Automatic control method and system based on microbiological detection

By combining frequency scanning technology and deep learning model, the problem of insufficient automation level of microbial detection technology is solved, and the accurate identification and quantification of microbial types and concentrations is achieved, the detection efficiency and accuracy are improved, and the multi-layer graph neural network and Fuzzy PID control optimizes the cultivation conditions, improving the intelligence and automation level of detection.

CN118932011AInactive Publication Date: 2024-11-12THE FIRST AFFILIATED HOSPITAL OF SHANTOU UNIV MEDICAL COLLEGE
View PDF 0 Cites 12 Cited by

Patent Information

Application Number
CN202411037867.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-31
Publication Date
2024-11-12
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The automation level of existing microbial detection technology is insufficient, the selection of environmental and state monitoring parameters is not comprehensive, the data lacks unified standards and real-time aggregation, the accuracy and generalization of colony image segmentation and recognition algorithms need to be improved, and the flexibility and coordination of robot control systems are insufficient , Expert experience is difficult to formally express, and knowledge-driven intelligent analysis capabilities are limited.

Method used

Frequency scanning technology is used to measure the impedance parameters of the culture medium, and the characteristic parameters of the impedance signal are obtained through dimensionality reduction and feature extraction, and input them into the pre-constructed microbial metabolic response model. Combining convolutional neural network and long and short-term memory network for microbial type and concentration prediction. , comprehensively considering the type of microbial and pathogenic probability, risk assessment is carried out through hierarchical analysis method, and adaptively adjusting the culture strategy, introducing multi-layer graph neural network and fuzzy PID control algorithm to optimize the culture conditions.

Benefits of technology

Accurate identification and quantification of microbial types and concentrations are achieved, the intelligence and automation of detection are improved, the adaptability to complex growth conditions is enhanced, and the detection efficiency and accuracy are improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118932011A_ABST
    Figure CN118932011A_ABST
Patent Text Reader

Abstract

The invention provides an automatic control method and system based on microbiological detection, and relates to the technical field of microbiological detection.The automatic control method comprises the steps that impedance parameters of a culture medium are measured through the frequency scanning technology at different frequencies, characteristic parameters of impedance signals are obtained and input into a pre-constructed microbiological metabolism response model, and a microbiological metabolism response model is obtained; predicting the type and concentration of microorganisms; judging whether the culture medium contains target microorganisms and whether the concentration of the target microorganisms exceeds the standard or not according to the prediction result of the microbial metabolism response model, and if the comprehensive risk assessment result shows that the culture medium contains the target microorganisms and the content of the target microorganisms exceeds the standard, determining the target microorganisms according to the type of the target microorganisms and the concentration of the target microorganisms. A microorganism culture strategy is adjusted in a self-adaptive mode, if the re-detected target microorganism concentration is reduced to be lower than a standard threshold value, detection data are fed back to a control system, a self-adaptive optimization algorithm is adopted to optimize the microorganism metabolism response model and the detection strategy, and microorganisms are automatically detected.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to microbial detection technology, and in particular to an automated control method and system based on microbial detection. Background Art

[0002] Microbial testing has widespread applications in food safety, environmental monitoring, clinical diagnosis, and other fields. Traditional microbial testing methods rely primarily on manual manipulation and empirical judgment, resulting in low efficiency, strong subjectivity, and poor reproducibility. To improve the automation and intelligence of microbial testing, researchers at home and abroad have conducted a series of explorations and practices.

[0003] At present, the automated control technology for microbial detection mainly includes the following aspects: First, the use of Internet of Things technology to realize real-time monitoring of the microbial culture environment and growth status, such as automatic collection and adjustment of parameters such as temperature, humidity, pH value, and dissolved oxygen content; second, the application of machine vision and image processing technology to automatically identify and count the morphology and number of microbial colonies; third, the use of robots and fluid control systems to realize the automation of sample preparation, inoculation, culture and other operations; fourth, based on big data analysis and knowledge base, microbial growth models and diagnostic rules are constructed to assist manual analysis and interpretation of results.

[0004] Although the above technologies have improved the automation level of microbial testing to a certain extent, some shortcomings still exist: First, the selection of parameters for environmental and state monitoring is incomplete, and data lacks unified standards and real-time aggregation. Second, the accuracy and generalization of colony image segmentation and recognition algorithms need to be improved, making it difficult to cope with complex and changing growth conditions. Third, the flexibility and coordination of robotic control systems are insufficient, making them unable to adapt to personalized testing needs. Fourth, expert experience is difficult to formalize, and the knowledge-driven intelligent analysis capabilities are limited. Therefore, there is an urgent need to develop a new generation of automated control methods and systems for microbial testing.

[0005] This invention aims to draw on the latest developments in the field of artificial intelligence, and to address the pain points and needs of automated control of microbial detection, innovatively proposes an automated control method and system based on microbial detection that integrates intelligent perception, intelligent analysis, intelligent decision-making, and intelligent control. Summary of the Invention

[0006] The embodiments of the present invention provide an automated control method and system based on microbial detection, which can solve the problems in the prior art.

[0007] According to a first aspect of the embodiments of the present invention,

[0008] Provide automated control methods based on microbial detection, including:

[0009] Frequency sweep technology is used to measure the impedance parameters of the culture medium at different frequencies to obtain complete impedance spectrum information. The obtained impedance spectrum information is then subjected to dimensionality reduction and feature extraction to obtain characteristic parameters of the impedance signal. The obtained characteristic parameters are input into a pre-built microbial metabolic response model to predict the type and concentration of microorganisms.

[0010] Based on the predictions from the microbial metabolic response model, a determination is made as to whether the culture medium contains the target microorganism and whether the concentration of the target microorganism exceeds the standard. During this determination, the microbial type and pathogenicity probability are comprehensively considered, and a comprehensive risk assessment result is derived using the analytic hierarchy process. If the comprehensive risk assessment result indicates that the culture medium contains the target microorganism and the concentration exceeds the standard, the microbial culture strategy is adaptively adjusted based on the target microorganism type and concentration, and the culture medium is re-cultured.

[0011] The target microorganism and target microorganism concentration detection steps are performed again on the re-cultured culture medium. If the re-detected target microorganism concentration has dropped below the standard threshold, the detection data is fed back to the control system, and an adaptive optimization algorithm is used to optimize the microbial metabolic response model and detection strategy to automatically detect the microorganisms.

[0012] In an alternative embodiment,

[0013] Frequency sweep technology is used to measure the impedance parameters of the culture medium at different frequencies to obtain complete impedance spectrum information. The obtained impedance spectrum information is then subjected to dimensionality reduction and feature extraction to obtain characteristic parameters of the impedance signal. The obtained characteristic parameters are input into a pre-built microbial metabolic response model to predict the type and concentration of microorganisms, including:

[0014] Preprocessing the obtained impedance spectrum information, and inputting the preprocessed impedance spectrum information into a pre-built microbial metabolic response model, wherein the microbial metabolic response model is composed of a convolutional neural network and a long short-term memory network;

[0015] A joint loss function is constructed that includes the microbial type identification loss and the microbial concentration estimation loss. By minimizing the joint loss function, the parameters of the microbial metabolic response model are updated. The iteration is repeated until the preset termination condition is met to obtain the final trained microbial metabolic response model.

[0016] Based on the trained microbial metabolic response model, a convolutional neural network is used to extract local patterns and key features of the impedance spectrum information. The feature dimension is then reduced through pooling operations to obtain low-dimensional key features.

[0017] The obtained low-dimensional key features are input into the long short-term memory network for sequence modeling to capture the temporal characteristics of the impedance spectrum information. Deep sequence features are constructed by combining multiple long short-term memory layers. Based on the deep sequence features, the probability distribution of microbial types and the predicted values ​​of concentration are output through the fully connected layer.

[0018] In an alternative embodiment,

[0019] The calculation formula for constructing a joint loss function that includes microbial type identification loss and microbial concentration estimation loss is as follows:

[0020]

[0021] Where L represents the joint loss function, A represents the number of microbial types, and α a represents the weight of microorganism type a, represents the probability of the predicted microbial type a, γ represents the regulatory factor, and y a represents the label of the actual microorganism type a, β represents the weight of the microorganism concentration estimation loss, M represents the number of samples of impedance spectrum information, K represents the number of mixed Gaussian components, ω mk The weight of the sample m representing the impedance spectrum information on the kth mixed Gaussian component, σ k represents the standard deviation of the kth Gaussian mixture component, r m The concentration of sample m, μ, represents the impedance spectrum information k Represents the mean of the kth Gaussian mixture component.

[0022] In an alternative embodiment,

[0023] Based on the prediction results of the microbial metabolic response model, it is determined whether the culture medium contains the target microorganism and whether the concentration of the target microorganism exceeds the standard. In the judgment process, the microbial type and pathogenicity probability are comprehensively considered. The comprehensive risk assessment results obtained through the hierarchical analysis method include:

[0024] Collect structured knowledge in the field of microorganisms and construct a microbial knowledge graph, where nodes represent sets of microbial entities and edges represent sets of relationships between entities;

[0025] Generating random walk sequences as training samples based on the microbial knowledge graph, and training the training samples using a graph embedding model to obtain a low-dimensional embedding representation of each node in the microbial knowledge graph as an initial feature of the node;

[0026] Construct a multi-layer graph neural network model, which includes graph convolutional layers, graph attention layers, and fully connected layers. The graph convolutional layers aggregate and update the initial features of nodes, the graph attention layers adaptively learn the importance weights between nodes, and the fully connected layers perform nonlinear transformations and risk prediction on node features.

[0027] The multi-layer graph neural network model is trained based on historical microbial detection data. By minimizing the error between the predicted risk value and the actual risk value, the model parameters are optimized using a gradient descent algorithm, and an early stopping mechanism and a learning rate scheduling strategy are introduced to ultimately obtain a trained multi-layer graph neural network model.

[0028] For the current test sample, the corresponding subgraph is matched in the microbial knowledge graph and input into the trained multi-layer graph neural network model. The risk score of each microbial node in the subgraph is calculated through forward propagation. The microbial type and pathogenicity probability are comprehensively considered, and the hierarchical analysis method is used to combine the risk scores of each microbial node to obtain a comprehensive risk assessment result.

[0029] In an alternative embodiment,

[0030] If the comprehensive risk assessment results indicate that the culture medium contains target microorganisms and the content exceeds the standard, the microbial culture strategy is adaptively adjusted based on the target microorganism type and target microorganism concentration. The culture medium is re-cultured, including:

[0031] A temperature control model is constructed based on the target microorganism type and target microorganism concentration, and a machine learning algorithm is used to train the temperature control strategy model. The trained temperature control strategy model is used to adaptively generate a temperature control curve and set value based on the real-time temperature of the culture medium and the target microorganism concentration.

[0032] According to the generated temperature control curve and set value, the fuzzy PID control algorithm is introduced. The temperature deviation, deviation change rate and deviation accumulation are comprehensively considered. Through fuzzy reasoning and defuzzification process, the parameters of the PID controller are dynamically adjusted and output as a control signal. The optimal temperature scheduling plan is generated according to the output control signal.

[0033] At the same time, a time scheduling optimization model is designed for multiple culture media, with minimizing the total culture time as the objective function. The culture time range and sequence relationship of each culture medium are defined as constraints. The time scheduling optimization model is solved by the target optimization algorithm to obtain the optimal time scheduling solution.

[0034] Combining the obtained optimal time scheduling plan and optimal temperature scheduling plan, the microbial cultivation strategy is adaptively adjusted and the culture medium is re-cultivated.

[0035] In an alternative embodiment,

[0036] According to the generated temperature control curve and set value, the fuzzy PID control algorithm is introduced. The temperature deviation, deviation change rate and deviation accumulation are comprehensively considered. Through the fuzzy reasoning and defuzzification process, the parameters of the PID controller are dynamically adjusted. The calculation formula for outputting the final control signal is as follows:

[0037]

[0038] Among them, U(t) represents the final control signal, K p represents the proportional coefficient, e(t) represents the temperature deviation at the current moment, K i Indicates the integral coefficient, e i (t) represents the rate of change of deviation, K d represents the differential coefficient, e c (t) represents the cumulative deviation;

[0039] At the same time, a time scheduling optimization model is designed for multiple culture media, with the minimization of total culture time as the objective function, and the culture time range and sequence relationship of each culture medium defined as constraints. The time scheduling optimization model is solved by the target optimization algorithm, including:

[0040] An objective function is constructed based on the amount of culture medium and the total culture time, while introducing a regularization parameter to balance the fitting error of the model and the smoothness of the parameters. The objective function is calculated as follows:

[0041]

[0042] Among them, J represents the objective function, θ represents the parameter vector, N represents the number of culture media, f(t i ; θ) represents the culture time of the culture medium determined by the parameter vector θ at time ti, f * (t i ) represents the target training time at time ti, λ1 represents the regularization coefficient of the equilibrium parameter vector, λ2 represents the regularization coefficient of the gradient of the equilibrium parameter vector, represents the gradient of the parameter vector, ||·|| 2 represents the square norm.

[0043] In an alternative embodiment,

[0044] The method further comprises:

[0045] Introducing smart sensors and IoT devices to collect real-time microbial growth environment parameters and status data, and preprocessing them. Feature extraction is performed on the preprocessed microbial growth environment parameters and status data to obtain microbial growth and metabolic characteristics.

[0046] Based on the growth and metabolic characteristics of microorganisms, and integrating pre-built mechanism models and data-driven models, a digital twin model of microbial growth and metabolism is constructed;

[0047] The obtained digital twin model of microbial growth and metabolism is trained, and the trained digital twin model of microbial growth and metabolism is used to perform real-time predictions of microbial growth trends, metabolites, and key indicators;

[0048] Based on the real-time prediction of microbial growth trends, metabolites, and key indicators, the system detects abnormal conditions during the microbial cultivation process. The detection results and real-time prediction results are fed back to the control system, and the optimal microbial cultivation control strategy is learned through the reinforcement learning algorithm.

[0049] The optimal microbial culture control strategy obtained is applied to the automatic control of microbial culture and detection processes, and human-computer interaction is combined to realize intelligent decision support for microbial detection.

[0050] According to a second aspect of the embodiments of the present invention,

[0051] Provide automated control systems based on microbial detection, including:

[0052] The first unit is used to measure the impedance parameters of the culture medium using frequency sweep technology at different frequencies to obtain complete impedance spectrum information. The obtained impedance spectrum information is then subjected to dimensionality reduction and feature extraction to obtain characteristic parameters of the impedance signal. The obtained characteristic parameters are input into a pre-built microbial metabolic response model to predict the type and concentration of microorganisms.

[0053] The second unit is used to determine whether the culture medium contains the target microorganism and whether the target microorganism concentration exceeds the standard based on the prediction results of the microbial metabolic response model. During the judgment process, the microorganism type and pathogenicity probability are comprehensively considered, and a comprehensive risk assessment result is obtained through the hierarchical analysis method. If the comprehensive risk assessment result indicates that the culture medium contains the target microorganism and the concentration exceeds the standard, the microbial culture strategy is adaptively adjusted according to the target microorganism type and target microorganism concentration, and the culture medium is re-cultured.

[0054] The third unit is used to re-execute the target microorganism and target microorganism concentration detection steps on the re-cultured culture medium. If the re-detected target microorganism concentration has dropped below the standard threshold, the detection data will be fed back to the control system, and an adaptive optimization algorithm will be used to optimize the microbial metabolic response model and detection strategy to perform automated detection of the microorganisms.

[0055] According to a third aspect of the embodiments of the present invention,

[0056] Provides electronic equipment, including:

[0057] processor;

[0058] a memory for storing processor-executable instructions;

[0059] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0060] According to a fourth aspect of the embodiments of the present invention,

[0061] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0062] In this embodiment, by combining impedance spectroscopy technology with a microbial metabolic response model, the types and concentration levels of microorganisms in the culture medium can be effectively identified and quantified. By correlating the type and concentration of microorganisms with the probability of pathogenicity, the comprehensive risk is assessed using the analytic hierarchy process to provide a basis for subsequent processing decisions. When the target microorganism is detected to be above the standard, the system can automatically adjust the culture conditions according to the type and concentration of the microorganism and re-cultivate the culture medium to reduce the microbial level. By feeding the detection data back to the control system, the system can use an adaptive optimization algorithm to continuously optimize the microbial metabolic response model and detection strategy, thereby realizing an automated and intelligent microbial detection process. By combining artificial intelligence technology with microbial detection, the intelligence and automation level of detection are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 Schematic diagram of the process of an automated control method based on microbial detection according to an embodiment of the present invention;

[0064] Figure 2 This is a schematic structural diagram of an automated control system based on microbial detection according to an embodiment of the present invention. DETAILED DESCRIPTION

[0065] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0066] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0067] Figure 1FIG. 1 is a flow chart of an automated control method based on microbial detection according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0068] S101. Use frequency scanning technology to measure the impedance parameters of the culture medium at different frequencies to obtain complete impedance spectrum information, and perform dimensionality reduction and feature extraction on the obtained impedance spectrum information to obtain characteristic parameters of the impedance signal. The obtained characteristic parameters are input into a pre-built microbial metabolic response model to predict the type and concentration of microorganisms.

[0069] For example, frequency sweep technology is used to measure the impedance parameters of the microbial culture medium at different frequencies. Impedance parameters include resistance, capacitance, and relaxation frequency. Alternatively, nanostructured metal electrode arrays can be fabricated using microelectromechanical systems and nanofabrication technologies. Surface modification techniques can be used to manipulate the electrode surface properties, improving the affinity and specific binding between the electrodes and microorganisms. A three-dimensional electrode structure is constructed to increase the effective contact area between the electrodes and the culture medium, thereby improving signal acquisition efficiency. During the measurement process, the electrode array is placed in a constant temperature incubator controlled at 37°C to simulate the environmental conditions for microbial growth. Impedance spectroscopy measurements are performed every 30 minutes for 24 hours to obtain dynamic impedance spectroscopy data during microbial growth.

[0070] The raw data output by an impedance analyzer includes the impedance modulus and phase angle at each frequency point, requiring data preprocessing and conversion. First, the impedance modulus and phase angle are converted into the real and imaginary parts of the impedance. Next, to eliminate dimensional differences between different frequency points, the real and imaginary parts of the impedance are normalized, resulting in the normalized impedance spectrum data.

[0071] Impedance spectroscopy data from microbial culture media is high-dimensional. Directly using raw impedance spectroscopy data for analysis and modeling is not only computationally complex but also susceptible to interference from noise and redundant information. To extract essential features from impedance spectroscopy data and reduce data dimensionality, principal component analysis (PCA) can be used to reduce the dimensionality of the data. PCA projects the raw high-dimensional data into a low-dimensional space through linear transformation, extracting key data variation and reducing noise and redundancy. In addition to PCA dimensionality reduction, several physically meaningful characteristic parameters can also be extracted from impedance spectroscopy data. For example, morphological features such as the slope and curvature of the impedance spectroscopy curve at different frequency bands, as well as the impedance values ​​corresponding to characteristic frequency points of the impedance spectrum (such as the maximum impedance point and relaxation frequency), all contain information about the metabolic activity and growth status of the microorganism. To extract these characteristic parameters, mathematical modeling and parameter fitting can be performed on the impedance spectroscopy curve. Commonly used impedance spectroscopy models include equivalent circuit models (such as the Randles circuit and the Voigt circuit) and empirical models (such as the Cole-Cole equation and the CPE). Through curve fitting, the parameter values ​​of equivalent circuit elements (such as resistance, capacitance, Warburg impedance, etc.) or the characteristic parameters of the empirical model (such as relaxation time constant, CPE index, etc.) can be obtained as another characteristic representation of the impedance spectrum data.

[0072] In an alternative embodiment,

[0073] Frequency sweep technology is used to measure the impedance parameters of the culture medium at different frequencies to obtain complete impedance spectrum information. The obtained impedance spectrum information is then subjected to dimensionality reduction and feature extraction to obtain characteristic parameters of the impedance signal. The obtained characteristic parameters are input into a pre-built microbial metabolic response model to predict the type and concentration of microorganisms, including:

[0074] Preprocessing the obtained impedance spectrum information, and inputting the preprocessed impedance spectrum information into a pre-built microbial metabolic response model, wherein the microbial metabolic response model is composed of a convolutional neural network and a long short-term memory network;

[0075] A joint loss function is constructed that includes the microbial type identification loss and the microbial concentration estimation loss. By minimizing the joint loss function, the parameters of the microbial metabolic response model are updated. The iteration is repeated until the preset termination condition is met to obtain the final trained microbial metabolic response model.

[0076] Based on the trained microbial metabolic response model, a convolutional neural network is used to extract local patterns and key features of the impedance spectrum information. The feature dimension is then reduced through pooling operations to obtain low-dimensional key features.

[0077] The obtained low-dimensional key features are input into the long short-term memory network for sequence modeling to capture the temporal characteristics of the impedance spectrum information. Deep sequence features are constructed by combining multiple long short-term memory layers. Based on the deep sequence features, the probability distribution of microbial types and the predicted values ​​of concentration are output through the fully connected layer.

[0078] For example, based on the characteristics of impedance spectroscopy data, a deep learning model architecture based on the combination of convolutional neural network (CNN) and long short-term memory network (LSTM) is designed for microbial type identification and concentration estimation tasks.

[0079] Specifically, the raw impedance spectroscopy data is preprocessed, including steps such as data cleaning, noise removal, data normalization, and feature scaling. First, outliers and invalid values ​​are removed from the impedance spectroscopy data to ensure data integrity and consistency. Then, methods such as median filtering and wavelet denoising are used to remove high-frequency noise and interference components from the impedance spectroscopy signal, thereby improving the signal-to-noise ratio. Next, the impedance spectroscopy data is subjected to maximum-minimum normalization or Z-score standardization to scale features of different dimensions to the same scale, accelerating model convergence. Finally, according to the model input requirements, the preprocessed impedance spectroscopy data is subjected to feature reorganization and format conversion, such as conversion into a fixed-length time series or spectrum image.

[0080] The preprocessed impedance spectrum time series data is first passed through a one-dimensional convolutional layer (1D-CNN) for local feature extraction. The 1D-CNN captures local patterns and key features of the impedance spectrum signal by sliding the convolution kernel along the time dimension. The convolutional layer is followed by a maximum pooling layer to reduce the feature dimension and extract the most significant feature representations. The pooled feature map is reshaped into a vector and passed to the LSTM layer for sequence modeling. The LSTM layer effectively processes time series data and captures the long-term dependencies and dynamic trends of the impedance spectrum signal. Multiple LSTM layers can be stacked to construct deeper sequence feature representations. The output of the LSTM layer passes through a fully connected layer and is mapped to the final microbial type and concentration prediction results. For the microbial type recognition task, the model output is multi-classified using the Softmax activation function; for the microbial concentration estimation task, the model output is a continuous value representing the predicted concentration level.

[0081] Among them, in order to train the CNN-LSTM model, it is necessary to construct a large-scale impedance spectroscopy dataset, including samples of different microbial types and concentration levels. The K-fold cross-validation method is used to divide the dataset into training set, validation set and test set. The model is trained using the training set, and the validation set is used for model selection and hyperparameter tuning. Finally, the generalization performance of the model is evaluated on the test set. During the training process, the early stopping strategy is adopted to monitor the performance indicators (such as accuracy, mean absolute error, etc.) on the validation set. When the performance no longer improves, the training is stopped in time to prevent the model from overfitting. At the same time, regularization techniques such as L1 / L2 regularization and Dropout are introduced to further improve the generalization ability of the model. Adaptive learning rate optimization algorithms such as Adam and AdaGrad are used to dynamically adjust the learning rate and accelerate model convergence.

[0082] Using a trained CNN-LSTM model, new impedance spectroscopy data is used to predict microbial type and concentration in real time. The preprocessed impedance spectroscopy data is fed into the model, and through a forward propagation process, a probability distribution of microbial type and an estimated concentration are obtained. For microbial type identification tasks, the category with the highest probability is selected as the final prediction result; for microbial concentration estimation tasks, the predicted concentration value is directly output. Optionally, to comprehensively evaluate model performance, multiple evaluation metrics are used, such as accuracy, precision, recall, and F1 value for microbial type identification tasks, and mean absolute error, root mean square error, and coefficient of determination for microbial concentration estimation tasks. By comparing the performance of different models and feature extraction methods, the optimal model configuration and parameter settings are selected.

[0083] In an alternative embodiment,

[0084] The calculation formula for constructing a joint loss function that includes microbial type identification loss and microbial concentration estimation loss is as follows:

[0085]

[0086] Where L represents the joint loss function, A represents the number of microbial types, and α a represents the weight of microorganism type a, represents the probability of the predicted microbial type a, γ represents the regulatory factor, and y a represents the label of the actual microorganism type a, β represents the weight of the microorganism concentration estimation loss, M represents the number of samples of impedance spectrum information, K represents the number of mixed Gaussian components, ω mk The weight of the sample m representing the impedance spectrum information on the kth mixed Gaussian component, σ k represents the standard deviation of the kth Gaussian mixture component, r m The concentration of sample m, μ, represents the impedance spectrum information kRepresents the mean of the kth Gaussian mixture component.

[0087] In this embodiment, a convolutional neural network can automatically learn the local patterns and key features of the impedance spectrum signal. Pooling operations can reduce feature dimensions and reduce computational complexity. By utilizing long-short-term memory networks, sequence data can be effectively modeled, capturing temporal dependencies. Through multi-level stacking, deep sequence feature representations can be learned. By constructing a joint loss function and simultaneously optimizing the two tasks of type recognition and concentration estimation, the final model can simultaneously output the probability distribution of microbial types and predicted concentration values, realizing intelligent analysis of complex impedance spectrum signals and high-precision microbial type recognition and concentration prediction, providing more accurate and intelligent technical support for microbial detection.

[0088] S102. Based on the prediction results of the microbial metabolic response model, determine whether the culture medium contains the target microorganism and whether the concentration of the target microorganism exceeds the standard. During the judgment process, comprehensively consider the type of microorganism and the probability of pathogenicity, and use the hierarchical analysis method to obtain a comprehensive risk assessment result. If the comprehensive risk assessment result indicates that the culture medium contains the target microorganism and the content exceeds the standard, then according to the type of target microorganism and the concentration of target microorganism, adaptively adjust the microbial culture strategy and re-cultivate the culture medium.

[0089] Target microorganisms are specific microbial species that require detection and attention. Based on the specific microbial types and concentration levels detected, the culture medium's culture conditions (such as temperature, humidity, and nutrients) are automatically adjusted to inhibit microbial growth. After adjusting the culture strategy, the culture medium is re-cultured to reduce the microbial concentration.

[0090] In an alternative embodiment,

[0091] Based on the prediction results of the microbial metabolic response model, it is determined whether the culture medium contains the target microorganism and whether the concentration of the target microorganism exceeds the standard. In the judgment process, the microbial type and pathogenicity probability are comprehensively considered. The comprehensive risk assessment results obtained through the hierarchical analysis method include:

[0092] Collect structured knowledge in the field of microorganisms and construct a microbial knowledge graph, where nodes represent sets of microbial entities and edges represent sets of relationships between entities;

[0093] Generating random walk sequences as training samples based on the microbial knowledge graph, and training the training samples using a graph embedding model to obtain a low-dimensional embedding representation of each node in the microbial knowledge graph as an initial feature of the node;

[0094] Construct a multi-layer graph neural network model, which includes graph convolutional layers, graph attention layers, and fully connected layers. The graph convolutional layers aggregate and update the initial features of nodes, the graph attention layers adaptively learn the importance weights between nodes, and the fully connected layers perform nonlinear transformations and risk prediction on node features.

[0095] The multi-layer graph neural network model is trained based on historical microbial detection data. By minimizing the error between the predicted risk value and the actual risk value, the model parameters are optimized using a gradient descent algorithm, and an early stopping mechanism and a learning rate scheduling strategy are introduced to ultimately obtain a trained multi-layer graph neural network model.

[0096] For the current test sample, the corresponding subgraph is matched in the microbial knowledge graph and input into the trained multi-layer graph neural network model. The risk score of each microbial node in the subgraph is calculated through forward propagation. The microbial type and pathogenicity probability are comprehensively considered, and the hierarchical analysis method is used to combine the risk scores of each microbial node to obtain a comprehensive risk assessment result.

[0097] For example, a microbial knowledge graph is first constructed based on the collected and organized structured knowledge in the field of microorganisms, in which nodes represent sets of microbial entities and edges represent sets of relationships between entities. A random walk is performed on the knowledge graph to generate a series of entity sequences as training samples. The random walk process takes into account the structural information and semantic relationships between entities to capture high-order interactions between entities. The training samples generated by the random walk are trained using a graph embedding model to learn a low-dimensional embedding representation of each entity. A low-dimensional embedding vector for each microbial entity is obtained through the graph embedding model as a semantic feature representation of the microorganism. The embedding vector captures the structural information and semantic relationships of the microorganism in the knowledge graph and can effectively characterize the characteristics of the microorganism.

[0098] After obtaining the graph embedding feature representation of microorganisms, a risk assessment model based on graph neural network is further constructed. Specifically, a multi-layer graph neural network model is designed, including a graph convolution layer, a graph attention layer, and a fully connected layer. The graph convolution layer aggregates and updates the node features, the graph attention layer adaptively learns the importance weights between nodes, and the fully connected layer performs nonlinear transformation and risk prediction on the node features. Based on the existing microbial risk annotation data, the graph neural network model is trained using supervised learning. The loss function uses cross entropy loss, and the optimization algorithm uses Adam optimizer. During the training process, the model parameters are updated through the backpropagation algorithm to minimize the difference between the predicted risk and the actual risk. For new microbial samples, their corresponding knowledge graph subgraphs are input into the trained graph neural network model, and the risk assessment results are obtained through forward propagation calculation.

[0099] Optionally, building on the existing microbial knowledge graph, entities and relationships related to risk transmission, such as transmission pathways, hosts, and vectors, can be introduced to construct a risk transmission knowledge graph. The risk transmission knowledge graph depicts the risk transmission network between microorganisms. A set of risk transmission inference rules is defined, leveraging the entities and relationships in the knowledge graph to discover potential risk transmission pathways through logical reasoning. For example, if microorganism A is airborne and microorganism B belongs to the same genus as A, it can be inferred that B is also airborne. Knowledge graph representation learning techniques, such as TransE and TransR, are used to learn low-dimensional embedding representations of entities and relationships. Through operations and transformations between embedded vectors, unknown risk transmission relationships are inferred. Based on the inferred risk transmission relationships, a risk transmission network diagram is constructed. Visualization techniques are used to intuitively display the risk transmission pathways and correlation strengths between microorganisms, assisting risk assessment and decision-making. The inferred risk transmission information is integrated into the risk assessment model, such as by incorporating risk transmission features into the graph neural network to update the risk assessment results. This approach comprehensively considers both microbial characteristics and risk transmission factors, improving the comprehensiveness and accuracy of risk assessments.

[0100] In practical applications, for new test samples, corresponding subgraphs are matched to the microbial knowledge graph based on their characteristic attributes. The subgraphs are then fed into a trained graph neural network model, and a forward propagation algorithm is used to calculate the risk score for each microbial node. Taking into account the prior pathogenicity probability of each microbial type, the risk scores are weighted and combined using the analytic hierarchy process to produce a comprehensive risk assessment.

[0101] Optionally, a fuzzy inference rule library can be constructed based on the risk assessment results and expert knowledge. The risk score is used as the input variable for fuzzy inference and mapped to a corresponding fuzzy set using a membership function. Fuzzy logic operations are then used to determine the risk level based on inference rules, resulting in a comprehensive risk assessment. Based on pre-set thresholds and decision rules, corresponding control measures and disposal recommendations are then provided.

[0102] Optionally, establish an incremental update mechanism for the knowledge graph, regularly acquiring the latest microbial knowledge from sources such as literature databases and laboratory reports. This new knowledge can be integrated into the existing graph through steps such as knowledge extraction and entity linking. Design a graph quality assessment and revision process, ensuring the accuracy and completeness of the knowledge graph through consistency checks and conflict resolution.

[0103] In this embodiment, by constructing a microbial knowledge graph, the entity knowledge and relationship knowledge in the microbial field are represented and stored in a structured manner to provide knowledge support for risk assessment. Using a graph embedding model, the nodes in the microbial knowledge graph are mapped to a low-dimensional continuous vector space as a feature representation of the node, preparing for subsequent graph neural network processing. Through the graph convolution layer and the graph attention layer, the topological structure information and relationship patterns between nodes can be automatically learned, and the potential correlation features between microbial entities can be fully explored. Feature extraction, relationship modeling and risk assessment are integrated into a multi-layer graph neural network model, and through end-to-end training, effective risk assessment patterns are automatically learned from data. By constructing a microbial knowledge graph, using graph embedding and graph neural network models, and closely combining domain knowledge and machine learning technology, intelligent, personalized and explainable microbial risk assessment can be achieved, which is of great significance for improving the accuracy and reliability of microbial detection.

[0104] In an alternative embodiment,

[0105] If the comprehensive risk assessment results indicate that the culture medium contains target microorganisms and the content exceeds the standard, the microbial culture strategy is adaptively adjusted based on the target microorganism type and target microorganism concentration. The culture medium is re-cultured, including:

[0106] A temperature control model is constructed based on the target microorganism type and target microorganism concentration, and a machine learning algorithm is used to train the temperature control strategy model. The trained temperature control strategy model is used to adaptively generate a temperature control curve and set value based on the real-time temperature of the culture medium and the target microorganism concentration.

[0107] According to the generated temperature control curve and set value, the fuzzy PID control algorithm is introduced. The temperature deviation, deviation change rate and deviation accumulation are comprehensively considered. Through fuzzy reasoning and defuzzification process, the parameters of the PID controller are dynamically adjusted and output as a control signal. The optimal temperature scheduling plan is generated according to the output control signal.

[0108] At the same time, a time scheduling optimization model is designed for multiple culture media, with minimizing the total culture time as the objective function. The culture time range and sequence relationship of each culture medium are defined as constraints. The time scheduling optimization model is solved by the target optimization algorithm to obtain the optimal time scheduling solution.

[0109] Combining the obtained optimal time scheduling plan and optimal temperature scheduling plan, the microbial cultivation strategy is adaptively adjusted and the culture medium is re-cultivated.

[0110] For example, if the evaluation results show that the detected sample contains target microorganisms and the content exceeds the standard, the control parameters of the temperature module and time module are adaptively adjusted according to the type and concentration of the microorganisms to achieve precise temperature control and time management; specifically, according to the type and concentration of the microorganisms, a temperature control strategy library is established, including parameters such as the optimal temperature range, temperature change gradient, and duration for the growth of different microorganisms.

[0111] The temperature control strategy model is trained using machine learning algorithms, such as decision trees and support vector machines. Based on historical data and expert knowledge, the model learns the mapping between microbial type and concentration and the optimal temperature control strategy. During the actual control process, the temperature control strategy model is dynamically invoked based on the real-time sample temperature and target microbial concentration, adaptively generating temperature control curves and setpoints.

[0112] The fuzzy PID control algorithm is introduced to comprehensively consider the temperature deviation, deviation change rate and deviation accumulation. Through fuzzy reasoning and defuzzification process, the parameters of the PID controller are dynamically adjusted to achieve flexible and stable temperature control.

[0113] At the same time, a time scheduling optimization model is designed to meet the cultivation needs of multiple samples. With the minimization of total cultivation time as the objective function and the cultivation time range and sequence relationship of each sample as constraints, an integer programming model is constructed. Heuristic search algorithms, such as genetic algorithms and ant colony algorithms, are used to solve the time scheduling optimization model. Through intelligent search and evolutionary iteration, the optimal scheduling plan that meets the constraints is quickly found. A real-time scheduling mechanism is introduced to dynamically adjust the scheduling plan based on the actual progress of the sample cultivation process and the results of risk assessment. A rolling time domain optimization strategy is adopted to regularly update the scheduling plan to deal with uncertainties in the cultivation process. Optionally, an intelligent scheduling dashboard is designed to display the cultivation progress, temperature curve and risk indicators of each sample in real time to support manual intervention and decision-making. Through visualization and human-computer interaction, the transparency and flexibility of the scheduling process are improved.

[0114] Furthermore, the obtained optimal time scheduling scheme and optimal temperature scheduling scheme are combined to adaptively adjust the microbial cultivation strategy and re-cultivate the culture medium.

[0115] Optionally, a knowledge base for microbial risk assessment can be established, including information on the pathogenic mechanisms, toxic effects, and hazardous concentration thresholds of various microorganisms. Ontology and semantic web technologies can be used to structure the representation and reasoning of risk knowledge. A rule-based risk warning model can be developed to convert risk assessment knowledge into a series of IF-THEN rules. Based on the real-time detection data of samples, potential risk factors and abnormal situations can be automatically identified through rule matching and reasoning. Machine learning methods, such as anomaly detection and time series prediction, can be introduced to analyze and model the historical data of samples. By learning the characteristic patterns of normal samples, early warning and trend prediction of abnormal samples can be achieved. A risk decision support system should be designed to comprehensively consider risk warning information, temperature control, and time scheduling plans, and provide risk response measures and optimization suggestions. For example, multi-criteria decision-making methods, such as hierarchical analysis method and fuzzy comprehensive evaluation, can be used to balance the trade-off between risk control and cultivation efficiency.

[0116] In an alternative embodiment,

[0117] According to the generated temperature control curve and set value, the fuzzy PID control algorithm is introduced. The temperature deviation, deviation change rate and deviation accumulation are comprehensively considered. Through the fuzzy reasoning and defuzzification process, the parameters of the PID controller are dynamically adjusted. The calculation formula for outputting the final control signal is as follows:

[0118]

[0119] Among them, U(t) represents the final control signal, K p represents the proportional coefficient, e(t) represents the temperature deviation at the current moment, K i Indicates the integral coefficient, e i (t) represents the rate of change of deviation, K d represents the differential coefficient, e c (t) represents the cumulative deviation;

[0120] At the same time, a time scheduling optimization model is designed for multiple culture media, with the minimization of total culture time as the objective function, and the culture time range and sequence relationship of each culture medium defined as constraints. The time scheduling optimization model is solved by the target optimization algorithm, including:

[0121] An objective function is constructed based on the amount of culture medium and the total culture time, while introducing a regularization parameter to balance the fitting error of the model and the smoothness of the parameters. The objective function is calculated as follows:

[0122]

[0123] Among them, J represents the objective function, θ represents the parameter vector, N represents the number of culture media, f(t i ; θ) represents the time ti The culture time of the culture medium is determined by the parameter vector θ, f * (t i ) indicates that at time t i The target training time under , λ1 represents the regularization coefficient of the equilibrium parameter vector, λ2 represents the regularization coefficient of the gradient of the equilibrium parameter vector, represents the gradient of the parameter vector, ||·|| 2 represents the square norm.

[0124] In this embodiment, the temperature control strategy model is trained by a machine learning algorithm, and the ideal temperature control curve and set value can be automatically generated according to the target microorganism type, concentration and real-time temperature information. The introduction of the fuzzy PID control algorithm can dynamically adjust the PID parameters to achieve precise temperature control. Through the time scheduling optimization model, the optimal time scheduling scheme is solved with the goal of minimizing the total culture time. Combined with the optimal temperature scheduling scheme, the temperature and time of multiple culture media can be collaboratively optimized to improve the culture efficiency. Through the comprehensive application of machine learning, fuzzy control, optimization modeling and other technologies, the adaptive dual optimization adjustment of culture temperature and time is achieved, and the intelligence and automation level of the culture strategy adjustment are improved, so that the growth of target microorganisms can be more efficiently inhibited, providing high-quality samples for microbial detection.

[0125] S103. The target microorganism and target microorganism concentration detection steps are performed again on the culture medium after re-cultivation. If the concentration of the re-detected target microorganism has dropped below the standard threshold, the detection data is fed back to the control system, and an adaptive optimization algorithm is used to optimize the microbial metabolic response model and detection strategy to perform automated detection of the microorganisms.

[0126] Among them, the detection data includes impedance spectrum information and microbial concentration.

[0127] In an alternative embodiment,

[0128] The method further comprises:

[0129] Introducing smart sensors and IoT devices to collect real-time microbial growth environment parameters and status data, and preprocessing them. Feature extraction is performed on the preprocessed microbial growth environment parameters and status data to obtain microbial growth and metabolic characteristics.

[0130] Based on the growth and metabolic characteristics of microorganisms, and integrating pre-built mechanism models and data-driven models, a digital twin model of microbial growth and metabolism is constructed;

[0131] The obtained digital twin model of microbial growth and metabolism is trained, and the trained digital twin model of microbial growth and metabolism is used to perform real-time predictions of microbial growth trends, metabolites, and key indicators;

[0132] Based on the real-time prediction of microbial growth trends, metabolites, and key indicators, the system detects abnormal conditions during the microbial cultivation process. The detection results and real-time prediction results are fed back to the control system, and the optimal microbial cultivation control strategy is learned through the reinforcement learning algorithm.

[0133] The optimal microbial culture control strategy obtained is applied to the automatic control of microbial culture and detection processes, and human-computer interaction is combined to realize intelligent decision support for microbial detection.

[0134] For example, smart sensors and IoT devices can be introduced during the microbial cultivation and detection process. Smart sensors can accurately measure various parameters of the microbial growth environment, such as temperature, humidity, pH, and dissolved oxygen, and transmit this data in real time to a data processing platform. IoT devices, such as wireless sensor networks and RFID tags, can automatically identify and track information such as the location, quantity, and status of microbial samples, enabling full-process monitoring of the microbial cultivation and detection process.

[0135] To ensure accurate and reliable data collection, smart sensors and IoT devices require calibration and optimization. Automatic calibration algorithms, such as Kalman filtering and Bayesian estimation, can be used to adaptively compensate for sensor drift and noise, improving measurement accuracy. Furthermore, energy optimization algorithms, such as data compression and sleep scheduling, can be employed to reduce device energy consumption and latency, extending battery life.

[0136] In terms of data transmission and processing, an architecture combining edge computing and cloud computing can be adopted. At the same time, the big data storage and computing capabilities of the cloud platform can be used to mine and analyze massive amounts of historical data, discover potential patterns and rules, and optimize the processes and strategies for microbial cultivation and detection.

[0137] Raw data collected by smart sensors and IoT devices often suffer from noise, missing data, and redundancy, requiring preprocessing operations such as data cleaning, denoising, and interpolation to improve data quality. Then, using techniques such as signal processing, image processing, and natural language processing, key features related to microbial growth and metabolism, such as morphological, spectral, and gene expression characteristics, are extracted from multi-source heterogeneous data to construct high-dimensional feature vectors.

[0138] Based on feature extraction, a digital twin model of microbial growth and metabolism can be constructed by combining mechanistic and data-driven models. Mechanistic models, based on prior knowledge from fields such as microbial physiology, biochemistry, and genetics, utilize mathematical tools such as differential equations and kinetic equations to describe the dynamic processes of microbial growth and metabolism, such as the Monod equation and the Luedeking-Piret equation. Data-driven models, based on big data analysis and machine learning techniques, such as support vector machines, random forests, and artificial neural networks, automatically learn the inherent laws of microbial growth and metabolism from historical data.

[0139] Furthermore, machine learning and deep learning algorithms are applied to train and optimize the model. For structured data, such as sensor readings and environmental parameters, traditional machine learning algorithms, such as support vector machines (SVMs), decision trees, and logistic regression, can be used to establish nonlinear mappings between microbial growth and metabolism and environmental factors. To avoid overfitting and improve generalization performance, cross-validation, regularization, and ensemble learning techniques can be used to optimize the model's hyperparameters and structure.

[0140] For unstructured data, such as microscopic images and expression profiles, deep learning algorithms, such as convolutional neural networks (CNNs) and long short-term memory networks (LSTMs), can be used to automatically learn the hierarchical features and temporal dependencies within the data. Through local connections and weight sharing, CNNs can effectively extract visual features such as texture and edges from images for analysis of microbial morphology and behavior. LSTMs, through gating mechanisms and memory cells, can model the dynamic changes and long-term dependencies in microbial growth and metabolism, enabling state prediction and anomaly detection.

[0141] During model training and optimization, strategies such as transfer learning and multi-task learning can be employed to fully leverage cross-domain and cross-platform data and knowledge, alleviating issues such as data sparsity and high annotation costs. At the same time, technologies such as attention mechanisms and adversarial learning can be introduced to enhance the interpretability and robustness of models and address challenges such as data noise and distribution drift.

[0142] Based on the optimized digital twin model of microbial growth and metabolism, real-time prediction and anomaly detection of microbial status can be achieved, providing decision support for the intelligent control of microbial cultivation and detection processes.

[0143] Real-time prediction involves using digital twin models to dynamically estimate microbial growth trends, metabolites, and key indicators over a period of time, based on current environmental conditions and microbial status. This allows for early detection of potential problems during microbial growth, such as reduced growth rate or low product concentration, providing a basis for adjusting culture conditions and optimizing control strategies. Common prediction methods include recursive neural networks, Kalman filters, and particle filters, which can be selected based on the prediction timescale and computational complexity.

[0144] Furthermore, digital twin models can be used for anomaly detection. For example, by establishing a probability distribution model for microbial state variables, such as Gaussian distribution or Poisson distribution, and using hypothesis testing and confidence interval principles, it can be determined whether the current state deviates significantly from the normal level. Alternatively, by measuring the distance or similarity between the current state and the normal state, such as Euclidean distance, Mahalanobis distance, or cosine similarity, it can be determined whether the current state is abnormal.

[0145] To achieve intelligent control of microbial cultivation and detection, prediction and detection results can be fed back to the control system, allowing it to autonomously optimize the control strategy through a reinforcement learning algorithm. Specifically, the control system is considered an intelligent agent, the microbial growth state is considered the environment, and the detection indicators are considered the rewards. By estimating the value function of the state-action pair, the action with the highest value is selected, resulting in the optimal microbial cultivation control strategy.

[0146] Furthermore, the optimized microbial culture control strategy is applied to the automatic control of microbial culture and detection processes, and a human-machine collaborative intelligent decision support system is constructed. By utilizing knowledge acquisition, cognitive interaction and collaborative optimization technologies, expert experience is embedded in the intelligent system to achieve barrier-free communication and collaborative judgment between humans and machines, providing explainable and trustworthy optimization suggestions for microbial detection.

[0147] In this embodiment, by introducing smart sensors and IoT devices, various parameters and status data of the microbial growth environment can be acquired in real time, providing data support for subsequent modeling and control. By combining mechanistic models with data-driven models, prior knowledge and actual data can be fully utilized to construct a high-precision digital twin model. This model can accurately simulate and predict the growth trends, metabolites, and key indicators of microorganisms. Based on the digital twin model, the growth status and metabolic status of microorganisms can be predicted in real time. By comparing with actual data, anomalies in the microbial cultivation process can be promptly detected. The real-time prediction results and anomaly detection results are fed back to the control system. Using a reinforcement learning algorithm, the system can autonomously learn and optimize the optimal microbial cultivation control strategy. The learned optimal control strategy is applied to the actual microbial cultivation and detection process, realizing real-time monitoring of the microbial growth environment, digital modeling and simulation, anomaly detection, automatic control, and intelligent decision support. This provides an integrated intelligent solution for microbial detection, which can significantly improve the accuracy, efficiency, and intelligence level of microbial detection.

[0148] Figure 2 FIG. 1 is a schematic diagram of the structure of an automated control system based on microbial detection according to an embodiment of the present invention. Figure 2 As shown, the system includes:

[0149] The first unit is used to measure the impedance parameters of the culture medium using frequency sweep technology at different frequencies to obtain complete impedance spectrum information. The obtained impedance spectrum information is then subjected to dimensionality reduction and feature extraction to obtain characteristic parameters of the impedance signal. The obtained characteristic parameters are input into a pre-built microbial metabolic response model to predict the type and concentration of microorganisms.

[0150] The second unit is used to determine whether the culture medium contains the target microorganism and whether the target microorganism concentration exceeds the standard based on the prediction results of the microbial metabolic response model. During the judgment process, the microorganism type and pathogenicity probability are comprehensively considered, and a comprehensive risk assessment result is obtained through the hierarchical analysis method. If the comprehensive risk assessment result indicates that the culture medium contains the target microorganism and the concentration exceeds the standard, the microbial culture strategy is adaptively adjusted according to the target microorganism type and target microorganism concentration, and the culture medium is re-cultured.

[0151] The third unit is used to re-execute the target microorganism and target microorganism concentration detection steps on the re-cultured culture medium. If the re-detected target microorganism concentration has dropped below the standard threshold, the detection data will be fed back to the control system, and an adaptive optimization algorithm will be used to optimize the microbial metabolic response model and detection strategy to perform automated detection of the microorganisms.

[0152] According to a third aspect of the embodiments of the present invention,

[0153] Provides electronic equipment, including:

[0154] processor;

[0155] a memory for storing processor-executable instructions;

[0156] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0157] According to a fourth aspect of the embodiments of the present invention,

[0158] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0159] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0160] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An automated control method based on microbial detection, characterized in that: include: The impedance parameters of the culture medium are measured using frequency scanning technology at different frequencies to obtain complete impedance spectrum information, and the obtained impedance spectrum information is subjected to dimensionality reduction and feature extraction to obtain characteristic parameters of the impedance signal. The obtained characteristic parameters are input into a pre-built microbial metabolic response model to predict the type and concentration of microorganisms. According to the prediction results of the microbial metabolic response model, it is judged whether the culture medium contains the target microorganism and whether the concentration of the target microorganism exceeds the standard. In the judgment process, the type of microorganism and the probability of pathogenicity are comprehensively considered, and the comprehensive risk assessment results are obtained through the hierarchical analysis method; If the comprehensive risk assessment results indicate that the culture medium contains target microorganisms and the content exceeds the standard, the microbial culture strategy is adaptively adjusted according to the target microorganism type and target microorganism concentration, and the culture medium is re-cultured; The target microorganism and target microorganism concentration detection steps are performed again on the re-cultured culture medium. If the re-detected target microorganism concentration has dropped below the standard threshold, the detection data is fed back to the control system, and an adaptive optimization algorithm is used to optimize the microbial metabolic response model and detection strategy to perform automated detection of the microorganisms.

2. The method according to claim 1, characterized in that The impedance parameters of the culture medium are measured using frequency scanning technology at different frequencies to obtain complete impedance spectrum information, and the obtained impedance spectrum information is subjected to dimensionality reduction and feature extraction to obtain characteristic parameters of the impedance signal. The obtained characteristic parameters are input into the pre-built microbial metabolic response model to predict the type and concentration of microorganisms, including: Preprocessing the obtained impedance spectrum information, and inputting the preprocessed impedance spectrum information into a pre-constructed microbial metabolism response model, wherein the microbial metabolism response model is composed of a convolutional neural network and a long short-term memory network; Construct a joint loss function that includes microbial type identification loss and microbial concentration estimation loss. By minimizing the joint loss function, updating the parameters of the microbial metabolic response model, and repeating the iteration until the preset termination condition is met, the final trained microbial metabolic response model is obtained; Based on the trained microbial metabolic response model, the local patterns and key features of the impedance spectrum information are extracted through a convolutional neural network, and the feature dimensions are reduced through a pooling operation to obtain low-dimensional key features; The obtained low-dimensional key features are input into the long short-term memory network for sequence modeling to capture the temporal characteristics of the impedance spectrum information, and deep sequence features are constructed by combining multiple long short-term memory layers. Based on the deep sequence features, the probability distribution of microbial types and the predicted values ​​of concentration are output through the fully connected layer.

3. The method according to claim 2, characterized in that The calculation formula for constructing a joint loss function that includes microbial type identification loss and microbial concentration estimation loss is as follows: Where L represents the joint loss function, A represents the number of microbial types, and α a represents the weight of microorganism type a, represents the probability of the predicted microbial type a, γ represents the regulatory factor, and y a represents the label of the actual microorganism type a, β represents the weight of the estimated loss of microorganism concentration, M represents the number of samples of impedance spectrum information, K represents the number of mixed Gaussian components, ω mk The weight of sample m representing the impedance spectrum information on the kth mixed Gaussian component, σ k represents the standard deviation of the kth Gaussian mixture component, r m The concentration of sample m, μ, represents the impedance spectrum information k Represents the mean of the kth Gaussian mixture component.

4. The method according to claim 1, characterized in that According to the prediction results of the microbial metabolic response model, it is judged whether the culture medium contains the target microorganism and whether the concentration of the target microorganism exceeds the standard. In the judgment process, the type of microorganism and the probability of pathogenicity are comprehensively considered, and the comprehensive risk assessment results are obtained through the hierarchical analysis method, including: collecting structured knowledge in the field of microorganisms and constructing a microbial knowledge graph, in which nodes represent microbial entity sets and edges represent relationship sets between entities; Based on the microbial knowledge graph, a random walk sequence is generated as a training sample, and a graph embedding model is used to train the training sample to obtain a low-dimensional embedding representation of each node in the microbial knowledge graph as an initial feature of the node; Construct a multi-layer graph neural network model, which includes graph convolution layer, graph attention layer and fully connected layer. The graph convolution layer aggregates and updates the initial features of the nodes, the graph attention layer adaptively learns the importance weights between nodes, and the fully connected layer performs nonlinear transformation and risk prediction on node features. The multi-layer graph neural network model is trained according to the historical microbial detection data, and the error between the predicted risk value and the actual risk value is minimized, the model parameters are optimized by using the gradient descent algorithm, and an early stopping mechanism and a learning rate scheduling strategy are introduced to finally obtain a trained multi-layer graph neural network model; For the current test sample, the corresponding subgraph is matched in the microbial knowledge graph and input into the trained multi-layer graph neural network model. The risk score of each microbial node in the subgraph is calculated through forward propagation. The microbial type and pathogenicity probability are comprehensively considered, and the risk scores of each microbial node are combined using the hierarchical analysis method to obtain a comprehensive risk assessment result.

5. The method according to claim 1, characterized in that If the comprehensive risk assessment results indicate that the culture medium contains target microorganisms and the content exceeds the standard, the microbial culture strategy is adaptively adjusted according to the target microorganism type and target microorganism concentration. The culture medium is re-cultured including: A temperature control model is constructed according to the target microorganism type and target microorganism concentration, and a machine learning algorithm is used to train the temperature control strategy model. The trained temperature control strategy model is used to adaptively generate a temperature control curve and set value according to the real-time temperature of the culture medium and the target microorganism concentration. According to the generated temperature control curve and set value, the fuzzy PID control algorithm is introduced, and the temperature deviation, deviation change rate and deviation accumulation are comprehensively considered. Through the fuzzy reasoning and defuzzification process, the parameters of the PID controller are dynamically adjusted and output as a control signal. The optimal temperature scheduling plan is generated according to the output control signal; At the same time, a time scheduling optimization model is designed for multiple culture media, with minimizing the total culture time as the objective function, defining the culture time range and sequence relationship of each culture medium as constraints, and solving the time scheduling optimization model through the target optimization algorithm to obtain the optimal time scheduling solution; The optimal time scheduling plan and the optimal temperature scheduling plan are combined to adaptively adjust the microbial cultivation strategy and re-cultivate the culture medium.

6. The method according to claim 5, characterized in that According to the generated temperature control curve and set value, the fuzzy PID control algorithm is introduced, and the temperature deviation, deviation change rate and deviation accumulation are comprehensively considered. Through the fuzzy reasoning and defuzzification process, the parameters of the PID controller are dynamically adjusted, and the calculation formula for outputting the final control signal is as follows: Among them, U(t) represents the final control signal, K p represents the proportionality coefficient, e(t) represents the temperature deviation at the current moment, K i Indicates the integral coefficient, e i (t) represents the rate of change of deviation, K d represents the differential coefficient, e c (t) represents the cumulative deviation; At the same time, for multiple culture media, a time scheduling optimization model is designed, with the minimization of the total culture time as the objective function, and the culture time range and sequence relationship of each culture medium are defined as constraints. The time scheduling optimization model is solved by the target optimization algorithm, including: The objective function is constructed based on the amount of culture medium and the total culture time, and the fitting error of the regularization parameter balance model and the smoothness of the parameters are introduced. The calculation formula of the objective function is as follows: Where J represents the objective function, θ represents the parameter vector, N represents the number of culture media, and f(t i ; θ) represents the time t i The culture time of the culture medium is determined by the parameter vector θ, f * (t i ) indicates that at time t i The target training time under , λ1 represents the regularization coefficient of the equilibrium parameter vector, λ2 represents the regularization coefficient of the gradient of the equilibrium parameter vector, represents the gradient of the parameter vector, ||·|| 2 Represents the square norm.

7. The method according to claim 1, characterized in that The method further comprises: Introduce smart sensors and IoT devices to obtain real-time microbial growth environment parameters and status data, and perform preprocessing. Perform feature extraction on the preprocessed microbial growth environment parameters and status data to obtain microbial growth and metabolic characteristics. Based on the growth and metabolic characteristics of microorganisms, and integrating pre-built mechanism models and data-driven models, a digital twin model of microbial growth and metabolism is constructed; The obtained digital twin model of microbial growth and metabolism is trained, and the trained digital twin model of microbial growth and metabolism is used to make real-time predictions of microbial growth trends, metabolites, and key indicators; According to the real-time predicted microbial growth trends, metabolites and key indicators, abnormal state detection of microbial cultivation process is carried out, and the obtained detection results and real-time prediction results are fed back to the control system, and the optimal microbial cultivation control strategy is learned through the reinforcement learning algorithm; The optimal microbial culture control strategy is applied to the automatic control of microbial culture and detection processes, and human-computer interaction is combined to realize intelligent decision support for microbial detection.

8. An automated control system based on microbial detection, used to implement the method according to any one of claims 1 to 7, characterized in that: include: The first unit is used to measure the impedance parameters of the culture medium using frequency scanning technology at different frequencies to obtain complete impedance spectrum information, and perform dimension reduction and feature extraction on the obtained impedance spectrum information to obtain characteristic parameters of the impedance signal. The obtained characteristic parameters are input into a pre-built microbial metabolic response model to predict the type and concentration of microorganisms; The second unit is used to determine whether the culture medium contains the target microorganism and whether the concentration of the target microorganism exceeds the standard based on the prediction results of the microbial metabolic response model. In the judgment process, the microbial type and pathogenicity probability are comprehensively considered, and the comprehensive risk assessment results are obtained through the hierarchical analysis method; If the comprehensive risk assessment results indicate that the culture medium contains target microorganisms and the content exceeds the standard, the microbial culture strategy is adaptively adjusted according to the target microorganism type and target microorganism concentration, and the culture medium is re-cultured; The third unit is used to re-execute the detection steps of target microorganisms and target microorganism concentrations on the culture medium after re-cultivation. If the concentration of the re-detected target microorganism has dropped below the standard threshold, the detection data will be fed back to the control system, and an adaptive optimization algorithm will be used to optimize the microbial metabolic response model and detection strategy to perform automated detection of the microorganisms.

9. An electronic device, characterized in that include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Citation Information

Cited By

  • Intelligent in-vivo component metabolism monitoring method and system

    CN119178873A

  • Microorganism data visualization chart generation system and method based on model parameters

    CN120388622A

  • Model parameter-based microbial data visualization chart generation system and method

    CN120388622B

  • Method for predicting metabolic characteristics of silicate decomposing bacteria based on deep learning

    CN120708723A

  • Silicate decomposing bacteria metabolic characteristic prediction method based on deep learning

    CN120708723B