technical Field

[0001] This application relates to the field of industrial control technology, and in particular to a method, equipment, medium and product for constructing a catalytic cracking prediction model. Background Technology

[0002] Fluid catalytic cracking (FCC) is a core process in petroleum refining that converts heavy oil into high-value-added light products. However, the FCC process is affected by many nonlinear factors and potential interactions, making it challenging to predict and optimize the FCC process in real industrial environments.

[0003] Related technologies typically model the FCC process using classical mechanistic models such as lumped kinetic models. However, determining the catalytic cracking kinetic parameters relies heavily on experimental work, which is not only cumbersome and inefficient, requiring significant resources and time to complete the necessary experiments, but also often limited by experimental conditions and measurement precision. This makes it difficult for mechanistic models to accurately understand and model the kinetic parameters of the catalytic cracking reaction, resulting in low efficiency and accuracy in kinetic parameter identification, thus significantly reducing the prediction efficiency and accuracy of the catalytic cracking process. While data-driven modules can capture the nonlinear characteristics of the catalytic cracking process, their performance is highly dependent on hyperparameter settings. Related technologies typically employ manual parameter tuning, which is not only inefficient and makes it difficult to accurately determine the globally optimal hyperparameter combination, but also results in data models lacking physical interpretability, making it difficult to guide the optimization of the FCC process.

[0004] Therefore, there is an urgent need for a method to construct a prediction model for catalytic cracking to improve the prediction efficiency and accuracy of catalytic cracking reaction processes. Summary of the Invention

[0005] This invention provides a method, apparatus, storage medium, and program product for constructing a catalytic cracking prediction model, which can improve the prediction efficiency and accuracy of catalytic cracking reaction processes.

[0006] In a first aspect, this application provides a method for constructing a catalytic cracking prediction model, the method comprising:

[0007] Acquire device data for the catalytic cracking riser, wherein the device data characterizes the operating status and reaction results of the catalytic cracking riser;

[0008] Based on a pre-defined multi-strategy improvement algorithm, the kinetic parameters of the mechanism model are determined; the mechanism model is used to describe the transformation relationship of the catalytic cracking reaction pathway; the multi-strategy improvement algorithm is determined by optimizing the dung beetle optimization algorithm based on the optimal point set algorithm, the osprey algorithm, and the adaptive t-distribution algorithm.

[0009] Based on a preset hyperparameter optimization algorithm, the hyperparameters of the initial data model are optimized, and a data-driven model is constructed based on the optimized hyperparameters; the data-driven model is used to fit the nonlinear mapping relationship between the device data.

[0010] Based on a preset weight adaptive strategy, the mechanism model and the data-driven model are fused to generate a hybrid model; the hybrid model is used to predict the product yield of the catalytic cracking process.

[0011] Optionally, the determination of the dynamic parameters of the mechanistic model based on the preset multi-strategy improvement algorithm includes:

[0012] Population initialization is performed based on a pre-defined optimal point set algorithm to determine the optimal individual in the current population;

[0013] Based on a preset dung beetle optimization algorithm, the positions of the current population and the best individual are iteratively updated; the dung beetle optimization algorithm is obtained by optimizing the osprey algorithm and the adaptive t-distribution algorithm.

[0014] When the iteration termination condition is met, the global optimal position is output, and the parameters corresponding to the global optimal position are used as the dynamic parameters.

[0015] Optionally, the optimization of the hyperparameters of the initial data model based on a preset hyperparameter optimization algorithm includes:

[0016] Based on the hyperparameter search space of the initial data model, the objective function is determined; the hyperparameter search space includes the learning rate interval, the leaf node number interval, and the regularization coefficient interval.

[0017] Based on the objective function, the hyperparameters are optimized.

[0018] Optionally, the process of fusing the mechanistic model and the data-driven model based on a preset weight adaptive strategy to generate a hybrid model includes:

[0019] Based on preset weight constraints, the initial weights of the mechanism model and the data-driven model are determined respectively.

[0020] Based on the predicted yield of the training set of the mechanistic model and the data-driven model, and combined with the initial weights, the predicted value of the hybrid model is calculated.

[0021] Based on a preset gradient descent strategy and the predicted value, the initial weights are iteratively updated until a preset termination condition is met, and the target weights are output.

[0022] The hybrid model is generated based on the target weights.

[0023] Optionally, the population initialization based on a preset optimal point set algorithm to determine the optimal individual in the current population includes:

[0024] The current population is determined based on a preset set of initialization parameters; the initialization parameter set includes the population size, the maximum number of iterations, and the search space for dynamic parameters.

[0025] Based on a preset fitness function, the fitness value of each individual in the current population is determined; the fitness function represents the root mean square error between the predicted value and the actual value.

[0026] Based on each fitness level, the optimal individual is determined.

[0027] Optionally, the step of iteratively updating the positions of the current population and the optimal individual based on a preset dung beetle optimization algorithm includes:

[0028] Based on the current population and the optimal individual, and combined with the osprey algorithm, the positions of the dung beetles in the current population are updated to obtain the first updated population; the positions of the dung beetles represent the search path in the solution space of the dynamic parameters;

[0029] Based on the first updated population and the corresponding optimal individuals, the brood ball positions are updated to obtain the second updated population; the brood ball positions represent the safe breeding region boundary in the solution space of dynamic parameters.

[0030] Based on the second updated population and the corresponding optimal individuals, and combined with the adaptive t-distribution algorithm, the foraging positions of dung beetles are updated to obtain the third updated population; the foraging positions of dung beetles represent the global random perturbation strategy of the solution space of dynamic parameters;

[0031] Based on the third updated population and the corresponding optimal individuals, the positions of the dung beetles are updated to obtain the fourth updated population; the positions of the dung beetles represent the diversity preservation mechanism of the solution space of dynamic parameters;

[0032] Based on the fourth updated population, the global optimal fitness value and global optimal position for this iteration are determined.

[0033] Optionally, the step of updating the positions of the dung beetles in the current population based on the current population and the best individual, combined with the osprey algorithm, to obtain the first updated population, includes:

[0034] Based on the individual's current behavior judgment value and the preset obstacle judgment threshold, the target state of the current population is determined;

[0035] When the target state is an obstacle state, the position of the dung beetle is updated based on a preset deviation coefficient;

[0036] When the target state is an unobstructed state, a guide individual is determined based on the osprey algorithm, and the position of the dung beetle is updated based on the guide individual.

[0037] In a second aspect, this application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a method for constructing any one of the catalytic cracking prediction models described in the first aspect above.

[0038] Thirdly, this application provides a computer storage medium storing computer program instructions, which are executed by a processor using any of the catalytic cracking prediction model construction methods described in the first aspect above.

[0039] Fourthly, an embodiment of this application provides a computer program product, including computer program instructions, which, when executed by a processor, implement the method for constructing any one of the catalytic cracking prediction models described in the first aspect above.

[0040] The beneficial effects of this invention are as follows:

[0041] This application provides a method, apparatus, medium, and product for constructing a catalytic cracking prediction model. The method acquires device data from the catalytic cracking riser and determines the kinetic parameters of the mechanistic model based on a preset multi-strategy improvement algorithm. It then optimizes the hyperparameters of the initial data model using a preset hyperparameter optimization algorithm and constructs a data-driven model based on the optimized hyperparameters. Finally, it fuses the mechanistic model and the data-driven model using a preset weighted adaptive strategy to generate a hybrid model. This hybrid model is then used to predict the product yield of the catalytic cracking process, thereby improving the prediction efficiency and accuracy of the catalytic cracking reaction process. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0043] Figure 1 A flowchart illustrating a method for constructing a catalytic cracking prediction model provided in an embodiment of this application;

[0044] Figure 2 This is a schematic diagram of partial device data for a catalytic cracking riser provided in an embodiment of this application;

[0045] Figure 3A schematic diagram of a multi-strategy improved dung beetle optimization algorithm provided in an embodiment of this application;

[0046] Figure 4 A schematic diagram of the ensemble partitioning of a mechanism model provided in this application embodiment;

[0047] Figure 5 A schematic diagram of a seventeen-lumped model reaction path provided for an embodiment of this application;

[0048] Figure 6 A schematic diagram of dynamic parameters provided for an embodiment of this application;

[0049] Figure 7 This is a schematic diagram of a hyperparameter optimization configuration provided in an embodiment of this application;

[0050] Figure 8 A schematic diagram illustrating the fusion process of a mechanistic model and a data-driven model provided in an embodiment of this application;

[0051] Figure 9 This application provides a schematic diagram comparing the performance indicators of different models in an embodiment.

[0052] Figure 10 A schematic diagram showing the comparison between predicted and actual yield values ​​of a mechanistic model and a hybrid model provided in an embodiment of this application;

[0053] Figure 11 This application provides a schematic diagram comparing the fitting curves of the predicted yield values ​​and the actual yield values ​​of a hybrid model.

[0054] Figure 12 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily combined with each other. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here.

[0056] The terms "first" and "second" in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the term "comprising" and any variations thereof are intended to cover non-exclusive protection. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices. The term "multiple" in this application can mean at least two, for example, two, three, or more, and this application does not impose limitations.

[0057] The term "and / or" in the embodiments of this application is merely a description of the association relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0058] It is understood that the following specific embodiments of this application involve data related to catalytic cracking processes. When the embodiments of this application are applied to specific products or technologies, relevant licenses or consents are required, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, relevant volunteers can be recruited and agreements can be signed to authorize their data, thereby enabling the implementation using the data of these volunteers; alternatively, implementation can be carried out within an authorized organization, using data from members of the organization to implement the following implementation methods for data management; or, in specific implementations, the relevant data used are all simulated data, such as simulated data generated in a virtual scenario.

[0059] The design concept of the embodiments of this application will be briefly introduced below.

[0060] Fluid catalytic cracking (FCC) is a core process in petroleum refining that converts heavy oil into high-value-added light products. It is a key chemical process in the oil and gas industry and an important pathway for renewable energy conversion. Currently, most diesel and gasoline on the market are produced using the FCC process. However, FCC is a highly complex process, influenced by many nonlinear factors and potential interactions, making prediction and optimization of the FCC process in real-world industrial environments extremely challenging. Therefore, developing new FCC prediction frameworks to shorten the R&D cycle of new processes, optimize unit design and operation, and achieve online optimization of the FCC process will help improve raw material utilization, product yield, and quality.

[0061] Currently, related technologies typically model FCC processes based on classical mechanistic models, with lumped kinetic models being the most commonly used method in existing FCC simulations. Lumped kinetic models classify all components in the reaction system into several virtual components with similar kinetic properties, effectively simplifying the complex reaction system. However, determining traditional catalytic cracking kinetic parameters relies heavily on experimental work, leading to cumbersome and inefficient operations. This requires significant resources and time to complete necessary experiments, and the accuracy of parameter estimation using traditional methods is often limited by experimental conditions and measurement precision. This makes it difficult for mechanistic models to accurately understand and model the kinetic parameters of catalytic cracking reactions, reducing the efficiency and accuracy of kinetic parameter identification, and consequently lowering the prediction efficiency and accuracy of catalytic cracking reaction processes. While data-driven modules can capture the nonlinear characteristics of catalytic cracking processes, their performance is highly dependent on hyperparameter settings. Related technologies typically employ manual parameter tuning, which is not only cumbersome and inefficient, making it difficult to accurately determine the globally optimal hyperparameter combination, but also lacks physical interpretability, making it difficult to guide the optimization of FCC processes.

[0062] In view of the above problems, embodiments of this application provide a method, apparatus, medium, and product for constructing a catalytic cracking prediction model. The method acquires device data from the catalytic cracking riser and determines the kinetic parameters of the mechanistic model according to a preset multi-strategy improvement algorithm. It then optimizes the hyperparameters of the initial data model according to a preset hyperparameter optimization algorithm and constructs a data-driven model based on the optimized hyperparameters. According to a preset weight adaptive strategy, the mechanistic model and the data-driven model are fused to generate a hybrid model, thereby predicting the product yield of the catalytic cracking process. Thus, embodiments of this application solve the kinetic parameters of the mechanistic model using a multi-strategy improved dung beetle optimization algorithm, which not only saves time and computational resources and improves parameter tuning efficiency but also enhances the predictive performance of the mechanistic model. Furthermore, the hybrid model of this application combines the stability of the mechanistic model with the accuracy of the data-driven model, significantly improving the prediction efficiency and accuracy of the catalytic cracking reaction process.

[0063] The method provided by the exemplary embodiments of this application will now be described with reference to the accompanying drawings. It should be noted that the above application scenarios are shown only to facilitate understanding of the spirit and principles of this application, and the embodiments of this application are not limited in any way.

[0064] Please refer to Figure 1 The following is a flowchart illustrating a method for constructing a catalytic cracking prediction model according to an embodiment of this application. The specific implementation process of this method is as follows:

[0065] Step 101: Obtain the device data for the catalytic cracking riser.

[0066] In this embodiment, the device data represents the operating status and reaction results of the catalytic cracking riser.

[0067] In one possible implementation, the embodiments of this application can obtain equipment data such as operating conditions, feedstock properties, and product yield of the catalytic cracking riser based on expert experience.

[0068] For details, please refer to Figure 2 The diagram shows partial device data for a catalytic cracking riser provided in this embodiment. This embodiment obtains device data corresponding to different operating conditions, feedstock properties, and product yields under four different operating conditions through empirical screening. The operating conditions include feed flow rate (t / h), reactor pressure (MPa), first reaction zone outlet temperature (°C), second reaction zone outlet temperature (°C), riser pre-lift dry gas (kg / h), riser pre-lift steam (kg / h), riser atomized steam 1 (kg / h), riser atomized steam 2 (kg / h), riser atomized steam 3 (kg / h), recycle oil reprocessing (t / h), catalyst stock in the first reaction zone (t), catalyst circulation rate (kg / h), and catalyst coke content (wt%). Feedstock properties include feed saturated hydrocarbons (wt%), feed aromatics (wt%), feed gums (wt%), feed asphaltenes (wt%), and feed feed density (20°C) (g / cm³). 3 The feedstock residual carbon (wt%), feedstock nitrogen content (ppm), and 50% distillation temperature (°C) are specified. Product yields include gasoline alkanes (wt%), gasoline olefins (wt%), gasoline cycloalkanes (wt%), and gasoline aromatics (wt%).

[0069] In one possible implementation, after obtaining the device data of the catalytic cracking riser, the embodiments of this application may further perform preprocessing on the initial device data, such as outlier handling, normalization, and dataset partitioning.

[0070] In one possible implementation, the embodiments of this application may employ the K-means algorithm to process outliers in the initial device data.

[0071] Specifically, in this embodiment, K data points are randomly selected from the initial device data as initial centroids. For each data point, its distance to all initial centroids is calculated, and it is assigned to the cluster represented by the nearest initial centroid. For each cluster, the average value of all data points assigned to that cluster is calculated, and this average value is used as the new centroid. The above two steps are repeated until the centroids no longer change significantly or a predetermined number of iterations is reached. The formula for calculating the centroid distance is as follows:

[0072]

[0073] Where, x i For data points, c j Let M be the centroid and M be the number of features of the data points.

[0074] The centroid of each cluster is updated to the average of all data points within that cluster. The formula for updating the centroid is as follows:

[0075]

[0076] Among them, C j For the j-th cluster, |C j | represents the number of data points in j clusters.

[0077] It is worth mentioning that the number of initial centroids in this embodiment is empirically set based on the feature dimensions and distribution characteristics of the data, aiming to achieve better clustering results and anomaly detection accuracy. Therefore, in this embodiment, the number of initial centroids is not a unique fixed value, but can be flexibly set. For example, in practical applications, different numbers of clusters can be evaluated based on indicators such as the elbow rule or silhouette coefficient to determine a better number of clusters. Other reasonable ranges of initial centroid numbers can also be selected, and this embodiment does not specifically limit this.

[0078] In one possible implementation, the embodiments of this application may employ a data shuffling and segmentation strategy to divide the dataset.

[0079] Specifically, this application embodiment reduces potential sequence dependencies between data by randomly shuffling the order of data in the dataset, thereby avoiding overfitting of the model to a specific data arrangement. For example, after shuffling the 307 sets of data collected in the aforementioned embodiment, the first 80% is used as the training set, and the K-means algorithm is used to handle outliers.

[0080] In one possible implementation, embodiments of this application may normalize all variables to the range 0 to 1.

[0081] Specifically, the normalization formula is as follows:

[0082]

[0083] Where x represents the original unit data point, characterizing the continuously changing actual operating parameters during catalytic cracking.

[0084] x min This represents the minimum value in the dataset and is used to define the lower limit of the dynamic range of parameters, eliminating the interference of negative values.

[0085] x maxIt represents the maximum value of the dataset, limits the upper limit of the dynamic range of parameters, avoids large values ​​dominating the dimensionless parameters after normalization, and thus resolves the conflict of dimensions in multi-source data.

[0086] Step 102: Determine the dynamic parameters of the mechanism model based on the preset multi-strategy improvement algorithm.

[0087] In this embodiment of the application, the mechanism model is used to describe the transformation relationship of the catalytic cracking reaction pathway, and the multi-strategy improved algorithm is determined by optimizing the multi-strategy improved dung betle optimizer (MSIDBO) algorithm using the best point set algorithm, the Osprey algorithm, and the adaptive t-distribution algorithm.

[0088] In one possible implementation, embodiments of this application can initialize the population using a preset optimal point set algorithm to determine the initial optimal individual. Then, based on the dung beetle optimization algorithm, iterative position updates are performed on the current population and the current optimal individual. When the iteration termination condition is met, the global optimal position is output, thereby using the parameters corresponding to the global optimal position as the dynamic parameters of the mechanistic model.

[0089] Specifically, in the embodiments of this application, the dung beetle optimization algorithm is obtained by optimizing the osprey algorithm and the adaptive t-distribution algorithm.

[0090] In one possible implementation, embodiments of this application can determine the current population based on a preset initialization parameter set, which includes the population size, maximum number of iterations, and a dynamic parameter search space. A preset fitness function, representing the root mean square error between the predicted and actual values, is used to determine the fitness value of each individual in the current population, thereby identifying the optimal individual based on each fitness value.

[0091] Specifically, the optimal point set in the embodiments of this application is as follows:

[0092]

[0093] Among them, P n (k) is the set of best points, r is the best point, and n is the number of points.

[0094] G D It is a unit cube in D-dimensional Euclidean space, 1≤k≤n, and its deviation is... C(r,ε) is a constant that depends only on r and ε (ε is an arbitrarily small positive number).

[0095] Pick 1≤k≤D, where p is a definite expression that satisfies The smallest prime number.

[0096] Specifically, the fitness calculation formula in this application embodiment is as follows:

[0097]

[0098] in, y is the model's predicted value. i The actual values ​​are used to select the individual with the best fitness as the optimal individual.

[0099] In one possible implementation, each iterative optimization operation of this application embodiment includes: updating the position of the dung beetle pushing ball based on the current population and the best individual in this iteration, using the osprey algorithm, to obtain a first updated population. Updating the position of the brooding ball based on the first updated population and the corresponding best individual, to obtain a second updated population. Updating the foraging position of the dung beetle based on the second updated population and the corresponding best individual, using the adaptive t-distribution algorithm, to obtain a third updated population. Updating the position of the dung beetle stealing ball based on the third updated population and the corresponding best individual, to obtain a fourth updated population, thereby determining the global optimal fitness value and the global optimal position based on the fourth updated population.

[0100] Specifically, in the embodiments of this application, each dung beetle represents a set of dynamic parameter combinations (e.g., pre-finger factor, activation energy, etc.). By updating the dung beetle position, parameter combinations with higher fitness values ​​are selected.

[0101] In one possible implementation, the dung beetle in this application represents a search path that finds the optimal solution as quickly as possible, i.e., a search path in the solution space of dynamic parameters. This application can determine the target state of the current population by comparing the individual's current behavior judgment value with a preset obstacle judgment threshold. When the target state is an obstacle-prone state, the dung beetle position is updated based on a preset deviation coefficient. Conversely, when the target state is an obstacle-free state, a guiding individual is determined based on the osprey algorithm, and the dung beetle position is updated based on the guiding individual.

[0102] Specifically, if an individual's current behavioral judgment value is greater than the set obstacle judgment threshold λ > γ, it is considered an obstacle state, meaning that the combination of dynamic parameters is trapped in a local optimum region, such as over-enhancing a certain reaction path, causing the model prediction error to be unable to be further reduced.

[0103] At this point, the updated position of the dung beetle is as follows:

[0104] X t (t+1)=X t (t)+tan(θ)|X t (t)-X t (t-1)|

[0105] Here, λ represents the individual's current behavioral judgment value (e.g., calculated based on indicators such as local search success rate and fitness fluctuation).

[0106] γ represents the set obstacle detection threshold.

[0107] θ∈[0,π] represents the deviation coefficient. When θ∈[0,π / 2,π], tan(θ) is meaningless and the position is not updated.

[0108] Conversely, in an unobstructed state, i.e., when the parameter space is not constrained by local optima, potential optimal regions can be freely explored. For example, the model can discover efficient response paths not covered by traditional methods. In this embodiment, the Osprey algorithm will be used to select a better dung beetle, i.e., a parameter combination with higher fitness values. The selection formula is as follows:

[0109] FP i ={X k |k∈{1,2,...,N}∧F k <F i}∪{X best}

[0110] Among them, FP i Let X be the set of locations of the i-th osprey. best The optimal position of the Osprey represents the parameter combination with the highest fitness value in the current iteration corresponding to the globally optimal position. Its position represents the parameter value in the parameter space that makes the model's prediction closest to the actual value.

[0111] At this point, the updated position of the dung beetle is as follows:

[0112] X t (t+1)=X t (t)+α×k×X t (t-1)+b×ΔX

[0113] in, η is a random number between (0,1).

[0114] λ is the probability that the dung beetle encounters an obstacle, λ∈(0,1).

[0115] k is the deviation coefficient, which can be set to 0.001, and b can be set to 0.5.

[0116] ΔX=|X i -X w | represents the light intensity, X w This represents the worst position in the population, γ = 0.8.

[0117] In one possible implementation, embodiments of this application can calculate a feasible region using the optimal individual and iteration progress, and update the brood ball position within the feasible region to obtain a second updated population. The brood ball represents a safe breeding region centered on the current optimal parameters, and its position represents the offspring generation parameters within the physically feasible domain, avoiding invalid solutions and thus ensuring parameter rationality, supporting the stable solution of the subsequent fourth-order Runge-Kutta method (ode45).

[0118] Specifically, the formula for updating the position of the brooding ball in this embodiment is as follows:

[0119] B i (t+1)=X * +b1×(B i (t)-Lb * )+b2×(B i (t)-Lb * )

[0120] Lb * =max(X * ×(1-R),Lb)

[0121] Ub * =min(X) * ×(1+R),Ub)

[0122] Where b1 and b2 are 1×D independent random vectors, and D is the dimension of the optimization problem, which can be set to 218 dimensions.

[0123] X * This represents the group's optimal position (local optimal position) in this iteration;

[0124] R is the ratio of the current iteration number t to the total number of iterations T, and T can be set to 3000;

[0125] Lb and Ub represent the lower and upper bounds of the feasible region, respectively, set to 0 and 100000; Ub * Lb * These are the upper and lower boundaries of the safe zone.

[0126] In one possible implementation, embodiments of this application generate a mutation perturbation factor based on the t-distribution probability density function of the adaptive t-distribution algorithm. This perturbation factor is used to mutate and perturb the dung beetle's foraging location, thereby updating the dung beetle's foraging location. In this application, the dung beetle represents an individual conducting a detailed search within a safe area, exploring extreme parameter combinations, thereby improving the model's predictive extrapolation ability for complex conditions (such as high-temperature coking).

[0127] Specifically, in this embodiment, the position of the dung beetle will be updated by using an adaptive t-distribution for mutation perturbation, as shown in the following formula:

[0128] X(t+1)=(X best -r×X rnd ×trnd(t))

[0129] The trnd(t) function calculates the probability density function of the t-distribution. The t-distribution exhibits significant fluctuations, which helps the algorithm escape local optima. Furthermore, the high randomness of the t-distribution facilitates global search.

[0130] In one possible implementation, embodiments of this application can generate a perturbation vector using the optimal individual and the globally optimal individual, and update the position of the dung beetle using a normally distributed random vector. In this application, the dung beetle represents an improvement on the suboptimal parameter combination through a plundering mechanism. Its position update simulates the competitive optimization of the parameter space, accelerating convergence to a high-precision region, thereby shortening the model calibration cycle and meeting the real-time optimization needs of industry.

[0131] Specifically, the update formula for the location of the dung beetle in this embodiment is as follows:

[0132] X i (t+1)=X b +S×g×(|X i (t)-X * |+|X i (t)-X b |)

[0133] Where g is a random vector of size 1×D that follows a normal distribution, and S represents a constant.

[0134] In summary, through the above iterative operations, the optimal fitness value and optimal position can be updated in each round. Based on the preset maximum number of iterations, it is determined whether the algorithm has reached the maximum number of iterations. If so, the iteration terminates, and the globally optimal position and corresponding solution are returned; otherwise, the next round of iterations continues.

[0135] In one possible implementation, refer to Figure 3The diagram illustrates a multi-strategy improved dung beetle optimization algorithm provided in this embodiment. This embodiment begins by setting the population size N, iteration count T, and other initial parameters. It constructs a uniformly initialized population using a set of optimal points and calculates the fitness values ​​of individual dung beetles. Next, it performs a dung beetle position update operation. A branch is selected based on the comparison between a random number λ and a threshold of 0.8. If λ < 0.8, it enters an unobstructed state, using the Osprey algorithm to select a dung beetle with better fitness to guide the position update; otherwise, it enters an obstructed state, randomly generating an integer θ ∈ [0, 180], and determining whether to update the dung beetle position based on the θ value (e.g., keeping the position unchanged or shifting direction). After the update, the current optimal position of the population is redefined. Then, the position of the small dung beetles is mutated and perturbed using an adaptive t-distribution probability density function, and the randomly perturbed position of the thieving dung beetles is updated simultaneously. Each iteration updates the optimal fitness value and optimal position for the current round. If the current iteration count t is less than a preset value T, the above process is repeated until the termination condition is met. Output the global optimal position and its corresponding dynamic parameter solution.

[0136] In one possible implementation, refer to Figure 4 The diagram shown is a schematic representation of a mechanism model provided in this application embodiment. This application embodiment divides the catalytic cracking riser reaction system into 17 components: feedstock (saturated components, aromatic components, asphaltenes, and gums), and products (diesel alkanes, diesel cycloalkanes, diesel olefins, diesel monocyclic aromatics, diesel polycyclic aromatics, gasoline olefins, gasoline aromatics, gasoline alkanes, gasoline cycloalkanes, liquefied petroleum gas, propylene, butene, dry gas, ethylene, and coke). A mechanism model for the catalytic cracking riser is established, and its expression is as follows:

[0137]

[0138] Among them, a j K represents the mass concentration of the lumped component. j S is the lumped reaction rate constant; ρ is the gas density, in kg / m³; S WH The actual weight-time space velocity is expressed in h⁻¹.

[0139] X is the dimensionless distance, and its expression is: X = x / L.

[0140] Where x represents the height of the riser tube in meters (m), and L represents the total length of the riser tube in meters (m).

[0141] The catalyst coking deactivation function is given by the following formula:

[0142]

[0143] Wherein, β is the catalyst coking deactivation factor; C c, where is the coke content of the catalyst, in wt%; M is the catalyst coking deactivation function exponent.

[0144] f(A) is the adsorption-deactivation function of heavy aromatic hydrocarbons, and its expression is:

[0145]

[0146] Where, k A It is a deactivation factor for heavy aromatic hydrocarbon adsorption; C A The residual carbon content of the feedstock is expressed in wt%; f(N) is the alkali-nitrogen adsorption-deactivation constant, expressed as follows:

[0147]

[0148] Where, k N It is an alkaline nitrogen adsorption deactivation factor; C N The alkali and nitrogen content of the raw oil is expressed in wt%; t c φ represents the catalyst residence time, measured in seconds. c / o This refers to the oil-to-reactant ratio.

[0149] Thus, based on the reaction network between each lumped group, a system of seventeen lumped reaction rate differential equations can be obtained, whose vector expression is:

[0150]

[0151] Where a is the lumped concentration vector, its expression is as follows:

[0152] a=[SS SA SRDPDNDASDAM GP GO GN GA C4=C3=LPG C2=DR C] T

[0153] K is the reaction rate constant matrix, and the reaction constants in the K matrix follow the Arrhenius equation, i.e.:

[0154]

[0155] Among them, K1 i,j For generating the pre-exponential factor in the lumped reaction rate constant of lumped i and lumped j; E i,j Let be the activation energy of the reaction in the rate constant of the lumped i-to-lumped j reaction; T is the reaction temperature in K; R is the molar gas constant, which can be set to 8.3145 in SI system.

[0156] In summary, for reference Figure 5The diagram shown is a schematic representation of a 17-lumped model reaction pathway provided in an embodiment of this application. The 17-lumped model reaction network encompasses 109 chemical reactions. The kinetic parameters of each reaction include the pre-exponential factor and activation energy, involving a total of 218 parameters that need to be solved. (Reference) Figure 6 The diagram shown is a schematic representation of dynamic parameters provided in an embodiment of this application. Figure 6 It includes the mechanistic model parameters obtained by combining MSIDBO and the fourth-order Runge-Kutta method in the embodiments of this application, including pre-exponential factor, activation energy, etc.

[0157] Step 103: Based on the preset hyperparameter optimization algorithm, optimize the hyperparameters of the initial data model, and construct a data-driven model based on the optimized hyperparameters.

[0158] In this embodiment, a hyperparameter optimization algorithm is used to optimize hyperparameters such as the learning rate, minimum data size of leaf nodes, and minimum weight of child nodes of the initial data model. The optimized hyperparameters are then used to train a data-driven model to fit the nonlinear mapping relationship between device data.

[0159] In one possible implementation, embodiments of this application may define a hyperparameter search space for an initial data model, and determine an objective function based on the hyperparameter search space, thereby optimizing the hyperparameters through the objective function.

[0160] Specifically, in this embodiment of the application, the hyperparameters of the Lightweight Gradient Boosting Machine (LightGBM) model can be optimized using the Bayesian Optimization Hyperband (BOHB) algorithm, and the optimization process is as follows:

[0161] First, define the hyperparameter search space of the LightGBM model, including the learning rate range, the leaf node number range, and the regularization coefficient range.

[0162] Define an objective function that accepts hyperparameter configuration and returns a performance metric for cross-validation. For example, the objective function can use the mean squared error (MSE), which is calculated as follows:

[0163]

[0164] Where is the sample size, y i Let i be the true value of the i-th sample. It is the predicted value of the i-th sample.

[0165] Configure the BOHB optimizer, define budget and optimization information, run the BOHB optimization process, monitor the optimization process, and save the best hyperparameter configuration.

[0166] For details, please refer to Figure 7 The diagram illustrates a BOHB-LightGBM hyperparameter optimization configuration provided in this embodiment of the application. It includes hyperparameters of different data types, such as learning rate, minimum data size of leaf nodes, minimum weight of child nodes, maximum number of bins, number of leaf nodes, L1 regularization coefficient, L2 regularization coefficient, and maximum tree depth, as well as the search space corresponding to each hyperparameter. Thus, this application can use the determined optimal hyperparameter configuration to train a LightGBM model and evaluate its performance on the test set.

[0167] Step 104: Based on the preset weight adaptive strategy, the mechanism model and the data-driven model are fused to generate a hybrid model.

[0168] In this embodiment of the application, a hybrid model is obtained by fusing the output values ​​of the mechanistic model and the data-driven model through a weighted adaptive strategy to predict the product yield of the catalytic cracking process.

[0169] In one possible implementation, embodiments of this application can determine the initial weights of the mechanistic model and the data-driven model through preset weight constraints, and calculate the predicted value of the hybrid model by combining the prediction yields of the training sets of the mechanistic model and the data-driven model. Then, according to a preset gradient descent strategy, the initial weights are iteratively updated based on the predicted values ​​until a preset termination condition is met, at which point the target weights are output. Thus, by using the target weights of the mechanistic model and the data-driven model, a hybrid model is generated through model mixing.

[0170] Specifically, in this application embodiment, adaptive weighting coefficients of the mechanism model and the data model can be obtained based on input data and training samples, and a hybrid model of data model and mechanism model can be obtained by parallel summation.

[0171] Specifically, in the embodiments of this application, the training set prediction yield of the mechanistic model and the data-driven model can be obtained through the MSIDBO algorithm and the BOHB-LightGBM algorithm.

[0172] Next, the initial weights W of the mechanistic model are initialized. M and the initial weights W of the data-driven model D Among them, W M and W D The values ​​should be between [0,1], and both should satisfy the following weight constraints:

[0173] W M =1-W D

[0174] Within a specified number of iterations, the weights are continuously updated. The predicted value of the hybrid model is calculated based on each weight. The gradient of the loss function with respect to each weight is calculated, and the weights are updated using gradient descent. To ensure that the sum of the weights is 1, the weights can also be normalized. Finally, the target weights are returned and used as the weight coefficients of the test set. The hybrid model prediction results of the test set are then calculated.

[0175] In one possible implementation, refer to Figure 8 The diagram illustrates the fusion process of a mechanistic model and a data-driven model according to an embodiment of this application. The original data of both the mechanistic model and the data-driven model are acquired. For the data from the data-driven model, a hyperparameter search space is defined, an objective function is defined, the BOHB optimizer is configured, BOHB is run and the optimal hyperparameters are saved, the LightGBM model is trained, and the prediction results are saved, thus achieving automated tuning and training of the LightGBM model. Simultaneously, for the data from the mechanistic model, parameter optimization and solution are completed through lumped partitioning, reaction network design, MSIDBO-ode45 solving, and saving the prediction results of the mechanistic model. Then, through adaptive weighting, the prediction outputs of the data-driven model and the mechanistic model are fused, ultimately generating a hybrid model with dual advantages, achieving high-precision prediction of the catalytic cracking process.

[0176] In one possible implementation, the embodiments of this application may also compare the fusion model with the mechanistic model, and use relative error, absolute error, and mean square error to evaluate the extrapolation and interpolation capabilities of the model.

[0177] Specifically, the embodiments of this application can comprehensively consider the accuracy and sensitivity of the model, construct new evaluation indicators, and compare the fusion model with the mechanistic model and the data-driven model to verify its extrapolation and interpolation capabilities.

[0178] For example, the Mean Relative Error (MRE), Mean Absolute Error (MAE), and MSE are used to evaluate the model, and their calculation formulas are as follows:

[0179]

[0180] Where n is the number of samples, y i Let i be the true value of the i-th sample. Let be the predicted value for the i-th sample.

[0181] refer to Figure 9The diagram shown illustrates a comparison of performance metrics for different models provided in this embodiment. In this embodiment, test set data is used to predict mechanistic and hybrid models respectively, and MAE, MRE, and MSE are used to evaluate the model performance metrics. Figure 9 It is evident that the three performance metrics of the hybrid model are closer to 0, and the model's accuracy is better than that of the mechanistic model. Furthermore, refer to... Figure 10 The diagram shown is a comparison of the predicted and actual yield values ​​of a mechanistic model and a hybrid model provided in this application embodiment. Figure 11 The figure shows a comparison of the fitting curves between the predicted and actual yield values ​​of the hybrid model. It can be seen that the mechanistic model has low prediction accuracy, while the fusion model has high prediction accuracy, which verifies the effectiveness and high accuracy of the hybrid model in predicting catalytic cracking yield.

[0182] Please see Figure 12 As shown, based on the same technical concept, this application also provides a computer device 120. In one embodiment, the computer device can be a device specifically used for predicting parameters of the ethylene production section, or it can be a control device used for overall control of the ethylene production process. The computer device is as follows... Figure 12 As shown, it includes a memory 1201, a communication module 1203, and one or more processors 1202.

[0183] The memory 1201 is used to store computer programs executed by the processor 1202. The memory 1201 may mainly include a program storage area and a data storage area. The program storage area may store the operating system and programs required to run instant messaging functions, etc.; the data storage area may store various instant messaging information and operation instruction sets, etc.

[0184] Memory 1201 may be volatile memory, such as random-access memory (RAM); memory 1201 may also be non-volatile memory, such as read-only memory, flash memory, hard disk drive (HDD), or solid-state drive (SSD); or memory 1201 may be any other medium capable of carrying or storing desired program code having the form of instructions or data structures and accessible by a computer, but is not limited thereto. Memory 1201 may be a combination of the above-described memories.

[0185] The processor 1202 may include one or more central processing units (CPUs) or digital processing units, etc. The processor 1202 is used to implement the above-described method for constructing the catalytic cracking prediction model when it calls the computer program stored in the memory 1201.

[0186] The communication module 1203 is used to communicate with the industrial control system.

[0187] This application embodiment does not limit the specific connection medium between the memory 1201, communication module 1203, and processor 1202. This application embodiment... Figure 12 The memory 1201 and the processor 1202 are connected via a bus 1204, and the bus 1204 is in Figure 12 The diagram uses thick lines to describe the connections between other components; these are for illustrative purposes only and should not be considered limiting. The 1204 bus can be divided into address bus, data bus, control bus, etc. For ease of description, Figure 12 It is described using only a thick line, but does not indicate that there is only one bus or one type of bus.

[0188] The memory 1201 stores a computer storage medium, which stores computer-executable instructions. The computer-executable instructions are used to implement the method for constructing the catalytic cracking prediction model of the present application embodiment. The processor 1202 is used to execute the method for constructing the catalytic cracking prediction model of the above embodiments.

[0189] Based on the same inventive concept, embodiments of this application also provide a storage medium storing a computer program that, when run on a computer, causes the computer to perform the steps in the method for constructing a catalytic cracking prediction model according to various exemplary embodiments of this application as described above.

[0190] In some possible implementations, various aspects of the method for constructing a catalytic cracking prediction model provided in this application can also be implemented in the form of a computer program product, which includes a computer program that, when run on a computer device, causes the computer device to perform the steps in the method for constructing a catalytic cracking prediction model according to various exemplary embodiments of this application as described above. For example, the computer device can perform the steps of the various embodiments.

[0191] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0192] The program product of the embodiments of this application may employ a portable compact disc read-only memory (CD-ROM) and include a computer program, and may run on a computer device. However, the program product of this application is not limited thereto. In this application, the readable storage medium may be any tangible medium that contains or stores a program, and the computer program included therein may be used by or in conjunction with a command execution system, apparatus, or device.

[0193] A readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying a readable computer program. This propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting a program for use by or in conjunction with a command execution system, apparatus, or device.

[0194] Computer programs contained on readable media may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0195] Computer programs for performing the operations of this application can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages.

[0196] It should be noted that although several units or sub-units of the device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of this application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided and embodied by multiple units.

[0197] Furthermore, although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0198] 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.

[0199] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0200] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for constructing a catalytic cracking prediction model, characterized in that, The method includes: Acquire device data for the catalytic cracking riser, wherein the device data characterizes the operating status and reaction results of the catalytic cracking riser; Based on a pre-defined multi-strategy improvement algorithm, the kinetic parameters of the mechanism model are determined; the mechanism model is used to describe the transformation relationship of the catalytic cracking reaction pathway; the multi-strategy improvement algorithm is determined by optimizing the dung beetle optimization algorithm based on the optimal point set algorithm, the osprey algorithm, and the adaptive t-distribution algorithm. Based on a preset hyperparameter optimization algorithm, the hyperparameters of the initial data model are optimized, and a data-driven model is constructed based on the optimized hyperparameters; the data-driven model is used to fit the nonlinear mapping relationship between the device data. Based on a preset weight adaptive strategy, the mechanism model and the data-driven model are fused to generate a hybrid model; the hybrid model is used to predict the product yield of the catalytic cracking process.

2. The method as described in claim 1, characterized in that, The preset multi-strategy improvement algorithm determines the dynamic parameters of the mechanism model, including: Population initialization is performed based on a pre-defined optimal point set algorithm to determine the optimal individual in the current population; Based on a preset dung beetle optimization algorithm, the positions of the current population and the best individual are iteratively updated; the dung beetle optimization algorithm is obtained by optimizing the osprey algorithm and the adaptive t-distribution algorithm. When the iteration termination condition is met, the global optimal position is output, and the parameters corresponding to the global optimal position are used as the dynamic parameters.

3. The method as described in claim 1, characterized in that, The preset hyperparameter optimization algorithm optimizes the hyperparameters of the initial data model, including: Based on the hyperparameter search space of the initial data model, the objective function is determined; the hyperparameter search space includes the learning rate interval, the leaf node number interval, and the regularization coefficient interval. Based on the objective function, the hyperparameters are optimized.

4. The method as described in claim 1, characterized in that, The method based on a preset weight adaptive strategy fuses the mechanistic model and the data-driven model to generate a hybrid model, including: Based on preset weight constraints, the initial weights of the mechanism model and the data-driven model are determined respectively. Based on the predicted yield of the training set of the mechanistic model and the data-driven model, and combined with the initial weights, the predicted value of the hybrid model is calculated. Based on a preset gradient descent strategy and the predicted value, the initial weights are iteratively updated until a preset termination condition is met, and the target weights are output. The hybrid model is generated based on the target weights.

5. The method as described in claim 2, characterized in that, The population initialization based on the preset optimal point set algorithm, which determines the optimal individual in the current population, includes: The current population is determined based on a preset set of initialization parameters; the initialization parameter set includes the population size, the maximum number of iterations, and the search space for dynamic parameters. Based on a preset fitness function, the fitness value of each individual in the current population is determined; the fitness function represents the root mean square error between the predicted value and the actual value. Based on each fitness level, the optimal individual is determined.

6. The method as described in claim 2, characterized in that, The preset dung beetle optimization algorithm iteratively updates the positions of the current population and the best individual, including: Based on the current population and the optimal individual, and combined with the osprey algorithm, the positions of the dung beetles in the current population are updated to obtain the first updated population; the positions of the dung beetles represent the search path in the solution space of the dynamic parameters; Based on the first updated population and the corresponding optimal individuals, the brood ball positions are updated to obtain the second updated population; the brood ball positions represent the safe breeding region boundary in the solution space of dynamic parameters. Based on the second updated population and the corresponding optimal individuals, and combined with the adaptive t-distribution algorithm, the foraging positions of dung beetles are updated to obtain the third updated population; the foraging positions of dung beetles represent the global random perturbation strategy of the solution space of dynamic parameters; Based on the third updated population and the corresponding optimal individuals, the positions of the dung beetles are updated to obtain the fourth updated population; the positions of the dung beetles represent the diversity preservation mechanism of the solution space of dynamic parameters; Based on the fourth updated population, the global optimal fitness value and global optimal position for this iteration are determined.

7. The method as described in claim 6, characterized in that, The process of updating the positions of the dung beetles in the current population based on the current population and the best individual, combined with the osprey algorithm, to obtain the first updated population includes: Based on the individual's current behavior judgment value and the preset obstacle judgment threshold, the target state of the current population is determined; When the target state is an obstacle state, the position of the dung beetle is updated based on a preset deviation coefficient; When the target state is an unobstructed state, a guide individual is determined based on the osprey algorithm, and the position of the dung beetle is updated based on the guide individual.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

9. A computer storage medium storing computer program instructions thereon, characterized in that, When executed by a processor, the computer program instructions implement the steps of the method according to any one of claims 1 to 7.

10. A computer program product comprising computer program instructions, characterized in that, When executed by a processor, the computer program instructions implement the steps of the method according to any one of claims 1 to 7.

Images