Optimization method for synthesis gas production based on multi-objective particle swarm optimization algorithm

Abstract

The invention discloses a synthesis gas production optimization method based on a multi-objective particle swarm optimization algorithm. The method comprises the following steps: decomposing a space into a plurality of sub-regions equal to a population number N to ensure diversity, classifying solutions, selecting an optimal non-dominated solution, classifying the remaining solutions in the sub-regions by using the optimal solution, and selecting a solution with the largest super volume value in the region. The method comprises the following steps: using an elite strategy to mix parent and child populations to compete and generate the next generation population. The method combines the crossover operation and domain correction of the particle swarm algorithm to improve the quality of the obtained solution by using domain information. The method assists the crossover operator in performing local search and global search by using a selection strategy, assigning a higher fitness value to particles with a sparser domain to improve the diversity of the result, and selecting a suitable solution to generate the next generation. The method controls the feed ratio of raw material O2 / CH4, the gas hourly space velocity and the temperature in the synthesis gas production process to achieve the highest CH4 conversion rate, the CO selectivity and the lowest H2 / CO ratio in the synthesis gas production.

Description

Technical Field

[0001] The present invention belongs to the technical field of synthesis gas production, and in particular relates to a synthesis gas production optimization method based on a multi-objective particle swarm optimization algorithm. Background Art

[0002] Synthesis gas is a raw gas used as a chemical raw material, with carbon monoxide and hydrogen as the main components. It is widely used in the production of methanol, ammonia, hydrogen and higher hydrocarbons, and is converted from carbon-containing compounds such as coal, oil, and natural gas. The commonly used routes are steam reforming of methane, carbon dioxide reforming of methane, partial oxidation of methane, and a combination of carbon dioxide reforming and partial oxidation of methane. According to the different sources, compositions and uses of synthesis gas, they can also be called coal gas, synthetic ammonia raw gas, methanol synthesis gas, etc.

[0003] The raw materials for making syngas contain different H / C molar ratios, such as about 1:1 H / C in coal and about 4:1 in natural gas, and the composition ratios of the syngas produced from these raw materials are also different. This application produces syngas by a combined method of carbon dioxide reforming and partial oxidation of natural gas over a Pt / γ-Al2O3 catalyst. This syngas production method produces a very low H2 / CO ratio, and because it is a combination of a highly exothermic reaction of methane oxidation and a moderately endothermic reaction of carbon dioxide reforming of methane, it is easier to control the reactor temperature.

[0004] This experiment hopes to obtain the highest CH4 conversion rate, the highest CO selectivity, and the lowest H2 / CO ratio. It is hoped that the results obtained in these three dimensions can be as optimal as possible. In the case of multiple objective functions and these objective functions conflict with each other, if a single-objective optimization technique is used to optimize the parameters, it is difficult to obtain a result that takes all parties into consideration. Therefore, multi-objective optimization technology can be used to better handle it. Unlike single-objective problems that will obtain an optimal solution, multi-objective optimization problems often obtain a series of non-dominated solutions, called Pareto optimal solutions. The surface formed by these solutions in space is called the Pareto front. The purpose of the multi-objective optimization algorithm is to achieve the following points as much as possible: (1) the solution obtained is as close to the optimal solution as possible, (2) the solution obtained can be distributed as evenly as possible on the Pareto front, and (3) the solution obtained covers the Pareto front as widely as possible.

[0005] A multi-objective optimization problem can be expressed as follows:

[0006]

[0007] in are called decision variables, and X is an n-dimensional decision space. i (x)(i=1,…,m) is the i-th objective to be minimized.i (x)(i=1,2,…,q) defines the i-th inequality constraint, h j (x)(j=1,2,…,p) defines the jth equality constraint. All constraints determine the set of feasible solutions, represented by Ω, and Y={F(x)|x∈Ω} represents the target space. The task of multi-objective optimization is to try to find a feasible solution x∈Ω so that each f i (x)(i=1,…,m) is as small as possible. The following are several definitions in multi-objective optimization problems:

[0008] 1) Pareto dominance: The Pareto dominance between two solutions x, z∈Ω is defined as follows if:

[0009]

[0010] Then x is said to dominate z, and is expressed as x>z.

[0011] 2) Pareto optimal: If Then the solution vector x is called the Pareto optimal solution of Ω.

[0012] 3) Pareto optimal set (PS): The set of Pareto optimal solutions:

[0013]

[0014] 4) Pareto frontier: The Pareto optimal solution set is mapped by the objective function to form the Pareto frontier of the problem:

[0015] PF={F(x)|x∈PS}.

[0016] Particle swarm optimization is a population-based random global evolutionary algorithm that is initialized as a group of random particles (random solutions) and then finds the optimal solution through iteration. Particle swarm optimization has the characteristics of easy implementation, effective storage, and the ability to effectively maintain the diversity of solutions, but it also has its disadvantages. Its primary problem is how to balance convergence and diversity. First, because the file size of multi-objective particle swarm optimization algorithms (MPSO) is not limited, the number of non-dominated solutions grows very fast, which increases the cost of updating the file. Second, the size of physical memory is always limited, so a solution is needed to screen these non-dominated solutions. In addition, the multi-objective particle swarm optimization problem also involves the problem of global optimal gbest and local optimal pbest. Because there is no absolute best solution, but a set of non-dominated solutions, it is necessary to consider how to select and update pbest and gbest.

[0017] In the Multi-objective Evolutionary Algorithm Based on Decomposition (MOEA / D) algorithm, the tasks close to the Pareto front are converted into multiple single-objective optimization problems. If the new solution is closer to the optimal solution than the old solution, the old solution is replaced. This algorithm has high search efficiency and is more deterministic. However, this method ignores the fact that the new solution may be too concentrated in the target space, which will result in poor diversity of the solution set and poor efficiency in facing unknown frontiers.

[0018] Hypervolume is mainly used to evaluate search results in multi-objective evolutionary algorithms. On the other hand, hypervolume is also incorporated into multi-objective evolutionary algorithms as an archiving strategy or selection strategy to guide the algorithm to search. It has good performance in estimating the diversity and convergence of solutions. The larger the hypervolume value, the better the overall performance of the algorithm. However, its disadvantage is that the computational complexity is large, so we only apply it to partial solutions, that is, the remaining solutions, to improve the efficiency of the algorithm. Summary of the invention

[0019] In order to solve the problems existing in the prior art, the present invention provides a synthesis gas production optimization method based on a multi-objective particle swarm optimization algorithm. In view of the respective shortcomings of the MOEA / D algorithm and the MPSO algorithm, the algorithm proposed in the present invention adopts a decomposition method to divide the target space into several sub-regions, classify the solutions according to the sub-regions, and select the optimal solution in each sub-region, and then classify the remaining solutions to select the solution that can maximize the hypervolume value of the sub-region. In this way, the hypervolume value of the next generation population is guaranteed to be as large as possible, so as to achieve the purpose of simultaneously improving diversity and convergence.

[0020] In order to achieve the above object, the technical solution adopted by the present invention is: a synthesis gas production optimization method based on a multi-objective particle swarm optimization algorithm, comprising the following steps:

[0021] S1, set the process parameters and objective function of syngas, the optimization objectives are as follows:

[0022] 1) Maximize CH4 conversion rate, f1;

[0023] 2) maximize CO selectivity, f2;

[0024] 3) Minimize the H2 / CO ratio, f3;

[0025] S2, initialization: Assume that the required population size is N, and take the O2 / CH4 feed molar ratio, gas hourly space velocity GV and temperature T as the three independent variables x = (x1, x2, x3) of the production process parameters. Randomly generate a group of 2N initial particles and a group of initial velocities within the range that meets the experimental requirements. Calculate the corresponding f(x) = {f i (x)|x∈Ω,i=1,2,3}, which is the target space, through a set of direction vectors (γ 1 ,γ 2 ,…γ N ), divide the target space into N sub-areas, and set up a reference point Z = (Z1,…,Z m ), where Z i =min{f i (x)|x∈Ω,i=1,2,3};

[0026] S3, decomposition and fitness calculation: decompose the population according to sub-regions, select the optimal solution in each sub-region; divide the remaining solutions according to the optimal solution, select the solution that can maximize the hypervolume value, and select a solution equal to the population size N, while assigning a higher fitness value to a solution with a sparser domain;

[0027] S4, generating a new solution: using the solution that can maximize the hypervolume value to perform a crossover operation and a domain correction to generate offspring, and to generate a new set of velocities;

[0028] S5, Update: If there exists j=1,…,m such that Z j >min{f j (x)|x∈Ω}, update the reference point, update the particles according to the fitness, and update pbest and gbest;

[0029] S6, mutation: mutate a part of particles;

[0030] S7, mix the parent generation and the child generation, compete to produce the next population, return to S3 to repeat the operation, and complete it after iterating to the set number of times, that is, the three production process parameters of O2 / CH4 feed molar ratio, gas hourly space velocity GV and temperature T are obtained when f1, f2 and f3 are optimized as much as possible.

[0031] The objective function is:

[0032] f1=-8.87×10 -6 ×(86.74+14.6(O2 / CH4)-3.06GV+18.82T+3.14(O2 / CH4)GV-6.91(O2 / CH4) 2 -13.31T 2 )

[0033] f2=-2.152×10 -9 ×(39.46+5.98(O2 / CH4)-2.4CV+13.06T+2.5(O2 / CH4)CV+1.64CVT-3.9(O2 / CH4) 2 -10.15T 2 -3.69GV 2 (O2 / CH4))+45.7

[0034] f3=4.425×10 -10 ×(1.29-0.45T-0.112(O2 / CH4)GV-0.142GVT+0.109(O2 / CH4) 2 +0.405T 2 +1.67T 2 GV)+0.18

[0035] f1=C CH4 , f2=S CO , f3 = H2 / COratio.

[0036] The population POP is calculated by the formula Divided into N categories (P 1 ,P 2 ,…,P N ),

[0037] The target space Y is calculated by the formula Divided into N sub-areas (Y1, Y2, ..., Y N ),

[0038] Δ(F(x),γ i ) is γ i The cosine of the angle between F(x)-Z and F(x)-Z. If there is a solution in a subregion, select that solution; if there are multiple solutions in a subregion, select Δ(F(x),γi ) The point with the largest value is taken as the optimal solution in this sub-region.

[0039] When performing decomposition and fitness calculation, the hypervolume value is used as the fitness function, and the optimal solution and the particle with the largest hypervolume value among the remaining solutions are selected.

[0040] In S3, for the selected optimal solutions, the set composed of them is defined as A = {x 1 , …, x n}, and the set composed of the remaining solutions is defined as B = {x n+1 ,..., x N}. For any solution in B its neighbor set is determined by the following formula where m is the size of set C:

[0041]

[0042] Then determine the upper bound of C and the lower bound where Then, from C U and C L determine set DF:

[0043]

[0044] Calculate the hypervolume value formed by each solution in DF and C U , and select the individual that can make the hypervolume value the largest in this sub-region and denote it as y 1 , let B = B - DF, and repeating the above operation can obtain a set of solutions y 1 , …, y K .

[0045] In S3, better fitness values are assigned to the dominant solutions in relatively sparser regions. The crowding distance is used as the fitness function, and N best particles are selected through binary tournament selection based on the crowding distance.

[0046] In S3, if the number of optimal solutions n ≥ N / 2, that is, the number of regions n + 1 divided according to the optimal solutions ≥ N / 2, and the solution k with the largest hypervolume value obtained ≥ N / 2, then they are sorted according to the size of their hypervolume values, and the ones with larger hypervolume values and set A are retained as the next-generation population; if the number of optimal solutions n < N / 2, then each time n + 1 solutions with the largest hypervolume values can be selected until the number of solutions lacking from the population size is less than n + 1. The remaining selected solutions are determined by sorting according to the size of the hypervolume values, and the next-generation population is composed of these parts.

[0047] In S4, the velocity of the k-th particle is vk =(v k1 ,…,v kn ), position x k =(x k1 ,…,x kn ), in each iteration, the particle updates the velocity and position of the particle through the previous global optimal value gbest and local optimal value pbest:

[0048]

[0049]

[0050] Search through the field of the current particle and use the neighborhood information to correct the current particle. The correction strategy of particle x is:

[0051] x new =x+0.5*(x′-x″)

[0052] Among them, x′ is the global optimum and x″ is the local optimum.

[0053] When selecting particle x, determine the local optimal value pbest and the global optimal value gbest of each selected particle, randomly select a solution as the global optimal value gbest or use the particle's adjacent solutions to determine its global optimal value gbest. The global optimal value gbest is the non-dominated solution with the smallest angle between the target vector and the center vector among the adjacent solutions.

[0054] Compared with the prior art, the present invention has at least the following beneficial effects: the diversity and convergence of the population are contradictory, and the method described in the present invention can use the fitness function of the hypervolume to balance the diversity and convergence of the algorithm; the algorithm complexity is high in the calculation of the hypervolume, and the present invention improves the efficiency of the algorithm by combining decomposition.

[0055] Furthermore, when calculating the decomposition and fitness, the hypervolume value is used as the fitness function. Selecting the optimal solution and the particles that maximize the hypervolume value among the remaining solutions can make the hypervolume value of the next generation population as large as possible, thereby improving the diversity and convergence of the solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 The present invention is an algorithm flow chart.

[0057] Figure 2 Schematic diagram of target space decomposition. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] refer to Figure 1 The present invention applies a new multi-objective particle swarm optimization algorithm based on decomposition and hypervolume, combines the multi-objective evolutionary algorithm based on decomposition (MOEA / D), classifies and selects particles with the largest hypervolume (Hypervolume, HV), and improves the multi-objective particle swarm optimization algorithm (MPSO).

[0060] The improved algorithm proposed in the present invention mainly includes four parts: spatial decomposition and population classification, updating of fitness function based on hypervolume, crossover operation and selection strategy. Among them, spatial decomposition and population classification and updating of fitness function based on hypervolume are used to maintain the diversity of solutions; crossover operation is used to search the decision space; selection strategy helps crossover operation to perform global search and local search.

[0061] (1) Spatial decomposition and population classification

[0062] The objective space Y of the multi-objective optimization problem is based on a set of direction vectors (γ 1 ,γ 2 ,…γ N ) is decomposed into a set of sub-regions Y1,Y2,…,Y N , and then classify the solution of population POP; spatial decomposition and population classification are implemented as follows:

[0063]

[0064]

[0065]

[0066] Z=(Z1,…,Z m ) is a reference point, where Z i =min{f i (x)|x∈Ω},Δ(F(x),γ i ) is γ iThe cosine of the angle between F(x)-Z and F(x)-Z. If there is a solution in a subregion, select that solution; if there are multiple solutions in a subregion, select Δ(F(x),γ i ) value is the optimal solution in the sub-region, that is, the angle should be as small as possible. The population POP is divided into N categories (P 1 ,P 2 ,…,P N ), the target space Y is divided into N sub-areas (Y1, Y2, …, Y N ), the decomposition method has the following characteristics: 1) the Pareto fronts of all sub-regions constitute the Pareto front of the multi-objective optimization problem, 2) because the Pareto front of each sub-region is only a small part, even if the entire Pareto front of the multi-objective optimization problem is a nonlinear set shape, the Pareto front of each sub-region can be approximated as linear, 3) this classification method does not require any aggregation method, only requires the user to select a set of direction vectors, so its computational complexity will be very small.

[0067] (2) Update based on the fitness function of the hypervolume

[0068] It is not necessary that there is a solution in every sub-region. The number of current optimal solutions must be less than or equal to the number of sub-regions. For the selected optimal solutions, the set consisting of them is defined as A = {x 1 , …, x n}, the set of remaining solutions is defined as B = {x n+1 , …, x N}, for any solution in B Its neighbor set is determined by the following formula Where m is the size of the set C.

[0069]

[0070] Then determine the upper bound of C and the lower bound in Then by C U and C L Determine the set DF,

[0071]

[0072] Calculate the hypervolume value formed by each solution in DF and CU, and select the individual in the sub-region that can maximize the hypervolume value and record it as y 1 , let B = B-DF, repeat the above operation to get a set of solutions y 1 , …, y K .

[0073] As shown Figure 2 in the figure, each point represents a solution in the target space, and the diamond-shaped points are the selected optimal solutions; the dashed lines are the uniform sub-regions obtained by decomposing the target space through the weight vector, and then divided by the optimal solutions; the regions divided by the optimal solutions are represented by solid lines, and the solution with the largest hypervolume value in each sub-region divided by the solid line is selected.

[0074] If the number of optimal solutions n ≥ N / 2, that is, the number of regions n + 1 divided by the optimal solutions ≥ N / 2, and the solution k with the largest hypervolume value obtained ≥ N / 2, then they are sorted according to the size of their hypervolume values, and the ones with large hypervolume values and set A are retained as the next-generation population; if the number of optimal solutions n < N / 2, then each time n + 1 solutions with the largest hypervolume values can be selected until the number of solutions lacking from the population size is less than n + 1, and the remaining selected solutions are determined by sorting according to the size of the hypervolume values, and the next-generation population is formed by these parts.

[0075] Through the elitist strategy, the offspring and the parent population are mixed, and they jointly compete to generate the next-generation population, which is conducive to maintaining the excellent individuals in the parent population and ensuring that those excellent individuals can be retained during the evolution process, thereby improving the population level. The elitist strategy usually can accelerate the convergence speed and improve the quality of the solutions. At the same time, we also need to maintain the diversity of the solutions.

[0076] (3) Crossover operation

[0077] There are two types of crossover operations adopted: the crossover operation of the particle swarm optimization algorithm and neighborhood correction. The global search of the algorithm is mainly realized through the crossover operation of the PSO algorithm, that is, through equations (6)(7). The velocity of the k-th particle is expressed as v k =(v k1 ,…,v kn ), and the position is expressed as x k =(x k1 ,…,x kn ). In each iteration, the particle updates its velocity and position through the previous global optimal value (gbest) and local optimal value (pbest):

[0078]

[0079]

[0080] Neighborhood correction searches through the neighborhood of the current particle and corrects the current particle using the neighborhood information, which can improve the quality of the obtained solutions and accelerate the convergence to the Pareto front. The correction strategy for particle x is:

[0081] x new=x+0.5*(x′-x″) (8)

[0082] Among them, x′ is the global optimum and x″ is the local optimum.

[0083] (4) Select strategy

[0084] The selection strategy has a great impact on the performance of the algorithm. The main purpose of the selection strategy is to help the crossover operator perform local and global searches and select appropriate solutions to generate offspring.

[0085] The first selection strategy in the present invention is used to select particle x, that is, x in equations (6) and (7) k and x in equation (8). However, there is a special case, as mentioned in (3), where a dominant solution coexists with a non-dominated solution that is more advantageous than it in a region, but the domain of this dominant solution is relatively sparse. Selecting this dominant solution can maintain better diversity. The sub-solutions generated by this dominant solution and its domain solution will be closer to the Pareto front of the sub-region corresponding to the dominant solution than the sub-solutions generated by other non-dominated solutions. Therefore, it is necessary to assign a higher fitness value to this dominant solution so that it is more likely to be selected to produce offspring. The present invention uses crowding distance to calculate the fitness value to better select such solutions with sparser domains. Then, the N best particles are selected through binary tournament selection based on crowding distance.

[0086] The second selection strategy is to determine the pbest and gbest of each selected particle. To save storage space, the pbest of a particle is its previous best neighboring particle, and it is randomly selected from its neighboring particles. In a single-objective problem, there will only be one global optimal gbest, but in a multi-objective problem, there will be a set of non-dominated solutions that can be used as gbest. There are two strategies that can be used to select gbest. The first is to randomly select a solution as the gbest. This strategy can help the crossover operation to perform a global search; the second is to use the particle's neighboring solutions to determine its gbest. The gbest is the non-dominated solution with the smallest angle between the target vector and the center vector among the neighboring solutions. This strategy can help the crossover operation to perform a local search.

[0087] The original experiment was conducted in a micro U-shaped tubular reactor made of quartz and placed with a catalyst. The reactants were natural gas consisting of 79% CH4, 17% C2H6, 4% C3H8, compressed dry air over molecular sieves consisting of 20% O2, 79% N2 and 1% Ar and ultrapure CO2 (99.99%), the O2 / CH4 feed molar ratio was [0.25, 0.55] gmol / gmol; the gas hourly space velocity GV was [10.000, 20.000] h -1 ; The experimental temperature T is [600,1100]℃.

[0088] The experimental reaction formula for CO2 reforming and CH4 partial oxidation is as follows:

[0089]

[0090] Try to do the following as much as possible:

[0091] 1) Maximize CH4 conversion rate, f1;

[0092] 2) maximize CO selectivity, f2;

[0093] 3) Minimize the H2 / CO ratio, f3;

[0094] Among them, the coefficients of the objective function are as follows:

[0095]

[0096] The objective function is:

[0097] f1=-8.87×10 -6 ×(86.74+14.6(O2 / CH4)-3.06GV+18.82T+3.14(O2 / CH4)GV-6.91(O2 / CH4) 2 -13.31T 2 )

[0098] f2=-2.152×10 -9 ×(39.46+5.98(O2 / CH4)-2.4GV+13.06T+2.5(O2 / CH4)GV+1.64GVT-3.9(O2 / CH4) 2 -10.15T 2 -3.69GV 2 (O2 / CH4))+45.7

[0099] f3=4.425×10 -10 ×(1.29-0.45T-0.112(O2 / CH4)GV-0.142GVT+0.109(O2 / CH4) 2 +0.405T 2 +1.67T 2 GV)+0.18

[0100] Algorithm parameter settings: population size is 1275, crossover probability is 1, mutation probability is 1 / 3, and iteration is 1000 times. The steps are as follows:

[0101] 1) Initialization: Through a set of direction vectors (γ 1 ,γ 2 ,…γ N), divide the target space into N sub-areas, randomly generate a set of initial individuals and a set of initial velocities. And set up a reference point Z = (Z1,…,Z m ) is a reference point, where Z i =min{f i (x)|x∈Ω}

[0102] 2) Decomposition and fitness calculation: Decompose the population according to sub-regions, select the optimal solution in each sub-region; divide the remaining solutions according to the optimal solution, select the solution that can maximize the hypervolume value, and select a solution equal to the population size N. And assign higher fitness values ​​to solutions with sparser domains.

[0103] 3) Generate new solutions: Use these selected solutions to perform crossover operations and domain correction to generate offspring and a new set of velocities.

[0104] 4) Update: If there exists j=1,…,m such that Z j >min{f j (x)|x∈Ω}, then update the reference point, update the particles according to the fitness, and update pbest and gbest.

[0105] 5) Mutation: Make some particles mutate to make the particle distribution more uniform and improve the diversity of the population.

[0106] 6) Mix the parent generation and the child generation, compete to produce the next population, return to step 2) and repeat the operation until the iteration is completed.

Claims

1. A synthesis gas production optimization method based on a multi-objective particle swarm optimization algorithm, characterized in that: It includes the following steps: S1. Set the process parameters and objective function of the syngas. The optimization objectives are as follows: 1) Maximize the CH4 conversion rate, f1; 2) Maximize the CO selectivity, f2; 3) Minimize the H2 / CO ratio, f3; S2, initialization: Assume that the required population size is N, and take the O2 / CH4 feed molar ratio, gas hourly space velocity GV and temperature T as the three independent variables x = (x1, x2, x3) of the production process parameters. Randomly generate a group of 2N initial particles and a group of initial velocities within the range that meets the experimental requirements. Calculate the corresponding f(x) = {f i (x)|x∈Ω,i=1,2,3}, which is the target space, through a set of direction vectors (γ 1 ,γ 2 ,…γ N ), divide the target space into N sub-areas, and set up a reference point Z = (Z1,…,Z m ), where Z i =min{f i (x)|x∈Ω,i=1,2,3}; S3. Decomposition and fitness calculation: Decompose the population according to sub-regions, and select the optimal solutions in each sub-region; re-divide the remaining solutions according to the optimal solutions, select the solution that can maximize the hypervolume value, and select the solutions equal to the population size N. At the same time, assign higher fitness values to the solutions in sparser regions; S4. Generate new solutions: Use the solution that can maximize the hypervolume value to perform crossover operations and neighborhood correction to generate offspring, and generate a new set of velocities; S5, Update: If there exists j=1,…,m such that Z j >min{f j (x)|x∈Ω}, update the reference point, update the particles according to the fitness, and update pbest and gbest; S6. Mutation: Make a part of the particles mutate; S7. Mix the parent generation and the offspring generation, compete to generate the next population, return to S3 to repeat the operation, and complete after iterating to the set number of times, that is, obtain the three production process parameters of the O2 / CH4 feed molar ratio, gas hourly space velocity GV, and temperature T that can make f1, f2, and f3 as optimal as possible.

2. The method for optimizing synthesis gas production based on a multi-objective particle swarm optimization algorithm according to claim 1, characterized in that: The objective function is: f1=-8.87×10 -6 ×(86.74+14.6(O2 / CH4)-3.06GV+18.82T+3.14(O2 / CH4)GV-6.91(O2 / CH4) 2 -13.31T 2 ) f2=-2.152×10 -9 ×(39.46+5.98(O2 / CH4)-2.4GV+13.06T+2.5(O2 / CH4)GV+1.64GVT-3.9(O2 / CH4) 2 -10.15T 2 -3.69GV 2 (O2 / CH4))+45.7 f3=4.425×10 -10 ×(1.29-0.45T-0.112(O2 / CH4)GV-0.142GVT+0.109(O2 / CH4) 2 +0.405T 2 +1.67T 2 GV)+0.18 <h2 style=";text-align:left;direction:ltr">f1 = C<h2 style=";text-align:left;direction:ltr"> CH4 <h2 style=";text-align:left;direction:ltr"> f2=S<h2 style=";text-align:left;direction:ltr"> CO <h2 style=";text-align:left;direction:ltr"> f3, H2 / COratio.

3. The method for optimizing synthesis gas production based on a multi-objective particle swarm optimization algorithm according to claim 1, characterized in that: The population POP is calculated by the formula Divided into N categories (P 1 ,P 2 ,…,P N ), The target space Y is calculated by the formula Divided into N sub-areas (Y1, Y2, ..., Y N ), Δ(F(x),γ i ) is γ i The cosine of the angle between F(x)-Z and F(x)-Z. If there is a solution in a subregion, select that solution; if there are multiple solutions in a subregion, select Δ(F(x),γ i ) value is the optimal solution in this sub-region.

4. The method for optimizing synthesis gas production based on a multi-objective particle swarm optimization algorithm according to claim 1, characterized in that: When performing decomposition and fitness calculation, use the hypervolume value as the fitness function, and select the optimal solution and the particle that can maximize the hypervolume value among the remaining solutions.

5. The method for optimizing synthesis gas production based on a multi-objective particle swarm optimization algorithm according to claim 4, characterized in that: In S3, for the selected optimal solution, the set consisting of it is defined as A = {x 1 , …, x n }, the set of remaining solutions is defined as B = {x n+1 , …, x N }, for any solution in B Its neighbor set is determined by the following formula Where m is the size of the set C: Then determine the upper bound of C and lower bound in Then by C U and C L Determine the set DF: Calculate the difference between each solution in DF and C U The hypervolume value is formed, and the individual that can make the hypervolume value the largest in this sub-region is selected and recorded as y 1 , let B = B-DF, repeat the above operation to get a set of solutions y 1 , …, y K .

6. The method for optimizing synthesis gas production based on a multi-objective particle swarm optimization algorithm according to claim 1, characterized in that: In S3, assign better fitness values to the dominant solutions in relatively sparser regions, use the crowding distance as the fitness function, and select N best particles through binary tournament selection based on the crowding distance.

7. The method for optimizing synthesis gas production based on a multi-objective particle swarm optimization algorithm according to claim 1, characterized in that: In S3, if the number of optimal solutions n≥N / 2, that is, the number of regions n + 1 divided according to the optimal solutions ≥N / 2, and the solution k that can maximize the hypervolume value ≥N / 2, then sort them according to the size of their hypervolume values, and keep the ones with larger hypervolume values and set A as the next generation population; if the number of optimal solutions n < N / 2, then each time select n + 1 solutions with the largest hypervolume values until the number of solutions lacking from the population size is less than n + 1. The remaining selected solutions are determined by sorting according to the size of the hypervolume value, and the next generation population is composed of these parts.

8. The method for optimizing synthesis gas production based on a multi-objective particle swarm optimization algorithm according to claim 1, characterized in that: In S4, the velocity of the kth particle is v k =(v k1 ,…,v kn ), position x k =(x k1 ,…,x kn ), in each iteration, the particle updates the velocity and position of the particle through the previous global optimal value gbest and local optimal value pbest: Search through the neighborhood of the current particle, and use the neighborhood information to correct the current particle. The correction strategy of particle x is: x new =x+0.5*(x′-x″) where x′ is the global optimum and x″ is the local optimum.

9. The method for optimizing synthesis gas production based on a multi-objective particle swarm optimization algorithm according to claim 1, characterized in that: When selecting particle x, determine the local optimum value pbest and the global optimum value gbest of each selected particle. Randomly select a solution as the global optimum value gbest or use the adjacent solutions of the particle to determine its global optimum value gbest. The global optimum value gbest is the non-dominated solution with the smallest angle between the objective vector and the central vector among the adjacent solutions.

Images