A smart monitoring system and method for biological reactions

Abstract

This invention discloses an intelligent biological reaction monitoring system and method, belonging to the field of intelligent monitoring. The system includes: a reaction stage segmentation unit for segmenting and determining multiple reaction stages; a variable relationship mining unit for determining dominant and auxiliary variables, mining linear relationships between variables, coordinating insensitive loss functions, and supervising the training of a soft monitoring module; a continuous monitoring and acquisition unit for configuring monitoring equipment groups, performing continuous monitoring, and determining reaction monitoring data; a data analysis and screening unit for preprocessing data, performing linear analysis and abnormal reaction localization, and determining valid monitoring data; a game equilibrium decision-making unit for performing game equilibrium decisions on abnormal reaction identifiers and determining reaction regulation strategies; and a reaction monitoring management unit for responding to regulation strategies, performing feedback monitoring analysis and reaction monitoring management, improving the accuracy of biological reaction monitoring, achieving real-time online monitoring, and intelligently regulating abnormal situations.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring, and specifically to an intelligent monitoring system and method for biological reactions. Background Technology

[0002] Due to the complexity and dynamic nature of biological reaction processes, real-time and precise monitoring and control are required. Current technologies for biological reaction monitoring primarily employ offline sampling and analysis, which suffers from low monitoring accuracy and poor real-time performance. Offline sampling and analysis requires manual operation, has a limited sampling frequency, and struggles to provide continuous real-time monitoring, leading to incomplete and delayed data. Furthermore, because biological reaction processes involve multiple parameters and indicators, existing offline analysis methods struggle to accurately identify and quantify key variables, affecting monitoring accuracy. In addition, abnormal operating conditions frequently occur during biological reactions due to factors such as raw material fluctuations, equipment malfunctions, and environmental disturbances, causing the reaction to deviate from its optimal state. Current technologies lack effective intelligent decision-making tools, making it difficult to quickly and accurately diagnose and control abnormal situations, thus impacting the biological reaction process. Therefore, current biological reaction monitoring technologies suffer from low accuracy, poor real-time performance, and difficulty in effectively controlling abnormal reactions. Summary of the Invention

[0003] This application provides an intelligent monitoring system and method for biological reactions, aiming to solve the technical problems of low accuracy and poor real-time performance in biological reaction monitoring in the prior art, which makes it difficult to effectively control abnormal reactions.

[0004] In view of the above problems, this application provides an intelligent monitoring system and method for biological reactions.

[0005] The first aspect disclosed in this application provides an intelligent monitoring system for biological reactions. This system includes: a reaction stage segmentation unit for interacting with the reaction process of the target biological reaction and segmenting and determining multiple reaction stages; a variable relationship mining unit for traversing multiple reaction stages, determining dominant and auxiliary variables, mining linear relationships between variables, employing a coordinating insensitive loss function, and supervising the training of a soft monitoring module; a continuous monitoring and acquisition unit for traversing dominant and auxiliary variables, configuring a monitoring device group for directly monitoring variables, continuously monitoring along with the reaction process, and determining reaction monitoring data; a data analysis and screening unit for preprocessing reaction monitoring data at the front end, transmitting it back to the soft monitoring module for linear analysis and abnormal reaction location, and determining valid monitoring data, wherein valid monitoring data includes stage variable data and abnormal reaction identifiers; a game equilibrium decision-making unit for traversing valid monitoring data, performing game equilibrium decisions on abnormal reaction identifiers, and determining reaction regulation strategies; and a reaction monitoring management unit for the reaction vessel to respond to the reaction regulation strategy and, in conjunction with the soft monitoring module, performing feedback monitoring analysis to manage the reaction monitoring of the target biological reaction.

[0006] Another aspect of this application discloses an intelligent monitoring method for biological reactions. This method includes: traversing the reaction process of the target biological reaction and defining multiple reaction stages; traversing the multiple reaction stages, identifying dominant and auxiliary variables, mining linear relationships between variables, employing a coordinating insensitive loss function, and supervising the training of a soft monitoring module; traversing the dominant and auxiliary variables, configuring a monitoring device group for directly monitoring variables, continuously monitoring along with the reaction process, and determining reaction monitoring data, which is marked with stage nodes; preprocessing the reaction monitoring data at the front end and transmitting it back to the soft monitoring module for linear analysis and abnormal reaction localization, determining valid monitoring data, which includes stage variable data and abnormal reaction markers; traversing the valid monitoring data, performing game-theoretic equilibrium decisions on the abnormal reaction markers, and determining a reaction regulation strategy; the reaction vessel responding to the reaction regulation strategy and combining it with the soft monitoring module for feedback monitoring and analysis, thereby performing reaction monitoring management of the target biological reaction.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] By employing a reaction process that divides the reaction into stages and interacts with the target biological reaction, multiple reaction stages are defined, providing a more refined description of the reaction process and laying the foundation for subsequent variable analysis and monitoring control. Through a variable relationship mining unit, multiple reaction stages are traversed to identify dominant and auxiliary variables, uncover linear relationships between variables, and, in conjunction with an insensitive loss function, supervise the training of the soft monitoring module. This data-driven approach reveals the intrinsic connections between key variables in the biological reaction process, improving monitoring accuracy. A continuous monitoring and acquisition unit traverses dominant and auxiliary variables, configures monitoring equipment groups for directly monitored variables, and continuously monitors the reaction process to determine reaction monitoring data, achieving real-time online monitoring of the biological reaction process and improving the timeliness and completeness of data acquisition. Finally, a data analysis and screening unit performs front-end preprocessing of the reaction monitoring data before transmitting it back to the soft monitoring module. The module performs linear analysis and abnormal response localization to determine effective monitoring data, including stage variable data and abnormal response identifiers. It automatically filters out high-quality monitoring data to provide a reliable basis for subsequent decision-making and control. Through a game equilibrium decision-making unit, it traverses the effective monitoring data, performs game equilibrium decisions on abnormal response identifiers, determines the response regulation strategy, and generates the optimal abnormal handling strategy through multi-objective optimization and equilibrium analysis, achieving intelligent response regulation. The response monitoring and management unit controls the response regulation strategy of the reaction vessel, and combines it with the soft monitoring module for feedback monitoring and analysis. This achieves a technical solution for full-process monitoring and management of the target biological reaction, solving the technical problems of low accuracy and poor real-time performance in existing biological reaction monitoring technologies, which make it difficult to effectively regulate abnormal reactions. It achieves the technical effects of improving the accuracy of biological reaction monitoring, realizing real-time online monitoring, and intelligent regulation for abnormal situations.

[0009] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0010] Figure 1 This application provides a schematic diagram of a biological reaction intelligent monitoring system.

[0011] Figure 2 This application provides a schematic flowchart of a method for intelligent monitoring of biological reactions.

[0012] Figure labeling: 11. Reaction stage division unit; 12. Variable relationship mining unit; 13. Continuous monitoring and acquisition unit; 14. Data analysis and screening unit; 15. Game equilibrium decision-making unit; 16. Reaction monitoring and management unit. Detailed Implementation

[0013] The overall concept of the technical solution provided in this application is as follows:

[0014] This application provides an intelligent monitoring system and method for biological reactions. First, a reaction stage segmentation unit is used to finely describe and divide the biological reaction process into stages. Second, a variable relationship mining unit is used to reveal the intrinsic connections between key variables and train a soft monitoring module. Third, a continuous monitoring and acquisition unit enables real-time online monitoring of the biological reaction process, and a data analysis and screening unit is used for intelligent data processing and anomaly localization. Furthermore, a game theory equilibrium decision-making unit is introduced to perform multi-objective optimization decisions on abnormal reactions, generating optimal control strategies. Finally, a reaction monitoring management unit is used to achieve closed-loop optimization control and full-process management of the biological reaction process, achieving the goals of improving the accuracy of biological reaction monitoring, realizing real-time online monitoring, and intelligent control in response to abnormal situations.

[0015] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0016] Example 1

[0017] like Figure 1 As shown in the figure, this application provides an intelligent monitoring system for biological responses, which includes:

[0018] Reaction stage division unit 11 is used for the reaction process of interactive target biological reaction to divide and determine multiple reaction stages.

[0019] Specifically, the reaction stage segmentation unit 11 first acquires and analyzes the reaction process of the target bioreaction. Based on the characteristics and patterns of the bioreaction at different stages, such as the dynamic changes in parameters like biomass, substrate concentration, and metabolite concentration, the complete reaction process is divided into multiple reaction stages. Each reaction stage corresponds to a specific time period and is associated with a specific step in the reaction process. By segmenting the target bioreaction into stages, the reaction stage segmentation unit 11 outputs the reaction stage segmentation results, i.e., multiple reaction stages, including the start and end times and duration of each reaction stage, laying the foundation for subsequent intelligent monitoring and control of each reaction stage.

[0020] By dividing the reaction into 11 units, the phased characteristics of biological reactions are fully considered, thereby enabling more refined and targeted phased monitoring and regulation, and improving the accuracy of biological reaction process optimization.

[0021] The variable relationship mining unit 12 is used to traverse multiple reaction stages, determine the dominant and auxiliary variables, mine the linear relationships between variables, coordinate the insensitive loss function, and supervise the training of the soft monitoring module.

[0022] Specifically, the variable relationship mining unit 12 identifies key influencing factors at each reaction stage by analyzing historical production data and expert experience, screening out a series of dominant variables, such as temperature, pH, and dissolved oxygen. Simultaneously, considering that some dominant variables may not be continuously monitored online due to limitations in detection methods, mathematical statistical methods, such as correlation analysis and regression analysis, are further utilized to discover auxiliary variables that have significant linear correlations with each dominant variable from numerous process parameters. Subsequently, the variable relationship mining unit 12 fully utilizes the linear relationships between the discovered dominant and auxiliary variables to construct a soft monitoring model. Through this soft monitoring model, the dominant variables that are difficult to monitor directly can be indirectly estimated and predicted using auxiliary variables that are easily monitored online, thereby achieving real-time monitoring of all dominant variables. In the process of constructing the soft monitoring model, the variable relationship mining unit 12 also introduces an insensitive loss function to describe the degree of impact on the accuracy of monitoring results when the measured values ​​of dominant or auxiliary variables deviate. By using the insensitive loss function in conjunction, the robustness and anti-interference ability of the soft monitoring model are further improved. Subsequently, using the identified dominant variable, auxiliary variable, linear relationships between variables, and an insensitive loss function, a soft monitoring module was constructed and subjected to supervised training. The optimized soft monitoring module, after training, can accurately predict the state of the dominant variable based on online-collected auxiliary variable data, and promptly reflect the dynamic changes in the biological reaction process.

[0023] Through the variable relationship mining unit 12, the coordinated monitoring of dominant and auxiliary variables was realized, overcoming the difficulty of continuous online monitoring of some key parameters. Furthermore, by introducing an insensitive loss function, the adaptability and reliability of the soft monitoring module were further enhanced, thereby effectively improving the accuracy and real-time performance of biological reaction process monitoring.

[0024] The continuous monitoring and acquisition unit 13 is used to traverse the dominant and auxiliary variables, configure the monitoring equipment group for the directly monitored variables, and continuously monitor the reaction process to determine the reaction monitoring data.

[0025] Specifically, the continuous monitoring and acquisition unit 13 selects appropriate sensors, such as temperature sensors, pH electrodes, and dissolved oxygen probes, based on the physicochemical characteristics of each directly monitored variable among the dominant and auxiliary variables. It also configures the installation positions and quantities of these sensors according to the structural layout of the reaction apparatus, forming a complete monitoring equipment group. During the bioreaction process, the continuous monitoring and acquisition unit 13 utilizes the configured monitoring equipment group to perform real-time, continuous online monitoring of each directly monitored variable. The raw data signals collected by the monitoring equipment group undergo preprocessing, such as filtering, amplification, and A / D conversion, to transform them into usable digital quantitative monitoring data. The continuous monitoring and acquisition unit 13 summarizes and organizes the digital quantitative monitoring data collected by the monitoring equipment group to obtain a series of reaction monitoring data reflecting the real-time state of the bioreaction process, presented in the form of a time series, with each time point corresponding to a set of measured values ​​of the directly monitored variables.

[0026] Through the continuous monitoring and acquisition unit 13, a complete monitoring equipment group is established for the direct monitoring variables in the bioreaction process, realizing the real-time continuous acquisition of process parameters, and providing reliable data support for subsequent data analysis and process optimization control.

[0027] The data analysis and screening unit 14 is used for front-end preprocessing of reaction monitoring data, which is then sent back to the soft monitoring module for linear analysis and abnormal reaction localization to determine valid monitoring data. Valid monitoring data includes stage variable data and abnormal reaction identifiers.

[0028] Specifically, the data analysis and screening unit 14 first performs front-end preprocessing on the received reaction monitoring data. This preprocessing includes a series of data quality control operations such as data cleaning, missing value imputation, noise removal, and data normalization to improve the completeness, accuracy, and consistency of the data. Then, the data analysis and screening unit 14 sends the preprocessed reaction monitoring data back to the soft monitoring module constructed by the variable relationship mining unit 12. The soft monitoring module uses the received reaction monitoring data to perform linear analysis and abnormal reaction localization on the dominant and auxiliary variables. By comparing the deviation between the actual monitored values ​​of each variable and the expected normal values, the soft monitoring module promptly detects abnormal fluctuations or situations exceeding the control range during the biological reaction process and generates corresponding abnormal reaction indicators.

[0029] After receiving the analysis results from the soft monitoring module, the data analysis and screening unit 14 performs a second screening of the data, taking into account both the quality of the reaction monitoring data and the reliability of the soft monitoring results, to determine the high-confidence, valid monitoring data. The valid monitoring data is encapsulated in a structured form, containing two types of key information: first, the variable data for each stage after being divided according to the reaction stage; and second, the abnormal reaction identifiers identified by the soft monitoring module.

[0030] Through the data analysis and screening unit 14, the monitoring data is effectively presented in a clearer, more accurate and efficient manner, providing key information and intelligent analysis results of the biological reaction process, thus laying a data foundation for intelligent monitoring and optimized control.

[0031] The game equilibrium decision unit 15 is used to traverse the effective monitoring data, make game equilibrium decisions on abnormal response indicators, and determine the response adjustment strategy.

[0032] Specifically, the game equilibrium decision-making unit 15 is connected after the data analysis and screening unit 14. It receives the valid monitoring data output by the data analysis and screening unit 14, performs intelligent decision analysis on the abnormal reaction indicators, and determines the reaction regulation strategy. The game equilibrium decision-making unit 15 traverses the valid monitoring data transmitted by the data analysis and screening unit 14 and extracts the abnormal reaction indicator information. Abnormal reaction indicators are usually associated with abnormal fluctuations in one or more key process parameters, characterizing potential problems or risks in the biological reaction process. For each abnormal reaction indicator, the game equilibrium decision-making unit 15 matches the abnormal situation with a preset strategy set, initially screening a series of possible response strategies. These strategies correct abnormal deviations and restore the reaction process to a normal state by adjusting relevant process parameters such as temperature, pH, stirring rate, and aeration rate. However, in complex biological reaction processes, there are often interactive influences and game-theoretic relationships between the various regulatory parameters. Simply adjusting one parameter may trigger secondary shifts in other parameters, leading to a worsening of the problem. Therefore, when selecting response strategies, the game equilibrium decision-making unit 15 introduces the idea of ​​game theory, regards the abnormal response process as a multi-party game, and seeks the optimal decision-making scheme of equilibrium by weighing the interests of all parties and analyzing the benefits and costs of strategy combinations, and determines the response adjustment strategy.

[0033] Through the game equilibrium decision-making unit 15, the abnormal reaction indicators are systematically analyzed and equilibrium is solved. The optimal decision scheme is quickly locked in from a large number of strategy combinations. It takes into account both the solution of local problems and the optimization of the global process, thereby improving the flexibility, efficiency and robustness of responding to anomalies.

[0034] The reaction monitoring and management unit 16 is used to monitor and manage the target biological reaction in response to the reaction vessel's reaction regulation strategy and, in conjunction with the soft monitoring module, to perform feedback monitoring and analysis.

[0035] Specifically, the reaction monitoring and management unit 16 is connected to the game equilibrium decision-making unit 15 and receives its output reaction regulation strategy. The reaction regulation strategy includes a series of optimized control instructions for abnormal reactions, specifying the key parameters that need to be adjusted, as well as the direction and magnitude of the adjustment. The reaction monitoring and management unit 16 parses the reaction regulation strategy into a series of executable control commands and processes them through interfaces such as digital / analog conversion, signal amplification, and isolation protection, distributing the control commands to various actuators in the bioreactor, such as heaters, coolers, pH adjustment pumps, and aeration devices. These actuators precisely adjust the relevant parameters of the reaction vessel according to the received control commands, ensuring the reaction process operates according to the guidance of the optimized strategy. Simultaneously with strategy execution, the reaction monitoring and management unit 16 collects various feedback signals from the reaction vessel in real time, such as temperature, pH value, and dissolved oxygen level, and transmits them to the soft monitoring module. The soft monitoring module uses this feedback data to dynamically update its internal mathematical model and state estimation, tracking and evaluating the effect of strategy execution. The soft monitoring module feeds back the evaluation results to the reaction monitoring and management unit 16, including the actual changes in each key parameter and the degree of elimination of abnormal states. Based on the feedback information, the reaction monitoring and management unit 16 judges the effectiveness and completion of the current strategy execution, ensuring that the reaction process has returned to normal and that the key quality indicators meet the predetermined requirements.

[0036] The reaction monitoring and management unit 16 enables real-time monitoring and analysis, dynamic strategy optimization, and continuous feedback adjustment of the target bioreaction, thereby improving the operational efficiency and product quality of the bioreaction process.

[0037] Furthermore, embodiments of this application also include:

[0038] By traversing the aforementioned reaction process, and based on the process steps, a first division result is determined;

[0039] By traversing the aforementioned reaction processes, and based on biological reaction patterns and growth and metabolic trends, a second division result is determined;

[0040] The first division result and the second division result are combined to determine the multi-reaction stage.

[0041] In one feasible implementation, the reaction stage segmentation unit 11 first traverses the complete reaction process of the target bioreaction. By analyzing each step in the process flow, such as feeding, heating, stirring, cooling, and sampling, it identifies the transition nodes and durations between different steps, thereby obtaining a preliminary segmentation of the bioreaction process and forming a first segmentation result. Then, the reaction stage segmentation unit 11 traverses the reaction process again, focusing on analyzing the inherent regularity of the bioreaction and the dynamic characteristics of microbial growth and metabolism. By reviewing relevant literature and historical data, the reaction stage segmentation unit 11 summarizes the general rules of this type of bioreaction, such as the lag phase, logarithmic growth phase, stationary phase, and decline phase, as well as the typical trends in the quantity, activity, and material transformation of microbial populations at different stages. Based on this information, the reaction stage segmentation unit 11 performs a second segmentation of the reaction process from a biological perspective, obtaining a second segmentation result. The second segmentation result reflects the inherent stage nature of the bioreaction process, and its segmentation nodes and durations may not completely overlap with the process steps. Finally, the reaction stage segmentation unit 11 comprehensively considers the first and second segmentation results, conducting a deeper comparison and integration of the two results. By analyzing the correspondence between process steps and biological reaction stages, the commonalities and differences between the two division results are identified. The reaction stage division unit 11 combines the reaction process with biological laws to determine a unified, multi-level reaction stage division scheme that includes multiple reaction stages.

[0042] By fully utilizing information from two different sources—process information and biological knowledge—this approach considers both the external operational characteristics of the reaction process and the intrinsic changing patterns of the biological system. Compared to single-dimensional classification methods, this integrated classification strategy yields more comprehensive, reasonable, and precise results in classifying reaction stages, laying a solid foundation for achieving intelligent phased monitoring and control.

[0043] Furthermore, embodiments of this application also include:

[0044] Based on the first reaction stage, the stage monitoring parameters were determined as the dominant variables, and a set of directly monitored dominant variables and two sets of indirectly monitored dominant variables were divided.

[0045] Traverse the two sets of dominant variables, determine auxiliary variables based on variable correlation, and explore the linear relationship between variables. The auxiliary variables include at least one item.

[0046] By integrating the first set of dominant variables with the linear relationship between the variables based on the second set of dominant variables, training samples are obtained, and supervised training for the first reaction stage is carried out based on the sample-driven approach.

[0047] In a preferred embodiment, the variable relationship mining unit 12 first analyzes the process characteristics and key influencing factors of the first reaction stage given by the reaction stage segmentation unit 11 to obtain key monitoring parameters that represent the biological reaction state of this stage. These parameters are defined as stage monitoring parameters and used as the dominant variables for establishing the soft monitoring model. Next, based on the physicochemical properties of each dominant variable and the available monitoring methods, the variable relationship mining unit 12 divides the dominant variables into two groups. One group consists of variables that can be directly monitored by online sensors or other devices, such as temperature, pH, and dissolved oxygen. The other group consists of variables that are difficult to monitor directly online, such as cell concentration, substrate concentration, and product concentration. For the two groups of dominant variables, since real-time monitoring data cannot be directly obtained, the variable relationship mining unit 12 iterates through each variable and analyzes its correlation with the first group of dominant variables and other process parameters. Parameters that have a significant linear correlation with the target variable are identified and used as auxiliary variables for that variable. These auxiliary variables can be elements from the first group of dominant variables or other process parameters that are easily monitored online. Unit 12, which mines the relationship between variables, uses mathematical statistics methods, such as partial correlation analysis and stepwise regression, to quantitatively characterize the linear relationship between the dominant variable and the auxiliary variable.

[0048] After obtaining real-time monitoring data for a set of dominant variables and the linear relationships between the two sets of dominant variables and their auxiliary variables, the variable relationship mining unit 12 integrates the two to construct a complete training sample for soft monitoring. Subsequently, based on the training sample, the variable relationship mining unit 12 uses appropriate machine learning algorithms, such as support vector machines and neural networks, to conduct supervised training of the soft monitoring model. Through the sample data-driven training process, the soft monitoring model can automatically learn and optimize the complex nonlinear relationships between each dominant variable and auxiliary variable, forming an intelligent measurement for the first reaction stage.

[0049] By training a soft monitoring model, the problem of difficulty in online monitoring of some key variables in biological reaction processes was solved, enabling online estimation of difficult-to-measure variables using easily measurable variables, thus expanding the online monitoring capabilities of biological processes.

[0050] Furthermore, embodiments of this application also include:

[0051] Based on the aforementioned set of dominant and auxiliary variables, determine the variable insensitivity coefficient;

[0052] Based on the insensitivity coefficient, a variable loss ratio is set, which is the threshold of the difference between the observed data and the predicted output data of each sample in the sample data. The insensitivity coefficient corresponds one-to-one with the variable.

[0053] If the difference between the observed data and the predicted output data of a sample is less than the variable loss ratio, the sample is determined to have no loss.

[0054] When the observed data and predicted output data of a sample are greater than or equal to the variable loss ratio, the vector value based on the variable loss ratio will be used as the loss compensation amount.

[0055] In a preferred embodiment, the variable relationship mining unit 12 introduces an insensitivity loss function when constructing the linear relationship between dominant and auxiliary variables. This function quantifies the impact of variable measurement errors on the model output, thereby improving the model's robustness. To obtain this insensitivity loss function, the variable relationship mining unit 12 first calculates the insensitivity coefficient for each variable based on a predetermined set of dominant and auxiliary variables. The insensitivity coefficient reflects the sensitivity of the model output to the measurement error of that variable; the larger the value, the stronger the model's dependence on that variable, and the greater the impact of the measurement error. Then, the variable relationship mining unit 12 sets a corresponding variable loss ratio based on the insensitivity coefficients of each variable. The variable loss ratio is defined as the threshold value of the ratio of the absolute value of the difference between the observed value of a variable and the model's predicted output value in each sample data to the observed amplitude of that variable. The variable loss ratio corresponds one-to-one with the insensitivity coefficient; the larger the insensitivity coefficient, the smaller the loss ratio is set, thus enabling focused monitoring of sensitive variables. During the model validation phase, the variable relationship mining unit 12 iterates through each sample data, calculates the percentage difference between the observed value and the model output value of each variable, and compares it with a preset variable loss ratio. If the percentage is less than the loss ratio, it is determined that the variable has no significant loss on that sample, and the corresponding loss function value is 0; if the percentage is greater than or equal to the loss ratio, the portion exceeding the loss ratio is used as the loss compensation amount for that variable, and the corresponding loss function value is a measure of this compensation amount, such as absolute value, squared value, etc. Through the obtained insensitive loss function, the errors of variables that have a significant impact on model performance can be accurately captured, and their loss contribution can be amplified through nonlinear transformation, while slight errors of insensitive variables are ignored. Embedding this loss function into the model training objective makes the model more robust in dealing with measurement noise and errors, reducing the impact of measurement errors of certain key variables on the model output.

[0056] Furthermore, embodiments of this application also include:

[0057] Traverse the dominant variables to determine the point monitoring variables and continuous monitoring variables, wherein the continuous monitoring variables are labeled with time zone identifiers;

[0058] The continuous monitoring variables are traversed, and the corresponding auxiliary variables are combined to perform long short-term memory compensation learning on the soft monitoring module.

[0059] In a preferred embodiment, after completing the linear relationship mining of dominant and auxiliary variables and training the soft monitoring module, the variable relationship mining unit 12 performs supplementary learning of the time-series characteristics of the trained soft monitoring module based on the dynamic characteristics of each variable. To this end, the variable relationship mining unit 12 first traverses all dominant variables and classifies them into two categories based on their measurement characteristics and response requirements: point monitoring variables and continuous monitoring variables. Point monitoring variables refer to those variables for which only instantaneous values ​​need to be obtained at certain key moments, such as temperature and pH values ​​during key process step transitions; continuous monitoring variables refer to those variables for which their dynamic trends need to be continuously tracked, such as biomass concentration and substrate consumption rate. For continuous monitoring variables, the variable relationship mining unit 12 labels their time zone, characterizing the variable's span characteristics on the time scale.

[0060] After determining the classification of the monitored variables, the variable relationship mining unit 12 further traverses all continuous monitored variables and, in conjunction with their corresponding auxiliary variables, performs long short-term memory compensation learning on the trained soft monitoring module. Specifically, the variable relationship mining unit 12 takes the sequence data of the continuous monitored variables and their auxiliary variables within a certain historical period as input, and trains the model to learn and extract the evolutionary patterns and related characteristics of the variables at different time scales by introducing a long short-term memory neural network unit into the model. Different types of continuous monitored variables have different time scales and memory periods. For example, for variables reflecting cell growth status, such as organism concentration, it is necessary to trace their trends over a longer period to accurately grasp the current state; while for process parameters with relatively short response periods, only historical data from the most recent period is needed. The variable relationship mining unit 12 matches the optimal memory time window for different continuous monitored variables using time zone identifiers, and then conducts targeted long short-term memory compensation training accordingly.

[0061] By performing long short-term memory compensation learning, a soft monitoring module that integrates temporal feature extraction capabilities is obtained. This module can not only accurately characterize the nonlinear mapping relationship between dominant and auxiliary variables, but also adaptively mine and utilize the dynamic temporal features of variables, exhibiting better predictive performance in continuous monitoring tasks.

[0062] Furthermore, embodiments of this application also include:

[0063] The abnormal response identifiers are traversed to determine the adjustment strategy set, which includes an initial strategy and an extended strategy.

[0064] By traversing the abnormal response identifiers and combining the correlation of abnormal response parameters with multi-level sub-divisions, a multi-level game tree is determined.

[0065] Based on the set of adjustment strategies and the multi-level game tree, the decision determines the reaction adjustment strategy.

[0066] In one feasible implementation, the game equilibrium decision-making unit 15 first iterates through the abnormal reaction identifiers sent by the data analysis and screening unit 14. Based on attributes such as identifier type and severity, it matches and filters a series of targeted adjustment strategies from a pre-set strategy library to form initial strategies, which are then incorporated into the adjustment strategy set. Simultaneously, the game equilibrium decision-making unit 15 expands and derives variant or combined strategies based on the characteristics of each initial strategy, and incorporates them into the adjustment strategy set, forming a complete candidate strategy space.

[0067] After obtaining the set of adjustment strategies, the game equilibrium decision unit 15 traverses the abnormal reaction identifiers, analyzes the dynamic correlation characteristics between the key parameters involved in the abnormal reaction, and performs multi-level subdivision of the abnormal reaction process accordingly. Specifically, the game equilibrium decision unit 15 extracts the first-level parameter group of the abnormal reaction based on the spatiotemporal correlation of the key parameters; then, within each first-level parameter group, it further explores higher-dimensional parameter correlation patterns to extract second-level, third-level, and even lower-level parameter groups. Through nested correlation deconstruction, the game equilibrium decision unit 15 divides the abnormal reaction into several multi-level, tree-like parameter game units, forming a complete multi-level game tree.

[0068] After generating the adjustment strategy set and the multi-level game tree, the game equilibrium decision unit 15 begins to make dynamic game decisions on the adjustment strategy based on both. Specifically, the game equilibrium decision unit 15 first maps each strategy in the adjustment strategy set to the corresponding node in the multi-level game tree, and assigns a corresponding utility function to each strategy node according to attributes such as strategy complexity and scope of influence. Then, the game equilibrium decision unit 15 performs layer-by-layer deduction and analysis on the multi-level game tree, examines the trade-offs of different strategy combinations in each game unit, and uses game equilibrium solving tools such as Nash equilibrium and Pareto optimality to determine the locally optimal strategy for each game unit. Afterwards, the game equilibrium decision unit 15 integrates the locally equilibrium strategies of each game unit, and through cross-level trade-offs and vertical coordination, finally determines the globally optimal response adjustment strategy.

[0069] By using intelligent regulation of abnormal responses based on dynamic game theory of strategy sets, and fully considering the multi-parameter interaction characteristics of abnormal responses, the regulation strategy achieves intelligent trade-offs among different temporal and spatial scales and different optimization objectives, thereby improving the effectiveness and global optimality of abnormal regulation.

[0070] Furthermore, embodiments of this application also include:

[0071] The multi-level game tree is initialized based on the set of adjustment strategies.

[0072] Based on the first-level game layer of the initialized multi-level game tree and combined with the revenue function, determine the first-level strategy points, where the first-level game layer is determined from bottom to top;

[0073] Taking the first-level strategy points as a benchmark, conduct multi-level games from bottom to top to determine the reaction adjustment strategy.

[0074] In a preferred embodiment, first, based on the generated adjustment strategy set, initialize and configure the multi-level game tree. Traverse each strategy in the adjustment strategy set and map it to different hierarchical nodes of the multi-level game tree according to its complexity and abstraction level. Among them, the more abstract and comprehensive overall strategies are placed on the higher-level strategy nodes, while the more specific and targeted sub-strategies are merged into the relatively lower-level strategy nodes. Through the hierarchical strategy configuration, the gradual expansion from coarse-grained regulation to fine-grained regulation is realized, and a hierarchical decision-making framework of the multi-level game tree is initially established. After completing the initialization of the multi-level game tree, start to deduce the game decision-making process from bottom to top. First, focus on the bottom layer of the game tree, that is, the first-level game layer. Within this layer, examine the specific regulation sub-strategies attached to each bottom node respectively, and based on the historical data and expert experience of the reaction process, set a revenue function for each sub-strategy node. The revenue function quantitatively evaluates the expected utility that can be achieved by different combinations of regulation behaviors and is the basis for carrying out strategy game analysis. By solving the Nash equilibrium points of the strategies of each node, determine the optimal strategy selection within each first-level game unit, and mark these local optimal solutions as first-level strategy points.

[0075] After determining the first-level strategy points, push forward the multi-level game decision-making layer by layer. For each high-level game node, regard it as a multi-party game problem composed of its subordinate sub-trees. By feedback-transmitting the first-level strategy points and their utility information at the lower layer to the current layer, accurately evaluate the comprehensive revenue of different overall strategies, and use methods such as Stackelberg game and analytic hierarchy process to determine the equilibrium regulation plan of each game node. Repeat this iteration until reaching the root node of the game tree, finally determine the globally optimal top-level regulation strategy, and expand it layer by layer from top to bottom with the equilibrium point strategies of each layer to form a complete multi-level linkage regulation plan as the reaction adjustment strategy.

[0076] Through the local equilibrium solution from bottom to top and the global collaborative optimization from top to bottom, the dynamic weighing and integrated decision-making of regulation behaviors are realized at multiple strategy granularity levels, fully utilizing the information and resources of different decision-making levels, achieving a system-level equilibrium between local revenue and overall effectiveness, and thus obtaining a comprehensive improvement in comprehensive performance.

[0077] In summary, a biological reaction intelligent monitoring system provided by an embodiment of the present application has the following technical effects:

[0078] The reaction stage segmentation unit is used to delineate and define multiple reaction stages in the interactive target biological reaction process, providing a more refined process description for subsequent variable analysis and monitoring control. The variable relationship mining unit is used to traverse multiple reaction stages, identify dominant and auxiliary variables, mine linear relationships between variables, coordinate insensitive loss functions, and supervise the training of the soft monitoring module to improve its accuracy. The continuous monitoring and acquisition unit is used to traverse dominant and auxiliary variables, configure monitoring equipment groups for directly monitored variables, and continuously monitor the reaction process to determine reaction monitoring data, improving the timeliness and completeness of data acquisition. The data analysis and screening unit is used for front-end preprocessing of reaction monitoring data, which is then fed back to the soft monitoring module for linear analysis and abnormal reaction localization to determine valid monitoring data. Valid monitoring data includes stage variable data and abnormal reaction identifiers; the unit filters out valid monitoring data containing both stage variable data and abnormal reaction identifiers, providing a reliable basis for subsequent decision-making and control. The game equilibrium decision-making unit is used to traverse valid monitoring data, perform game equilibrium decisions on abnormal reaction identifiers, determine reaction regulation strategies, and achieve intelligent abnormal reaction control through multi-objective optimization and equilibrium analysis. The reaction monitoring and management unit is used to respond to the reaction vessel's reaction adjustment strategy and, in conjunction with the soft monitoring module, performs feedback monitoring and analysis to monitor and manage the target biological reaction. This enables full-process monitoring and management of the target biological reaction, improves the accuracy of biological reaction monitoring, achieves real-time online monitoring, and provides intelligent control for abnormal situations.

[0079] Example 2

[0080] Based on the same inventive concept as the intelligent biological reaction monitoring system in the foregoing embodiments, such as Figure 2 As shown in the embodiment of this application, a method for intelligent monitoring of biological responses is provided, the method comprising:

[0081] The reaction process of interactive target biological reactions is divided and defined into multiple reaction stages;

[0082] The process involves traversing the multiple reaction stages, identifying dominant and auxiliary variables, exploring linear relationships between variables, coordinating insensitive loss functions, and supervising the training of the soft monitoring module.

[0083] Traverse the dominant variable and the auxiliary variable, configure the monitoring device group for the directly monitored variable, continuously monitor it along with the reaction process, determine the reaction monitoring data, and identify the stage nodes in the reaction monitoring data;

[0084] The front-end preprocesses the reaction monitoring data and sends it back to the soft monitoring module for linear analysis and abnormal reaction localization to determine the valid monitoring data, which includes stage variable data and abnormal reaction identifiers.

[0085] By traversing the valid monitoring data, a game-theoretic equilibrium decision is made on the abnormal reaction indicators to determine the reaction regulation strategy;

[0086] The reaction vessel responds to the reaction regulation strategy and, in conjunction with the soft monitoring module, performs feedback monitoring and analysis to monitor and manage the target biological reaction.

[0087] Furthermore, the division to determine multiple reaction stages includes:

[0088] By traversing the aforementioned reaction process, and based on the process steps, a first division result is determined;

[0089] By traversing the aforementioned reaction processes, and based on biological reaction patterns and growth and metabolic trends, a second division result is determined;

[0090] The first division result and the second division result are combined to determine the multi-reaction stage.

[0091] Furthermore, the supervised training soft monitoring module includes:

[0092] Based on the first reaction stage, the stage monitoring parameters were determined as the dominant variables, and a set of directly monitored dominant variables and two sets of indirectly monitored dominant variables were divided.

[0093] Traverse the two sets of dominant variables, determine auxiliary variables based on variable correlation, and explore the linear relationship between variables. The auxiliary variables include at least one item.

[0094] By integrating the first set of dominant variables with the linear relationship between the variables based on the second set of dominant variables, training samples are obtained, and supervised training for the first reaction stage is carried out based on the sample-driven approach.

[0095] Furthermore, obtaining an insensitive loss function includes:

[0096] Based on the aforementioned set of dominant and auxiliary variables, determine the variable insensitivity coefficient;

[0097] Based on the insensitivity coefficient, a variable loss ratio is set, which is the threshold of the difference between the observed data and the predicted output data of each sample in the sample data. The insensitivity coefficient corresponds one-to-one with the variable.

[0098] If the difference between the observed data and the predicted output data of a sample is less than the variable loss ratio, the sample is determined to have no loss.

[0099] When the observed data and predicted output data of a sample are greater than or equal to the variable loss ratio, the vector value based on the variable loss ratio will be used as the loss compensation amount.

[0100] Furthermore, after the supervised training soft monitoring module, the system includes:

[0101] Traverse the dominant variables to determine the point monitoring variables and continuous monitoring variables, wherein the continuous monitoring variables are labeled with time zone identifiers;

[0102] The continuous monitoring variables are traversed, and the corresponding auxiliary variables are combined to perform long short-term memory compensation learning on the soft monitoring module.

[0103] Furthermore, determining the response regulation strategy includes:

[0104] The abnormal response identifiers are traversed to determine the adjustment strategy set, which includes an initial strategy and an extended strategy.

[0105] By traversing the abnormal response identifiers and combining the correlation of abnormal response parameters with multi-level sub-divisions, a multi-level game tree is determined.

[0106] Based on the set of adjustment strategies and the multi-level game tree, the decision determines the reaction adjustment strategy.

[0107] Furthermore, based on the set of adjustment strategies and the multi-level game tree, the decision to determine the response adjustment strategy includes:

[0108] The multi-level game tree is initialized based on the set of adjustment strategies.

[0109] Based on the first-level game layer of the initialized multi-level game tree, and combined with the payoff function, the first-level strategy point is determined, wherein the first-level game layer is determined from bottom to top.

[0110] Using the first-level strategy point as a benchmark, a multi-level game is conducted from bottom to top to determine the reaction adjustment strategy.

[0111] In summary, any step of the method described above can be stored as a computer instruction or program in an unrestricted computer memory, and can be called and identified by an unrestricted computer processor to implement any method in the embodiments of this application, without any additional restrictions.

[0112] Furthermore, the "first" or "second" mentioned above may not only represent a sequential relationship, but may also represent a specific concept, and / or refer to the individual or collective selection of multiple elements. Clearly, those skilled in the art can make various modifications and variations to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A biological reaction intelligent monitoring system, characterized in that, The system includes: Reaction stage division unit, used in reaction processes for interactive target biological reactions, to divide and define multiple reaction stages; The variable relationship mining unit is used to traverse multiple reaction stages, determine the dominant and auxiliary variables, mine the linear relationships between variables, coordinate the insensitive loss function, and supervise the training of the soft monitoring module. The continuous monitoring and acquisition unit is used to traverse the dominant and auxiliary variables, configure the monitoring equipment group for the directly monitored variables, and continuously monitor the reaction process to determine the reaction monitoring data. The data analysis and filtering unit is used to preprocess the reaction monitoring data at the front end and send it back to the soft monitoring module for linear analysis and abnormal reaction localization to determine the effective monitoring data. The effective monitoring data includes stage variable data and abnormal reaction identifiers. The game equilibrium decision unit is used to traverse the effective monitoring data, make game equilibrium decisions on abnormal response indicators, and determine the response adjustment strategy. The reaction monitoring and management unit is used to monitor and manage the target biological reaction in response to the reaction vessel's response to the reaction regulation strategy, and is combined with the soft monitoring module to perform feedback monitoring and analysis. The determined response regulation strategy includes: The abnormal response identifiers are traversed to determine the adjustment strategy set, which includes an initial strategy and an extended strategy. By traversing the abnormal response identifiers and combining the correlation of abnormal response parameters with multi-level sub-divisions, a multi-level game tree is determined. Based on the set of adjustment strategies and the multi-level game tree, the decision is made to determine the reaction adjustment strategy; Based on the set of adjustment strategies and the multi-level game tree, the decision-making process for determining the response adjustment strategy includes: The multi-level game tree is initialized based on the set of adjustment strategies. Based on the first-level game layer of the initialized multi-level game tree, and combined with the payoff function, the first-level strategy point is determined, wherein the first-level game layer is determined from bottom to top. Using the first-level strategy point as a benchmark, a multi-level game is conducted from bottom to top to determine the reaction adjustment strategy.

2. The intelligent biological reaction monitoring system as described in claim 1, characterized in that, The division determines multiple reaction stages, including: By traversing the aforementioned reaction process, and based on the process steps, a first division result is determined; By traversing the aforementioned reaction processes, and based on biological reaction patterns and growth and metabolic trends, a second division result is determined; The first division result and the second division result are combined to determine the multi-reaction stage.

3. The intelligent biological reaction monitoring system as described in claim 1, characterized in that, The supervised training soft monitoring module includes: Based on the first reaction stage, the stage monitoring parameters were determined as the dominant variables, and a set of directly monitored dominant variables and two sets of indirectly monitored dominant variables were divided. Traverse the two sets of dominant variables, determine auxiliary variables based on variable correlation, and explore the linear relationship between variables. The auxiliary variables include at least one item. By integrating the first set of dominant variables with the linear relationship between the variables based on the second set of dominant variables, training samples are obtained, and supervised training for the first reaction stage is carried out based on the sample-driven approach.

4. The intelligent biological reaction monitoring system as described in claim 3, characterized in that, Obtaining an insensitive loss function includes: Based on the aforementioned set of dominant and auxiliary variables, determine the variable insensitivity coefficient; Based on the insensitivity coefficient, a variable loss ratio is set. The variable loss ratio is a threshold value that is the ratio of the absolute value of the difference between the observed data and the predicted output data of each sample in the sample data to the observed amplitude of the variable. The insensitivity coefficient corresponds one-to-one with the variable. If the percentage difference between the observed data and the predicted output data of the sample is less than the variable loss ratio, the sample is determined to have no loss. When the percentage difference between the observed data and the predicted output data of the sample is greater than or equal to the variable loss ratio, the vector value based on the variable loss ratio will be used as the loss compensation amount.

5. The intelligent biological reaction monitoring system as described in claim 1, characterized in that, Following the supervised training soft monitoring module, the system includes: Traverse the dominant variables to determine the point monitoring variables and continuous monitoring variables, wherein the continuous monitoring variables are labeled with time zone identifiers; The continuous monitoring variables are traversed, and the corresponding auxiliary variables are combined to perform long short-term memory compensation learning on the soft monitoring module.

6. A method for intelligent monitoring of biological reactions, characterized in that, The method includes: The reaction process of interactive target biological reactions is divided and defined into multiple reaction stages; The process involves traversing the multiple reaction stages, identifying dominant and auxiliary variables, exploring linear relationships between variables, coordinating insensitive loss functions, and supervising the training of the soft monitoring module. Traverse the dominant variable and the auxiliary variable, configure the monitoring device group for the directly monitored variable, continuously monitor it along with the reaction process, determine the reaction monitoring data, and identify the stage nodes in the reaction monitoring data; The front-end preprocesses the reaction monitoring data and sends it back to the soft monitoring module for linear analysis and abnormal reaction localization to determine the valid monitoring data, which includes stage variable data and abnormal reaction identifiers. By traversing the valid monitoring data, a game-theoretic equilibrium decision is made on the abnormal reaction indicators to determine the reaction regulation strategy; The reaction vessel responds to the reaction regulation strategy and, in conjunction with the soft monitoring module, performs feedback monitoring and analysis to monitor and manage the target biological reaction. The abnormal response identifiers are traversed to determine the adjustment strategy set, which includes an initial strategy and an extended strategy. By traversing the abnormal response identifiers and combining the correlation of abnormal response parameters with multi-level sub-divisions, a multi-level game tree is determined. Based on the set of adjustment strategies and the multi-level game tree, the decision is made to determine the reaction adjustment strategy; Based on the set of adjustment strategies and the multi-level game tree, the decision-making process for determining the response adjustment strategy includes: The multi-level game tree is initialized based on the set of adjustment strategies. Based on the first-level game layer of the initialized multi-level game tree, and combined with the payoff function, the first-level strategy point is determined, wherein the first-level game layer is determined from bottom to top. Using the first-level strategy point as a benchmark, a multi-level game is conducted from bottom to top to determine the reaction adjustment strategy.

Images