A method and system for controlling the stability of pigment produced by Monascus

Through deep learning technology, data preprocessing and inverse residual calculations are used to adjust the production parameters of pigments produced by Aspergillus Rhodian, the problems of unstable pigment yield and high energy consumption in the existing technology are solved, and more efficient pigmentation is achieved Production.

CN118378064BActive Publication Date: 2024-11-22TIBET RED YEAST BIOLOGICAL CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410407789.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-07
Publication Date
2024-11-22
Estimated Expiration
2044-04-07

AI Technical Summary

Technical Problem

The existing stability control methods for the production of Aspergillus Aromatic pigments have problems such as unstable yield, excessive energy consumption and long production cycle, which limits the efficiency and sustainability of Aspergillus Aromatic pigment production.

Method used

Deep learning technology is adopted to collect data from the pigment production process of Aspergillus Rhodobacteria for data preprocessing and adaptive filtering, and the production parameters are adjusted using inverse residual calculation and reinforcement learning model to stabilize pigment yield. The specific steps include converting production features to frequency domain space, performing Hadamard product operation and self-attention layer processing, splitting them into vector elements mixing, and inputting inverse residual modules and feedforward neural networks for prediction and adjustment.

Benefits of technology

It improves the control ability of Aspergillus pigment production, reduces control costs, and improves production efficiency and sustainability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118378064B_ABST
    Figure CN118378064B_ABST
Patent Text Reader

Abstract

The present application discloses a stability control method and system for Monascus pigment production. The method comprises: collecting data during the current pigment production process of Monascus to obtain production parameters; performing data preprocessing on the production parameters to obtain production characteristics; performing adaptive filtering on the production characteristics, and using the results of the processing to perform inverse residual calculation for rapid reasoning; and adjusting the production parameters based on the results of the rapid reasoning to stabilize the pigment production of the Monascus. The present application scheme can improve control capabilities and reduce control costs.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and in particular to a method and system for controlling the stability of pigment produced by Monascus. Background Art

[0002] Monascus is a fungus that produces red pigments and is commonly used in food processing, such as fermented tofu, rice vinegar and rice wine. Monascus pigments not only give food a unique color and flavor, but also have antioxidant, antibacterial and lipid-lowering functions, so they play an important role in the development of modern science and technology. The stability control of the pigment produced by Monascus means taking measures to ensure the stability of the pigment production and quality during its cultivation and fermentation process, and to resist the influence of external environmental factors on the pigment. This control is crucial in the food industry because it is directly related to the quality and stability of the final product, and also affects the market competitiveness of the product.

[0003] Currently, widely used methods for controlling the stability of Monascus pigment production include optimizing culture medium formula, controlling fermentation conditions, and using engineered strains. Optimizing culture medium formula can improve pigment production and stability by adding appropriate carbon, nitrogen sources, and trace elements; controlling fermentation conditions includes adjusting parameters such as temperature, pH, and oxygen supply to create an environment that is most suitable for Monascus growth and pigment production; using engineered strains is to improve strains through genetic engineering technology to have higher pigment production and stability. However, these methods still face some challenges, such as unstable production, excessive energy consumption, and long production cycles, which limit the efficiency and sustainability of Monascus pigment production.

[0004] Therefore, there is an urgent need for a technical solution that can improve control capabilities and reduce control costs. Summary of the invention

[0005] In order to solve the deficiencies of the prior art, the present invention provides a method and system for controlling the stability of pigment produced by Monascus spp. The present invention solves the technical problems of the prior art control methods, such as poor performance and high cost.

[0006] A method for controlling the stability of pigment produced by Monascus, comprising: collecting data during the current pigment production process of Monascus to obtain production parameters; wherein the data include temperature, pH value, oxygen supply, nutrient concentration and current pigment output; performing data preprocessing on the production parameters to obtain production characteristics; wherein the data preprocessing method comprises: Among them, y i represents the production parameters after processing at position i, k represents the window size, x i+jrepresents a production parameter, j represents a step size; the production feature is subjected to adaptive filtering processing, and the result of the processing is used for inverse residual calculation for fast reasoning; and based on the result of the fast reasoning, the production parameter is adjusted to stabilize the pigment yield of the Monascus.

[0007] In a possible implementation, the production features are subjected to adaptive filtering processing, and the results of the processing are used to perform inverse residual calculation for fast reasoning, including: converting the production features to frequency domain space to obtain frequency domain features; performing Hadamard product operations on the frequency domain features and the production features in the spatial domain, and the results of the operations are used to input the self-attention layer to obtain adaptive features; splitting the adaptive features into multiple vector elements for vector element mixing; and inputting the results of the vector element mixing into an inverse residual module, and the output of the module is used to input a feedforward neural network to infer the predicted pigment production at the target time.

[0008] In a possible implementation, converting the production feature into a frequency domain space to obtain a frequency domain feature includes: Among them, f(t) represents the production characteristics, W f (a,b) represents the frequency domain features, a represents the scale parameter, b represents the location parameter, and ψ() represents the basis function.

[0009] In one possible implementation, splitting the adaptive feature into multiple vector elements for vector element mixing includes: calculating each vector element using a target embedding function, and using the result of the calculation to weight with the embedding weight to obtain an embedded feature; and mixing the embedded features based on the weight of the vector element.

[0010] In one possible implementation, each vector element is calculated using a target embedding function, and the result of the calculation is used to weight the embedding weight to obtain an embedding feature including: Among them, q represents the embedded feature, represents the target embedding function, ω i represents the embedding weight, p represents the vector element, N(p) represents the set of vector elements p, and i represents the index in the set.

[0011] In a possible implementation, mixing the embedded features based on the weights of the vector elements includes: in, represents the mixed result, h represents the length of the embedded feature in the first direction, w represents the length of the embedded feature in the second direction, K(x) represents the weight of the vector element, and x(m,n) represents the vector element corresponding to the embedded feature at the (m,n) position.

[0012] In one possible implementation, the structure of the anti-residual module includes: F(·)=Conv(Conv(MHSA(·))+(·)), where F() represents an anti-residual module, Conv() represents a convolutional neural network, MHSA() represents a cascaded multi-head attention layer, and (·) represents the input of the module.

[0013] In one possible implementation, before the result of the vector element mixing is input into the anti-residual module, the output of the module is used to input the feedforward neural network to infer the predicted pigment production at the target time, it also includes: taking the numerical values ​​of temperature, pH value, oxygen supply and nutrient concentration as actions; taking the current pigment production as a reward; and constructing a reinforcement learning model based on the actions and rewards to train the anti-residual module.

[0014] The embodiment of the present application also provides a stability control system for Monascus pigment production, including: an acquisition module, used to collect data in the current pigment production process of Monascus to obtain production parameters; wherein the data includes temperature, pH value, oxygen supply, nutrient concentration and current pigment production; a data preprocessing module, used to perform data preprocessing on the production parameters to obtain production characteristics; an inference module, used to perform adaptive filtering on the production characteristics, and the results of the processing are used to perform inverse residual calculation for fast reasoning; and a stability control module, used to adjust the production parameters based on the results of the fast reasoning to stabilize the pigment production of the Monascus.

[0015] An embodiment of the present application also provides a device for controlling the stability of pigment produced by Monascus, comprising: a processor, a memory, and a system bus; wherein the processor and the memory are connected via the system bus; the memory is used to store one or more programs, and the one or more programs include instructions, which, when executed by the processor, enable the processor to execute any one of the systems or methods described above.

[0016] Through the above-provided method and system for controlling the stability of pigment produced by Monascus, the embodiments of the present application can improve the control capability and reduce the control cost by combining deep learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1A schematic diagram of a method for controlling the stability of pigment produced by Monascus provided in an embodiment of the present application;

[0019] Figure 2 An exemplary flowchart of rapid reasoning provided for an embodiment of the present application. DETAILED DESCRIPTION

[0020] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present disclosure unless otherwise specifically stated.

[0021] Those skilled in the art can understand that the terms "first", "second" and the like in the embodiments of the present disclosure are only used to distinguish different steps, devices or modules, etc., and neither represent any specific technical meaning nor represent the necessary logical order between them. It should also be understood that in the embodiments of the present disclosure, "multiple" can refer to two or more, and "at least one" can refer to one, two or more. It should also be understood that for any component, data or structure mentioned in the embodiments of the present disclosure, in the absence of explicit limitation or contrary revelation given in the context, it can generally be understood as one or more. In addition, the term "and / or" in the present disclosure is only a description of the association relationship of the associated objects, indicating that there can be three relationships, for example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in the present disclosure generally indicates that the associated objects before and after are an "or" relationship. It should also be understood that the description of each embodiment in the present disclosure emphasizes the differences between the embodiments, and the same or similar parts can refer to each other. For the sake of brevity, they will not be repeated one by one.

[0022] At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present disclosure and its application or use. The techniques, methods and devices known to ordinary technicians in the relevant fields may not be discussed in detail, but where appropriate, the techniques, methods and devices should be considered part of the specification. It should be noted that similar numbers and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0023] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0024] Figure 1 This is a schematic diagram of a method for controlling the stability of pigment produced by Monascus provided in an embodiment of the present application. Figure 1 As shown, at step S101, data of the current pigment production process of Monascus is collected to obtain production parameters; wherein the data includes temperature, pH value, oxygen supply, nutrient concentration and current pigment production. It should be understood that Monascus is an important microbial species that can produce a variety of beneficial substances, including pigments, during the fermentation process. The production of pigments is affected by many factors, such as temperature, pH value, oxygen supply, nutrient concentration, etc. These factors can be described by mathematical models to describe their influence on pigment production, thereby optimizing and controlling production parameters.

[0025] In step S102, data preprocessing is performed on the production parameters to obtain production characteristics; wherein the data preprocessing method includes:

[0026]

[0027] Among them, y i represents the production parameters after processing at position i, k represents the window size, x i+j Represents the production parameters, and j represents the step size. In practical applications, choosing the right window size and step size is crucial to the effect of data preprocessing. A larger window size can capture a wider range of contextual information, but may lead to information confusion and over-smoothing; while a smaller window size may lead to the loss of detail information. Therefore, in specific applications, parameter selection and tuning are required based on actual conditions. The advantage of this data preprocessing method is that it can effectively reduce noise and unnecessary fluctuations in the data and highlight the potential features in the production parameters.

[0028] At step S103, the production features are subjected to adaptive filtering processing, and the processing result is used to perform inverse residual calculation for fast reasoning (such as Figure 2 The specific method will be shown in Figure 2). Adaptive filtering is a signal processing technique that adjusts the filter parameters according to the local characteristics of the signal to better adapt to the dynamic changes of the signal. Inverse residual calculation refers to obtaining additional information about the signal by calculating and processing the residual of the signal.

[0029] It should be noted that before the production characteristics are subjected to adaptive filtering processing and the results of the processing are used for anti-residual calculation for fast reasoning, it also includes: taking the numerical values ​​of temperature, pH value, oxygen supply and nutrient concentration as actions; taking the current pigment production as a reward; and constructing a reinforcement learning model based on the actions and rewards to train the anti-residual module.

[0030] In this process, the first thing to consider is the step of adaptive filtering the production features. Adaptive filtering is a signal processing technology that aims to adjust the parameters of the filter according to the statistical characteristics of the signal itself to achieve effective signal processing and denoising. Specifically, in the production features mentioned in this article, there may be various noises and interferences. Adaptive filtering can help remove these interferences, making the final processing results more accurate and reliable. The core of adaptive filtering is to continuously update the weights of the filter to adapt to the changing characteristics of the signal and improve the filtering effect.

[0031] Next, the processing results are used for reverse residual calculation. Reverse residual calculation refers to adjusting model parameters to achieve better prediction results by calculating and back-propagating the residual between the processing results and the expected values. Here, various feedback control algorithms or neural network structures can be used to achieve accurate modeling and prediction of production characteristics.

[0032] In the process of constructing a reinforcement learning model, it is first necessary to define the state space, action space, and reward function. The state space can be composed of production characteristics such as temperature, pH, oxygen supply, and nutrient concentration, the action space includes possible operations to adjust these parameters, and the reward function evaluates the performance of the model based on the increase or decrease of the current pigment production. The goal of the reinforcement learning model is to find the optimal strategy through interactive learning with the environment to maximize the long-term cumulative reward. In this process, the model can use various reinforcement learning algorithms, such as Q learning, deep reinforcement learning, etc., to optimize and train the anti-residual module.

[0033] At step S104, based on the result of the rapid reasoning, the production parameters are adjusted to stabilize the pigment yield of the Monascus.

[0034] Figure 2An exemplary flowchart of fast reasoning provided for an embodiment of the present application. It should be understood that the method shown in the figure is exemplary and not restrictive. This means that the module architecture involved is not limited to a specific form or design, but is presented as an example. In other words, the architecture shown in the figure can be regarded as a way of expression to clearly describe related concepts and relationships, and does not exclude other forms of architecture. Therefore, when interpreting the architecture in the picture, it should be understood that the module has flexibility and diversity, and its purpose is to provide an exemplary description rather than a restrictive provision on a specific form.

[0035] like Figure 2 As shown, at step S201, the production feature is converted into a frequency domain space to obtain a frequency domain feature. In one embodiment, it may include:

[0036]

[0037] Among them, f(t) represents the production characteristics, W f (a,b) represents frequency domain features, a represents scale parameters, b represents location parameters, and ψ() represents basis functions. It should be understood that frequency domain analysis is a method in signal processing. By converting the signal from the time domain to the frequency domain, the frequency components and amplitude information of the signal can be revealed, which helps to analyze and process the signal more deeply. Through the above frequency domain feature extraction process, the original production features can be converted into the frequency domain space, thereby revealing the energy distribution corresponding to different frequency components. This helps to model and analyze the signal in the frequency domain, and then achieve a deep understanding and utilization of the signal characteristics. Through the extraction of frequency domain features, richer information and more effective feature representation can be provided for subsequent signal processing tasks, thereby improving the performance and robustness of the entire system.

[0038] At step S202, a Hadamard product operation is performed on the frequency domain features and the production features in the spatial domain, and the result of the operation is used to input the self-attention layer to obtain adaptive features. It should be understood that the frequency domain features and the production features in the spatial domain refer to two different representation methods commonly used in the fields of signal processing and image processing. The frequency domain feature refers to the performance of the signal or image in the frequency domain, which is usually obtained by performing a Fourier transform on the signal or image. The production feature in the spatial domain refers to the original representation of the signal or image in the time domain, that is, the distribution of its pixel values ​​in space.

[0039] The Hadamard product operation is an element-wise multiplication operation, usually represented by the symbol ⊙. Given two matrices or tensors A and B of the same size, each element c[i][j] of their Hadamard product C is equal to the product of the elements of A and B at the corresponding position, that is, c[i][j] = a[i][j]*b[i][j].

[0040] The self-attention layer is a mechanism commonly used in deep learning models to capture the correlation and dependency between positions in the input sequence. The basic principle is to strengthen or suppress the representation of different positions by calculating the attention weights of each position in the input sequence to obtain better feature representation. The self-attention mechanism is usually implemented by calculating the attention score, applying the softmax function and weighted summation.

[0041] In the method described in this article, the purpose of performing Hadamard product operations on the frequency domain features and the production features in the spatial domain is to fuse their feature information and generate adaptive features for input into the self-attention layer. This feature fusion and conversion can effectively combine the information in the frequency domain and the spatial domain, and improve the model's ability to understand and characterize input data. Specifically, assuming that the frequency domain feature is represented by F and the production feature in the spatial domain is represented by P, the feature representation obtained after the Hadamard product operation is H = F⊙P.

[0042] Next, the obtained adaptive feature H is input into the self-attention layer for processing. The self-attention layer calculates the correlation of the features at each position in the input sequence and generates attention weights for each position. These weights can reflect the importance and correlation between each position in the input sequence, thereby helping the model to better understand and utilize the information of the input data. Ultimately, through the processing of the self-attention layer, a richer and more accurate feature representation can be obtained for subsequent tasks and applications.

[0043] At step S203, the adaptive feature is split into multiple vector elements for vector element mixing. Specifically, first, each vector element is calculated using the target embedding function, and the result of the calculation is used to weight the embedding weight to obtain the embedded feature. In one embodiment:

[0044]

[0045] Among them, q represents the embedded feature, represents the target embedding function, ω i represents the embedding weight, p represents the vector element, N(p) represents the set of vector elements p, and i represents the index in the set. By applying the target embedding function to each element in the vector element set and performing weighted summation according to the corresponding embedding weight, the embedding feature representing the entire set is finally obtained. This method can effectively capture the relationship and characteristics between vector elements, providing strong support for subsequent data processing and model application.

[0046] In practical applications, the selection of target embedding functions and the adjustment of weights are crucial, as they directly affect the quality and expressiveness of the final embedded features. By rationally designing the target embedding function and carefully setting the weight parameters, the embedded features can better reflect the characteristics of the original data, thereby improving the performance of the model on various tasks.

[0047] Then, based on the weights of the vector elements, the embedded features are mixed. It should be understood that embedded features refer to the technology of mapping high-dimensional data to low-dimensional space, which is usually used to represent the key features of the data. For example, in image processing, embedded features can be used to describe abstract information in the image, such as texture, shape, and structure. However, in practical applications, due to the complexity and diversity of data, a single embedded feature is often difficult to fully express the rich information of the data, so it is necessary to mix multiple embedded features to obtain a more comprehensive and accurate data representation. However, the existing embedded feature mixing methods often have the following problems: first, the weight allocation of embedded features is not flexible and accurate enough, and it is difficult to effectively capture the intrinsic characteristics of the data; second, when mixing embedded features, the computational complexity is high, resulting in low mixing efficiency; third, it is difficult for the existing technology to fully consider the mutual influence and correlation between embedded features, and it is difficult to achieve global modeling of data. In one embodiment:

[0048]

[0049] in, represents the mixed result, h represents the length of the embedded feature in the first direction, w represents the length of the embedded feature in the second direction, K(x) represents the weight of the vector element, and x(m,n) represents the vector element corresponding to the embedded feature at the (m,n) position. Through the method of the present application, it is possible to achieve flexible and accurate mixing of embedded features, thereby obtaining a more comprehensive and accurate data representation. And the computational complexity is low, the mixing efficiency is high, and it is suitable for processing large-scale data. In addition, it is also possible to fully consider the mutual influence and correlation between embedded features to achieve global modeling of data.

[0050] At step S204, the result of the vector element mixing is input into the anti-residual module, and the output of the module is used to input the feedforward neural network to infer the predicted pigment production at the target time. The structure of the anti-residual module includes:

[0051] F(·)=Conv(Conv(MHSA(·))+(·)),

[0052] Among them, F() represents an anti-residual module, Conv() represents a convolutional neural network, MHSA() represents a cascaded multi-head attention layer, and (·) represents the input of the module. It should be noted that the process of vector element mixing refers to mixing and integrating element vectors of different dimensions, usually using linear transformation or nonlinear transformation. This step aims to obtain a more comprehensive and global feature representation to better support subsequent model learning and reasoning.

[0053] Here, the design of the anti-residual module adopts a combination of cascaded multi-head attention layers and convolutional neural networks. The multi-head attention mechanism is a mechanism for capturing global dependencies in sequence information. It obtains a more global and comprehensive sequence representation by modeling the interactions and associations at different positions in the sequence. At the same time, convolutional neural networks have excellent expression capabilities for local features, and the convolution operation can effectively capture local patterns and features in the sequence.

[0054] Specifically, in the structure of the anti-residual module F(·) = Conv(Conv(MHSA(·))+(·)), the input is first processed by the multi-head attention mechanism through MHSA() to obtain a more global and rich sequence feature representation. Subsequently, the representation is further feature extracted and transformed through the convolution operation Conv(MHSA(·)) to obtain a more advanced and abstract feature representation. Finally, the output of the convolution operation is residually connected to the input (·), and the final module output F(·) is obtained again through the convolution operation Conv(Conv(MHSA(·))+(·)).

[0055] The anti-residual module designed in this way combines the advantages of multi-head attention mechanism and convolution operation to a certain extent, so that the module can better capture the global dependencies and local patterns in the sequence, and provide a more comprehensive and effective feature representation for the subsequent feedforward neural network. At the same time, the design of this module also takes into account the balance between computational efficiency and model parameter quantity in practical applications.

[0056] Furthermore, an embodiment of the present application also provides a device for controlling the stability of pigment produced by Monascus, comprising: a processor, a memory, and a system bus; the processor and the memory are connected via the system bus; the memory is used to store one or more programs, and the one or more programs include instructions, which, when executed by the processor, enable the processor to execute any one of the implementation methods of the above-mentioned processing method.

[0057] Furthermore, an embodiment of the present application also provides a computer program product, which, when executed on a terminal device, enables the terminal device to execute any one of the implementation methods of the above-mentioned processing method.

[0058] It can be known from the description of the above implementation mode that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a storage medium such as ROM / RAM, a disk, an optical disk, etc., including several instructions for enabling a computer device (which can be a personal computer, a server, or a network communication device such as a media gateway, etc.) to execute the methods described in the various embodiments of the present application or certain parts of the embodiments.

[0059] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0060] It should also be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0061] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for controlling the stability of pigment produced by Monascus, characterized in that: include: Collect data on the current pigment production process of Monascus to obtain production parameters; The data include temperature, pH, oxygen supply, nutrient concentration and current pigment production; Performing data preprocessing on the production parameters to obtain production characteristics; The data preprocessing method comprises: Among them, y i represents the production parameters after processing at position i, k represents the window size, x i+j represents the production parameter, j represents the step length; The production features are subjected to adaptive filtering processing, and the processing results are used for inverse residual calculation to perform fast reasoning; The result of the processing is used to perform inverse residual calculation to perform fast reasoning including: Converting the production features into a frequency domain space to obtain frequency domain features; Performing a Hadamard product operation on the frequency domain feature and the production feature in the spatial domain, and the result of the operation is used to input a self-attention layer to obtain an adaptive feature; Splitting the adaptive feature into a plurality of vector elements for vector element mixing; and The result of the vector element mixing is input into an anti-residual module, and the output of the module is used to input a feed-forward neural network to infer the predicted pigment production at the target time; The structure of the anti-residual module includes: F(·)=Conv(Conv(MHSA(·))+(·)), Among them, F() represents an anti-residual module, Conv() represents a convolutional neural network, MHSA() represents a cascaded multi-head attention layer, and (·) represents the input of the module; The output of the module is used to input the feedforward neural network to infer the predicted pigment production at the target time, and also includes: Use values ​​of temperature, pH, oxygen supply, and nutrient concentration as actions; Current pigment production as a reward; and constructing a reinforcement learning model based on the action and the reward to train the anti-residual module; and Based on the result of the rapid reasoning, the production parameters are adjusted to stabilize the pigment yield of the Monascus.

2. The method according to claim 1, characterized in that The production characteristics are converted into the frequency domain space to obtain the frequency domain characteristics, which includes: Among them, f(t) represents the production characteristics, W f (a,b) represents the frequency domain features, a represents the scale parameter, b represents the location parameter, and ψ() represents the basis function.

3. The method according to claim 1, characterized in that The step of splitting the adaptive feature into a plurality of vector elements for vector element mixing includes: Calculating each vector element using the target embedding function, wherein the calculated result is used to be weighted with the embedding weight to obtain an embedding feature; and The embedded features are mixed based on the weights of the vector elements.

4. The method according to claim 3, characterized in that The target embedding function is used to calculate each vector element, and the result of the calculation is used to weight the embedding weight to obtain the embedding features including: Among them, q represents the embedded feature, represents the target embedding function, ω i represents the embedding weight, p represents the vector element, N(p) represents the set of vector elements p, and i represents the index in the set.

5. The method according to claim 3, characterized in that: Wherein, mixing the embedded features based on the weights of the vector elements comprises: in, represents the mixed result, h represents the length of the embedded feature in the first direction, w represents the length of the embedded feature in the second direction, K(x) represents the weight of the vector element, and x(m,n) represents the vector element corresponding to the embedded feature at the (m,n) position.

6. A stability control system for pigment produced by Monascus, characterized in that: include: The collection module is used to collect data on the pigment production process of Monascus spp. to obtain production parameters; The data include temperature, pH, oxygen supply, nutrient concentration and current pigment production; A data preprocessing module, used for performing data preprocessing on the production parameters to obtain production characteristics; An inference module, used for performing adaptive filtering processing on the production features, and the processing result is used for performing inverse residual calculation for fast inference; The result of the processing is used to perform inverse residual calculation to perform fast reasoning including: Converting the production features into a frequency domain space to obtain frequency domain features; Performing a Hadamard product operation on the frequency domain feature and the production feature in the spatial domain, and the result of the operation is used to input a self-attention layer to obtain an adaptive feature; Splitting the adaptive feature into a plurality of vector elements for vector element mixing; and The result of the vector element mixing is input into an anti-residual module, and the output of the module is used to input a feed-forward neural network to infer the predicted pigment production at the target time; The structure of the anti-residual module includes: F(·)=Conv(Conv(MHSA(·))+(·)), Among them, F() represents an anti-residual module, Conv() represents a convolutional neural network, MHSA() represents a cascaded multi-head attention layer, and (·) represents the input of the module; The output of the module is used to input the feedforward neural network to infer the predicted pigment production at the target time, and also includes: Use values ​​of temperature, pH, oxygen supply, and nutrient concentration as actions; Current pigment production as a reward; and constructing a reinforcement learning model based on the action and the reward to train the anti-residual module; and The stabilization control module is used to adjust the production parameters based on the result of the rapid reasoning to stabilize the pigment yield of the Monascus purpurogenous bacteria.