Fermentation process prediction method, system and terminal based on generative algorithm

Abstract

The present invention provides a fermentation process prediction method, system and terminal based on a generative algorithm. First, various fermentation parameters in the entire life cycle of biological fermentation are comprehensively collected and finely preprocessed to construct a sample data set. Subsequently, for the constructed sample data set, the domain features closely related to the fermentation process are deeply explored, and a model construction data set is established based on this. On this basis, one or more generative algorithms are used to construct a phased generative model based on the model construction data set. Finally, with the help of the constructed phased generative model, the fermentation process is accurately predicted according to the input fermentation parameters, thereby obtaining reliable prediction results. On the basis of ensuring that the effectiveness of the data set division and the stability of the model training are not affected, the present invention effectively incorporates domain features into the modeling process, significantly improving the prediction accuracy and generalization ability of the model.

Description

Technical Field

[0001] The present invention relates to the technical field of biological fermentation process prediction, and in particular to a fermentation process prediction method, system and terminal based on a generative algorithm. Background Art

[0002] Traditional fermentation process modeling methods mainly rely on statistical methods or machine learning models to analyze and model data. However, these methods often ignore the domain characteristics that play a key role in the fermentation process, such as differences in microbial species, the exquisite formulation of culture media, and the unique design and structure of reactors. Examining existing technical paths, most focus on modeling based on the data itself, but lack the efficient integration and application of domain expertise. This deficiency directly leads to the lack of predictive accuracy of the constructed models. When faced with complex and changing actual scenarios, the generalization ability is also poor, making it difficult to adapt to diverse application needs.

[0003] With the rapid development of big data and machine learning technologies, combining domain and data features for modeling has become key to improving the accuracy of fermentation process predictions and optimizing results. However, efficiently integrating domain knowledge into models without sacrificing the effectiveness of dataset partitioning and the stability of model training remains a technical challenge in this field. Summary of the Invention

[0004] In view of the shortcomings of the prior art described above, the purpose of the present invention is to provide a fermentation process prediction method, system and terminal based on a generative algorithm, which is used to solve technical problems such as the prior art neglecting the integration of domain features in the fermentation process, resulting in insufficient model prediction accuracy and generalization ability.

[0005] To achieve the above-mentioned purpose and other related purposes, the present invention provides a fermentation process prediction method based on a generative algorithm, the method comprising: performing fermentation parameter collection and preprocessing operations to construct a sample data set, which comprises: collecting fermentation parameters throughout the entire life cycle of biological fermentation, and then preprocessing these fermentation parameters to construct a sample data set; wherein these fermentation parameters can be subdivided into pre-fermentation parameters, fermentation process parameters and fermentation result parameters; performing feature extraction of the sample data set and model data set establishment operations, which comprises: extracting domain features related to the fermentation process from the constructed sample data set, and then establishing a model construction data set; wherein each domain feature includes fermentation static features, fermentation dynamic features and fermentation stage features; utilizing one or more generative algorithms to construct a stage-by-stage generation model according to the model construction data set; based on the constructed stage-by-stage generation model, predicting the subsequent fermentation process according to the input fermentation parameters, and obtaining corresponding prediction results.

[0006] In one embodiment of the present invention, the extracting of domain features related to the fermentation process from the constructed sample data set and then establishing a model construction data set includes: extracting fermentation static features, fermentation dynamic features and fermentation stage features respectively according to the pre-fermentation parameters, fermentation process parameters and fermentation result parameters in the sample data set; performing corresponding feature processing operations on the extracted fermentation static features, fermentation dynamic features and fermentation stage features with the help of the fermentation domain feature processing mechanism; and fusing the features after the feature processing operations to obtain a model construction data set.

[0007] In one embodiment of the present invention, the fermentation field feature processing mechanism includes: for non-one-dimensional features, a specific encoding method is used to convert them into low-dimensional embedding vectors; for unquantized features, domain knowledge is used to perform quantization operations; for time series data features, special time series feature processing and encoding operations are performed.

[0008] In one embodiment of the present invention, the features after the feature processing operation are fused to obtain the model construction data set, including: using feature engineering to perform feature dimensionality upgrading operations on the processed fermentation static features, fermentation dynamic features, and fermentation stage features to enhance the correlation between features in various fields; wherein, the feature dimensionality upgrading operation includes: for continuous variables, using polynomial expansion and logarithmic transformation to upgrade the dimension; for discrete variables, using unique hot encoding and combined features to upgrade the dimension; after completing the feature dimensionality upgrading operation, the upgraded features are fused to obtain a unified feature representation, and finally the model construction data set is obtained.

[0009] In one embodiment of the present invention, the use of one or more generative algorithms to construct a phased generative model based on the model construction data set includes: dividing the model construction data set according to different stages of the fermentation process into data sets for the fermentation startup stage, the main fermentation stage, and the fermentation end stage; wherein the data set for each stage is divided into a training set, a test set, and a validation set; based on the data sets of each stage, using one or more generative algorithms to construct generative models for the fermentation startup stage, the main fermentation stage, and the fermentation end stage.

[0010] In one embodiment of the present invention, the generative model of each stage is composed of a plurality of sub-stage model structures using the same or different generative algorithms connected in sequence; for the first sub-stage model structure, it is used to take the features within a time period as input and output the prediction results of the next time period after the time period; for the non-first sub-stage model structure, it is used to take the features input by the connected previous sub-stage model structure and the prediction results output by the previous sub-stage model structure as input, and output the prediction results of the next time period after the time period predicted by the previous sub-stage model.

[0011] To achieve the above-mentioned purpose and other related purposes, the present invention provides a fermentation process prediction system based on a generative algorithm, the system comprising: a parameter collection module, for performing fermentation parameter collection and preprocessing operations to construct a sample data set, which comprises: collecting fermentation parameters throughout the entire life cycle of biological fermentation, and then preprocessing these fermentation parameters to construct a sample data set; wherein these fermentation parameters can be subdivided into pre-fermentation parameters, fermentation process parameters and fermentation result parameters; a domain feature extraction module, connected to the parameter collection module, for performing feature extraction of the sample data set and model data set establishment operations, which comprises: extracting domain features related to the fermentation process from the constructed sample data set, and then establishing a model construction data set; wherein each domain feature includes fermentation static features, fermentation dynamic features and fermentation stage features; a model construction module, connected to the domain feature extraction module, for using one or more generative algorithms to construct a stage-by-stage generation model based on the model construction data set; a fermentation process prediction module, connected to the model construction module, for predicting the subsequent fermentation process based on the input fermentation parameters based on the constructed stage-by-stage generation model, and obtaining corresponding prediction results.

[0012] To achieve the above-mentioned objectives and other related objectives, the present invention provides an electronic terminal comprising: one or more memories and one or more processors; the one or more memories are used to store computer programs; the one or more processors are connected to the memories and are used to run the computer programs to execute the fermentation process prediction method based on the generative algorithm.

[0013] As described above, the present invention is a fermentation process prediction method, system and terminal based on a generative algorithm, which has the following beneficial effects: the present invention first comprehensively collects various fermentation parameters throughout the life cycle of biological fermentation, and performs fine preprocessing on them, thereby constructing a sample data set. Subsequently, for the constructed sample data set, the domain features closely related to the fermentation process are deeply explored, and a model construction data set is established based on this. On this basis, one or more generative algorithms are used to construct a phased generative model based on the model construction data set. Finally, with the help of the constructed phased generative model, the fermentation process is accurately predicted according to the input fermentation parameters, and reliable prediction results are obtained. On the basis of ensuring that the effectiveness of the data set division and the stability of the model training are not affected, the present invention effectively incorporates domain features into the modeling process, significantly improving the prediction accuracy and generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 Shown is a schematic flow chart of a fermentation process prediction method based on a generative algorithm in one embodiment of the present invention.

[0015] Figure 2 Shown is a schematic diagram of domain feature extraction and processing in one embodiment of the present invention.

[0016] Figure 3 Shown is a schematic structural diagram of a phased generation model in one embodiment of the present invention.

[0017] Figure 4 Shown is a schematic flow chart of a fermentation process prediction method based on a generative algorithm in one embodiment of the present invention.

[0018] Figure 5 Shown is a schematic flow chart of a fermentation process prediction method based on a generative algorithm in one embodiment of the present invention.

[0019] Figure 6 Shown is a structural schematic diagram of a fermentation process prediction system based on a generative algorithm in one embodiment of the present invention.

[0020] Figure 7 Shown is a schematic structural diagram of an electronic terminal in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The following describes the embodiments of the present invention through specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. The details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other unless they conflict.

[0022] It should be noted that in the following description, reference is made to the accompanying drawings, which describe several embodiments of the present invention. It should be understood that other embodiments may also be used, and that mechanical, structural, electrical, and operational changes may be made without departing from the spirit and scope of the present invention. The following detailed description should not be considered restrictive, and the scope of the embodiments of the present invention is limited only by the claims of the published patents. The terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention. Spatially related terms, such as "upper", "lower", "left", "right", "below", "below", "lower", "above", "upper", etc., may be used in the text to facilitate the description of the relationship between one element or feature shown in the figure and another element or feature.

[0023] Throughout this specification, when a part is said to be "connected" to another part, this includes not only "direct connection" but also "indirect connection" with other elements interposed therebetween. Furthermore, when a part is said to "include" a certain component, unless otherwise stated, this does not exclude the inclusion of such other components but rather implies that the part may include such other components.

[0024] The terms "first," "second," and "third" are used to describe various parts, components, regions, layers, and / or segments, but are not intended to be limiting. These terms are used solely to distinguish one part, component, region, layer, or segment from another. Therefore, a reference to a first part, component, region, layer, or segment below may also refer to a second part, component, region, layer, or segment without departing from the scope of the present invention.

[0025] Furthermore, as used herein, the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context indicates otherwise. It should be further understood that the terms "comprise", "include" indicate the presence of the described features, operations, elements, components, items, kinds, and / or groups, but do not exclude the presence, occurrence, or addition of one or more other features, operations, elements, components, items, kinds, and / or groups. The terms "or" and "and / or" used herein are interpreted as inclusive, or mean any one or any combination. Thus, "A, B, or C" or "A, B, and / or C" means "any of the following: A; B; C; A and B; A and C; B and C; A, B, and C". Exceptions to this definition occur only when the combination of elements, functions, or operations is inherently mutually exclusive in some way.

[0026] The present invention provides a fermentation process prediction method based on a generative algorithm. First, various fermentation parameters in the entire life cycle of biological fermentation are comprehensively collected and finely preprocessed to construct a sample data set. Subsequently, for the constructed sample data set, the domain features closely related to the fermentation process are deeply excavated, and a model construction data set is established accordingly. On this basis, one or more generative algorithms are used to construct a phased generative model based on the model construction data set. Finally, with the help of the constructed phased generative model, the fermentation process is accurately predicted according to the input fermentation parameters, thereby obtaining reliable prediction results. On the basis of ensuring that the effectiveness of the data set division and the stability of the model training are not affected, the present invention effectively incorporates domain features into the modeling process, significantly improving the prediction accuracy and generalization ability of the model.

[0027] The following is a detailed description of the embodiments of the present invention with reference to the accompanying drawings so that those skilled in the art can easily implement the present invention. The present invention can be embodied in many different forms and is not limited to the embodiments described herein.

[0028] like Figure 1 A schematic flow chart showing a fermentation process prediction method based on a generative algorithm in an embodiment of the present invention.

[0029] The method comprises:

[0030] Step S1: performing fermentation parameter collection and preprocessing operations to construct a sample data set.

[0031] In detail, the fermentation parameter collection and preprocessing operations include: collecting fermentation parameters throughout the entire life cycle of biological fermentation, and then preprocessing these fermentation parameters to construct a sample data set; all fermentation parameters throughout the entire life cycle of biological fermentation, including pre-fermentation parameters, fermentation process parameters and fermentation result parameters, which run through the entire biological fermentation period before, during and after fermentation.

[0032] In one embodiment, the pre-fermentation parameters mainly focus on the preparation and initial conditions before fermentation. These parameters include:

[0033] 1. Raw material characteristic parameters, including:

[0034] Composition analysis includes the content of carbohydrates (such as glucose, sucrose, and starch), protein, fat, minerals, and vitamins in the raw materials. For example, before fermentation, the starch content in the malt needs to be determined, as this affects the sugars produced during subsequent fermentation and, in turn, the alcohol yield.

[0035] Particle size: The size of raw material particles affects the contact area between microorganisms and substrates. For example, in solid-state fermentation to produce enzyme preparations, smaller raw material particles can provide a larger surface area, which is conducive to microbial attachment and enzymatic hydrolysis. However, too small a particle size may lead to poor aeration.

[0036] Moisture content: The right moisture content in raw materials is crucial for the growth and metabolism of microorganisms. Before wine fermentation, the moisture content of grapes affects the sugar concentration and the growth environment of microorganisms. Generally, the moisture content of fresh grapes is around 70%-80%.

[0037] 2. Microbiological parameters, including:

[0038] Different microbial strains determine the fermentation products and pathways. For example, Saccharomyces cerevisiae is used for alcoholic fermentation, while Lactobacillus is used for lactic acid fermentation. Before fermentation, ensure that the selected strain is a pure strain that meets the production objectives to prevent contamination by other bacteria that may affect the fermentation results.

[0039] The initial concentration of microorganisms introduced into the fermentation system has a significant impact on the fermentation start-up speed and fermentation cycle. If the initial concentration is too low, the fermentation lag phase may be prolonged; while too high a concentration may result in intense competition for nutrients from the microorganisms in the early stages of fermentation.

[0040] 3. Environmental parameters, including:

[0041] Temperature: Before fermentation, the appropriate ambient temperature needs to be set, which affects the growth of microorganisms and the fermentation reaction rate. For example, before yogurt fermentation, milk is usually preheated to around 40-45°C, which is conducive to the growth and fermentation of lactic acid bacteria.

[0042] pH value: A suitable pH value can provide a suitable growth environment for microorganisms and affect enzyme activity. Before citric acid fermentation, the pH of the fermentation medium is generally adjusted to around 3-7, which is the optimal pH range for Aspergillus niger growth and citric acid production.

[0043] The fermentation process parameters are key indicators reflecting the growth and metabolic status of microorganisms during the fermentation process. These parameters include:

[0044] 1. Physical parameters, including:

[0045] Temperature. During the fermentation process, temperature must be strictly controlled. Different microorganisms have their optimal growth and fermentation temperatures. For example, during beer fermentation, the temperature is generally controlled at 10-12°C during the main fermentation phase of yeast, and the temperature is lowered to around 0-4°C during the secondary fermentation phase. This helps control the fermentation rate and the formation of beer flavor.

[0046] Pressure control is important in some anaerobic fermentations or fermentations involving gas production. For example, in biogas fermentation, appropriate pressure helps maintain the anaerobic environment of the fermentation system and promotes the growth and gas production of methanogens.

[0047] Agitation speed (if applicable). For liquid fermentation, agitation allows microorganisms to fully contact nutrients and increases dissolved oxygen (if it is aerobic fermentation). For example, in the fermentation process of antibiotics, agitation is used to evenly mix the mycelium with the nutrients in the culture medium and also facilitates the transfer of oxygen. The agitation speed is generally set at around 100-300 rpm, depending on the size of the fermenter and the needs of the microorganisms.

[0048] 2. Chemical parameters, including:

[0049] pH changes. Microorganisms produce acidic or alkaline metabolites during fermentation, which can cause pH changes. For example, during lactic acid fermentation, lactic acid bacteria produce lactic acid, which lowers the pH of the fermentation broth. Buffers or acid-base regulators are needed to maintain the pH within an appropriate range to ensure normal microbial growth and fermentation.

[0050] Dissolved oxygen (if fermenting aerobic). For aerobic microbial fermentation, the dissolved oxygen concentration directly affects the respiration and metabolic rate of the microorganisms. For example, during amino acid fermentation, the appropriate dissolved oxygen level is maintained by controlling the ventilation volume and agitation speed. Generally, the dissolved oxygen concentration should be kept at around 20% - 30% saturation.

[0051] Substrate concentration changes. As fermentation progresses, substrates (such as sugars and proteins) are utilized by microorganisms, and their concentration gradually decreases. During alcohol fermentation, monitoring changes in glucose concentration can provide insights into the progress of fermentation. When glucose concentration drops to a certain level, fermentation may gradually cease or enter the post-fermentation stage.

[0052] Metabolite concentration changes: During the fermentation process, the concentration of metabolites (such as alcohol, organic acids, and antibiotics) gradually increases. For example, during penicillin fermentation, the penicillin content needs to be regularly monitored and the fermentation conditions adjusted based on the concentration changes to optimize the process.

[0053] 3. Microbiological parameters

[0054] The growth status of microorganisms can be determined by observing changes in their morphology, size, and number under a microscope. For example, during yeast fermentation, the presence of budding and cell activity can be observed.

[0055] Microbial metabolic activity is measured by testing indicators such as respiration intensity and enzyme activity. For example, during fermentation, the activity of enzymes such as amylase and protease can be tested. These enzymes are key enzymes for microbial substrate decomposition, and changes in enzyme activity reflect the level of microbial metabolic activity.

[0056] Fermentation result parameters mainly focus on the quality and yield of the product after fermentation, as well as the growth and metabolism of microorganisms during the fermentation process. These parameters include:

[0057] 1. Product quality parameters, including:

[0058] Purity is a crucial indicator for fermentation products (such as enzymes, organic acids, and antibiotics). For example, in the production of industrial enzyme preparations, high-purity enzymes can improve their effectiveness and reduce the adverse effects of impurities. Through various separation and purification technologies, enzyme purity can be achieved to over 90%.

[0059] Activity. For biologically active fermentation products (such as enzymes and bioactive peptides), the level of activity determines their quality. For example, after protease fermentation, the activity of the protease needs to be tested. This is usually measured by the amount of product produced per minute by decomposing the substrate. Higher activity indicates better product quality.

[0060] Physical and chemical properties, including the product's color, odor, solubility, density, etc. For example, after the citric acid fermentation is completed, the citric acid product should be a colorless or white crystalline powder with no odor and good water solubility.

[0061] 2. Production parameters, including:

[0062] Product yield refers to the ratio of the actual amount of target product obtained to the theoretical amount of product calculated based on substrate consumption. For example, in alcohol fermentation, the theoretical alcohol yield is calculated based on the amount of glucose input. The ratio of the actual amount of alcohol obtained to the theoretical yield is the alcohol yield, which reflects the efficiency of the fermentation process.

[0063] Volumetric productivity. For liquid fermentation, volumetric productivity is the amount of product produced per unit volume of fermentation broth per unit time. For example, in biodiesel production in a fermenter, volumetric productivity is measured by calculating the amount of biodiesel produced per liter of fermentation broth per day.

[0064] 3. Microbiological parameters, including:

[0065] The number of surviving microorganisms is important after fermentation. For example, in active dry yeast production, the number of surviving yeast cells must be tested after fermentation. Typically, a certain number of yeast cells per gram of product is required to ensure the product's fermentation activity during subsequent use.

[0066] Changes in the physiological state of microorganisms: Observe the physiological state of microorganisms after fermentation, such as whether they enter a dormant state or whether there is cell damage. For example, after high-temperature fermentation, some microorganisms may experience physiological changes such as cell membrane damage due to high temperature stress, which will affect the storage and subsequent use of the microorganisms.

[0067] In one embodiment, the method of preprocessing the fermentation parameters includes: performing data cleaning and normalization processing on all fermentation parameters respectively to obtain a sample data set that meets the data processing requirements.

[0068] Specifically, the data cleaning process covers the handling of missing values ​​and outliers. Missing values ​​refer to situations where some observations in the data set are missing or not recorded, which may have a negative impact on the accuracy and reliability of data analysis in fermentation parameter data. To this end, we prefer to use the single linear fitting filling method. This method performs linear fitting based on the data before and after the missing value, thereby making a reasonable estimate of the missing value. Outliers are values ​​that are significantly different from most data points, and their causes may include measurement errors, data entry errors, etc. These outliers will interfere with the data analysis results. Therefore, we prefer to use the quartile method to record outliers so that they can be replaced in subsequent preprocessing.

[0069] After data cleaning is completed, the normalization stage begins. The purpose of normalization is to scale the data range to between 0 and 1, which not only helps speed up the convergence of the model but also improves the overall performance of the model. First, each feature of the dataset is normalized. Then, a random floating amount that conforms to the Gaussian distribution (normal distribution) is generated and added to the mean of each feature to simulate the random fluctuations of the data. Subsequently, the new data with the added Gaussian floating amount is normalized again to ensure that the data range remains between 0 and 1. Finally, through the shearing operation, all normalized values ​​are forced to be within the range of 0 to 1 to ensure data consistency and obtain a sample dataset that meets the requirements.

[0070] Step S2: Execute feature extraction of the sample data set and establishment of the model data set.

[0071] Specifically, the sample dataset feature extraction and model dataset construction operations involve extracting domain features relevant to the fermentation process from the constructed sample dataset, thereby establishing a model-building dataset. These fermentation parameters can be further categorized into static fermentation features, dynamic fermentation features, and stage-specific fermentation features. Specifically, static fermentation features primarily originate from the preparatory phase of fermentation. These static features set the tone for the fermentation process and have a profound impact on the microbial growth environment and subsequent metabolic activity. By carefully analyzing these features, we can more comprehensively understand the initial state of the fermentation system, providing a solid foundation for model construction. Dynamic fermentation features focus on real-time changes during the fermentation process. By continuously monitoring and recording these dynamic parameters, we can capture the temporal evolution of the fermentation system and reveal the underlying patterns of microbial metabolic activity. These dynamic features are crucial for accurately predicting fermentation progress, optimizing fermentation conditions, and improving product purity. Stage-specific fermentation features often exhibit distinct phase-specific characteristics, with microbial metabolic activity, product production rates, and resource utilization efficiency varying at each stage. By identifying and distinguishing these stage-specific features, we can more accurately identify key nodes in the fermentation process, providing a strong basis for developing targeted regulatory strategies.

[0072] In one embodiment, if Figure 2 , step S2 includes:

[0073] Based on the pre-fermentation parameters, fermentation process parameters, and fermentation result parameters in the sample dataset, static fermentation features, dynamic fermentation features, and stage-specific fermentation features are extracted, respectively. Specifically, static fermentation features are extracted based on the pre-fermentation parameters in the sample dataset. These pre-fermentation parameters often include relatively fixed data such as the composition of the fermentation raw materials, input amounts, and initial environmental conditions (basic settings such as temperature and humidity). Static features that reflect the characteristics of the initial fermentation state are extracted from these parameters. Dynamic features that reflect the changing state of the fermentation process are extracted based on the fermentation process parameters in the sample dataset. Fermentation process parameters are data that change in real time during the actual fermentation process, such as temperature curves, pH fluctuations, and microbial growth rates. Fermentation result parameters include various indicators of the fermentation end product (such as yield, quality indicators, and component ratios), as well as state parameters at the end of the entire fermentation process. By analyzing the fermentation result parameters, stage-specific features that characterize the characteristics of different fermentation stages are summarized.

[0074] After extracting various features, we perform corresponding feature processing operations on each fermentation static, dynamic, and phased feature based on the fermentation domain's feature processing mechanism to optimize feature representation or improve feature quality. This step aims to optimize feature representation or improve feature quality, laying a solid foundation for subsequent feature fusion and model building.

[0075] We then fuse the processed static, dynamic, and stage-specific fermentation characteristics to generate a model-building dataset that effectively supports subsequent model development. This dataset integrates information from multiple aspects of the fermentation process, fully reflecting the complexity and dynamics of the fermentation system. By utilizing this dataset for model training and validation, we can build more accurate and reliable fermentation process prediction and optimization models, providing strong support for the intelligent transformation of the biomanufacturing industry.

[0076] In one embodiment, the fermentation static characteristics, fermentation dynamic characteristics, and fermentation stage characteristics can be extracted in different ways.

[0077] Specifically, given the temporal variations in fermentation process parameters, a layer of LSTM (Long Short-Term Memory) can be used to extract temporal features for these dynamically changing parameters. LSTM effectively captures long-term dependencies in time series. It can effectively learn the dynamic patterns and regularities of this evolving fermentation data, thereby extracting valuable fermentation dynamic features.

[0078] To extract fermentation stage features, a Transformer layer is applied within each stage to extract these features from the fermentation result parameters. The Transformer's powerful self-attention mechanism enables global modeling of data within each stage, capturing the relationships between data elements and effectively extracting stage-specific features that reflect the unique characteristics of each fermentation stage.

[0079] As for the static characteristics of fermentation, since the pre-fermentation parameters are determined at the initial stage of fermentation and are relatively stable during the subsequent fermentation process, these pre-fermentation parameters can be directly used as the static characteristics of fermentation.

[0080] In one embodiment, in the fermentation field, in order to more effectively utilize data for model building and analysis, it is necessary to perform targeted processing on various features. This processing mechanism mainly involves performing one or more of the following operations on each feature:

[0081] Processing Non-One-Dimensional Features: In the fermentation field, many complex non-one-dimensional features exist. For example, microbial metabolic pathways involve the interconversion of multiple substances and a series of biochemical reactions, forming a multidimensional, complex system. Another example is reactor design, which includes multi-dimensional parameter descriptions such as reactor shape, size, and internal structure. These non-one-dimensional features are converted into low-dimensional embedding vectors using specific encoding methods, such as embedding layers and word embeddings. Because non-one-dimensional features may have high dimensionality, which increases computational complexity and model burden, encoding them into low-dimensional embedding vectors can reduce dimensionality while preserving important feature information, facilitating subsequent processing and computation. In the fermentation field, we can view the various elements of microbial metabolic pathways or reactor designs as "words." Through appropriate training, these non-one-dimensional features can be converted into low-dimensional embedding vectors. This not only reduces data dimensionality and computational complexity, but also better extracts key information from the features, facilitating subsequent data analysis and model building.

[0082] Non-quantitative feature processing: In the fermentation field, some features may only be qualitative descriptions and lack quantitative numerical representations. These features can be converted into quantifiable forms through domain knowledge so that they can better participate in subsequent calculations and model training.

[0083] For example, we can use metabolic flux analysis (MFA) or growth models to quantitatively describe this. Metabolic flux analysis, through quantitative analysis of the flow of substances during microbial metabolism, can reveal the intensity and direction of microbial metabolic activity under different conditions, thereby providing a quantitative basis for the dynamic characteristics of microbial growth. Growth models can quantitatively simulate microbial growth rates and biomass accumulation based on microbial growth patterns and environmental factors (such as nutrient concentration and temperature). These domain-knowledge-based methods convert previously non-quantifiable features into specific numerical values, making the data more standardized and easier to process, which helps improve the accuracy of data analysis and the reliability of the model.

[0084] Time Series Data Feature Processing: During the fermentation process, these time series data features contain rich information about the fermentation process. These features undergo specialized time series feature processing and encoding. Because time series data has temporal order and inherent correlations, specific processing and encoding methods are required to extract and utilize the temporal information contained within, enabling the model to better understand and learn the characteristics of this data. First, in feature processing, we may perform data smoothing to remove noise and enhance trends. We may also calculate statistical features such as mean, variance, and slope to reflect changes in the data over time. For encoding, recurrent neural networks (RNNs) and their variants, such as long short-term memory (LSTM) and gated recurrent units (GRU), can be used. These models effectively capture the temporal information and long-term dependencies in time series data. By encoding the time series data into the model at time steps, key features are extracted, providing strong support for analysis and prediction of the fermentation process.

[0085] In one embodiment, the step of fusing the processed fermentation static features, fermentation dynamic features, and fermentation stage features to obtain a model construction dataset includes:

[0086] First, feature engineering was used to increase the dimensionality of the three processed domain features. This process aims to enhance the correlation between features and uncover more valuable information hidden in the data, enabling the subsequent model to better capture the inherent patterns and relationships within the fermentation process.

[0087] Specific dimensionality-raising operations include:

[0088] For continuous variables, polynomial expansion and logarithmic transformation are used to achieve feature dimensionality enhancement. For example, in the fermentation process, continuous variables such as fermenter temperature and substrate concentration can be expanded through polynomial expansion to generate new variables with higher powers, such as quadratic and cubic terms, based on the original variables. This approach has the advantage of capturing potential nonlinear relationships between variables, as many factors in actual fermentation processes affect fermentation results in non-linear relationships. Logarithmic transformation is equally important. For example, for some concentration data, logarithmic transformation can adjust the range of values ​​to make the distribution more consistent with the model's assumptions and help alleviate potential heteroscedasticity in the data. Polynomial expansion can generate higher-order features through polynomial combination of original features, exploring nonlinear relationships between variables; while logarithmic transformation can stretch or compress the data, changing its distribution and making it more consistent with the model's assumptions.

[0089] For discrete variables, one-hot encoding and combined features are used to increase dimensionality. One-hot encoding is a method that converts categorical variables into a numerical form that can be understood by machine learning models. The basic idea is to create a new binary column (i.e., feature) for each category. If the sample belongs to that category, the value of the corresponding column is 1, otherwise it is 0. Combined features are used to explore the correlation between different discrete variables. For example, considering the two discrete variables of microbial species and fermentation stage, by combining them, new features such as "a certain microorganism in a specific fermentation stage" can be generated, thereby more comprehensively reflecting the interaction between different discrete factors in the fermentation process. One-hot encoding converts each category of the discrete variable into a binary vector, avoiding the bias caused by the size relationship between categories; combined features can explore the combination relationship between different discrete variables and increase the expressive power of the features.

[0090] After upgrading the features of various variables, the upgraded static, dynamic, and stage-specific fermentation features are fused. Specifically, all features are summed across the dimensions to create a unified feature representation. This constitutes the model construction dataset, providing comprehensive data support for subsequent model construction. This dataset integrates comprehensive information about the fermentation process, from its initial state and dynamic changes to the characteristics of each stage, providing a rich data foundation for subsequent model training and analysis.

[0091] In order to scientifically and effectively evaluate the performance and generalization ability of the model, the processed dataset needs to be divided into training set, validation set and test set based on the data volume. For example, the training set: validation set: test set can be divided in a ratio of 8:1:1.

[0092] Step S3: Utilize one or more generative algorithms to construct a phased generative model based on the model construction dataset.

[0093] In one embodiment, constructing a phased generative model based on the model building dataset based on one or more generative algorithms includes:

[0094] First, the constructed model-building dataset was divided according to the different stages of the fermentation process: the startup phase, the main fermentation phase, and the final phase. This division is due to the unique characteristics, dynamics, and influencing factors of each stage of the fermentation process. By processing the data in stages, we can more accurately capture the internal mechanisms of each stage and build a model that is more accurate to the actual fermentation situation.

[0095] The data sets for each stage include training set, test set, and validation set. This is done to enable targeted model training, testing, and validation of the fermentation process at different stages, ensuring the performance and effectiveness of the model at different stages.

[0096] Subsequently, based on one or more generative algorithms, generative models for the startup, main fermentation, and final stages are constructed for each dataset. The primary function of these generative models is to predict the characteristics of each fermentation stage, thereby achieving a refined prediction of the entire fermentation process. This phased model construction approach better captures the unique characteristics and changing patterns of each fermentation stage, improving the accuracy and reliability of predictions.

[0097] It should be noted that the types of generative algorithms used include naive Bayes model and its derivative models, Gaussian mixture model and its derivative models, hidden Markov model and its derivative models, generative adversarial network and its derivative models, variational autoencoder and its derivative models, deep Boltzmann machine and its derivative models, Transformer model and its derivative models, Mamba model and its derivative models, LSTM model and its derivative models, NBEATS model and its derivative models, etc.

[0098] In one embodiment, if Figure 3 The generative model for each stage is not a single structure, but rather a series of connected sub-stage model structures, each employing the same or different generative algorithms. This hierarchical, chain-like structure is designed to leverage the strengths of different algorithms and information transfer between sub-stage models, enabling more accurate predictions of the fermentation process over multiple consecutive time periods within the corresponding stage. Furthermore, the number of stage model structures within each stage's generative model can be the same or different.

[0099] The first sub-stage model structure is Figure 3The sub-stage 1 model structure uses features from a specific time period as input, and then, based on its internal algorithm and parameter settings, outputs predictions for the next time period following that time period. For example, during the fermentation startup phase, if the first sub-stage model uses various fermentation characteristics collected within the initial hour (such as temperature, pH, and microbial concentration) as input, it can output predictions for the relevant fermentation characteristics for the hour immediately following that hour. This provides a foundation for further predictions by the subsequent sub-stage model structures. During training, features from a specific time period in the dataset are used as input as the initial time period, and predictions for the next time period following the initial time period are output. Simultaneously, actual data corresponding to the predicted time period is extracted from the dataset. By carefully comparing the model structure's predictions with actual data, the model structure's parameters are adjusted, and the first sub-stage model structure is constructed. This approach enables a step-by-step prediction of the fermentation process over a time series, providing initial prediction information for subsequent fermentation processes.

[0100] The non-first sub-stage model structure, that is Figure 3 From the sub-stage 2 model structure to the sub-stage N model structure, its input not only includes the features of the input of the connected previous sub-stage model structure, but also covers the prediction results output by the previous sub-stage model structure. By integrating this information, it can output the prediction results for the next time period after the time period predicted by the previous sub-stage model structure. In other words, the subsequent sub-stage models will continuously and progressively predict the subsequent time periods based on the prediction results of the previous sub-stage and combined with the original input features. This design utilizes the information of the previous sub-stage to form a coherent time series prediction chain, making the prediction of the entire fermentation process more coherent and accurate, and better reflecting the dynamic changes and mutual influences between the various stages in the fermentation process.

[0101] When training a non-first sub-stage model structure, the input features and output prediction results of the previous sub-stage model structure are used as input, and the prediction results for the next time period after the previous time period are output. At the same time, the actual data corresponding to the predicted time period is extracted from the dataset. By carefully comparing the model structure's prediction results with the actual data, the model structure parameters are adjusted and the sub-stage model structure is constructed.

[0102] Step S4: Based on the constructed stage-by-stage generation model, the subsequent fermentation process is predicted according to the input fermentation parameters, and the corresponding prediction results are obtained.

[0103] In one embodiment, step S4 includes:

[0104] Preprocess the input fermentation parameters to be predicted. This preprocessing process is similar to the preprocessing of the model and dataset. It may include data cleaning (such as handling missing values ​​and outliers) and normalization. The purpose is to ensure that the input data is reliable and formatted in a standardized manner to meet the requirements of subsequent analysis and model operations.

[0105] After preprocessing, domain features related to the fermentation process are extracted. Static, dynamic, and stage-specific fermentation features are extracted from the input fermentation parameters. After extracting domain features, these features are processed accordingly based on the fermentation domain feature processing mechanism.

[0106] The processed domain features are input into the constructed phased generation model to predict the fermentation process and output the corresponding prediction results;

[0107] It is important to note that the range of predictions available varies depending on the type of fermentation parameter being input. When pre-fermentation parameters are input, since they reflect the initial state of fermentation, the phased generation model allows predictions of characteristic parameters for subsequent time periods during the startup phase. For example, the changes in parameters such as temperature and microbial count within the fermenter over the next few hours can be predicted. Furthermore, characteristic parameters for the main fermentation phase and the final phase can be predicted, allowing for accurate assessment of the parameters of the entire fermentation process from the very beginning.

[0108] If the input is fermentation process parameters, because these parameters have already reflected some dynamic changes in the fermentation process, the model can be used to predict the characteristic parameters of subsequent time periods in the main fermentation stage. For example, in the main fermentation stage, based on parameters such as the substrate consumption rate in the current period, it is possible to predict at what rate the substrate will be consumed and how the product generation will change in the next period of time. At the same time, the characteristic parameters of the end stage can also be predicted, so that various situations that may occur in the end stage can be known in advance so that corresponding regulatory measures can be taken in time.

[0109] When the input is the fermentation result parameters, the characteristic parameters of the subsequent time period within the end stage are mainly predicted based on the existing fermentation result information. For example, it is judged that after the fermentation is completed, the changes in product quality in subsequent storage, transportation and other links may occur, or it is predicted whether some further post-processing measures are needed to ensure the stability of the product.

[0110] Through such a complete and detailed process, based on different types of fermentation parameter inputs, the characteristic parameters of different stages and time periods of the fermentation process can be predicted with the help of a phased generation model.

[0111] In one embodiment, if Figure 4The method further includes: step S5 executed after step S4: performing fermentation parameter sensitivity analysis on the prediction results output by the constructed phased generation model, and optimizing the corresponding characteristic parameters in the prediction results based on the fermentation parameter sensitivity analysis results to obtain a prediction optimization result.

[0112] Specifically, the prediction results output by the staged generation model are used to carry out fermentation parameter sensitivity analysis, and the sensitivity of each characteristic parameter is ranked, so as to accurately identify the key parameters that have the greatest impact on the fermentation path. This process is one of the important innovations of the present invention, which provides a key basis for the subsequent optimization of the prediction results. Based on the above-mentioned fermentation parameter sensitivity ranking results, the corresponding characteristic parameters in the prediction results are targeted for optimization. For example, for those parameters that are identified as critical and have a significant impact on the fermentation results, the prediction results are made more in line with the actual fermentation needs by adjusting their value ranges, optimizing their control strategies, etc. After this optimization process, the prediction optimization results are finally obtained, which significantly improves the accuracy and practicality of the fermentation process prediction, and provides strong support for the improvement of the fermentation process and the improvement of production efficiency.

[0113] This is another key invention point of the present invention.

[0114] In order to better describe the fermentation process prediction method based on the generative algorithm, it is now explained in conjunction with the following specific examples.

[0115] Example 1: A fermentation process prediction method based on a generative algorithm. Figure 5 Schematic diagram of the fermentation process prediction method based on the generative algorithm in this embodiment.

[0116] The method comprises:

[0117] Step 1: Collect all parameters in the entire life cycle of biological fermentation, including but not limited to raw material parameters, process parameters and result parameters.

[0118] Among them, the raw material parameters represent the various index parameters of the raw materials used in biological fermentation. The raw material parameters used in this example include more than 1,000 items such as carbon content, moisture content, wheat content, pH buffer, etc.

[0119] Process parameters represent various indicators and parameters in the biofermentation process. In this example, we used over 100 parameters, including temperature, pH, dissolved oxygen concentration, agitation rate, ventilation volume, fermentation time, reactor type and volume, exhaust gas emissions, and pressure. Process parameters also include control parameters, which represent indicator parameters that can be controlled during the biofermentation process. In this example, we used over 20 control parameters, including pH control, temperature control, dissolved oxygen control, ventilation control, substrate dosage control, and pressure control.

[0120] The outcome parameters represent the quality, yield, and purity of the final product in the biofermentation process, including but not limited to product concentration, product yield, product yield, metabolites, product purity, cell concentration, byproduct inhibition, pH change, gas production, etc.

[0121] Step 2: Perform data cleaning and normalization on all parameters.

[0122] First, data cleaning is performed on all parameters. Data cleaning methods include, but are not limited to, fitting missing values ​​and correcting or removing outliers from the sample dataset. Missing values ​​are filled using a single linear fit. Outliers are recorded using quartiles to facilitate replacement during subsequent preprocessing.

[0123] Next, the cleaned data is normalized. A random float with a Gaussian distribution is generated and added to the mean. The new fill values ​​are then normalized to the range of 0 to 1, ensuring that the filled values ​​do not exceed 0 or 1. A shear operation is then performed to obtain the sample dataset.

[0124] Step 3: Perform feature extraction based on all processed parameters.

[0125] Based on all the processed parameters, fermentation static characteristics, fermentation dynamic characteristics, fermentation stage characteristics and curve characteristics are extracted respectively;

[0126] For dynamic features, a layer of LSTM is used to extract temporal features.

[0127] For stage-specific features, only one layer of Transformer module is used within each stage to extract stage-specific features.

[0128] Graphs are drawn for all features to extract curve features. The curve features used in this example include: frequency domain, tangent slope, normal slope, tangent vector, normal vector, simple moving average, weighted moving average, exponential moving average, single exponential smoothing, double exponential smoothing, triple exponential smoothing, volatility, amplitude, absolute rate of change, relative rate of change, autocorrelation, sliding correlation, skewness, kurtosis, Hurst index, coefficient of variation, and cumulative distribution function. The extracted curve features are also filled with missing values, processed for outliers, and normalized.

[0129] Step 4: Perform feature dimensionality upgrade on all extracted features and fuse the upgraded features to obtain a unified feature representation and obtain the model construction dataset.

[0130] Use feature engineering to increase the dimensionality of all features, including:

[0131] (1) Perform polynomial expansion and logarithmic transformation on continuous variables;

[0132] (2) Perform one-hot encoding and combined features on discrete variables.

[0133] All the features after dimensionality increase are fused and all the features are added together in the dimension to obtain a unified feature representation to form a data set.

[0134] The processed data set is divided into training set, validation set, and test set, and the division ratio corresponds to 8:1:1 for training set: validation set: test set.

[0135] Step 5: Design a phased generative model and use one or more generative algorithms to predict and generate fermentation parameters.

[0136] Use the dataset prepared in step 4 to design a phased generative model. Use one or more generative algorithms, including but not limited to the Transformer model and its derivatives, the Mamba model and its derivatives, the LSTM model and its derivatives, the NBEATS model and its derivatives, to train, verify, and test the model.

[0137] This example uses the Mamba sequence model as a generative algorithm. The training data set, validation data set, and test data set are divided into three stages: the startup stage, the main fermentation stage, and the end stage. The data of each stage is input into the Mamba sequence model of each stage, and model training, validation, and testing are performed to obtain three generative models.

[0138] Each generative model also includes: sub-stage 1 model, sub-stage 2 model and sub-stage 3 model. The training process of these three models includes:

[0139] The input of the sub-stage 1 model is the data X1 for the time period 0-x, and the output is the prediction result Y1' for the time period x-2x; where x is in hours;

[0140] The input of the sub-stage 2 model is the input X1 of the stage 1 model plus the prediction result Y1 of the stage 1 model, X2 = [X1, Y1'], and the total input length is 2x. The output data Y2' is for the time period 2x-4x.

[0141] The input of the sub-stage 3 model is the input X2 of the stage 2 model plus the prediction result Y2 of the stage 2 model, X3 = [X2, Y2'], and the total input length is 2x. The output data Y3' is for the time period 4x-6x.

[0142] Specifically, the loss function formula is as follows:

[0143] ; (1)

[0144] Where: N represents the total number of sub-stages, L total Represents the total loss, Represents the loss function of the i-th stage, measuring the true value and predicted value The error between them is calculated using cosine similarity. The final model accuracy reaches 99.6%.

[0145] It should be noted that the NBEATS model can also be used as a generative algorithm in this example. The training, validation, and test datasets are first divided into three phases: startup, main fermentation, and finalization. The data from each phase is then fed into the NBEATS model for training, validation, and testing, resulting in three generative models. Specifically, the model training method is the same as that used for the Mamba sequence model, with the loss function switched from cosine similarity to mean square error (MSE). The model accuracy ultimately reached 95.7%-97.5%.

[0146] Step 6: Based on the constructed phased generation model, the fermentation process is predicted according to the input fermentation parameters to obtain the corresponding prediction results.

[0147] Step 7: Using the prediction results from the model generated in step 6, perform a parameter sensitivity analysis to identify the key parameters that have the greatest impact on the fermentation pathway. Based on the results of the parameter sensitivity analysis, optimize the corresponding parameters in the prediction results to obtain the optimal prediction results.

[0148] Similar to the principles of the above embodiments, the present invention provides a fermentation process prediction system based on a generative algorithm.

[0149] The following provides specific embodiments in conjunction with the accompanying drawings:

[0150] Such as Figure 6 A schematic structural diagram of a fermentation process prediction system based on a generative algorithm in an embodiment of the present invention is shown.

[0151] The system comprises:

[0152] Parameter collection module 1 is used to perform fermentation parameter collection and preprocessing operations to construct a sample data set, which includes: collecting fermentation parameters throughout the entire life cycle of the biological fermentation, and then preprocessing these fermentation parameters to construct a sample data set; wherein these fermentation parameters can be further divided into pre-fermentation parameters, fermentation process parameters, and fermentation result parameters;

[0153] The domain feature extraction module 2 is connected to the parameter collection module 1 and is used to perform feature extraction of the sample data set and establish a model data set, which includes: extracting domain features related to the fermentation process from the constructed sample data set, and then establishing a model construction data set; wherein each domain feature includes fermentation static features, fermentation dynamic features, and fermentation stage features;

[0154] A model construction module 3, connected to the domain feature extraction module 2, is used to use one or more generative algorithms to construct a phased generative model based on the model construction dataset;

[0155] The fermentation process prediction module 4 is connected to the model construction module 3 and is used to predict the subsequent fermentation process based on the constructed stage-by-stage generation model and the input fermentation parameters, and obtain corresponding prediction results.

[0156] Since the implementation principle of the fermentation process prediction method based on the generative algorithm has been described in the above embodiments, it will not be repeated here.

[0157] The fermentation process prediction method based on the generative algorithm provided in the embodiment of the present invention can be implemented on the terminal side or the server side. As for the hardware structure of the electronic terminal, please refer to Figure 7 , is an optional hardware structure diagram of an electronic terminal 1000 provided in an embodiment of the present invention. The terminal 1000 can be a mobile phone, a computer device, a tablet device, etc. The terminal 1000 includes: at least one processor 1001, a memory 1002, at least one network interface 10010 and a user interface 1009. The various components in the device are coupled together through a bus system 1005. It can be understood that the bus system 1005 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 1005 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, Figure 7 Various buses are labeled as bus systems.

[0158] The user interface 1009 may include a display, a keyboard, a mouse, keys, buttons, a touch panel or a touch screen.

[0159] It will be appreciated that the memory 1002 may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM) or a programmable read-only memory (PROM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory (SRAM) and synchronous static random access memory (SSRAM). The memory described in the embodiments of the present invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0160] The memory 1002 in the embodiment of the present invention is used to store various categories of data to support the operation of the terminal 1000. Examples of these data include: any executable program for operating on the terminal 1000, such as an operating system 10021 and an application 10022; the operating system 10021 includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks. The application 10022 can include various applications, such as a media player (MediaPlayer), a browser (Browser), etc., for implementing various application services. The fermentation process prediction method based on the generative algorithm provided in the embodiment of the present invention can be included in the application 10022.

[0161] The methods disclosed in the above embodiments of the present invention can be applied to or implemented by processor 1001. Processor 1001 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by hardware integrated logic circuits in processor 1001 or by software instructions. The above processor 1001 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 1001 can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor 1001 may be a microprocessor or any conventional processor. The steps of the accessory optimization method provided in the embodiments of the present invention can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium located in a memory. The processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above method.

[0162] In an exemplary embodiment, the terminal 1000 may be used by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), and complex programmable logic devices (CPLDs) to execute the aforementioned method.

[0163] Those skilled in the art will appreciate that all or part of the steps in the above-described method embodiments can be implemented using hardware associated with a computer program. The aforementioned computer program can be stored in a computer-readable storage medium. When executed, the program performs the steps in the above-described method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0164] In the embodiments provided herein, the computer-readable and writable storage medium may include read-only memory, random access memory, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage device, flash memory, USB flash drive, removable hard drive, or any other medium capable of storing desired program code in the form of instructions or data structures and capable of being accessed by a computer. In addition, any connection may be appropriately referred to as a computer-readable medium. For example, if instructions are sent from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwaves, the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwaves are included in the definition of the medium. However, it should be understood that computer-readable and writable storage medium and data storage medium do not include connections, carrier waves, signals, or other temporary media, but are intended to refer to non-temporary, tangible storage media. Disk and disc, as used in this application, includes compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and Blu-ray disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers.

[0165] In summary, the fermentation process prediction method, system and terminal based on the generative algorithm of the present invention first comprehensively collect various fermentation parameters in the entire life cycle of biological fermentation, and perform fine preprocessing on them, thereby constructing a sample data set. Subsequently, for the constructed sample data set, the domain features closely related to the fermentation process are deeply excavated, and a model construction data set is established based on this. On this basis, one or more generative algorithms are used to construct a phased generation model based on the model construction data set. Finally, with the help of the constructed phased generation model, the fermentation process is accurately predicted according to the input fermentation parameters, and reliable prediction results are obtained. On the basis of ensuring that the effectiveness of the data set division and the stability of the model training are not affected, the present invention effectively incorporates domain features into the modeling process, significantly improving the prediction accuracy and generalization ability of the model. Therefore, the present invention effectively overcomes the various shortcomings in the prior art and has a high industrial utilization value.

[0166] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Anyone skilled in the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, any equivalent modifications or alterations made by one of ordinary skill in the art without departing from the spirit and technical principles disclosed herein are intended to be covered by the claims of the present invention.

Claims

1. A fermentation process prediction method based on a generative algorithm, characterized in that: The method comprises: Performing fermentation parameter collection and preprocessing operations to construct a sample data set, which includes: collecting fermentation parameters throughout the entire life cycle of the biological fermentation, and then preprocessing the fermentation parameters to construct the sample data set; wherein the fermentation parameters are subdivided into pre-fermentation parameters, fermentation process parameters, and fermentation result parameters; Performing feature extraction of the sample data set and model data set establishment operations, which includes: extracting domain features related to the fermentation process from the constructed sample data set, and then establishing a model construction data set; wherein the domain features include fermentation static features, fermentation dynamic features, and fermentation stage features; constructing a staged generative model based on the model-building dataset using one or more generative algorithms; Based on the constructed stage-by-stage generation model, the subsequent fermentation process is predicted according to the input fermentation parameters, and the corresponding prediction results are obtained; The extracting of domain features related to the fermentation process from the constructed sample data set and then establishing the model construction data set includes: extracting fermentation static features, fermentation dynamic features, and fermentation stage features according to pre-fermentation parameters, fermentation process parameters, and fermentation result parameters in the sample data set; performing corresponding feature processing operations on the extracted fermentation static features, fermentation dynamic features, and fermentation stage features by means of a fermentation domain feature processing mechanism; and fusing the features after the feature processing operations to obtain the model construction data set; The fermentation domain feature processing mechanism includes: for non-one-dimensional features, a specific encoding method is used to convert them into low-dimensional embedding vectors; for unquantized features, domain knowledge is used to perform quantization operations; for time series data features, special time series feature processing and encoding operations are performed; The method of using one or more generative algorithms to construct a phased generative model based on the model construction dataset includes: dividing the model construction dataset into datasets for a fermentation startup phase, a main fermentation phase, and a fermentation end phase according to different stages of the fermentation process; wherein the dataset for each stage is divided into a training set, a test set, and a validation set; and using one or more generative algorithms based on the datasets for each stage to construct generative models for the fermentation startup phase, the main fermentation phase, and the fermentation end phase; wherein types of generative algorithms include: Transformer models and their derivative models, Mamba models, and their derivative models; The method of fusing the features after the feature processing operation to obtain the model construction data set includes: using feature engineering to perform feature dimensionality upgrading operations on the processed fermentation static features, fermentation dynamic features and fermentation stage features to enhance the correlation between features in various fields; wherein, the feature dimensionality upgrading operations include: for continuous variables, using polynomial expansion and logarithmic transformation to perform dimensionality upgrading; for discrete variables, using unique hot encoding and combined features to perform dimensionality upgrading; after completing the feature dimensionality upgrading operation, fusing the upgraded features to obtain a unified feature representation, and finally obtaining the model construction data set.

2. The fermentation process prediction method based on the generative algorithm according to claim 1, characterized in that: The generative model of each stage is composed of multiple sub-stage model structures that use the same or different generative algorithms connected in sequence; The first sub-stage model structure is used to take the features within a time period as input and output the prediction results for the next time period after the time period; For non-first sub-stage model structures, it is used to take the input features of the connected previous sub-stage model structure and the prediction results output by the previous sub-stage model structure as input, and output the prediction results for the next time period after the time period predicted by the previous sub-stage model.

3. A fermentation process prediction system based on a generative algorithm, characterized in that: The system comprises: A parameter collection module is used to perform fermentation parameter collection and preprocessing operations to construct a sample data set, which includes: collecting fermentation parameters throughout the entire life cycle of the biological fermentation, and then preprocessing these fermentation parameters to construct a sample data set; wherein these fermentation parameters are subdivided into pre-fermentation parameters, fermentation process parameters, and fermentation result parameters; A domain feature extraction module, connected to the parameter collection module, is used to perform feature extraction of the sample data set and establishment of the model data set, which includes: extracting domain features related to the fermentation process from the constructed sample data set, and then establishing the model construction data set; wherein each domain feature includes fermentation static features, fermentation dynamic features, and fermentation stage features; A model construction module, connected to the domain feature extraction module, for constructing a phased generative model based on the model construction dataset using one or more generative algorithms; A fermentation process prediction module, connected to the model building module, is used to predict the subsequent fermentation process based on the constructed stage-by-stage generation model and the input fermentation parameters, and obtain corresponding prediction results; The extracting of domain features related to the fermentation process from the constructed sample data set and then establishing the model construction data set includes: extracting fermentation static features, fermentation dynamic features, and fermentation stage features according to pre-fermentation parameters, fermentation process parameters, and fermentation result parameters in the sample data set; performing corresponding feature processing operations on the extracted fermentation static features, fermentation dynamic features, and fermentation stage features by means of a fermentation domain feature processing mechanism; and fusing the features after the feature processing operations to obtain the model construction data set; The fermentation domain feature processing mechanism includes: for non-one-dimensional features, a specific encoding method is used to convert them into low-dimensional embedding vectors; for unquantized features, domain knowledge is used to perform quantization operations; for time series data features, special time series feature processing and encoding operations are performed; The method of using one or more generative algorithms to construct a phased generative model based on the model construction dataset includes: dividing the model construction dataset into datasets for a fermentation startup phase, a main fermentation phase, and a fermentation end phase according to different stages of the fermentation process; wherein the dataset for each stage is divided into a training set, a test set, and a validation set; and using one or more generative algorithms based on the datasets for each stage to construct generative models for the fermentation startup phase, the main fermentation phase, and the fermentation end phase; wherein types of generative algorithms include: Transformer models and their derivative models, Mamba models, and their derivative models; The method of fusing the features after the feature processing operation to obtain the model construction data set includes: using feature engineering to perform feature dimensionality upgrading operations on the processed fermentation static features, fermentation dynamic features and fermentation stage features to enhance the correlation between features in various fields; wherein, the feature dimensionality upgrading operations include: for continuous variables, using polynomial expansion and logarithmic transformation to perform dimensionality upgrading; for discrete variables, using unique hot encoding and combined features to perform dimensionality upgrading; after completing the feature dimensionality upgrading operation, fusing the upgraded features to obtain a unified feature representation, and finally obtaining the model construction data set.

4. An electronic terminal, characterized in that: include: one or more memories and one or more processors; The one or more memories are used to store computer programs; The one or more processors are connected to the memory and are configured to run the computer program to perform the method according to claim 1 or 2.

Images