A smart monitoring and control system for cell culture environment

Abstract

This invention discloses an intelligent monitoring and control system for cell culture environments, relating to the field of environmental monitoring technology. It includes: an environmental perception module that generates a standardized environmental feature set; a feature fusion and enhancement module that, based on the standardized environmental feature set, performs multimodal fusion and cross-dimensional feature association and enhancement to generate high-dimensional fused feature data; an intelligent state assessment module that performs quantitative assessment and type classification of cell growth states to obtain a cell growth state assessment report; and a precise control strategy module that, based on the cell growth state assessment report, constructs a personalized control model and outputs targeted combinations of culture environment control parameters. This invention, through a distributed micro-sensor array combined with regional difference analysis technology, solves the problems of single-point acquisition failing to capture micro-regional environmental differences within the culture container and neglecting regional environmental inhomogeneity, thus improving control accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of environmental monitoring, in particular to a cell culture environment intelligent monitoring and control system. BACKGROUND

[0002] Cell culture is a core technical means in the fields of biomedical research, biopharmaceuticals, cell therapy, etc., and its culture effect directly determines the quality and efficiency of subsequent research or production. Cell growth is highly sensitive to environmental conditions, and subtle changes in multiple environmental factors such as temperature, humidity, pH value, dissolved oxygen, nutrient concentration, and contamination bacteria content can cause abnormal cell growth, morphological variation, or even death.

[0003] The cell culture environment monitoring and control technology generally has the following shortcomings: first, the data collection dimension is single and lacks regional specificity, and multiple sensors are usually used to collect overall environmental data, making it difficult to capture environmental differences in different micro-regions within the culture container, resulting in the problem of regionalized uneven environment being ignored; second, the feature processing capability is weak, and the fusion of multi-dimensional heterogeneous environmental data is mostly limited to simple splicing or weighted processing, which cannot mine the deep semantic association between different dimensional data, and the feature representation accuracy is insufficient. Therefore, developing a cell culture environment intelligent monitoring and control system that can perform multi-dimensional micro-region accurate perception and deep feature fusion intelligent monitoring and control system has become a key requirement to solve the pain points of existing technology and improve the quality and efficiency of cell culture. SUMMARY

[0004] To solve the above technical problems, a cell culture environment intelligent monitoring and control system is provided, which solves the above problems.

[0005] To achieve the above purposes, the technical scheme adopted by the present application is: A cell culture environment intelligent monitoring and control system, comprising: an environment perception module, a feature fusion enhancement module, an intelligent state evaluation module, and a precise control strategy module; The environment perception module is used to collect multi-dimensional heterogeneous data of the cell culture environment and generate a standardized environment feature set. The feature fusion enhancement module is electrically connected with the environment perception module, and the feature fusion enhancement module is used to perform multi-modal fusion and cross-dimensional feature association and enhancement based on the standardized environment feature set, and generate high-dimensional fusion feature data. The intelligent state evaluation module is electrically connected with the feature fusion enhancement module, and the intelligent state evaluation module is used to perform quantitative evaluation and type division of the cell growth state based on the high-dimensional fusion feature data, and obtain a cell growth state evaluation report. The precise regulation strategy module is electrically connected to the intelligent state assessment module. The precise regulation strategy module is used to construct a personalized regulation model based on the cell growth state assessment report and output a targeted combination of culture environment regulation parameters.

[0006] Preferably, the environment sensing module includes: The data acquisition unit collects multi-dimensional heterogeneous environmental data based on the needs of cell culture scenarios. The multi-dimensional heterogeneous environmental data includes physical environmental data, chemical environmental data, and biological environmental data. The physical environmental data includes temperature, humidity, and air pressure. The chemical environmental data includes pH value, dissolved oxygen, and nutrient concentration. The biological environmental data includes the content of polluting bacteria and the concentration of cell metabolic byproducts. The data preprocessing unit uses an adaptive filtering algorithm to reduce noise for the collected multi-dimensional heterogeneous environmental data, and maps data of different dimensions to a unified interval through the min-max normalization method to generate a standardized environmental feature set. The data quality verification unit constructs data validity evaluation indicators, performs quality verification on the preprocessed standardized environmental feature set, removes invalid data and supplements missing data, and outputs a high-quality standardized environmental feature set.

[0007] Preferably, the environment sensing module further includes: The micro-region acquisition unit deploys a distributed array of micro sensors inside the culture container to collect environmental data from different micro-regions at a preset sampling frequency, generating a subset of micro-region environmental data. The regional difference analysis unit calculates the parameter deviations between each micro-region and the overall environment based on a subset of micro-region environmental data, and constructs a heat map of micro-region environmental differences. The data fusion output unit associates the micro-regional environmental difference heatmap with a high-quality standardized environmental feature set, supplements regional dimensional information, forms a complete standardized environmental feature set, and outputs it to the feature fusion enhancement module.

[0008] Preferably, the feature fusion enhancement module includes: The multimodal feature mapping unit, based on the standardized environmental feature set output by the environmental perception module, establishes corresponding feature encoders according to data types, and converts physical, chemical, and biological features into feature vectors respectively, thus obtaining a multi-dimensional feature vector set; The cross-dimensional fusion unit employs a Transformer-based cross-modal cross-attention fusion algorithm. It performs deep semantic fusion of multi-modal features by modeling intra-dimensional feature dependencies through self-attention and inter-dimensional feature interactions through cross-attention, resulting in globally fused feature data. The specific method is as follows: Linear embedding is performed on the feature vectors of the three dimensions of physics, chemistry and biology respectively. At the same time, positional encoding is added to preserve the temporal information of the data. The feature data after embedding and encoding is used as the input for subsequent processing. The embedded and encoded feature data of each dimension are input into the self-attention layer. The self-attention layer calculates the dependency weights within the single-dimensional feature, thereby obtaining the enhanced feature data of each dimension. Using augmented feature data from any one dimension as query information, the augmented feature data from the other two dimensions are concatenated to serve as key and value information. The interaction weights between features of different dimensions are calculated through cross-attention. This cross-fusion process is iterated three times to obtain intermediate fused feature data. Then, the intermediate fused feature data is input into the feedforward neural network, and finally, the global fused feature data is output. This algorithm captures feature association information at different scales in parallel through a multi-head mechanism, enabling adaptive semantic fusion of multi-dimensional features and effectively improving the discriminativeness and robustness of feature representation. Among them, the multi-head self-attention function is used to achieve parallel capture of multi-scale feature associations, the feature concatenation function is used to complete the combination of enhanced feature data of different dimensions, and the feedforward neural network is used to further process the intermediate fused feature data, and finally output the global fused feature data. The feature enhancement unit, based on globally fused feature data, uses an attention mechanism to strengthen the weights of key features, suppress redundant information, and generate high-dimensional fused feature data.

[0009] Preferably, the feature fusion enhancement module further includes: The feature dimensionality reduction unit reduces the dimensionality of high-dimensional fusion feature data based on principal component analysis, retains principal components that exceed the cumulative contribution rate threshold, and obtains low-dimensional optimized fusion features. The feature verification unit compares the low-dimensional optimized fusion features with historical feature data in the cell growth baseline database, calculates the feature matching degree, and outputs high-dimensional fusion feature data if the matching degree is greater than the matching degree threshold; otherwise, it returns to the cross-dimensional fusion unit for re-optimization. The feature time series association unit associates high-dimensional fusion feature data of continuous time series, supplements time dimension information, forms time-series high-dimensional fusion feature data, and outputs it to the intelligent state assessment module.

[0010] Preferably, the intelligent state assessment module includes: The evaluation index construction unit, based on the time-series high-dimensional fusion feature data output by the feature fusion enhancement module, constructs a cell growth status evaluation index system, including four core indicators: growth rate, morphological integrity, metabolic activity, and environmental adaptability. The quantitative evaluation unit uses the analytic hierarchy process (AHP) to determine the weight of each evaluation indicator, and combines the standard thresholds in the cell growth baseline database to calculate the quantitative score of each indicator and obtain the comprehensive evaluation value. The state clustering unit, based on the comprehensive evaluation value, adopts an improved clustering algorithm that introduces cell generation weight coefficients to divide cell growth states into four categories: optimal growth state, adaptive state to be regulated, critical abnormal state, and severely abnormal state. The update formula of the improved clustering algorithm is as follows:

[0011] In the formula, For the first Cluster centers after the next iteration For the first The generational weighting coefficients of individual cell samples. For the first The overall assessment value of each cell sample.

[0012] Preferably, the intelligent state assessment module further includes: The trend prediction unit, based on time-series high-dimensional fused feature data, trains an LSTM deep learning network. It takes historical feature sequences and corresponding growth states as input and outputs the trend of cell growth state changes and the probability of abnormal risks for the next 24 hours. The core hidden state update formula of the LSTM is as follows:

[0013] In the formula, for LSTM hidden state at time step for Output gate output value at any time, The hyperbolic tangent activation function is used. for Cellular state at all times It is the Sigmoid activation function. This is the output gate weight matrix. for Hide your status at all times. for Temporal high-dimensional fusion features of input at any time; This is the output gate bias term; The evaluation and calibration unit calls up historical monitoring and control data, compares the actual growth status with the evaluation results, and corrects the weights of the evaluation indicators and the clustering thresholds. The report generation unit integrates quantitative assessment results, state classification, and trend prediction information to generate a cell growth status assessment report that includes state level, risk warning, and key influencing factors, and outputs it to the precision regulation strategy module.

[0014] Preferably, the precise control strategy module includes: The regulatory factor extraction unit extracts key regulatory factors affecting cell growth status based on the cell growth status assessment report output by the intelligent status assessment module, including temperature regulation, pH calibration, nutrient supplementation, and pollution control. The factor priority ranking unit uses the entropy weight method combined with the analytic hierarchy process to comprehensively consider the sensitivity of the regulatory factors to different growth states, the speed of regulatory response, and energy consumption costs to determine the priority weight of each regulatory factor. The regulation model construction unit builds personalized regulation models for different growth states, inputs key influencing factors and regulation factor priorities from the evaluation report, and outputs the initial regulation parameter range.

[0015] Preferably, the precise control strategy module further includes: The regional targeted regulation unit, combined with the micro-region environmental difference heat map output by the environmental sensing module, performs differential correction on the initial regulation parameter range based on the environmental deviation and cell growth status of different micro-regions, and obtains regional regulation parameters. The pre-verification unit for the regulation effect constructs a virtual simulation model of cell culture based on digital twin technology. It inputs regionalized regulation parameters and current environmental characteristics to simulate the cell growth state after regulation. If the simulation optimization rate is greater than the optimization rate threshold, it enters the parameter output stage; otherwise, it readjusts the weights of the regulation factors. The parameter optimization unit uses the particle swarm optimization algorithm to finely optimize the validated regional control parameters, reduce the range of parameter fluctuations, and obtain a subset of precise control parameters.

[0016] Preferably, the precise control strategy module further includes: The control command generation unit divides the precise control parameter subset according to the control factor type and micro-region, generates structured control commands, and clarifies the operating parameters, execution timing and duration of each control device; The parameter feedback unit records environmental data and changes in cell growth status after the execution of the regulation command, and feeds them back to the intelligent state assessment module for evaluation index calibration and regulation model optimization. The regulatory database update unit stores the evaluation report, regulatory parameters, and effect data of this regulatory process into the system database, enriches the historical samples, improves the adaptability of subsequent regulatory models, and finally outputs a targeted combination of culture environment regulatory parameters.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention proposes a method that combines a distributed micro-sensor array to simultaneously collect multi-dimensional physical, chemical, and biological environmental data, as well as environmental difference data from different micro-regions. Combined with regional difference analysis and data fusion, this overcomes the shortcomings of existing technologies, such as single-point data collection and neglect of regional differences. It provides comprehensive and accurate data source support for subsequent precise regulation. By employing a Transformer-based cross-modal cross-attention fusion algorithm, which models intra-dimensional feature dependencies through self-attention and inter-dimensional feature interactions through cross-attention, combined with feature dimensionality reduction, validation, and time-series correlation processing, it effectively mines deep semantic relationships in multi-dimensional heterogeneous data, improving the discriminative power and robustness of feature representation. This solves the problems of simple feature fusion and insufficient representational capabilities. Attached Figure Description

[0018] Figure 1 This is a system framework diagram of the present invention; Figure 2 This is a diagram showing the internal system framework of the environmental perception module in this invention. Figure 3 This is a diagram showing the internal system framework of the feature fusion enhancement module in this invention. Figure 4 This is an internal system framework diagram of the intelligent state assessment module in this invention; Figure 5 This is an internal system framework diagram of the precise control strategy module in this invention. Detailed Implementation

[0019] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0020] Reference Figure 1 As shown, an intelligent monitoring and control system for cell culture environment includes: Environmental perception module, feature fusion enhancement module, intelligent state assessment module, and precise control strategy module; The environmental perception module is used to collect multi-dimensional heterogeneous data of the cell culture environment and generate a standardized environmental feature set. The feature fusion enhancement module is electrically connected to the environment perception module. The feature fusion enhancement module is used to perform multimodal fusion and cross-dimensional feature association and enhancement based on the standardized environmental feature set to generate high-dimensional fusion feature data. The intelligent state assessment module is electrically connected to the feature fusion enhancement module. The intelligent state assessment module is used to perform quantitative assessment and type classification of cell growth status based on the high-dimensional fusion feature data to obtain a cell growth status assessment report. The precise regulation strategy module is electrically connected to the intelligent state assessment module. The precise regulation strategy module is used to construct a personalized regulation model based on the cell growth state assessment report and output a targeted combination of culture environment regulation parameters.

[0021] The environmental perception module is implemented as follows: Data Acquisition Unit: Based on the environmental monitoring needs of the cell culture scenario, a sensor deployment scheme is planned to cover environmental parameters across three dimensions: physical, chemical, and biological. The physical dimension focuses on fundamental environmental indicators affecting cell growth, the chemical dimension focuses on chemical indicators related to cell metabolism, and the biological dimension focuses on indicators related to pollution control and metabolic byproducts. All sensors collect data synchronously according to a unified time sequence and transmit it to the data preprocessing unit via a pre-set communication bus to ensure the synchronicity and integrity of the data acquisition.

[0022] Data preprocessing unit: For the collected multi-dimensional heterogeneous data, noise reduction processing is first performed, and an appropriate filtering algorithm is used to eliminate invalid information caused by environmental interference and sensor errors. Then, normalization processing is performed, and the influence of data with different dimensions is eliminated by a preset normalization method, mapping the data of each dimension to a unified range and generating a standardized environmental feature set. The entire preprocessing process builds a data cleaning pipeline to ensure that the data format is uniform and the information is effective.

[0023] Data Quality Verification Unit: Constructs a three-dimensional validity evaluation index system including completeness, accuracy, and timeliness. For each of the preprocessed standardized environmental feature sets, quality verification is carried out. For data that does not meet the completeness requirements, an appropriate interpolation method is used to supplement the missing information. For data with abnormal accuracy, it is judged and removed based on historical data trends. For data that is outdated in timeliness, a re-collection command is triggered. After verification, a high-quality standardized environmental feature set is output to provide reliable data support for subsequent feature processing.

[0024] Micro-region acquisition unit: Based on the structural characteristics of the culture container, the deployment positions of the distributed micro-sensor array are planned to ensure coverage of different micro-regions within the container. Environmental data of each micro-region is collected according to a preset time sequence, forming multiple sets of micro-region environmental data subsets. This enables precise capture of local environmental differences within the culture container, compensating for the shortcomings of global acquisition in reflecting local environmental conditions.

[0025] Regional Difference Analysis Unit: Based on a subset of micro-regional environmental data, an adapted difference calculation method is used to compare the degree of deviation between each micro-region and the overall environmental parameters. Combined with data visualization tools, a micro-regional environmental difference heat map is constructed to intuitively present the environmental distribution and deviation level of each micro-region, providing an intuitive basis for subsequent regional targeted regulation.

[0026] Data fusion output unit: Constructs a feature association mapping mechanism to fuse the spatial distribution information contained in the micro-region environmental difference heat map with a high-quality standardized environmental feature set, supplementing the feature information of the regional dimension. The fusion process fully considers the weight ratio of global features and local features to form a complete standardized environmental feature set, which is output to the feature fusion enhancement module through a preset communication method.

[0027] The feature fusion enhancement module is implemented as follows: Multimodal Feature Mapping Unit: Based on the standardized environmental feature set output by the environmental perception module, an adapted feature encoder is constructed according to data type classification. Differentiated feature extraction logic is designed for the data characteristics of different dimensions such as physical, chemical and biological data. The features of each dimension are converted into feature vectors of a unified dimension and integrated to form a multi-dimensional feature vector set, laying the foundation for cross-dimensional fusion.

[0028] Cross-dimensional fusion unit: A multi-dimensional feature fusion model is constructed by adopting a Transformer-based cross-modal attention fusion algorithm. The self-attention mechanism is used to mine the internal dependencies of single-dimensional features and enhance the representation ability of single-dimensional features. Then, the interaction relationship between different-dimensional features is constructed through the cross-attention mechanism to achieve deep semantic fusion of multi-modal features. The fusion process is executed according to a preset iterative logic, ultimately outputting global fused feature data. Linear embedding is performed on the feature vectors of each dimension to achieve a unified mapping of feature dimensions. Simultaneously, positional encoding information is added to preserve the temporal characteristics of the data, ensuring that temporal information can be fully utilized in subsequent fusion processes, thus improving the integrity of feature fusion. The embedded and encoded feature data of each dimension are input into the attention layer. By calculating the dependency weights of each element within a single-dimensional feature, the representation of key information is strengthened, and redundant information is suppressed, resulting in enhanced feature data for each dimension. Following a preset query, key, and value matching logic, the enhanced features of each dimension are used sequentially as query information, and the enhanced features of the other two dimensions are concatenated as key and value information. Cross-attention calculation is performed, and this process is executed iteratively to fully explore the interactive relationships between features of different dimensions, obtaining intermediate fused feature data. This intermediate fused feature data is then input into a feedforward neural network for optimization, ultimately outputting global fused feature data.

[0029] Feature Enhancement Unit: Based on globally fused feature data, an attention mechanism is introduced to construct a feature enhancement model. By calculating the importance weights of each feature channel, key features that significantly affect cell growth status assessment are strengthened, interference from redundant feature information is suppressed, and the discriminative power and robustness of feature representation are further improved, generating high-dimensional fused feature data.

[0030] Feature Dimensionality Reduction Unit: To address the dimensionality redundancy problem in high-dimensional fused feature data, principal component analysis is used to perform dimensionality reduction. By calculating the feature covariance matrix, principal components whose cumulative contribution rate meets the preset requirements are selected, and invalid redundant dimensions are eliminated to obtain low-dimensional optimized fused features, thereby improving the efficiency of subsequent model processing.

[0031] Feature Validation Unit: Constructs a feature matching validation mechanism to compare the low-dimensional optimized fused features with historical feature data in the cell growth baseline database and calculate the feature matching degree. If the matching degree meets the preset threshold, it indicates that the current fused feature is effective, and high-dimensional fused feature data is output; if the matching degree does not meet the requirements, it returns to the cross-dimensional fusion unit, adjusts the fusion model parameters, and re-optimizes until the feature matching degree meets the standard.

[0032] Feature Time Series Association Unit: Introduces a time series feature extraction model to associate high-dimensional fusion feature data of continuous time series. By extracting historical time series feature information, it supplements the time dimension attributes of features to form time series high-dimensional fusion feature data, providing time series support for subsequent trend prediction of cell growth status, and outputting it to the intelligent state assessment module.

[0033] The intelligent status assessment module is implemented as follows: Evaluation Index Construction Unit: Based on the time-series high-dimensional fused feature data output by the feature fusion enhancement module, and combined with the physiological characteristics of cell growth, a multi-dimensional core evaluation index system is constructed. This index system covers cell growth rate, morphological integrity, metabolic activity, and environmental adaptability, comprehensively reflecting the core state of cell growth and providing index support for quantitative evaluation.

[0034] Quantitative evaluation unit: The weight of each evaluation indicator is determined by the analytic hierarchy process (AHP), and a judgment matrix is ​​constructed by combining the experience of domain experts. After consistency testing, the final weight is determined. Based on the standard thresholds in the cell growth baseline database, an appropriate scoring method is used to calculate the quantitative score of each indicator. Finally, the comprehensive evaluation value is calculated by combining the indicator weights to achieve a quantitative representation of cell growth status.

[0035] State Clustering Unit: An improved clustering algorithm with cell generation weight coefficients is adopted to perform cluster analysis on the comprehensive evaluation value. Combining the influence of cell generation on growth state, differentiated generation weight coefficients are set, and cluster centers are initialized according to the preset number of clusters. By iteratively updating the cluster centers, the cell growth state is divided into four categories: optimal growth state, adaptive state to be regulated, critical abnormal state, and severe abnormal state, so as to achieve fine classification of growth state. The improved clustering algorithm update formula is as follows:

[0036] In the formula, For the first Cluster centers after the next iteration For the first The generational weighting coefficients of individual cell samples. For the first The overall assessment value of each cell sample.

[0037] Trend prediction unit: Based on time-series high-dimensional fusion feature data, a time-series prediction model is trained. The model parameters are optimized using historical feature sequences and corresponding growth states as training samples, so that the model has the ability to extract time-series features and predict trends. The input is the historical feature sequence to be predicted, and the output is the trend of cell growth state changes and the probability of abnormal risks within a preset time period, so as to achieve a forward-looking prediction of growth state. The formula for updating the hidden state of the LSTM core is as follows:

[0038] In the formula, for LSTM hidden state at time step for Output gate output value at any time, The hyperbolic tangent activation function is used. for Cellular state at all times It is the Sigmoid activation function. This is the output gate weight matrix. for Hide your status at all times. for Temporal high-dimensional fusion features of input at any time; This is the output gate bias term; Evaluation and calibration unit: Constructs a historical data traceability and evaluation optimization mechanism, calls historical monitoring and control data within a preset time period, compares the actual growth status with the evaluation results, calculates the evaluation error, and if the evaluation error exceeds the preset range, uses an appropriate optimization algorithm to correct the evaluation index weights and clustering thresholds, and iterates and calibrates repeatedly until the evaluation error meets the requirements, thereby improving the accuracy of the evaluation results.

[0039] Report generation unit: Integrates quantitative assessment results, status classification, trend prediction and other information to build a standardized assessment report generation mechanism. The report includes status level labeling, abnormal risk warning and key influencing factor analysis. It adopts an intuitive presentation format to label the status level and risk level, and outputs to the precision control strategy module through a preset interface to provide a comprehensive basis for the formulation of control strategies.

[0040] The specific implementation process for the precise control strategy module is as follows: Regulatory Factor Extraction Unit: Based on the cell growth status assessment report output by the intelligent status assessment module, the unit uses an association rule mining algorithm to extract key regulatory factors that affect cell growth status. Combining the physiological needs of cell growth and environmental adaptation characteristics, the unit determines the categories of core regulatory factors, covering key areas such as temperature regulation, pH calibration, nutrient supplementation, and pollution control.

[0041] Factor Priority Ranking Unit: A combined weighting strategy of entropy weighting and analytic hierarchy process (AHP) is adopted to determine the priority weight of each regulatory factor. Entropy weighting calculates objective weights based on historical regulatory data, reflecting the information entropy characteristics of the data itself; AHP calculates subjective weights based on the experience of domain experts, reflecting the regulatory needs in actual applications. The final priority weights are obtained through weighted fusion, clarifying the regulatory priority of each regulatory factor.

[0042] Regulation Model Construction Unit: For different growth states, a personalized regulation model is constructed. Based on the key influencing factors and regulation factor priorities in the evaluation report, a differentiated regulation logic is designed to match the appropriate regulation direction and initial regulation parameter range for each growth state, ensuring the relevance and adaptability of the regulation model.

[0043] Regional Targeted Regulation Unit: Combining the micro-regional environmental difference heatmap output by the environmental sensing module, a regional differentiated regulation mechanism is constructed. Based on the degree of environmental deviation and the corresponding cell growth state in different micro-regions, the initial regulation parameter range is specifically modified to form regional regulation parameters, thereby achieving precise regulation with a one-region-one-policy approach and solving the problem of inaccurate regulation caused by uneven local environment.

[0044] Pre-validation unit for regulatory effects: A virtual simulation model of cell culture is constructed based on digital twin technology to reproduce the environmental characteristics and cell growth patterns of actual cell culture. Regionalized regulatory parameters and current environmental characteristics are input into the simulation model to simulate the cell growth state within a preset time after regulation, and the regulation simulation optimization rate is calculated. If the optimization rate meets the preset requirements, the parameter output stage begins; if the optimization rate does not meet the requirements, the weights of the regulatory factors are readjusted, and the verification is repeated until the requirements are met.

[0045] Parameter optimization unit: The unit uses an intelligent optimization algorithm to finely optimize the verified regional control parameters. The objective function is set as the optimal control effect. Through iterative optimization, the range of parameter fluctuations is narrowed, and invalid parameter combinations are eliminated to obtain a subset of precise control parameters, thereby further improving the accuracy and stability of the control parameters.

[0046] Control command generation unit: The precise control parameter subset is structured and organized according to the control factor type and micro-region division to generate standardized control commands. The commands clearly specify the operation requirements, execution timing and duration of each control device. The commands are sent to the field control devices using an appropriate industrial communication protocol to ensure the accurate execution of the control commands.

[0047] Parameter Feedback Unit: Constructs a regulatory effect feedback mechanism. Through the environmental perception module, it collects environmental data and cell growth status data in real time after the execution of regulatory commands, calculates the regulatory effect evaluation value, and transmits the feedback data to the intelligent status evaluation module for evaluation index calibration, regulatory model optimization, and iterative improvement of subsequent regulatory strategies, forming a closed-loop optimization.

[0048] The regulation database update unit constructs a historical data storage and update mechanism, storing evaluation reports, regulation parameters, and effect data from this regulation process into the system database according to preset naming rules, enriching the historical sample library. When the number of new samples added to the database reaches a preset threshold, incremental training is performed on the regulation model to continuously improve the adaptability and accuracy of the regulation model and ensure the stability of subsequent regulation effects.

[0049] In summary, the advantages of this invention are as follows: Through the multi-dimensional data acquisition unit and micro-area acquisition unit of the environmental sensing module, combined with the distributed micro-sensor array, physical, chemical and biological multi-dimensional environmental data and environmental difference data of different micro-areas can be collected simultaneously. With the help of regional difference analysis and data fusion, it makes up for the shortcomings of existing technologies that focus on single-point acquisition and ignore regional differences, and provides comprehensive and accurate data source support for subsequent precise regulation. The feature fusion enhancement module adopts a Transformer-based cross-modal cross-attention fusion algorithm. It models the intra-dimensional dependency of single-dimensional features through self-attention and the interaction between multi-dimensional features through cross-attention. Combined with feature dimensionality reduction, verification and time-series correlation processing, it effectively mines the deep semantic association of multi-dimensional heterogeneous data, improves the discriminativeness and robustness of feature representation, and solves the problems of simple feature fusion and insufficient representation ability in existing technologies. The intelligent state assessment module constructs a multi-core indicator system and adopts an improved clustering algorithm that introduces cell generation weight coefficients to achieve refined classification of cell growth status. Combined with the LSTM deep learning network, it can accurately predict the trend of cell growth status changes and the probability of abnormal risks in the next 24 hours. With the assessment and calibration mechanism, it further improves the accuracy of assessment and prediction, and provides a scientific basis for precise regulation. The precision regulation strategy module constructs a personalized regulation model based on cell growth status assessment reports and micro-regional environmental differences, achieving differentiated and precise regulation through regional targeted regulation. Combined with the pre-validation of regulation effects using digital twin technology and the fine-tuning of parameters using particle swarm optimization algorithms, it can effectively avoid over- or under-regulation and improve the effectiveness of regulation. At the same time, through parameter feedback and database update mechanisms, it enables continuous iterative optimization of the regulation model, ensuring the stability and consistency of cell culture. This system achieves intelligent processing of the entire process from environmental perception, feature processing, state assessment to control execution through the collaborative work of various modules. It significantly reduces the cost of manual intervention, reduces human error, can be widely adapted to different cell culture scenarios, and significantly improves the quality and efficiency of cell culture. It has important practical application value and promotion prospects. The regulatory database update unit stores the evaluation report, regulatory parameters, and effect data of this regulatory process into the system database, enriches the historical samples, improves the adaptability of subsequent regulatory models, and finally outputs a targeted combination of culture environment regulatory parameters.

[0050] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A smart monitoring and control system for cell culture environment, characterized in that, include: Environmental perception module, feature fusion enhancement module, intelligent state assessment module, and precise control strategy module; The environmental perception module is used to collect multi-dimensional heterogeneous data of the cell culture environment and generate a standardized environmental feature set. The feature fusion enhancement module is electrically connected to the environment perception module. The feature fusion enhancement module is used to perform multimodal fusion and cross-dimensional feature association and enhancement based on the standardized environmental feature set to generate high-dimensional fusion feature data. The intelligent state assessment module is electrically connected to the feature fusion enhancement module. The intelligent state assessment module is used to perform quantitative assessment and type classification of cell growth status based on the high-dimensional fusion feature data to obtain a cell growth status assessment report. The precise regulation strategy module is electrically connected to the intelligent state assessment module. The precise regulation strategy module is used to construct a personalized regulation model based on the cell growth state assessment report and output a targeted combination of culture environment regulation parameters.

2. The intelligent monitoring and control system for cell culture environment according to claim 1, characterized in that, The environment sensing module includes: The data acquisition unit collects multi-dimensional heterogeneous environment data based on the needs of cell culture scenarios. The multi-dimensional heterogeneous environment data includes physical environment data, chemical environment data, and biological environment data. The data preprocessing unit uses an adaptive filtering algorithm to reduce noise in the collected multi-dimensional heterogeneous environmental data, mapping data of different dimensions to a unified range and generating a standardized environmental feature set. The data quality verification unit constructs data validity evaluation indicators, performs quality verification on the preprocessed standardized environmental feature set, removes invalid data and supplements missing data, and outputs a high-quality standardized environmental feature set.

3. The intelligent monitoring and control system for cell culture environment according to claim 2, characterized in that, The environmental sensing module also includes: The micro-region acquisition unit deploys a distributed array of micro sensors within the culture container to collect environmental data from different micro-regions at a preset sampling frequency, generating a subset of micro-region environmental data. The regional difference analysis unit calculates the parameter deviations between each micro-region and the overall environment based on a subset of micro-region environmental data, and constructs a heat map of micro-region environmental differences. The data fusion output unit associates the micro-regional environmental difference heatmap with a high-quality standardized environmental feature set, supplements regional dimensional information, forms a complete standardized environmental feature set, and outputs it to the feature fusion enhancement module.

4. The intelligent monitoring and control system for cell culture environment according to claim 3, characterized in that, The feature fusion enhancement module includes: The multimodal feature mapping unit, based on the standardized environmental feature set output by the environmental perception module, establishes corresponding feature encoders according to data types, and converts physical, chemical, and biological features into feature vectors respectively, thus obtaining a multi-dimensional feature vector set; The cross-dimensional fusion unit uses a Transformer-based cross-modal cross-attention fusion algorithm. It performs deep semantic fusion of multi-modal features by modeling intra-dimensional feature dependencies through self-attention and cross-attention modeling interactions between multi-dimensional features, and obtains global fused feature data. The feature enhancement unit, based on globally fused feature data, uses an attention mechanism to strengthen the weights of key features, suppress redundant information, and generate high-dimensional fused feature data.

5. The intelligent monitoring and control system for cell culture environment according to claim 4, characterized in that, The feature fusion enhancement module also includes: The feature dimensionality reduction unit reduces the dimensionality of high-dimensional fusion feature data based on principal component analysis, retains principal components that exceed the cumulative contribution rate threshold, and obtains low-dimensional optimized fusion features. The feature verification unit compares the low-dimensional optimized fusion features with historical feature data in the cell growth baseline database, calculates the feature matching degree, and outputs high-dimensional fusion feature data if the matching degree is greater than the matching degree threshold; otherwise, it returns to the cross-dimensional fusion unit for re-optimization. The feature time series association unit associates high-dimensional fusion feature data of continuous time series, supplements time dimension information, forms time-series high-dimensional fusion feature data, and outputs it to the intelligent state assessment module.

6. The intelligent monitoring and control system for cell culture environment according to claim 5, characterized in that, The intelligent state assessment module includes: The evaluation index construction unit, based on the time-series high-dimensional fusion feature data output by the feature fusion enhancement module, constructs a cell growth status evaluation index system, including four core indicators: growth rate, morphological integrity, metabolic activity, and environmental adaptability. The quantitative evaluation unit uses the analytic hierarchy process (AHP) to determine the weight of each evaluation indicator, and combines the standard thresholds in the cell growth baseline database to calculate the quantitative score of each indicator and obtain the comprehensive evaluation value. The state clustering unit, based on the comprehensive evaluation value, adopts an improved clustering algorithm that introduces cell generation weight coefficients to divide cell growth states into four categories: optimal growth state, adaptive state to be regulated, critical abnormal state, and severely abnormal state. The update formula of the improved clustering algorithm is as follows: ， In the formula, For the first Cluster centers after the next iteration For the first The generational weighting coefficients of individual cell samples. For the first The overall assessment value of each cell sample.

7. The intelligent monitoring and control system for cell culture environment according to claim 6, characterized in that, The intelligent state assessment module also includes: The trend prediction unit, based on time-series high-dimensional fused feature data, trains an LSTM deep learning network. It takes historical feature sequences and corresponding growth states as input and outputs the trend of cell growth state changes and the probability of abnormal risks for the next 24 hours. The core hidden state update formula of the LSTM is as follows: ， In the formula, for LSTM hidden state at time step for Output gate output value at any time, The hyperbolic tangent activation function is used. for Cellular state at all times It is the Sigmoid activation function. This is the output gate weight matrix. for Hide your status at all times. for Temporal high-dimensional fusion features of input at any time; This is the output gate bias term; The evaluation and calibration unit calls up historical monitoring and control data, compares the actual growth status with the evaluation results, and corrects the weights of the evaluation indicators and the clustering thresholds. The report generation unit integrates quantitative assessment results, state classification, and trend prediction information to generate a cell growth status assessment report that includes state level, risk warning, and key influencing factors, and outputs it to the precision regulation strategy module.

8. The intelligent monitoring and control system for cell culture environment according to claim 7, characterized in that, The precise control strategy module includes: The regulatory factor extraction unit extracts key regulatory factors affecting cell growth status based on the cell growth status assessment report output by the intelligent status assessment module, including temperature regulation, pH calibration, nutrient supplementation, and pollution control. The factor priority ranking unit uses the entropy weight method combined with the analytic hierarchy process to comprehensively consider the sensitivity of the regulatory factors to different growth states, the speed of regulatory response, and energy consumption costs to determine the priority weight of each regulatory factor. The regulation model construction unit builds personalized regulation models for different growth states, inputs key influencing factors and regulation factor priorities from the evaluation report, and outputs the initial regulation parameter range.

9. The intelligent monitoring and control system for cell culture environment according to claim 8, characterized in that, The precise control strategy module also includes: The regional targeted regulation unit, combined with the micro-region environmental difference heat map output by the environmental sensing module, performs differential correction on the initial regulation parameter range based on the environmental deviation and cell growth status of different micro-regions, and obtains regional regulation parameters. The pre-verification unit for the regulation effect constructs a virtual simulation model of cell culture based on digital twin technology. It inputs regionalized regulation parameters and current environmental characteristics to simulate the cell growth state after regulation. If the simulation optimization rate is greater than the optimization rate threshold, it enters the parameter output stage; otherwise, it readjusts the weights of the regulation factors. The parameter optimization unit uses the particle swarm optimization algorithm to finely optimize the validated regional control parameters, reduce the range of parameter fluctuations, and obtain a subset of precise control parameters.

10. The intelligent monitoring and control system for cell culture environment according to claim 9, characterized in that, The precise control strategy module also includes: The control command generation unit divides the precise control parameter subset according to the control factor type and micro-region, generates structured control commands, and clarifies the operating parameters, execution timing and duration of each control device; The parameter feedback unit records environmental data and changes in cell growth status after the execution of the regulation command, and feeds them back to the intelligent state assessment module for evaluation index calibration and regulation model optimization. The regulatory database update unit stores the evaluation report, regulatory parameters, and effect data of this regulatory process into the system database, enriches the historical samples, improves the adaptability of subsequent regulatory models, and finally outputs a targeted combination of culture environment regulatory parameters.

Images