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

Abstract

The present application relates to the technical field of visual chart generation, in particular to a microorganism data visualization chart generation system and method based on model parameters. The present application combines PCA dimension reduction and weighted LSTM modeling to improve the processing efficiency and prediction accuracy of microorganism culture experiment data. Through PCA dimension reduction, the key features of the culture environment, microorganism characteristics and culture process are extracted, and redundant information is reduced, so that the model has better generalization ability. Based on the comparison between the prediction of the weighted LSTM model and the actual statistical microorganism amount, the experiment trend is visualized to help optimize the culture conditions. The real-time error calculation and abnormal detection mechanism can dynamically monitor the microorganism growth, accurately identify the abnormal time point, and through the LOF algorithm, the key abnormal components are analyzed in depth to find out the abnormal experiment parameters, and finally the visualization image is generated to intuitively show the abnormal source and influencing factors.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of visualization chart generation, in particular to a microorganism data visualization chart generation system and method based on model parameters. BACKGROUND

[0002] In the process of microorganism cultivation, small changes in factors such as cultivation environment, microorganism characteristics and operating conditions may lead to abnormal cultivation results, thereby affecting the accuracy of experiments and the stability of production. Therefore, it is crucial to find and analyze abnormal data in real time during the cultivation process for optimizing cultivation conditions, improving experimental repeatability and ensuring production quality. Traditional anomaly detection methods often rely on manual detection or simple threshold judgment, which cannot track the dynamic growth of microorganisms in real time, and may lead to the failure to discover abnormalities in time, thereby affecting experimental and production decisions.

[0003] Currently, microorganism cultivation data analysis mainly adopts static statistical methods such as mean analysis, variance analysis and traditional regression models. Although these methods can provide certain reference information, they have many limitations. First, static analysis methods are usually based on data at fixed time points, which cannot effectively reflect the complex dynamic changes in the process of microorganism cultivation. Second, existing analysis methods lack intuitive data visualization presentation methods, making it difficult for researchers to quickly understand data trends and abnormal points. In addition, a single statistical analysis method cannot fully exploit the potential information of experimental data, limiting the optimization space of the cultivation process. Therefore, there is an urgent need for a system that can detect abnormalities in real time, dynamically model and analyze, and visually display the changes in cultivation data, in order to improve the intelligent level of the microorganism cultivation process.

[0004] To this end, a microorganism data visualization chart generation system and method based on model parameters are proposed. SUMMARY

[0005] The present application aims to provide a microorganism data visualization chart generation system and method based on model parameters. Through PCA dimension reduction, weighted LSTM modeling and local outlier factor (LOF) algorithm, the quality, prediction accuracy and abnormal detection capability of microorganism cultivation data are improved. First, PCA dimension reduction optimizes the processing of historical experimental data, removes redundant information, retains key features, and improves computational efficiency and data consistency. Second, based on the weighted LSTM model, the experimental data after dimension reduction is accurately modeled, the core features of the cultivation environment, microorganism characteristics and cultivation process are extracted using different LSTM sub-modules, and the information fusion is optimized through dynamic weight distribution, realizing high-precision microorganism quantity prediction. Finally, the LOF algorithm combines KNN density estimation to analyze abnormal experimental parameters, accurately identifies key abnormal components, and visually displays the abnormal factors, providing a scientific basis for experimental optimization, abnormal source tracing and data correction.

[0006] To achieve the above object, the present application provides the following technical solutions:

[0007] The microorganism data visualization chart generation system based on model parameters comprises:

[0008] The principal component parameter acquisition module is configured to acquire historical experimental data of microorganism culture, perform dimension reduction on the historical experimental data by using a PCA method, and obtain reduced dimension experimental data; the historical experimental data comprises historical experimental parameters and statistical microorganism amount;

[0009] The contrast data visualization module is configured to model the reduced dimension experimental data by using a weighted LSTM model, obtain a pre-trained weighted LSTM model, record test microorganism amount generated by a test set, and compare the test microorganism amount with the statistical microorganism amount to generate a contrast visualization image;

[0010] The error visualization module is configured to record an initial microorganism amount, perform microorganism amount prediction and microorganism amount collection every first time interval, obtain predicted microorganism amount and real-time microorganism amount, calculate real-time error rate between the predicted microorganism amount and the real-time microorganism amount, and generate an error visualization image;

[0011] The anomaly detection module is configured to judge a relationship between the real-time error rate and an error threshold value, mark a calculation time of the real-time error rate as an anomaly point if the real-time error rate is greater than the error threshold value, and mark the anomaly point in the error visualization image;

[0012] The anomaly data visualization module is configured to acquire abnormal experimental parameters at the anomaly point, calculate LOF values of each component of the abnormal experimental parameters by using a local outlier factor algorithm, acquire key abnormal components of the abnormal experimental parameters, and output a visualization image of the key abnormal components.

[0013] Further, the dimension reduction on the historical experimental data by using the PCA method comprises:

[0014] Performing missing value filling, abnormal value removal and standardization processing on historical experimental parameters of the historical experimental data to obtain standardized experimental parameters;

[0015] Calculating a covariance matrix of the standardized experimental parameters, solving eigenvalues and corresponding eigenvectors of the covariance matrix, and acquiring principal component variation directions;

[0016] Selecting a number of principal component spaces according to cumulative variance contribution rates, and projecting the standardized experimental parameters to the selected principal component spaces to obtain principal component parameters.

[0017] Further, the dimensionality reduction experimental data comprises principal component parameters and statistical microbial quantities, and the principal component parameters comprise culture environment parameters, microbial characteristic parameters and culture process parameters.

[0018] Further, the weighted LSTM model comprises:

[0019] a data preprocessing module configured to clean and standardize input data;

[0020] an LSTM module comprising three LSTM sub-modules configured to process culture environment parameters, microbial characteristic parameters and culture process parameters, respectively;

[0021] a weight distribution module configured to distribute weights to outputs of each LSTM sub-module;

[0022] a weighted fusion module configured to sum outputs of different LSTM sub-modules to obtain a weighted hidden state;

[0023] an output module configured to predict the weighted hidden state and output a predicted microbial quantity;

[0024] The weighted hidden state calculation formula is:

[0025] h final = ω env *h env + ω micro *h micro + ω process *h process ;

[0026] wherein h final represents the weighted hidden state, h env represents an environment hidden state, h micro represents a characteristic hidden state, h process represents a process hidden state, ω env represents an environment weight coefficient, ω micro represents a characteristic weight coefficient, and ω process represents a process weight coefficient.

[0027] Further, obtaining an abnormal component of an abnormal experimental parameter comprises:

[0028] obtaining an abnormal experimental parameter, and merging the historical experimental parameter and the abnormal experimental parameter into a complete data set;

[0029] calculating a LOF value of each parameter component of the complete data set by KNN density estimation to determine a local abnormal degree of each parameter component;

[0030] setting an abnormal threshold, screening out abnormal components with a LOF value greater than the abnormal threshold, and constructing an abnormal component set.

[0031] Analyze the set of abnormal components to find key abnormal components that cause the experimental data to be abnormal.

[0032] The application also proposes a model parameter-based microbial data visualization chart generation method, comprising:

[0033] Obtain historical experimental data of microbial culture, use PCA method to reduce dimension of the historical experimental data, and obtain reduced dimension experimental data; the historical experimental data includes historical experimental parameters and statistical microbial quantity;

[0034] Model the reduced dimension experimental data through a weighted LSTM model to obtain a pre-trained weighted LSTM model, record the test microbial quantity generated by the test set, and compare it with the statistical microbial quantity to generate a comparison visualization image;

[0035] Record the initial microbial quantity, perform microbial quantity prediction and microbial quantity collection every first time interval, obtain predicted microbial quantity and real-time microbial quantity, calculate the real-time error rate between them, and generate an error visualization image;

[0036] Determine the relationship between the real-time error rate and the error threshold, if the real-time error rate is greater than the error threshold, mark the calculation time of the real-time error rate as an abnormal point, and mark the abnormal point in the error visualization image;

[0037] Obtain abnormal experimental parameters at the abnormal point, calculate the LOF value of each component of the abnormal experimental parameters through a local outlier factor algorithm, obtain key abnormal components of the abnormal experimental parameters, and output a visualization image of the key abnormal components.

[0038] Further, using the PCA method to reduce dimension of the historical experimental data comprises:

[0039] Fill in missing values, remove outliers, and standardize the historical experimental parameters of the historical experimental data to obtain standardized experimental parameters;

[0040] Calculate the covariance matrix of the standardized experimental parameters, solve the eigenvalues and corresponding eigenvectors of the covariance matrix, and obtain the principal component variation direction;

[0041] Select the number of principal component spaces according to the cumulative variance contribution rate, and project the standardized experimental parameters to the selected principal component spaces to obtain principal component parameters.

[0042] Further, the reduced dimension experimental data includes principal component parameters and statistical microbial quantity, and the principal component parameters include culture environment parameters, microbial characteristic parameters, and culture process parameters.

[0043] Further, the weighted LSTM model comprises:

[0044] a data preprocessing module for cleaning and standardizing input data;

[0045] an LSTM module comprising three LSTM sub-modules for processing culture environment parameters, microbial characteristic parameters and culture process parameters, respectively;

[0046] a weight distribution module for distributing weights to the outputs of each LSTM sub-module;

[0047] a weighted fusion module for weighted summation of the outputs of different LSTM sub-modules to obtain a weighted hidden state;

[0048] an output module for predicting the weighted hidden state to output a predicted microbial amount;

[0049] The weighted hidden state calculation formula is:

[0050] h final =ω env *h env +ω micro *h micro +ω process *h process ;

[0051] wherein h final represents the weighted hidden state, h env represents the environment hidden state, h micro represents the characteristic hidden state, h process represents the process hidden state, ω env represents the environment weight coefficient, ω micro represents the characteristic weight coefficient, and ω process represents the process weight coefficient.

[0052] Further, obtaining the abnormal components of the abnormal experimental parameters comprises:

[0053] obtaining abnormal experimental parameters, and merging the historical experimental parameters and the abnormal experimental parameters into a complete data set;

[0054] calculating the LOF value of each parameter component of the complete data set by KNN density estimation to determine the local abnormal degree of each parameter component;

[0055] setting an abnormal threshold, screening out abnormal components with LOF values greater than the abnormal threshold, and constructing an abnormal component set;

[0056] analyzing the abnormal component set to find key abnormal components causing abnormal experimental data.

[0057] Compared with the prior art, the present application has the following beneficial effects:

[0058] 1. The historical experimental data of microbial culture is optimized by PCA dimension reduction, improving the data quality and calculation efficiency. First, missing value filling and outlier removal ensure data integrity, and standardization processing eliminates the dimensional difference of different parameters, enhancing data consistency. Second, covariance matrix analysis combined with eigenvalue and eigenvector calculation can accurately identify the principal component change direction of experimental parameters, and extract the most representative feature information. Finally, by selecting the optimal principal component dimension through cumulative variance contribution rate, the key information is retained while the data is reduced, and the redundancy is reduced.

[0059] 2. The experimental data after dimension reduction is modeled by weighted LSTM model, improving the prediction accuracy and data utilization efficiency of microbial culture process. Through data preprocessing to ensure input quality, LSTM submodule extracts key features of culture environment, microbial characteristics and culture process, and weight distribution module dynamically adjusts the contribution of each module to optimize information fusion. Finally, the model accurately predicts the number of microorganisms, and the prediction results and actual statistical data are compared through visual images, providing efficient and accurate data support for optimization and scientific decision-making of microbial culture process.

[0060] 3. The key abnormal components of abnormal experimental parameters are accurately identified by local outlier factor algorithm, improving the accuracy of anomaly detection. First, combined with historical experimental data, a complete data set is constructed to ensure the comprehensiveness of anomaly analysis. Then, the LOF value of each parameter component is calculated by KNN density estimation, quantifying its local anomaly degree. By setting an abnormal threshold, abnormal components are selected to construct an abnormal component set, accurately locking the key factors causing experimental anomalies. Finally, the key abnormal components are intuitively displayed through visual images, providing scientific basis for experimental optimization, anomaly tracing and data correction. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 The structure diagram of the microbial data visualization chart generation system based on model parameters is provided for the present application;

[0062] Figure 2 The structure diagram of the weighted LSTM model is provided for the present application;

[0063] Figure 3 The flow chart of the microbial data visualization chart generation method based on model parameters is provided for the present application. DETAILED DESCRIPTION

[0064] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] Example 1

[0066] Microbial data visualization chart generation system based on model parameters, such as Figure 1 As shown, it includes:

[0067] The principal component parameter acquisition module is used to acquire historical experimental data of microbial culture, and to perform dimensionality reduction on the historical experimental data using the PCA method to obtain dimensionality-reduced experimental data; the historical experimental data includes historical experimental parameters and statistical microbial biomass.

[0068] Specifically, the historical experimental data is time-series data, with each time interval being the first time interval. The historical experimental data for each time interval includes historical experimental parameters and statistical microbial biomass. In this embodiment, the first time interval is 6 hours, and the historical experimental parameters include parameters such as temperature, pH value, dissolved oxygen concentration, humidity, light conditions, microbial activity, and metabolite concentration.

[0069] Furthermore, the dimensionality reduction of the historical experimental data using the PCA method includes:

[0070] The historical experimental parameters of the historical experimental data are filled with missing values, removed outliers, and standardized to obtain standardized experimental parameters.

[0071] Calculate the covariance matrix of the standardized experimental parameters, solve for the eigenvalues ​​and corresponding eigenvectors of the covariance matrix, and obtain the direction of change of the principal components;

[0072] The number of principal component spaces is selected based on the cumulative variance contribution rate, and the standardized experimental parameters are projected onto the selected principal component spaces to obtain the principal component parameters.

[0073] Specifically, the mean imputation method was used to fill in the missing values ​​in all historical experimental parameters. The Z-score algorithm was used to detect outliers for each parameter. Values ​​exceeding 3 times the standard deviation were considered outliers and removed. All historical experimental parameters were standardized to have zero mean and unit variance, which facilitates subsequent PCA dimensionality reduction.

[0074] The historical experimental data of microbial cultivation is reduced in dimension by the PCA method, which can improve the data quality, reduce noise, eliminate the influence of dimension by standardization processing, and improve the stability and calculation efficiency of the model. Covariance matrix analysis ensures that the key features are retained, and cumulative variance contribution rate optimizes principal component selection, so that the reduced data reduces redundancy while maintaining main information, avoiding overfitting.

[0075] Further, the reduced experimental data includes principal component parameters and statistical microbial quantities, and the principal component parameters include cultivation environment parameters, microbial characteristic parameters and cultivation process parameters.

[0076] Specifically, the reduced experimental data is obtained by reducing the historical experimental data, in which the historical experimental parameters as independent variables are reduced to principal component parameters, and the statistical microbial quantities as dependent variables remain unchanged. Part of the reduced experimental data under a certain time sequence is shown in Table 1.

[0077] Table 1 Part of the reduced experimental data under a certain time sequence

[0078] Experiment number Cultivation environment parameters Microorganism property parameters Cultivation process parameters Statistical microorganism quantity 1 0.32 0.45 -0.18 500 2 0.38 0.52 -0.22 510 3 0.25 0.40 -0.10 480 4 0.59 0.65 -0.25 520 5 0.54 0.63 -0.20 530

[0079] Through the reduced experimental data, the redundant information can be reduced, the calculation efficiency can be improved, the key features can be retained, and the data interpretability can be enhanced. The division of principal component parameters enables the cultivation environment, microbial characteristics and cultivation process factors to be refined, which helps to identify the core variables affecting microbial growth.

[0080] The comparative data visualization module is used to model the reduced experimental data by the weighted LSTM model, obtain a pre-trained weighted LSTM model, record the test microbial quantity generated by the test set, and compare it with the statistical microbial quantity to generate a comparative visualization image.

[0081] Further, the weighted LSTM model is as shown in Figure 2 , which includes:

[0082] The data preprocessing module is used to clean and standardize the input data.

[0083] The LSTM module includes three LSTM sub-modules, which are respectively used to process the cultivation environment parameters, microbial characteristic parameters and cultivation process parameters.

[0084] The weight distribution module is used to distribute weights to the output of each LSTM sub-module.

[0085] The weighted fusion module: the outputs of different LSTM sub-modules are weighted and summed to obtain the weighted hidden state.

[0086] The output module is used to predict the weighted hidden state and output the predicted microbial quantity.

[0087] The weighted hidden state calculation formula is:

[0088] h final = ω env *h env + ω micro *h micro + ω process *h process ;

[0089] wherein h final represents a weighted hidden state, h env represents an environmental hidden state, h micro represents a characteristic hidden state, h process represents a process hidden state, ω env represents an environmental weight coefficient, ω micro represents a characteristic weight coefficient, and ω process represents a process weight coefficient.

[0090] Specifically, the three LSTM sub-modules of the LSTM module respectively output h env , h micro and h process , which can be represented as:

[0091] h env = LSTM(X env , θ env ) ;

[0092] h micro = LSTM(X micro , θ micro ) ;

[0093] h process = LSTM(X process , θ process ) ;

[0094] wherein LSTM() represents an LSTM sub-module, X env represents a culture environment parameter, X micro represents a microorganism characteristic parameter, and X process represents a culture process parameter, and θ env , θ micro and θ process respectively represent the weights and biases of the LSTM sub-modules; the weight distribution module assigns weight coefficients to the outputs of each LSTM sub-module, wherein the calculation formula of the weight coefficients can be represented as:

[0095]

[0096] Wherein, exp() represents an exponential function, f() represents a scoring function, which can be a simple feedforward neural network or other learning mechanism, and the weighted average function calculated by the attention mechanism is adopted in the embodiment; the final output module outputs a prediction result, and the process can be represented as:

[0097]

[0098] Wherein, W f represents a weight matrix of the output module, and b f represents a bias term of the output module.

[0099] The weighted LSTM model is used to model the dimensionality reduction experimental data, which can not only effectively extract the time sequence characteristics of the key factors in the microbial culture process, but also process the culture environment, microbial characteristics and culture process data through different LSTM sub-modules to improve the pertinence and accuracy of feature extraction. The weight distribution module dynamically adjusts the influence weight of different sub-modules, so that the model can adaptively focus on the most important influencing factors, thereby improving the prediction accuracy and robustness. The weighted fusion module further optimizes information integration, so that the contribution of different factors is more reasonable, and finally the microbial quantity is accurately predicted through the output module, and a comparison visualization image is generated to intuitively display the prediction effect.

[0100] The error visualization module is used to record the initial microbial quantity, and the microbial quantity is predicted and collected every first time interval to obtain the predicted microbial quantity and the real-time microbial quantity, calculate the real-time error rate between them, and generate an error visualization image.

[0101] Specifically, the calculation formula of the real-time error rate is as follows:

[0102]

[0103] Wherein, E represents the real-time error rate, y represents the real-time microbial quantity, and y p represents the predicted microbial quantity.

[0104] The anomaly detection module is used to determine the relationship between the real-time error rate and the error threshold. If the real-time error rate is greater than the error threshold, the calculation time of the real-time error rate is marked as an anomaly point, and the anomaly point is marked in the error visualization image. In the embodiment, the error threshold is 0.2.

[0105] The abnormal data visualization module is used to obtain abnormal experimental parameters at the anomaly point, calculate the LOF value of each component of the abnormal experimental parameters through the local outlier factor algorithm, obtain the key abnormal components of the abnormal experimental parameters, and output a visualization image of the key abnormal components.

[0106] Further, the abnormal component of the abnormal experimental parameter comprises:

[0107] The abnormal experimental parameter is obtained, and the historical experimental parameter and the abnormal experimental parameter are combined into a complete data set;

[0108] The LOF value of each parameter component of the complete data set is calculated by using KNN density estimation, and the local abnormal degree of each parameter component is determined;

[0109] An abnormal threshold is set, and abnormal components with LOF values greater than the abnormal threshold are screened out to construct an abnormal component set;

[0110] The abnormal component set is analyzed to find key abnormal components causing abnormal experimental data.

[0111] Specifically, the abnormal experimental parameter at the abnormal point, i.e. the abnormal experimental data when the abnormality occurs, is obtained, wherein the abnormal experimental parameter is a 1×N dimensional vector, wherein N represents N components of the experimental parameter, including temperature, pH value, dissolved oxygen concentration, humidity, light condition, microbial activity and metabolic product concentration, etc. The historical experimental parameter is a M×N dimensional vector, wherein M represents the number of groups of experiments, and therefore the complete data set X is a (M+1)×N dimensional vector, which can be represented as X={x1,x2,...,x N}; The calculation formula of the LOF value of each parameter component can be represented as:

[0112]

[0113] Wherein, LOF represents the LOF value of the parameter component, k represents the number of nearest points of the KNN algorithm, N k (x i ) represents the k nearest points of point x i , and lrd(x i ) represents the local reachable density of point x i , which can be represented as:

[0114]

[0115] Wherein, rd(x i ,x j ) represents the reachable distance between point x i and point x j , which can be represented as:

[0116] rd(x i ,x j )=max(d(x i ,x j ),k_dist(x j ));

[0117] wherein max() represents a maximum value function, d(x i ,x j ) represents the Euclidean distance between two points, and k_dist(x j ) represents the distance from the point x j to its nearest neighbor; the abnormal threshold is set to 1.5, that is, when LOF>1.5, the point is considered to be an abnormal component, and an abnormal component set is constructed; the method for analyzing the abnormal component set can directly take the maximum value in the abnormal component set as the key abnormal component, or use independent component analysis method to analyze the relationship between each abnormal component to determine the key abnormal component, and the embodiment directly takes the maximum value in the abnormal component set as the key abnormal component. Table 2 is a LOF data table of part of experimental data.

[0118] Through the fine analysis of each component of the abnormal experimental parameter by the local outlier factor algorithm, the key factors causing the experimental abnormality can be effectively identified. By combining the historical experimental parameters to construct a complete data set and using KNN density estimation to calculate the LOF value, the abnormality detection is more robust and reliable. By setting an abnormal threshold, the most abnormal component can be accurately screened out to construct an abnormal component set, so as to clearly determine the core influencing factors of the experimental abnormality.

[0119] Table 2 is a LOF data table of part of experimental data.

[0120] Experiment number Parameter component 1 Parameter component 2 Parameter component 3 LOF1 LOF2 LOF3 1 25.4 3.2 45.6 0.75 1.20 0.80 2 30.2 2.7 47.9 1.45 0.98 1.05 3 32.1 3.9 50.8 1.65 1.30 1.55 4 26.5 3.1 49.0 0.90 1.10 0.92

[0121] The system combines PCA dimension reduction and weighted LSTM modeling to improve the processing efficiency and prediction accuracy of microbial culture experimental data. Through PCA dimension reduction, the key features of the culture environment, microbial characteristics and culture process are extracted, and redundant information is reduced, so that the model has better generalization ability. Based on the prediction of the weighted LSTM model and the actual statistical microbial amount, the experimental trend is visualized to help optimize the culture conditions. The real-time error calculation and abnormality detection mechanism can dynamically monitor the microbial growth, accurately identify the abnormal time point, and further analyze the abnormal experimental parameters through the LOF algorithm to find out the key abnormal components, and finally generate a visual image to intuitively show the abnormal source and influencing factors.

[0122] Embodiment Two

[0123] The method for generating a microbial data visualization chart based on model parameters, as shown in Figure 3 , comprises:

[0124] Obtain historical experimental data of microbial culture, use PCA method to reduce dimension of the historical experimental data to obtain reduced dimension experimental data; the historical experimental data includes historical experimental parameters and statistical microbial amount;

[0125] modeling the dimensionality reduction experiment data by using a weighted LSTM model to obtain a pre-training weighted LSTM model, recording the test microbial amount generated by a test set, and comparing the test microbial amount with the statistical microbial amount to generate a comparison visualization image;

[0126] recording an initial microbial quantity, performing microbial quantity prediction and microbial quantity collection every first time interval to obtain predicted microbial quantity and real-time microbial quantity and calculate a real-time error rate therebetween to generate an error visualization image;

[0127] judging a relationship between the real-time error rate and an error threshold, if the real-time error rate is greater than the error threshold, marking a calculation time of the real-time error rate as an abnormal point, and marking the abnormal point in the error visualization image;

[0128] obtaining abnormal experiment parameters at the abnormal point, calculating a LOF value of each component of the abnormal experiment parameters by using a local outlier factor algorithm, obtaining a key abnormal component of the abnormal experiment parameters, and outputting a visualization image of the key abnormal component.

[0129] Further, the PCA method is used to reduce the dimensionality of the historical experiment data, including:

[0130] performing missing value filling, abnormal value removal and standardization processing on historical experiment parameters of the historical experiment data to obtain standardized experiment parameters;

[0131] calculating a covariance matrix of the standardized experiment parameters, solving eigenvalues and corresponding eigenvectors of the covariance matrix, and obtaining principal component variation directions;

[0132] selecting a number of principal component spaces according to cumulative variance contribution rates, and projecting the standardized experiment parameters to the selected principal component spaces to obtain principal component parameters.

[0133] Further, the dimensionality reduction experiment data includes principal component parameters and statistical microbial quantity, and the principal component parameters include culture environment parameters, microbial characteristic parameters and culture process parameters.

[0134] Further, the weighted LSTM model includes:

[0135] a data preprocessing module for cleaning and standardizing input data;

[0136] an LSTM module including three LSTM sub-modules for processing culture environment parameters, microbial characteristic parameters and culture process parameters, respectively;

[0137] a weight distribution module for distributing weights to outputs of each LSTM sub-module;

[0138] The weighted fusion module: the outputs of different LSTM sub-modules are weighted and summed to obtain a weighted hidden state;

[0139] The output module is configured to predict the weighted hidden state and output a predicted microbial amount.

[0140] The weighted hidden state calculation formula is:

[0141] h final = ω env *h env + ω micro *h micro + ω process *h process ;

[0142] wherein h final represents the weighted hidden state, h env represents the environmental hidden state, h micro represents the characteristic hidden state, h process represents the process hidden state, ω env represents the environmental weight coefficient, ω micro represents the characteristic weight coefficient, and ω process represents the process weight coefficient.

[0143] Further, the abnormal component of the abnormal experimental parameter comprises:

[0144] The abnormal experimental parameter is obtained, and the historical experimental parameter and the abnormal experimental parameter are combined into a complete data set;

[0145] The LOF value of each parameter component of the complete data set is calculated by KNN density estimation, and the local abnormal degree of each parameter component is determined;

[0146] An abnormal threshold is set, and abnormal components with an LOF value greater than the abnormal threshold are screened out to construct an abnormal component set;

[0147] The abnormal component set is analyzed to find a key abnormal component causing abnormal experimental data.

[0148] Although embodiments of the present application have been shown and described, it is to be understood that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A model parameter-based microorganism data visualization chart generation system, characterized by, The method comprises the following steps: The principal component parameter acquisition module is configured to acquire historical experimental data of microbial culture, perform dimension reduction on the historical experimental data by using a PCA method, and obtain reduced dimension experimental data; the historical experimental data comprises historical experimental parameters and statistical microbial quantity; the reduced dimension experimental data comprises principal component parameters and statistical microbial quantity, and the principal component parameters comprise culture environment parameters, microbial characteristic parameters, and culture process parameters; The comparative data visualization module is used for modeling the dimensionality reduction experimental data by a weighted LSTM model to obtain a pre-trained weighted LSTM model, recording the test microbial amount generated by the test set, and comparing the test microbial amount with the statistical microbial amount to generate a comparative visualization image. 、 and The three LSTM sub-modules of the LSTM module output ； ； ； wherein denotes an LSTM submodule, denotes a cultivation environment parameter, denotes a microorganism property parameter, denotes a cultivation process parameter, , and denote weights and biases of the LSTM submodule, respectively. The error visualization module is configured to record an initial microbial quantity, perform microbial quantity prediction and microbial quantity collection every first time interval, obtain predicted microbial quantity and real-time microbial quantity, calculate a real-time error rate between the predicted microbial quantity and the real-time microbial quantity, and generate an error visualization image; The anomaly detection module is configured to determine a relationship between the real-time error rate and an error threshold value, and if the real-time error rate is greater than the error threshold value, mark a calculation time of the real-time error rate as an anomaly point, and mark the anomaly point in the error visualization image; The anomaly data visualization module is configured to acquire abnormal experimental parameters at the anomaly point, calculate LOF values of each component of the abnormal experimental parameters by using a local outlier factor algorithm, acquire key abnormal components of the abnormal experimental parameters, and output a visualization image of the key abnormal components; the acquisition of abnormal components of the abnormal experimental parameters comprises the following steps: acquiring abnormal experimental parameters, merging the historical experimental parameters and the abnormal experimental parameters into a complete data set; calculating LOF values of each parameter component of the complete data set by using KNN density estimation, and determining a local anomaly degree of each parameter component; setting an abnormal threshold value, screening abnormal components with LOF values greater than the abnormal threshold value, and constructing an abnormal component set; analyzing the abnormal component set, and finding key abnormal components causing abnormal experimental data; and a calculation formula of the LOF value of each parameter component can be expressed as: ； wherein, LOF value of a parameter component, number of nearest points of a KNN algorithm, nearest points of a point local reachable density of a point local reachable density of a point​​ 2. The model parameter based microorganism data visualization chart generating system of claim 1, wherein, The PCA method is used to perform dimension reduction on the historical experimental data, which comprises the following steps: The historical experimental parameters of the historical experimental data are subjected to missing value filling, abnormal value removal, and standardization processing, to obtain standardized experimental parameters; A covariance matrix of the standardized experimental parameters is calculated, eigenvalues and corresponding eigenvectors of the covariance matrix are solved, and a principal component variation direction is acquired; The number of principal component spaces is selected according to a cumulative variance contribution rate, and the standardized experimental parameters are projected into the selected principal component spaces, to obtain principal component parameters.

3. The model parameter based microorganism data visualization graph generation system of claim 1, wherein, The weighted LSTM model comprises: A data preprocessing module is configured to clean and standardize input data; A weight distribution module is configured to distribute weights to outputs of each LSTM submodule; A weighted fusion module is configured to sum the outputs of different LSTM submodules by using weights, to obtain a weighted hidden state; An output module is configured to perform prediction on the weighted hidden state, and output predicted microbial quantity; The calculation formula of the weighted hidden state is: ； wherein, represents a weighted hidden state, represents an environmental hidden state, represents a characteristic hidden state, represents a process hidden state, represents an environmental weight coefficient, represents a characteristic weight coefficient, represents a process weight coefficient.

4. A method for generating a visualization chart of microbial data based on model parameters, characterized by, The method comprises the following steps: The historical experimental data of the microbial culture is acquired, and a PCA method is used to reduce the dimension of the historical experimental data to obtain reduced dimension experimental data; the historical experimental data includes historical experimental parameters and statistical microbial quantity; the reduced dimension experimental data includes principal component parameters and statistical microbial quantity, and the principal component parameters include culture environment parameters, microbial characteristic parameters and culture process parameters; The dimensionality reduction experiment data is modeled by a weighted LSTM model to obtain a pre-trained weighted LSTM model, the test microbial amount generated by a test set is recorded, and the statistical microbial amount is compared to generate a comparison visualization image; the LSTM module includes three LSTM sub-modules for processing culture environment parameters, microbial characteristic parameters and culture process parameters respectively; the three LSTM sub-modules of the LSTM module output , and , which can be represented as: ； ； ； wherein denotes an LSTM submodule, denotes a cultivation environment parameter, denotes a microorganism property parameter, denotes a cultivation process parameter, , and denote weights and biases of the LSTM submodule, respectively. An initial microbial quantity is recorded, microbial quantity prediction and microbial quantity collection are performed every first time interval, predicted microbial quantity and real-time microbial quantity are obtained, a real-time error rate between the predicted microbial quantity and the real-time microbial quantity is calculated, and an error visualization image is generated; A relationship between the real-time error rate and an error threshold is judged, if the real-time error rate is greater than the error threshold, a calculation time of the real-time error rate is marked as an abnormal point, and the abnormal point is marked in the error visualization image; Abnormal experimental parameters at the abnormal point are acquired, a LOF value of each component of the abnormal experimental parameters is calculated through a local outlier factor algorithm, a key abnormal component of the abnormal experimental parameters is acquired, and a visualization image of the key abnormal component is output; acquiring the abnormal component of the abnormal experimental parameters includes: acquiring the abnormal experimental parameters, and merging the historical experimental parameters and the abnormal experimental parameters into a complete data set; a KNN density estimation is used to calculate the LOF value of each parameter component of the complete data set, and the local abnormal degree of each parameter component is determined; an abnormal threshold is set, abnormal components with a LOF value greater than the abnormal threshold are screened out to construct an abnormal component set; the abnormal component set is analyzed to find a key abnormal component causing the abnormality of the experimental data; and a calculation formula of the LOF value of each parameter component can be expressed as: ； wherein, LOF value of a parameter component, number of nearest points of a KNN algorithm, nearest points of a point local reachable density of a point local reachable density of a point​​ 5. The model parameter based microorganism data visualization chart generating method of claim 4, wherein, The PCA method is used to reduce the dimension of the historical experimental data, including: The historical experimental parameters of the historical experimental data are subjected to missing value filling, abnormal value removal and standardization processing to obtain standardized experimental parameters; A covariance matrix of the standardized experimental parameters is calculated, eigenvalues and corresponding eigenvectors of the covariance matrix are solved, and a principal component variation direction is acquired; The number of principal component spaces is selected according to a cumulative variance contribution rate, and the standardized experimental parameters are projected into the selected principal component spaces to obtain principal component parameters.

6. The model parameter based microorganism data visualization chart generating method of claim 4, wherein, The weighted LSTM model includes: A data preprocessing module for cleaning and standardizing input data; An LSTM module including three LSTM sub-modules for processing culture environment parameters, microbial characteristic parameters and culture process parameters respectively; A weight distribution module for distributing weights to outputs of each LSTM sub-module; A weighted fusion module for weighted summation of outputs of different LSTM sub-modules to obtain a weighted hidden state; An output module for predicting the weighted hidden state to output predicted microbial quantity; The calculation formula of the weighted hidden state is: ； wherein, represents a weighted hidden state, represents an environmental hidden state, represents a characteristic hidden state, represents a process hidden state, represents an environmental weight coefficient, represents a characteristic weight coefficient, represents a process weight coefficient.

Images