An intelligent optimization method for biological fermentation based on multi-objective optimization

Abstract

The present invention discloses an intelligent optimization method for biological fermentation based on multi-objective optimization, comprising the following steps: 1) Collecting the whole-process parameters of the biological fermentation process, including but not limited to raw material parameters, process parameters, regulation parameters and result parameters; 2) Conducting data cleaning and preprocessing on the collected parameters, and using a generative adversarial network (GAN) for data augmentation to solve the problem of insufficient samples and improve data representation; The present invention significantly improves the efficiency and benefits of the biological fermentation process by integrating a variety of advanced optimization algorithms and data augmentation technologies. At the same time, various control parameters are optimized to ensure the stability and accuracy of the process. In addition, through Pareto front analysis, the best balance between multiple objectives can be achieved, further optimizing production efficiency and cost, demonstrating the significant advantages of this technical solution in improving yield, reducing costs and simplifying the process.

Description

Technical Field

[0001] The present invention belongs to the fields of artificial intelligence and biological fermentation, and specifically relates to an intelligent optimization method for biological fermentation based on multi-objective optimization. Background Art

[0002] Traditional biological fermentation technologies mainly rely on manual experience and trial-and-error methods to optimize the production process. Such methods have defects such as insufficient data utilization, high manual dependence, high regulation complexity, insufficient samples, poor data representativeness, lack of multi-objective optimization, and insufficient consideration of actual constraint conditions. These deficiencies limit the yield improvement and efficiency optimization of the fermentation process, while also increasing costs and operational complexity.

[0003] This patent effectively overcomes the limitations of traditional methods by adopting advanced optimization algorithms, data augmentation techniques, and metabolic pathway optimization. It uses means such as generative adversarial networks to enhance data, multi-objective optimization algorithms to balance different objectives, and graph neural networks to optimize metabolic pathways, significantly improving the yield and efficiency of biological fermentation, while reducing costs and operational complexity, and promoting the industrial application of biological fermentation technology. Summary of the Invention

[0004] The present invention aims at the characteristics of the biological fermentation field: the high complexity of biological processes, incomplete data, multiple objectives, and the diversity of process parameters and controls. Through data augmentation techniques, the problem of insufficient samples is solved; through a multi-objective optimization algorithm framework, the balance between multiple objectives is found to optimize the yield, volume, conversion efficiency, and cost while ensuring constraint conditions. Constraint optimization is added to constrain the regulation parameters to avoid solutions that do not meet production conditions. The Pareto front analysis technique is applied to find the optimal solution among multiple optimization objectives, improving the flexibility and practicality of the algorithm.

[0005] To achieve the above object, the present invention provides the following technical solution: an intelligent optimization method for biological fermentation based on multi-objective optimization, comprising the following steps:

[0006] 1) Collect the whole-process parameters of the biological fermentation process, including but not limited to raw material parameters, process parameters, regulation parameters, and result parameters;

[0007] 2) Clean and preprocess the collected parameters, and use a generative adversarial network (GAN) for data augmentation to solve the problem of insufficient samples and improve data representativeness;

[0008] 3) Perform feature engineering on the preprocessed data, including polynomial expansion and logarithmic transformation of continuous variables, and one-hot encoding and combined features of discrete variables;

[0009] 4) Optimize the fermentation parameters using a multi-objective optimization algorithm, including maximizing yield, volume, and chemical efficiency, minimizing cost, and screening for optimized solutions through Pareto front analysis;

[0010] 5) Based on the optimized solution set, optimize the strain's metabolic pathway. Use graph neural networks to model and identify key metabolites, key enzymes, and metabolic pathways, and design gene editing strategies. The optimization of the metabolic pathway includes: converting the microbial metabolic network into a graph model, where nodes represent metabolites, enzymes, or metabolic fluxes, and edges represent metabolic reaction relationships; using graph neural networks to analyze the weights of nodes and edges in the metabolic network to identify key enzymes and key metabolic pathways; designing gene editing strategies based on the analysis results of key pathways and enzymes to optimize the metabolic performance of the strain.

[0011] The parameters collected in step 1) include:

[0012] Raw material parameters: water-soluble fiber content, amino acid composition, mineral content, salt additives;

[0013] Process parameters: stirring method, temperature, pH change, dissolved oxygen concentration, reactor pressure, carbon dioxide concentration;

[0014] Regulation parameters: sugar source addition amount, nitrogen source adjustment, gas flow rate control, pH regulator dosage;

[0015] Result parameters: product concentration, metabolite content, by-product accumulation, protein purity.

[0016] Data cleaning in step 2) includes outlier detection and normalization. Specifically: use the quartile method to detect outliers; process the detected outliers through mean correction or replacement; normalize all parameters to adjust the numerical range to between 0 and 1.

[0017] In step 2), a generative adversarial network (GAN) is used for data augmentation, including: the generator takes fermentation parameters other than the target variable as input; the discriminator provides feedback training to the generator by comparing the statistical characteristics of the generated data and the real data; use the trained generative adversarial network to generate a new augmented data set.

[0018] In step 4), the multi-objective optimization algorithm uses the differential evolution algorithm. The optimization objectives include: maximizing fermentation yield; maximizing fermentation volume; maximizing chemical efficiency; minimizing production cost; the weights of the objective functions are dynamically adjusted according to actual production requirements.

[0019] Use the penalty function method to improve the objective function and constrain the optimization process. The constraint conditions include but are not limited to: fermentation temperature range; pH value range; dissolved oxygen concentration range; reactor pressure range.

[0020] The Pareto front analysis in step 4) includes: comparing the advantages and disadvantages of each optimized solution on different objectives, and screening out the non-dominated solutions; forming the Pareto front through the non-dominated solutions;

[0021] Select the optimal solution in the Pareto front according to actual needs, including maximizing output and chemical efficiency while avoiding a significant increase in cost.

[0022] The optimization algorithms described in step 4) include but are not limited to: gradient-based optimization algorithms, including gradient descent, Adam, RMSProp; population-based optimization algorithms, including genetic algorithms, particle swarm optimization; heuristic-based optimization algorithms, including tabu search, local search; reinforcement learning algorithms or Bayesian optimization algorithms.

[0023] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0024] By integrating a variety of advanced optimization algorithms and data augmentation techniques, the present invention significantly improves the efficiency and effectiveness of the biological fermentation process. At the same time, various control parameters are optimized to ensure the stability and accuracy of the process. In addition, through Pareto front analysis, the best balance between multiple objectives can be achieved, further optimizing production efficiency and cost, demonstrating the significant advantages of this technical solution in improving productivity, reducing costs, and simplifying the process. Brief Description of the Drawings

[0025] Figure 1 It is a schematic flow diagram of the present invention. Detailed Embodiments

[0026] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0027] As Figure 1 shown, a multi-objective optimization-based intelligent optimization method for biological fermentation includes the following steps:

[0028] 1) Collect the whole-process parameters of the biological fermentation process, including but not limited to raw material parameters, process parameters, regulation parameters, and result parameters;

[0029] Among them, the raw material parameters represent various index parameters of the raw materials used in biological fermentation, including but not limited to water-soluble fiber content, mineral content, acidity regulator, amino acid composition, etc. In the present invention, the involved raw material characteristic parameters include cereal content, amino acid content, mineral composition, water-soluble fiber, salt additives, etc., covering more than 1000 items.

[0030] The process parameters represent various index parameters in the biological fermentation process, including but not limited to stirring method, temperature, pH change, dissolved oxygen, reactor pressure, carbon dioxide concentration, reactor type and volume, etc. These parameters are crucial for microbial growth, metabolic pathways, and the formation of the final product. In the examples of the present invention, the involved process control parameters include stirring speed, temperature control, dissolved oxygen level, gas concentration change, reactor type, reaction pressure, fermentation time, etc., covering about 100 parameters.

[0031] The regulation parameters represent the index parameters that can be considered regulated in the biological fermentation process, including but not limited to the amount of sugar supplement, nitrogen source dosage, gas flow rate, dissolved oxygen regulation, pH regulator dosage, etc. Through these regulations, the fermentation process can be better controlled, and the yield and quality of the product can be improved. The regulation parameters involved in the present invention include sugar source addition amount, nitrogen source adjustment, gas flow rate control, pH regulation, etc., totaling about 20 items.

[0032] The result parameters represent the quality, yield, and purity of the final product in the biological fermentation process, including but not limited to product purity, post-fermentation pH value, gas release amount, product concentration, total yield, protein content, cell concentration, accumulation of by-products, etc. These indicators help to evaluate the efficiency of the fermentation process and the quality of the product. In the present invention, the involved result parameters include product concentration, metabolite content, gas generation rate, protein purity, by-product accumulation, pH value after fermentation, etc.

[0033] 2) Clean and preprocess the collected parameters, and use a generative adversarial network (GAN) for data augmentation to solve the problem of insufficient samples and improve data representativeness;

[0034] Clean all the parameters, use a generative adversarial network (GAN) to augment the data, solve the problem of insufficient samples, and improve the representativeness of the data.

[0035] Specifically, take the dissolved oxygen concentration in the original dataset as the data to be filled, and other data as the training set.

[0036] Subsequently, for the generator of the GAN, input other fermentation process parameters except the dissolved oxygen concentration (such as temperature, pH value, etc.), and the generator generates a new sample of dissolved oxygen concentration.

[0037] Subsequently, for the discriminator of the GAN, the generated dissolved oxygen concentration data is compared with the true dissolved oxygen concentration data, and through training, the generated data is made to be consistent with the true data in terms of statistical characteristics.

[0038] Subsequently, the GAN model after training is used to generate dissolved oxygen concentration data, which is combined with the original dataset to generate a new dataset containing more dissolved oxygen concentration data.

[0039] In addition, outliers are recorded using quartiles, which facilitates their replacement during subsequent preprocessing.

[0040] 3) Perform feature engineering on the preprocessed data, including polynomial expansion and logarithmic transformation of continuous variables, one-hot encoding and combined features of discrete variables.

[0041] Preprocess the data in the sample dataset. Among them, the outlier data recorded in S1 is normalized, and the normalization formula is:

[0042]

[0043] Subsequently, a random floating quantity z (with a mean of 0 and a standard deviation of σ) that conforms to the Gaussian distribution is generated. The Gaussian distribution formula is:

[0044] z = RandomNormal(0, σ)

[0045] Subsequently, the Gaussian floating quantity is added to the mean. Assuming the filled mean is μ, the new filled value is:

[0046] x new = μ + z

[0047] Subsequently, the new filled value is normalized to the interval from 0 to 1:

[0048]

[0049] Subsequently, to ensure that the filled value does not exceed 0 and 1, a clipping operation is used:

[0050]

[0051] In addition, feature engineering is used to perform feature dimensionality increase on all features. In the example, it includes: polynomial expansion and logarithmic transformation for continuous variables; one-hot encoding and combined features for discrete variables.

[0052] 4) Use a multi-objective optimization algorithm to optimize the fermentation parameters, including maximizing yield, volume, and chemical efficiency, minimizing cost, and screening the optimal solutions through Pareto front analysis.

[0053] Using a differential evolution as the optimization algorithm and a multi-objective optimization method, the maximization of production, volume, conversion efficiency, and the minimization of cost are set as the objective functions. The specific formulas are as follows:

[0054] Objective 1: Maximize production: f1(x) = - production

[0055] Objective 2: Maximize volume: f2(x) = - volume

[0056] Objective 3: Maximize conversion efficiency: f3(x) = - conversion efficiency

[0057] Objective 4: Minimize cost: f4(x) = cost

[0058] At this time, the objective function is:

[0059] f(x) = α1 * f1(x) + α2 * f2(x) + α3 * f3(x) - α4 * f4(x)

[0060] Where:

[0061] α1, α2, α3, and α4 are the weights of each objective, determining the priority between objectives.

[0062] Set the constraint optimization function and incorporate the constraint conditions into the optimization model. Use the penalty function method to improve the original objective function.

[0063] Determine the constraint conditions:

[0064] (1) Temperature range: 32.0 ≤ T ≤ 33.5

[0065] (2) pH value range: 6.9 ≤ pH ≤ 7.8

[0066] For the constraint g(x), the penalty function is defined as:

[0067] f all (x) = f(x) + γ * P(x)

[0068] Where:

[0069] (1) P(x) is a penalty term, indicating the degree of violation of the constraint:

[0070]

[0071] (2) g i (x) is each constraint function. If g i (x) > 0, it means the constraint is violated;

[0072] (3) γ is the penalty coefficient, controlling the influence degree of the penalty function.

[0073] Multiple solutions obtained by iterating the above algorithm multiple times are used for Pareto front analysis to calculate the dominance relationship.

[0074] Some of the solutions obtained by the differential algorithm are shown in the following table:

[0075]

[0076]

[0077] When calculating the dominance relationship, it can be obtained that:

[0078] (1) When comparing Solution No. 1 with Solution No. 2, Solution No. 1 has a higher yield, but Solution No. 2 has a higher volume. If Solution No. 1 has no obvious disadvantages in other objectives, then Solution No. 1 may dominate Solution No. 2.

[0079] (2) When comparing Solution No. 1 with Solution No. 5, Solution No. 1 has a higher cost, while Solution No. 5 has a higher yield and lower cost. Therefore, Solution No. 5 may dominate Solution No. 1.

[0080] Therefore, it can be obtained that Solution No. 3 and Solution No. 4 are non-dominated, which means they each have advantages in some objectives.

[0081] Therefore, Solution No. 3, Solution No. 4, and Solution No. 5 form the Pareto front.

[0082] In addition, when selecting the optimal solution, according to the actual application requirements, assuming that we hope to maximize the yield and efficiency without significantly increasing the cost. At this time, Solution No. 3 or Solution No. 4 can be selected as the final optimized solution.

[0083] 5) Based on the optimized solution set, optimize the strain metabolic pathway, use graph neural network modeling to identify key metabolites, key enzymes, and metabolic pathways, and design gene editing strategies.

[0084] Specifically, convert the metabolic network of microorganisms into a graph form, where nodes represent metabolites or enzymes, including metabolite concentrations, enzyme expression levels, metabolic fluxes, etc., and edges represent the metabolic reactions or transformation relationships between them, including enzyme activities, reaction rates, reaction equilibria, etc. Use graph neural network for modeling and optimization, analyze the weights of nodes and edges, identify key enzymes and pathways, and help design gene editing strategies.

[0085] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-objective optimization-based intelligent optimization method for biological fermentation, characterized in that, Including the following steps: 1) Collect the whole-process parameters of the biological fermentation process, including but not limited to raw material parameters, process parameters, regulation parameters, and result parameters; 2) Conduct data cleaning and preprocessing on the collected parameters, and use a generative adversarial network (GAN) for data augmentation to solve the problem of insufficient samples and improve data representativeness; 3) Perform feature engineering on the preprocessed data, including polynomial expansion and logarithmic transformation of continuous variables, one-hot encoding and combined features of discrete variables; 4) Use a multi-objective optimization algorithm to optimize the fermentation parameters, including maximizing yield, volume, and chemical efficiency, minimizing cost, and screening the optimal solutions through Pareto front analysis; 5) Based on the optimized solution set, optimize the strain metabolic pathway. Use graph neural network modeling to identify key metabolites, key enzymes, and metabolic pathways, and design gene editing strategies. The optimization of the metabolic pathway includes: converting the microbial metabolic network into a graph model, where nodes represent metabolites, enzymes, or metabolic fluxes, and edges represent metabolic reaction relationships; Use graph neural network to analyze the node and edge weights in the metabolic network, identify key enzymes and key metabolic pathways; design gene editing strategies based on the analysis results of key pathways and enzymes to optimize the metabolic performance of the strain.

2. The intelligent optimization method for biological fermentation based on multi-objective optimization according to claim 1, characterized in that: The parameters collected in step 1) include: Raw material parameters: water-soluble fiber content, amino acid composition, mineral content, salt additives; Process parameters: stirring method, temperature, pH change, dissolved oxygen concentration, reactor pressure, carbon dioxide concentration; Regulation parameters: sugar source addition amount, nitrogen source adjustment, gas flow rate control, pH regulator dosage; Result parameters: product concentration, metabolite content, by-product accumulation, protein purity.

3. A multi-objective optimization-based intelligent optimization method for biological fermentation according to claim 1, characterized in that: In step 2), data cleaning includes outlier detection and normalization. Specifically: use the quartile method to detect outliers; process the detected outliers by mean correction or replacement; normalize all parameters to adjust the numerical range to between 0 and 1.

4. A multi-objective optimization-based intelligent optimization method for biological fermentation according to claim 1, characterized in that: In step 2), a generative adversarial network (GAN) is used for data augmentation, including: the generator takes fermentation parameters other than the target variable as input; the discriminator feeds back and trains the generator by comparing the statistical characteristics of the generated data and the real data; use the trained generative adversarial network to generate a new augmented data set.

5. A multi-objective optimization-based intelligent optimization method for biological fermentation according to claim 1, characterized in that: In step 4), the multi-objective optimization algorithm uses the differential evolution algorithm, and the optimization objectives include: maximizing fermentation yield; maximizing fermentation volume; maximizing chemical efficiency; minimizing production cost; the weights of the objective functions are dynamically adjusted according to actual production requirements.

6. The intelligent optimization method for biological fermentation based on multi-objective optimization according to claim 5, characterized in that: Use the penalty function method to improve the objective function and constrain the optimization process. The constraint conditions include but are not limited to: fermentation temperature range; pH value range; dissolved oxygen concentration range; reactor pressure range.

7. A bio-fermentation intelligent optimization method based on multi-objective optimization according to claim 1, characterized in that: The Pareto front analysis in step 4) includes: comparing the superiority and inferiority relationships of each optimized solution on different objectives, screening out the non-dominated solutions; forming the Pareto front through the non-dominated solutions; selecting the optimal solution in the Pareto front according to actual needs, including maximizing yield and chemical efficiency while avoiding a significant increase in cost.

8. A multi-objective optimization-based intelligent optimization method for biological fermentation according to claim 1, characterized in that: The optimization algorithms described in step 4) include, but are not limited to: gradient-based optimization algorithms, including gradient descent, Adam, RMSProp; population-based optimization algorithms, including genetic algorithms, particle swarm optimization; heuristic-based optimization algorithms, including tabu search, local search; reinforcement learning algorithms or Bayesian optimization algorithms.

Images