Ethylene cracking furnace optimization control method based on adaptive multi-objective evolutionary algorithm

Abstract

The invention discloses an ethylene cracking furnace optimization control method based on a self-adaptive multi-objective evolutionary algorithm. Comprising the following steps: firstly, obtaining historical operation parameters of the ethylene cracking furnace and constructing a cracking furnace multi-objective optimization model; then, solving the cracking furnace multi-objective optimization model by using an adaptive multi-objective evolutionary algorithm until solving is completed, and obtaining a Pareto optimal solution set; and determining an optimal cracking furnace outlet pipe temperature control scheme according to the Pareto optimal solution set to realize control of the ethylene cracking furnace. The method provided by the invention can effectively balance the increase of the ethylene yield and the decrease of the propylene yield in the ethylene cracking furnace, and provides a new scheme for operation optimization.

Description

Technical Field

[0001] This invention belongs to the field of chemical production process optimization technology, specifically involving an optimization control method for ethylene cracking furnace based on the Adaptive Multi-Objective Evolutionary Algorithm (AMOEA). Background Technology

[0002] As one of the most important basic raw materials in the petrochemical industry, the optimization and control of its production process is of great significance to improving the economic efficiency of enterprises. In the ethylene production process, the ethylene cracking furnace is the most important production unit, and many scholars both domestically and internationally have conducted extensive research on it.

[0003] In the operation of an ethylene cracking furnace, there are two conflicting objectives: to increase ethylene production to improve economic efficiency, and to minimize the decline in propylene production to maintain a reasonable product structure. During the operation cycle of the ethylene cracking furnace, the problem of coking also arises. As the number of production days increases, a layer of by-products adheres to the inside of the reaction tube, insulating the heat transfer and causing a decrease in the internal temperature of the reaction tube, thus affecting ethylene production.

[0004] Some domestic experts have proposed many solutions for the parameter control of ethylene cracking furnaces, including using metaheuristic optimization algorithms such as NSGA-II and PSO, but the optimization results are not very good. Summary of the Invention

[0005] The purpose of this invention is to provide an optimization control method for ethylene cracking furnaces based on an adaptive multi-objective evolutionary algorithm, so as to solve the problems of poor convergence, insufficient population diversity, and easy getting trapped in local optima in the multi-objective optimization of ethylene cracking furnaces in the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] I. An Optimization Control Method for Ethylene Cracking Furnaces Based on Adaptive Multi-Objective Evolutionary Algorithm

[0008] S1: Obtain historical operating parameters of the ethylene cracking furnace and construct a multi-objective optimization model for the cracking furnace;

[0009] S2: Use an adaptive multi-objective evolutionary algorithm to solve the multi-objective optimization model of the cracking furnace until the solution is completed, and obtain the Pareto optimal solution set; determine the optimal temperature control scheme of the cracking furnace outlet pipe based on the Pareto optimal solution set to realize the control of the ethylene cracking furnace.

[0010] In S1, constructing the multi-objective optimization model for the cracking furnace includes:

[0011] The operation cycle of the ethylene cracking furnace is divided into several stages, and the outlet pipe temperature of the cracking furnace in each stage constitutes the decision variable. Based on the obtained historical operating parameters, temperature range constraints and temperature difference constraints between adjacent stages are set. An optimization objective is constructed, the first objective of which is to maximize the ethylene yield and the second objective is to minimize the decrease in propylene yield, thereby obtaining a multi-objective optimization model for the cracking furnace.

[0012] In S2, the adaptive multi-objective evolutionary algorithm is an improvement on the multi-objective evolutionary algorithm. Specifically, it includes one or more of the following: adaptive mutation strategy, diversity enhancement mechanism, and stagnation detection and restart strategy. The adaptive mutation strategy is used to dynamically adjust the mutation intensity according to Cauchy mutation and polynomial mutation. The diversity enhancement mechanism is used to maintain population diversity according to the target space matrix. The stagnation detection and restart strategy is used to monitor the stagnation state and restart the population according to the hypervolume.

[0013] The adaptive mutation strategy is used to dynamically adjust the mutation intensity based on Cauchy mutation and polynomial mutation, including:

[0014] In the early stages of evolution, each decision variable is dominated by Cauchy mutation; in the later stages of evolution, it is dominated by multinomial mutation.

[0015] The diversity enhancement mechanism is used to maintain population diversity according to the target space matrix, including:

[0016] Every few generations, effective individuals in the current population are selected, and a target space matrix of the population is constructed based on all effective individuals. Then, based on the target space matrix of the population, the distance between each effective individual and other effective individuals is calculated to obtain all distances of the current effective individual. The second smallest distance among the distances between each individual and other individuals is taken as the distance of the current individual. After traversing and processing other effective individuals, the distances of other effective individuals are obtained. Then, the distances of all effective individuals are sorted in ascending order to obtain an ascending distance sequence. The distance corresponding to the preset percentage quantile in the ascending distance sequence is taken as the sparsity threshold of the current population. Effective individuals with a distance less than or equal to the sparsity threshold are mutated according to the mutation method in the current iteration round to obtain mutated individuals. A new population is formed by individuals in the current population whose distance is greater than the sparsity threshold and the mutated individuals.

[0017] The stagnation detection and restart strategy is used to monitor the stagnation state based on the overvolume and restart the population, including:

[0018] After each generation of evolution is completed, the hypervolume of the population is calculated based on the objective function values ​​of all non-dominated solutions in the current generation. A sliding window is used to obtain the current generation and the previous K-1 generations. The change in hypervolume between adjacent generations within the window is calculated and the average of all hypervolume changes is obtained to obtain the average change in hypervolume of the current generation. If the average change in hypervolume of the current generation is less than a preset change threshold, the population restart mechanism is triggered; otherwise, it is not triggered.

[0019] The population restart mechanism includes:

[0020] Retain a preset proportion of individuals in the current population and designate them as elite individuals. Then, randomly select values ​​for the decision variables within the feasible range and generate other proportions of individuals, so that the number of individuals in the population is the same as the population size before the restart. In subsequent generations, the current elite individuals are always retained.

[0021] II. An Optimized Control System for Ethylene Cracking Furnace Based on Adaptive Multi-Objective Evolutionary Algorithm

[0022] The data acquisition unit is used to acquire historical operating parameters of the ethylene cracking furnace;

[0023] The model building unit is used to build a multi-objective optimization model for the pyrolysis furnace.

[0024] The solution unit is used to solve the multi-objective optimization model of the pyrolysis furnace using an adaptive multi-objective evolutionary algorithm to obtain the Pareto optimal solution set;

[0025] The output unit is used to determine the optimal temperature control scheme for the pyrolysis furnace outlet pipe based on the Pareto optimal solution set.

[0026] III. A computer device

[0027] It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the ethylene cracking furnace optimization control method based on an adaptive multi-objective evolutionary algorithm.

[0028] IV. A computer-readable storage medium

[0029] The medium stores a computer program, which, when executed by a processor, implements the steps of the ethylene cracking furnace optimization control method based on an adaptive multi-objective evolutionary algorithm.

[0030] The beneficial effects of this invention are as follows:

[0031] This invention constructs decision variables by setting the outlet pipe temperature of the cracking furnace in stages during the operating cycle; then, it proposes the AMOEA method and uses this algorithm to solve for the decision variables composed of the multi-stage outlet pipe temperatures of the cracking furnace. The resulting cracking furnace outlet pipe temperature control scheme can slow down the rate of decline in ethylene production and improve the production efficiency of the ethylene cracking furnace by periodically adjusting the outlet pipe temperature. Attached Figure Description

[0032] Figure 1 This is a flow chart of an ethylene cracking furnace production unit.

[0033] Figure 2 This is a flowchart of the method of the present invention.

[0034] Figure 3 This is a Pareto front distribution diagram of AMOEA on the ZDT3 problem.

[0035] Figure 4 This is a diagram showing the Pareto front distribution of AMOEA on the DTLZ2 problem.

[0036] Figure 5 This is a diagram showing the Pareto front distribution of AMOEA on the WFG2 problem.

[0037] Figure 6 The Pareto front plot is obtained by applying the AMOEA method to the optimization of an ethylene cracking furnace. Detailed Implementation

[0038] To enable those skilled in the art to better understand the present invention and to more clearly define the scope of protection of the present invention, the present invention will be described in detail below with reference to certain specific embodiments and accompanying drawings. It should be noted that the following are only some specific embodiments of the present invention, and are merely a part of the embodiments of the present invention. The specific and direct descriptions of related structures are only for the convenience of understanding the present invention, and the specific features do not necessarily or directly limit the scope of the present invention. Conventional selections and substitutions made by those skilled in the art under the guidance of the present invention, as well as reasonable arrangements and combinations of several technical features under the guidance of the present invention, should all be considered within the scope of protection of the present invention.

[0039] Figure 1This is a process flow diagram for ethylene production, where 1 represents the feed inlet; 2 represents the cracking furnace; 3 represents the quench oil tower; 4 represents the acid gas outlet; 5 represents the gas measuring device; 6 represents the quench water tower; 7 represents the desiccant; 8 represents the propane removal tower; 9 represents the butane removal tower; 10 represents the acetylene feed inlet; 11 represents the hydrogen processor; 12 represents the cold box; 13 represents the fuel; 14 represents the methanation removal tower; 15 represents the ethane removal tower; 16 represents the mixed feedstock processing section; 17 represents ethylene distillation; and 18 represents the ethylene distillation tower.

[0040] like Figure 2 As shown, the ethylene cracking furnace optimization control method based on an adaptive multi-objective evolutionary algorithm proposed in this invention specifically includes the following steps:

[0041] S1: Collect and acquire historical operating parameters of the ethylene cracking furnace, including process variables affecting ethylene and propylene yields such as cracking furnace outlet pipe temperature, feed properties, gas-liquid ratio, naphtha flow rate (kg / h), outlet pressure (kPa), and coking thickness (cm), as well as corresponding ethylene and propylene yield data; and construct a multi-objective optimization model for the cracking furnace based on the historical operating parameters of the ethylene cracking furnace.

[0042] In one feasible implementation, constructing a multi-objective optimization model for the pyrolysis furnace includes:

[0043] The operating cycle of an ethylene cracking furnace is divided into several stages, with the furnace outlet pipe temperature at each stage constituting a decision variable. For example, if the operating cycle of an ethylene cracking furnace is 63 days, the furnace outlet pipe temperature over those 63 days is divided into nine stages, each lasting seven days. This invention periodically increases the furnace temperature in seven-day cycles to slow the rate of decline in ethylene production. An ethylene cracking furnace simulation system is constructed based on acquired historical operating parameters to establish a quantitative relationship between furnace operating conditions (especially outlet pipe temperature) and product distribution (ethylene and propylene yields). This system is crucial for solving the multi-objective optimization model.

[0044] In one feasible implementation, the specific construction process of the ethylene cracking furnace simulation system is as follows:

[0045] Based on a large amount of historical operating data, a mechanistic modeling method was used to train a model capable of accurately predicting ethylene and propylene yields under specific operating conditions. The model's inputs include the cracking furnace outlet pipe temperature T (which includes the outlet pipe temperatures T0 for each stage). i The parameters include raw material properties (such as density and PIONA composition), gas-liquid ratio, feed flow rate, and outlet pressure. The output is the predicted ethylene yield and propylene yield.

[0046] During the optimization process, for each set of given decision variables (i.e., a temperature control scheme), the ethylene cracking furnace simulation system can calculate the predicted ethylene and propylene yields at each stage under the temperature control scheme, and then directly substitute the calculated predicted yield values ​​into the optimization objective function.

[0047] Temperature range constraints and temperature difference constraints between adjacent stages are set; optimization objectives are constructed, with the first objective being to maximize ethylene yield and the second objective being to minimize the decrease in propylene yield, thus obtaining a multi-objective optimization model for the cracking furnace. The specific formula is as follows:

[0048] Ethylene yield f1 satisfies in Let be the ethylene yield in stage i, where i = 1, ..., 9.

[0049] The propylene yield decreases, f2 satisfies in The initial propylene yield, Let be the propylene yield in stage i.

[0050] Temperature range constraint is T min ≤T i ≤T max T i T represents the temperature of the pyrolysis furnace outlet pipe in stage i. min and T max These represent the lower and upper limits of the outlet pipe temperature of the pyrolysis furnace, respectively. min and T max Set according to process requirements, such as T min =1122K and T max =1142K.

[0051] The temperature difference between adjacent stages is constrained to be |ΔT i |≤ΔT max , where ΔT i =T i -T i-1 ΔT max The maximum permissible temperature difference, || represents absolute value operation, T i-1 This indicates the temperature of the pyrolysis furnace outlet pipe in stage i-1.

[0052] Temperature range constraints and temperature difference constraints between adjacent stages directly affect the decision variables themselves, ensuring that the optimized temperature scheme is within the safe and operational range allowed by the process.

[0053] S2: The Adaptive Multi-Objective Evolutionary Algorithm (AMOEA) is used to solve the multi-objective optimization model of the cracking furnace until the solution is complete, obtaining the Pareto optimal solution set. Based on the Pareto optimal solution set, the optimal temperature control scheme for the cracking furnace outlet pipe is determined to achieve the control of the ethylene cracking furnace and complete the coordinated optimization of ethylene and propylene yields. In the actual production of the ethylene cracking furnace, the optimal temperature control scheme for the cracking furnace outlet pipe is selected from the Pareto optimal solution set according to different production needs.

[0054] In one feasible implementation, the adaptive multi-objective evolutionary algorithm is an improvement on the multi-objective evolutionary algorithm (MOEA), specifically including one or more of the following: adaptive mutation strategy, diversity enhancement mechanism, and stagnation detection and restart strategy. The adaptive mutation strategy is used to dynamically adjust the mutation intensity according to Cauchy mutation and polynomial mutation, the diversity enhancement mechanism is used to maintain population diversity according to the target space matrix, and the stagnation detection and restart strategy is used to monitor the stagnation state and restart the population according to the hypervolume.

[0055] Adaptive mutation strategies are used to dynamically adjust the mutation intensity based on Cauchy mutation and polynomial mutation, including:

[0056] Each decision variable has a 50% probability of mutation. In the early stages of evolution, Cauchy mutation is dominant, meaning the probability of choosing Cauchy mutation is greater than that of choosing polynomial mutation. In the later stages of evolution, polynomial mutation is dominant, meaning the probability of choosing polynomial mutation is greater than that of choosing Cauchy mutation. For example, in an adaptive multi-objective evolutionary algorithm with a fixed number of iterations of 10,000, the first 5,000 iterations are considered the early evolutionary stage, and the last 5,000 iterations are considered the late evolutionary stage. In the early evolutionary stage, Cauchy mutation is chosen with a 70% probability; in the late evolutionary stage, polynomial mutation is chosen with an 80% probability.

[0057] The formula for Cauchy variation is as follows:

[0058] x i ′=x i +γ·C(0,1)·(U i -L i );

[0059] Where, x i 'x' represents the mutated decision variable. i Let U be the decision variable before mutation, with a scaling factor γ = 0.1, C(0,1) be a standard Cauchy random variable, and U be the decision variable before mutation. i and L i These are the upper and lower bounds of the decision variable, respectively;

[0060] The formula for polynomial variation is as follows:

[0061] x i ′=xi +δ·(U i -L i );

[0062]

[0063] Where u is a uniform random number in [0,1], η = 30 is the distribution index, δ is the variance, δ1 is the first auxiliary quantity, and δ2 is the second auxiliary quantity.

[0064] This invention, for each individual, dynamically adjusts the selection tendency of Cauchy mutation and polynomial mutation according to the core needs of the current evolutionary stage, rather than using a fixed probability.

[0065] In the early stages of evolution, the core objective of the algorithm is "global exploration," which requires covering the decision space as much as possible and identifying potential high-quality regions. Therefore, the mutation mechanism tends to be Cauchy mutation, which uses a large perturbation to quickly disperse individuals in the population to different search areas, avoiding blind spots caused by the initial population distribution limitations, and laying a broad foundation for the subsequent discovery of high-quality solutions.

[0066] In the later stages of evolution, the algorithm has already identified several high-quality search regions through early exploration. The core objective shifts to "local development," requiring fine-tuning of variable values ​​within these high-quality regions to improve solution accuracy. Therefore, the mutation mechanism tends to favor polynomial mutation, optimizing existing high-quality solutions through small, targeted perturbations. For example, within the optimal range of ethylene cracking furnace temperature parameters, the temperature value is fine-tuned to further balance the synergistic optimization effect of ethylene and propylene yields, avoiding fluctuations in solution quality due to large perturbations.

[0067] Diversity enhancement mechanisms are used to maintain population diversity based on a target spatial matrix, including:

[0068] Every few generations, effective individuals in the current population are selected, and a target space matrix for the population is constructed based on all effective individuals. Each row of the target space matrix consists of two target values ​​for each effective individual, and the target space matrix is ​​stored in the form of a KDTree. Next, based on the target space matrix of the population, the distance between each effective individual and other effective individuals is calculated to obtain all distances of the current effective individual. Then, the second smallest distance among the distances between each individual and other individuals is taken as the distance of the current individual. After traversing and processing other effective individuals, the distances of other effective individuals are obtained. Then, the distances of all effective individuals are sorted in ascending order to obtain an ascending distance sequence. The distance corresponding to the preset percentage (e.g., 70%) quantile in the ascending distance sequence is taken as the sparsity threshold of the current population. Effective individuals with distances less than or equal to the sparsity threshold are mutated according to the mutation method in the current iteration round to obtain mutated individuals. A new population is formed by individuals in the current population whose distances are greater than the sparsity threshold and the mutated individuals.

[0069] Stagnation detection and restart strategies are used to detect stagnant states based on overvolume and restart the population, including:

[0070] After each generation of evolution (i.e., after the population completes one crossover, mutation, and selection operation), the hypervolume HV of the population is calculated based on the objective function values ​​of all non-dominated solutions in the current generation. t The sliding window is K in size. It captures the current generation and the previous K-1 generations. The change in hypervolume between adjacent generations within the window is calculated, and the average of all hypervolume changes is obtained to determine the average hypervolume change for the current generation. If the average hypervolume change for the current generation is less than a preset threshold (e.g., 0.01), it indicates that the HV index has shown almost no improvement in the last 10 generations, the algorithm has stagnated, and a population restart mechanism is triggered. Otherwise, it indicates that the HV index has still shown effective improvement in the last 10 generations, the algorithm has not stagnated, and continues to evolve normally, thus the population restart mechanism is not triggered. This invention uses a sliding window to cover the hypervolume changes of more generations, which can avoid misjudgments caused by accidental fluctuations in data from a single generation and ensure the stability of the detection results.

[0071] In one feasible implementation, the population restart mechanism includes:

[0072] Retain a predetermined proportion (e.g., 20%) of individuals in the current population and designate them as elite individuals. Then, randomly select values ​​for the decision variables within the feasible range and generate other proportions of individuals, so that the number of individuals in the population is the same as the population size before the restart. In the subsequent generations (e.g., 5 generations), the current elite individuals are always retained. That is, even if the crowding distance of the elite individuals is small, they will not be eliminated, allowing them enough time to recombine with new individuals and improve the overall quality of the subsequent solution set.

[0073] In one feasible implementation, for all individuals in the current generation (i.e., non-dominated solutions), elite individuals are selected according to the rule of "non-dominated sorting + crowding distance calculation":

[0074] First, classify all individuals in the solution set according to whether they are dominated by other individuals (those not dominated by any individual are in level 1, those dominated by level 1 are in level 2, and so on). Prioritize non-dominated solutions in level 1. For non-dominated solutions in level 1, calculate the distance between each individual and its neighboring individuals in the target space (the larger the distance, the more "sparse" the region where the individual is located, and the better the distribution of the solution set), and sort them from largest to smallest crowding distance. From the sorted non-dominated solutions in level 1, select the top 20% of individuals as "elite individuals" to retain (for example, if the current population size is 100, then retain 20 elite individuals). These individuals carry the optimal genes that the algorithm has explored and should be avoided from being lost.

[0075] In one feasible implementation, randomly selecting values ​​for the decision variables within a feasible range and generating individuals in other proportions specifically involves randomly generating temperature parameters for new individuals within the pyrolysis furnace temperature range, with the number of new individuals being 80. Optionally, the pyrolysis furnace temperature range is 1050K to 1150K. This invention, by randomly generating new individuals, injects entirely new gene combinations into the population, helping the algorithm escape the current local optimum, restart global exploration, and find a better solution set.

[0076] To verify the effectiveness of the AMOEA method proposed in this invention for optimizing the outlet pipe temperature of a pyrolysis furnace, a benchmark test function was first used. Seven classic test functions were selected in this embodiment, as shown in Table 1. The parameters of the AMOEA method are shown in Table 2. The Pareto front distribution obtained by the AMOEA method on different test function sets is illustrated in the following figures. Figure 3 , Figure 4 and Figure 5 As shown. Figure 3 The figure shows the Pareto front distribution of the AMOEA method on the ZDT3 problem. This front has the characteristics of being discontinuous and piecewise continuous. As can be seen from the figure, the AMOEA method can approximate the real Pareto front well and maintain the distribution of the solution. Figure 4 The Pareto front distribution of the AMOEA method on the DTLZ2 problem is shown. This problem has a high-dimensional objective space (3 objectives in this embodiment) and its Pareto front is a continuous sphere. The AMOEA method also shows good performance on this problem. Figure 5 The distribution of the Pareto front of the AMOEA method on the WFG2 problem, which has a mixed, piecewise continuous and non-convex Pareto front, is shown. The AMOEA method can effectively handle its complexity.

[0077] Table 1 Comparison of characteristics of test function sets

[0078]

[0079] Table 2 AMOEA Algorithm Parameter Configuration

[0080]

[0081] This embodiment selects representative optimization algorithms from various metaheuristic optimization algorithms for comparison, including NSGA-II (Non-dominated Sorting Genetic Algorithm II), MOEA / D (Multi-Objective Evolutionary Algorithm based on Decomposition), and SPEA2 (Strength Pareto Evolutionary Algorithm 2). The performance of the algorithms is determined by comparing the optimal values ​​of the test functions obtained by each algorithm. The comparison results are shown in Table 3.

[0082] Table 3 Comparison of optimization results for benchmark test functions

[0083]

[0084]

[0085] As can be seen from Table 3, compared with other optimization algorithms, the AMOEA method has the smallest optimization accuracy and standard deviation, which means that the AMOEA method has significantly improved optimization accuracy and stability.

[0086] In the AMOEA method, the adaptive mutation strategy combines Cauchy mutation and multinomial mutation and dynamically adjusts the mutation intensity. The diversity enhancement mechanism is based on KDTree to calculate the distance distribution between individuals to maintain population diversity. The stagnation detection and restart strategy judges the stagnation state by monitoring changes in hypervolume index and restarts part of the population.

[0087] This invention applies the AMOEA method to the production application of ethylene cracking furnaces. To verify the effectiveness of the AMOEA optimization scheme, two comparative experiments were set up:

[0088] Constant temperature experiment: Throughout the entire operation cycle, the temperature of the cracking furnace outlet pipe is kept constant at 1122K (this temperature is a commonly used reference temperature in traditional operation).

[0089] AMOEA Optimization Experiment: A multi-objective optimization model was constructed based on historical operating data and solved using the AMOEA method. The feasible range for the decision variables (temperature at each stage) searched by the AMOEA method was set to [1112K, 1142K]. This range represents the operational space reserved for temperature adjustment under the premise of process safety.

[0090] The optimization and comparative analysis process is as follows:

[0091] First, the AMOEA method is run, ultimately yielding a series of Pareto optimal solutions. Each solution corresponds to a control scheme consisting of nine stage temperature values ​​and its corresponding optimization objective (i.e., the increase in ethylene yield and the decrease in propylene yield). These solutions constitute the Pareto optimal solution set, and their distribution in the frontier objective space is as follows: Figure 6 As shown. Figure 6 The Pareto front clearly demonstrates the trade-off between the conflicting objectives of increasing ethylene yield and decreasing propylene yield. Appropriate operating points (i.e., temperature control schemes) can be selected from this front based on actual production needs (such as prioritizing increased ethylene production or requiring greater propylene stability).

[0092] Then, three representative optimization schemes (A, B, and C) were selected from the Pareto optimal solution set for detailed simulation and effect analysis, and the results were compared with those of the constant temperature scheme. These three schemes are located at different positions on the Pareto front, representing different optimization tendencies. Each scheme specifies the exact outlet pipe temperature setpoint for each of the nine stages.

[0093] The cumulative total yield (or average yield) obtained by simulating the operation of each optimization scheme (A, B, C) over the entire 63-day operating cycle is compared with the cumulative total yield (or average yield) of the constant temperature scheme (1122K) over the same period to calculate the corresponding percentage change, which is the yield change, as shown in Table 4.

[0094] Table 4 Comparison of ethylene and propylene yields

[0095]

[0096] The following can be analyzed from Table 4:

[0097] Optimized Scheme A (Compromise Scheme): Its temperature control strategy generally follows a "rise first, then fall" pattern. That is, the temperature is appropriately increased in the early stages of operation to improve ethylene yield, while in the later stages, the temperature is appropriately decreased to mitigate excessive propylene cracking and yield decline. This scheme achieved a 2.7% increase in ethylene production while effectively controlling the decline in propylene production to within 0.7%, achieving a good balance between the two objectives.

[0098] Optimization Scheme B: Its temperature control strategy exhibits the characteristic of "periodic fluctuations". This scheme tends to explore the comprehensive impact of periodic temperature changes on yield, achieving a higher increase in ethylene production (+4.4%), but the decrease in propylene production also increased (-1.9%).

[0099] Optimization Scheme C: Its temperature control strategy is characterized by a "gradual increase" pattern. In order to maximize the ethylene yield (+6.2%), this scheme continuously increases the reaction temperature throughout the operating cycle, but this also leads to an increased degree of propylene cracking, resulting in a significant decrease in its yield (-3.1%).

[0100] Temperature adjustment strategies indicate that moderate and targeted periodic temperature increases can effectively promote ethylene production and compensate for the efficiency decline caused by coking; while timely and appropriate cooling helps to slow down excessive propylene cracking and maintain a reasonable product structure.

[0101] In actual production, the most suitable solution can be selected from the Pareto optimal solution set based on current market demand (such as the price ratio of ethylene to propylene), production plan, and equipment status. For example, solution A can be selected as a preferred compromise, which successfully controls the decrease in propylene production to a low level (-0.7%) while significantly increasing ethylene production (+2.7%). This invention calculates the final ethylene production under the same output and conditions, specifically when the outlet pipe temperature is constant and when the outlet pipe temperature is periodically adjusted, thereby calculating the rate of increase in ethylene production under periodic adjustment compared to the constant temperature condition. This invention verifies the effectiveness of the AMOEA method for optimizing ethylene cracking furnaces; its adaptive mechanism and diverse control capabilities ensure the practicality and stability of the optimized solution. Simultaneously, experimental results demonstrate that this algorithm can guide production departments to rationally find outlet pipe temperature parameter adjustment schemes to improve ethylene production yield.

[0102] This invention also proposes an optimized control system for an ethylene cracking furnace based on an adaptive multi-objective evolutionary algorithm. The system includes:

[0103] The data acquisition unit is used to acquire historical operating parameters of the ethylene cracking furnace;

[0104] The model building unit is used to build a multi-objective optimization model for the pyrolysis furnace.

[0105] The solution unit is used to solve the multi-objective optimization model of the pyrolysis furnace using an adaptive multi-objective evolutionary algorithm to obtain the Pareto optimal solution set;

[0106] The output unit is used to determine the optimal temperature control scheme for the pyrolysis furnace outlet pipe based on the Pareto optimal solution set.

[0107] To address the coking problem in ethylene cracking furnaces during production, this invention proposes the AMOEA method. This algorithm is used to calculate the outlet pipe temperature of the cracking furnace, and the rate of decline in ethylene production can be slowed down by periodically adjusting the outlet pipe temperature, thereby improving the production efficiency of the ethylene cracking furnace.

[0108] The present invention also proposes a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of an ethylene cracking furnace optimization control method based on an adaptive multi-objective evolutionary algorithm.

[0109] The present invention also proposes a computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of an ethylene cracking furnace optimization control method based on an adaptive multi-objective evolutionary algorithm.

[0110] The present invention also proposes a computer program product, including a computer program / instruction that, when executed by a processor, implements the steps of an ethylene cracking furnace optimization control method based on an adaptive multi-objective evolutionary algorithm.

Claims

1. A method for optimizing the control of an ethylene cracking furnace based on an adaptive multi-objective evolutionary algorithm, characterized in that, Includes the following steps: S1: Obtain historical operating parameters of the ethylene cracking furnace and construct a multi-objective optimization model for the cracking furnace; S2: Use an adaptive multi-objective evolutionary algorithm to solve the multi-objective optimization model of the cracking furnace until the solution is completed, and obtain the Pareto optimal solution set; determine the optimal temperature control scheme of the cracking furnace outlet pipe based on the Pareto optimal solution set to realize the control of the ethylene cracking furnace.

2. The ethylene cracking furnace optimization control method based on an adaptive multi-objective evolutionary algorithm according to claim 1, characterized in that, In S1, constructing the multi-objective optimization model for the cracking furnace includes: The operation cycle of the ethylene cracking furnace is divided into several stages, and the outlet pipe temperature of the cracking furnace in each stage constitutes the decision variable. Based on the obtained historical operating parameters, temperature range constraints and temperature difference constraints between adjacent stages are set. An optimization objective is constructed, the first objective of which is to maximize the ethylene yield and the second objective is to minimize the decrease in propylene yield, thereby obtaining a multi-objective optimization model for the cracking furnace.

3. The ethylene cracking furnace optimization control method based on an adaptive multi-objective evolutionary algorithm according to claim 1, characterized in that, In S2, the adaptive multi-objective evolutionary algorithm is an improvement on the multi-objective evolutionary algorithm. Specifically, it includes one or more of the following: adaptive mutation strategy, diversity enhancement mechanism, and stagnation detection and restart strategy. The adaptive mutation strategy is used to dynamically adjust the mutation intensity according to Cauchy mutation and polynomial mutation. The diversity enhancement mechanism is used to maintain population diversity according to the target space matrix. The stagnation detection and restart strategy is used to monitor the stagnation state and restart the population according to the hypervolume.

4. The ethylene cracking furnace optimization control method based on an adaptive multi-objective evolutionary algorithm according to claim 3, characterized in that, The adaptive mutation strategy is used to dynamically adjust the mutation intensity based on Cauchy mutation and polynomial mutation, including: In the early stages of evolution, each decision variable is dominated by Cauchy mutation; in the later stages of evolution, it is dominated by multinomial mutation.

5. The ethylene cracking furnace optimization control method based on an adaptive multi-objective evolutionary algorithm according to claim 3, characterized in that, The diversity enhancement mechanism is used to maintain population diversity according to the target space matrix, including: Every few generations, effective individuals in the current population are selected, and a target space matrix of the population is constructed based on all effective individuals. Then, based on the target space matrix of the population, the distance between each effective individual and other effective individuals is calculated to obtain all distances of the current effective individual. The second smallest distance among the distances between each individual and other individuals is taken as the distance of the current individual. After traversing and processing other effective individuals, the distances of other effective individuals are obtained. Then, the distances of all effective individuals are sorted in ascending order to obtain an ascending distance sequence. The distance corresponding to the preset percentage quantile in the ascending distance sequence is taken as the sparsity threshold of the current population. Effective individuals with a distance less than or equal to the sparsity threshold are mutated according to the mutation method in the current iteration round to obtain mutated individuals. A new population is formed by individuals in the current population whose distance is greater than the sparsity threshold and the mutated individuals.

6. The ethylene cracking furnace optimization control method based on an adaptive multi-objective evolutionary algorithm according to claim 3, characterized in that, The stagnation detection and restart strategy is used to monitor the stagnation state based on the overvolume and restart the population, including: After each generation of evolution is completed, the hypervolume of the population is calculated based on the objective function values ​​of all non-dominated solutions in the current generation. A sliding window is used to obtain the current generation and the previous K-1 generations. The change in hypervolume between adjacent generations within the window is calculated and the average of all hypervolume changes is obtained to obtain the average change in hypervolume of the current generation. If the average change in hypervolume of the current generation is less than a preset change threshold, the population restart mechanism is triggered; otherwise, it is not triggered.

7. The ethylene cracking furnace optimization control method based on an adaptive multi-objective evolutionary algorithm according to claim 6, characterized in that, The population restart mechanism includes: Retain a preset proportion of individuals in the current population and designate them as elite individuals. Then, randomly select values ​​for the decision variables within the feasible range and generate other proportions of individuals, so that the number of individuals in the population is the same as the population size before the restart. In subsequent generations, the current elite individuals are always retained.

8. An optimized control system for an ethylene cracking furnace based on an adaptive multi-objective evolutionary algorithm, characterized in that, include: The data acquisition unit is used to acquire historical operating parameters of the ethylene cracking furnace; The model building unit is used to build a multi-objective optimization model for the pyrolysis furnace. The solution unit is used to solve the multi-objective optimization model of the pyrolysis furnace using an adaptive multi-objective evolutionary algorithm to obtain the Pareto optimal solution set; The output unit is used to determine the optimal temperature control scheme for the pyrolysis furnace outlet pipe based on the Pareto optimal solution set.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the ethylene cracking furnace optimization control method based on the adaptive multi-objective evolutionary algorithm as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the ethylene cracking furnace optimization control method based on the adaptive multi-objective evolutionary algorithm as described in any one of claims 1 to 7.

Images