Multi-probe data fusion construction method and system for fermentation tank concentration spatial distribution diagram
By constructing a multi-probe data fusion method, the problem of uneven concentration distribution in the fermenter was solved, enabling high-precision real-time monitoring and process optimization, automatic identification of mixing dead zones and concentration gradient anomalies, and support for the generation of concentration spatial distribution maps updated in minutes.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-16
- Publication Date
- 2026-03-27
AI Technical Summary
Traditional fermenter concentration monitoring methods cannot accurately reflect the spatial distribution within the tank, nor can they identify areas of uneven mixing and dead zones, thus affecting reaction efficiency. Furthermore, existing multi-probe data is not fully utilized, making it difficult to achieve real-time online monitoring and process optimization.
By constructing a multi-probe data fusion method, combining process knowledge and physical mechanisms, and employing a dynamic weighting mechanism, it automatically identifies mixing dead zones and abnormal concentration gradient regions, generating a high-precision three-dimensional concentration spatial distribution map that supports updates at the minute or even second level.
It represents a leap from point monitoring to field perception, providing real-time and accurate basis for process optimization and fault early warning, and improving the system's fault tolerance and model generalization ability.
Smart Images
Figure CN121744192A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method and system for constructing multi-probe data fusion of spatial distribution maps of concentration in fermenters. Background Technology
[0002] In the biofermentation industry, the uniformity of concentration distribution of chemical substances (such as substrates, products, metabolic intermediates, dissolved oxygen, pH, etc.) directly affects microbial growth and metabolism, product yield, and production efficiency. Traditional fermentation process monitoring typically uses single or a few fixed-position probes for measurement, a method with significant limitations:
[0003] 1. Insufficient spatial representativeness: Fermentation tanks, especially large ones, have complex fluid dynamics, and the flow field created by stirring leads to uneven distribution of chemical substances. Measurements at single or a few points cannot reflect the true spatial distribution within the tank.
[0004] 2. Inability to identify "dead zones" or unevenly mixed areas: Poor local mixing can create "dead zones" with excessively large concentration gradients, affecting reaction efficiency and even leading to the accumulation of byproducts. Traditional monitoring methods struggle to detect and locate these areas.
[0005] 3. Insufficient basis for process optimization: Precise process control (such as feeding, pH adjustment, and regulation of stirring and aeration) requires information on global concentration distribution rather than local point data.
[0006] Existing technologies include methods for predicting concentration fields using computational fluid dynamics (CFD) simulations. However, these methods heavily rely on model accuracy, have high computational complexity, and are difficult to apply in real-time online. Some studies have attempted to increase the number of probes, but these methods merely average or display data from multiple points independently, failing to fully utilize the spatiotemporal correlations of multi-probe data and thus unable to reconstruct the entire spatial field with high accuracy. Summary of the Invention
[0007] To overcome the shortcomings of existing technologies, this invention provides a multi-probe data fusion construction method and system for the spatial distribution map of concentration in fermenters. By integrating process knowledge with limited multi-point probe data, it reconstructs the three-dimensional spatial distribution of chemical substance concentrations within the fermenter with high precision, achieving a leap from point monitoring to field perception. Compared to offline CFD simulation, this method has high computational efficiency and can meet the requirements of online real-time monitoring (updating at the minute or even second level). It automatically identifies mixing dead zones and abnormal concentration gradient areas, and provides quantitative indicators, offering intuitive and accurate basis for process optimization and fault early warning. The dynamic weighting mechanism of data layer fusion improves the system's fault tolerance to individual probe failures; feature layer fusion incorporates physical mechanisms, enhancing the model's generalization ability; and the system's self-optimization module continuously improves model accuracy.
[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for constructing a multi-probe data fusion model of concentration spatial distribution in a fermenter, comprising the following steps:
[0009] Step S1, Multi-probe data acquisition: Deploy chemical substance concentration detection probes at multiple preset spatial monitoring points in the fermenter. The number of detection probes is no less than 3 and they are distributed non-collinearly. Each probe synchronously collects real-time chemical substance concentration data at the corresponding point according to a preset sampling frequency. At the same time, it also collects the installation coordinate information of each probe, the geometric parameters of the fermenter, and the auxiliary process parameters such as stirring rate, temperature, and pressure during the fermentation process.
[0010] Step S2, Raw Data Preprocessing: Outlier identification and removal, and missing value completion processing are performed on the collected concentration data from each probe; and the geometric parameters of the tank and auxiliary process parameters are standardized to unify the data units.
[0011] Step S3, Spatiotemporal Registration of Multi-Source Data: Establish a three-dimensional spatial coordinate system with the center of the fermenter as the origin, map the installation coordinate information of each probe to this coordinate system, and align the time dimension of concentration data and auxiliary process parameters of different probes based on timestamps to form a spatiotemporally unified multi-source dataset.
[0012] Step S4, Multi-probe data fusion calculation: Construct a fusion algorithm model, which includes a data layer fusion module, a feature layer fusion module, and a decision layer fusion module connected in sequence;
[0013] The data layer fusion module is used to perform weighted average initial fusion of the concentration data of each probe after spatiotemporal registration to obtain initial fused concentration data. Its weighting coefficient is dynamically adjusted according to the historical detection accuracy and real-time working status of the probe.
[0014] The feature layer fusion module is used to extract the quantitative distribution features of concentration gradient and spatial autocorrelation from the initial fused concentration data, and calculate the correlation features with concentration changes in combination with the auxiliary process parameters. The quantitative distribution features and correlation features are fused through an improved Bayesian network model to output the fused concentration feature matrix. The decision layer fusion module is used to input the concentration feature matrix into a Kriging space interpolation model that combines the flow field characteristics in the fermenter to calculate the estimated chemical substance concentration data in the entire space of the fermenter.
[0015] Step S5: Concentration Spatial Distribution Map Generation and Anomaly Identification: The estimated chemical concentration data of the entire space is visualized to generate spatial distribution maps of chemical concentration in fermenters in different dimensions; the spatial distribution maps are analyzed by using preset gradient thresholds and distribution uniformity evaluation indicators to identify areas of uneven mixing of chemical substances in the tank or abnormal areas with significant concentration gradients, and the corresponding abnormal locations and gradient quantification data are output.
[0016] As a preferred technical solution of the present invention, in step S1, the detection probes are deployed in a layered manner. The fermenter is divided into at least three monitoring layers in the height direction: upper layer, middle layer, and lower layer. The probes in each monitoring layer are evenly distributed in a ring, and the probes in adjacent monitoring layers are staggered.
[0017] As a preferred technical solution of the present invention, in step S4, when the data layer fusion module dynamically adjusts the weighting coefficient, the judgment indicators of the real-time working status of the probe include the probe's response time, data fluctuation amplitude, and calibration deviation value. When any of the judgment indicators exceeds its corresponding preset threshold, the weighting coefficient of the corresponding probe data is automatically reduced.
[0018] As a preferred technical solution of the present invention, in step S4, the construction process of the improved Bayesian network model is as follows: the concentration data and auxiliary process parameters of each probe monitoring point are used as observation nodes, the concentration of unmonitored points are used as hidden variable nodes, and constraints describing the physical mechanism of mass transfer in the fermenter are introduced as priors for the network structure to optimize the Bayesian network.
[0019] As a preferred technical solution of the present invention, in step S4, the construction process of the Kriging spatial interpolation model improved by combining the flow field characteristics in the fermenter is as follows: the velocity field data of the fermenter stirring flow field associated with the stirring rate is introduced as a covariate, the influence factor of the flow field on the diffusion and transport of chemical substances is incorporated into the semivariance function of Kriging interpolation, and the spatial correlation calculation results are corrected.
[0020] As a preferred technical solution of the present invention, in step S5, the concentration spatial distribution map includes at least two of the following: a three-dimensional distribution map, a two-dimensional distribution map of cross-sections at different heights, a radial concentration gradient curve, and an axial concentration gradient curve, and supports real-time dynamic updating of the distribution map.
[0021] As a preferred technical solution of the present invention, in step S5, the distribution uniformity evaluation index is a combination of concentration variation coefficient and spatial distribution entropy value. When the concentration variation coefficient exceeds a preset threshold and the spatial distribution entropy value is lower than a preset threshold, it is determined that there is a problem of uneven mixing.
[0022] A system for constructing a multi-probe data fusion method for spatial distribution maps of fermenter concentrations includes:
[0023] The multi-probe data acquisition unit consists of multiple chemical substance concentration detection probes, a coordinate positioning module, and process parameter sensors. It is used to collect concentration data, probe coordinate data, and auxiliary process parameters at various points in the fermenter.
[0024] The data preprocessing unit is communicatively connected to the multi-probe data acquisition unit and is used to perform outlier removal, missing value completion, and data standardization processing of the raw data.
[0025] The spatiotemporal registration unit is used to establish a three-dimensional spatial coordinate system for the fermenter, enabling spatiotemporal dimension alignment of multi-source data.
[0026] The data fusion computing unit is equipped with the fusion algorithm model as described in claim 1 and the improved Kriging space interpolation model combined with the flow field characteristics in the fermenter, and is used to complete the multi-level fusion of multi-probe data and the prediction of global concentration data.
[0027] The visualization and anomaly detection unit is used to generate spatial distribution maps of chemical substance concentrations and to perform the identification and quantification of mixing inhomogeneities and concentration gradient anomalies.
[0028] The storage and output unit is used to store all collected data, intermediate calculation data, distribution map data and final results, and output anomaly identification result report.
[0029] As a preferred embodiment of the present invention, the data fusion calculation unit further includes a model self-optimization module. The model self-optimization module can periodically correct the weight coefficients, the improved Bayesian network model, and the parameters of the improved Kriging space interpolation model in the fusion algorithm model based on historical monitoring data and batch data of the fermentation process, so as to improve the accuracy of concentration prediction.
[0030] The present invention also provides a technical solution, which further includes a control and early warning unit. The control and early warning unit is connected to the visualization and anomaly identification unit and is used to generate stirring rate adjustment suggestions, feeding strategy suggestions or early warning information based on the identified uneven mixing area or significant concentration gradient anomaly, and send them to the fermentation process control system or user interface.
[0031] Compared with existing technologies, the beneficial effects of this invention are as follows: by integrating limited multi-point probe data with process knowledge, the three-dimensional spatial distribution of chemical substance concentrations in the fermenter can be reconstructed with high precision, achieving a leap from point monitoring to field perception. Compared with offline CFD simulation, this method has high computational efficiency and can meet the requirements of online real-time monitoring (minute-level or even second-level updates). It automatically identifies mixing dead zones and abnormal concentration gradient regions, and provides quantitative indicators, providing intuitive and accurate basis for process optimization and fault early warning. The dynamic weighting mechanism of data layer fusion improves the system's fault tolerance to single probe failures; the feature layer fusion combines physical mechanisms to enhance the model's generalization ability; and the system's self-optimization module can continuously improve model accuracy. Attached Figure Description
[0032] Figure 1 This is a flowchart illustrating the overall technical solution of the present invention;
[0033] Figure 2 This is a three-dimensional schematic diagram of the fermenter probe deployment according to the present invention;
[0034] Figure 3 This is a block diagram of the original data preprocessing logic of the present invention;
[0035] Figure 4 This is a schematic diagram of the multi-source data spatiotemporal registration coordinate system and alignment according to the present invention;
[0036] Figure 5 This is a diagram of the multi-level data fusion model architecture of the present invention;
[0037] Figure 6 This is an example of a concentration spatial distribution map and an anomaly identification diagram of the present invention;
[0038] Figure 7 This is a diagram of the system hardware architecture of the present invention;
[0039] Figure 8 This is a schematic diagram of the model self-optimization module of the present invention. Detailed Implementation
[0040] To make the technical means, creative features, and achieved objectives and effects of this invention easier to understand, the invention is further described below with reference to specific embodiments. However, the following embodiments are merely preferred embodiments of this invention and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments described herein without creative effort are all within the protection scope of this invention. Unless otherwise specified, the experimental methods in the following embodiments are conventional methods, and the materials and reagents used in the following embodiments are commercially available unless otherwise specified.
[0041] Step S1: Multi-probe data acquisition. Chemical substance concentration detection probes are deployed at multiple pre-set spatial monitoring points within the fermenter. The number of probes is no less than three and they are non-collinearly distributed to ensure the collected data includes three-dimensional spatial information. Each probe synchronously collects real-time chemical substance concentration data at its corresponding point according to a pre-set sampling frequency (e.g., once per second or once per minute). Simultaneously, the precise installation coordinates (x, y, z) of each probe are acquired through an integrated or external positioning device. In addition, the fermenter's geometric parameters (such as diameter, height, baffle size, etc.) and key auxiliary process parameters during fermentation, especially stirring rate, temperature, and pressure, also need to be collected. The detection probes are deployed in a layered manner. Specifically, the fermenter is divided into at least three monitoring layers along its height: upper, middle, and lower, to capture axial concentration gradients. Probes within each monitoring layer are evenly distributed in a ring to capture radial concentration gradients. To obtain more comprehensive spatial information, probes in adjacent monitoring layers are preferably staggered to avoid monitoring blind spots.
[0042] Step S2, Raw Data Preprocessing: The collected raw data is cleaned and organized to provide a high-quality data foundation for subsequent fusion calculations. Outlier Removal: For the concentration data collected by each probe, outliers caused by probe momentary failures, signal interference, or bubble interference are identified and removed. A composite identification method based on statistics and trend analysis is adopted, which integrates the static statistical characteristics and dynamic change characteristics of the data. First, the 3σ criterion (Laida criterion) is used to identify outliers (Type I outliers) that exceed the normal statistical distribution. Second, the gradient mutation identification method is used to calculate the gradient of the concentration data over time. When the gradient value exceeds a preset multiple of the historical normal fluctuation range, it is determined to be an abnormal mutation point (Type II outlier), even if the value of the point does not exceed the 3σ range. Finally, these two types of outliers are removed. Missing Value Completion: For data missing due to outlier removal or brief communication interruptions, a spatiotemporally weighted interpolation method is used for completion. Specifically, for data missing at time t, the imputation value is determined by the weighted average of data from the same probe at adjacent times (t-1, t+1) and the weighted average of data from probes at neighboring spatial locations at the same time t. The weights can be dynamically set based on the spatial distance between probes and the correlation of historical data. Data standardization: The tank's geometric parameters and auxiliary process parameters of different dimensions (such as stirring rate in rpm, temperature in °C, and pressure in MPa) are standardized (e.g., Z-score standardization or Min-Max normalization) to the same or dimensionless dimension, facilitating subsequent model calculations.
[0043] Step S3, Spatiotemporal Registration of Multi-Source Data: To achieve effective fusion of multi-source data, a unified spatiotemporal reference system needs to be established. Spatial registration: A three-dimensional Cartesian coordinate system is established with the geometric center (or bottom center) of the fermenter as the origin O (e.g., the Z-axis coincides with the central axis of the tank, with upward being positive; the X and Y axes are in the horizontal plane). The installation coordinates of all probes acquired in Step S1 are mapped to this unified coordinate system. Temporal registration: Based on the timestamps of a high-precision synchronous clock, the concentration data sequences and auxiliary process parameter data sequences such as stirring rate acquired by all probes are time-aligned. Data with different sampling frequencies are unified to the same time series through interpolation or resampling, forming a multi-source spatiotemporal dataset with strict spatiotemporal alignment.
[0044] Step S4, Multi-probe data fusion calculation: Data layer fusion module: Performs first-level fusion on the original concentration data C_i(t) (i=1,2,...,N) of each probe after spatiotemporal registration. The initial fused concentration data C_fused(t) at time t is calculated using a dynamic weighted average method. C_fused(t) = Σ [w_i(t) * C_i(t)] / Σ w_i(t). The weight w_i(t) is dynamically adjusted based on two factors: a) Historical detection accuracy: Probes with small calibration errors and long-term stability are given higher weights. b) Real-time working status: This is determined by monitoring indicators such as probe response time, data fluctuation amplitude, and recent calibration deviation. When any indicator exceeds a preset threshold, the system automatically reduces the weight of the probe data, or even temporarily excludes it from the fusion calculation, thereby enhancing the robustness of the system. Feature layer fusion module: Extracts high-level features from the initial fused data C_fused(t) and auxiliary process parameters, and performs correlation analysis. Feature Extraction: Extracting concentration gradient features of C_fused(t) (e.g., concentration change rate in all spatial directions) and spatial distribution features (e.g., spatial autocorrelation index calculated based on known probe point data); extracting correlation features between auxiliary process parameters (especially stirring rate) and concentration changes (e.g., cross-correlation function values of stirring rate change and local concentration change rate). Feature Fusion: Inputting the above features into an improved Bayesian network model for correlation fusion. The model construction process is as follows: using the concentration data of each actual probe point and auxiliary process parameters as observable nodes, and the concentration of unmonitored points to be determined as latent variable nodes. The key improvement is the introduction of constraints describing the physical mechanism of mass transfer in the fermenter (e.g., the basic form of the convection-diffusion equation) as prior knowledge for network structure connections, thereby combining data-driven and mechanism models to optimize the network topology and conditional probability table. The model outputs a concentration feature matrix F that integrates spatial correlation and process dynamics. Decision Layer Fusion Module: Based on the feature matrix F, generating the final global concentration prediction data. This module employs an improved Kriging spatial interpolation model that incorporates the flow field characteristics within the fermenter. Standard Kriging interpolation relies on a spatial variogram (semivariance function) and only considers geographic distance. The improvement of this invention lies in introducing flow field velocity data (which can be derived from a pre-stored CFD simulation database or a simplified flow field model) associated with the current stirring rate as covariates. When calculating the correlation between two points in space, not only Euclidean distance is considered, but also the influence of flow direction and velocity on mass transport. For example, in the axial flow direction formed by strong stirring, the concentration influence of downstream points on upstream points increases. By incorporating flow field influence factors (such as velocity vector dot product, streamline distance, etc.) into the calculation of the semivariogram function, the estimation results of spatial correlation are corrected, making the interpolation results more consistent with the physical reality of fluid mixing.By inputting the feature matrix F into this improved model, the estimated chemical concentration of all discrete grid points (the entire domain) within the fermenter can be calculated.
[0045] Step S5: Concentration Spatial Distribution Map Generation and Anomaly Identification: Visualization Generation: The concentration estimates of each grid point in the entire spatial domain output by the decision-making layer fusion module are visualized and rendered using computer graphics technology. This can generate, but is not limited to: 3D isostatic distribution maps, 2D cloud maps of different horizontal / vertical sections, radial (from center to tank wall) concentration gradient curves, and axial (from tank bottom to liquid surface) concentration gradient curves. These graphics support real-time dynamic updates, forming dynamic concentration field animations. Anomaly Identification and Quantification: The system automatically analyzes the generated distribution maps to identify areas of uneven mixing or abnormalities. Gradient Threshold Judgment: The concentration gradient magnitude at each location in space is calculated and compared with a preset safety gradient threshold to mark areas exceeding the limit. Distribution Uniformity Evaluation: The concentration variation coefficient and spatial distribution entropy of the entire concentration data are calculated. The concentration variation coefficient reflects the dispersion of the data; a larger value indicates greater unevenness. Spatial distribution entropy reflects the spatial order of the concentration; the entropy value is highest when the distribution is completely uniform, and decreases when there is significant concentration aggregation. When the concentration variation coefficient exceeds a preset threshold and the spatial distribution entropy value is lower than a preset threshold, the system determines that there is a significant mixing inhomogeneity problem within the tank. Finally, the system outputs the spatial location (coordinate range) of the abnormal area and gradient quantization data (maximum gradient value, average gradient value, etc.), forming an anomaly report.
[0046] On the other hand, the system for constructing a multi-probe data fusion method for the spatial distribution map of fermenter concentration includes: a multi-probe data acquisition unit, composed of multiple chemical substance concentration detection probes (such as online spectral probes, electrochemical sensors, etc.), coordinate positioning modules (such as laser rangefinders, encoders), and process parameter sensors (stirring motor power / speed sensors, temperature sensors, pressure sensors), used to synchronously acquire raw data; a data preprocessing unit, communicating with the acquisition unit, with a built-in processor and storage, running preprocessing algorithms to perform outlier removal, missing value completion, and standardization; a spatiotemporal registration unit, used to establish a unified three-dimensional coordinate system and align the timestamps of all data streams; and a data fusion calculation unit, the core calculation unit, which incorporates the aforementioned fusion algorithm model (including data layer, feature layer, and decision layer fusion modules) and an improved Kriging interpolation model. Furthermore, a model self-optimization module is preferably included. This module can automatically correct the weight coefficients, Bayesian network parameters, and Kriging model semivariance function parameters in the fusion algorithm periodically (e.g., after each batch is completed) based on long-term accumulated historical monitoring data and process data from different batches of fermentation, allowing the model to continuously optimize with use and improve prediction accuracy. Visualization and Anomaly Detection Unit: Used for graphic rendering, distribution map generation, and automatic analysis and identification of anomalies. Storage and Output Unit: Used to store all raw data, intermediate results, historical distribution maps, and reports. Control and Early Warning Unit (Preferred): Connected to the anomaly detection unit. When an anomaly is detected, it can automatically generate operational suggestions, such as increasing the stirring rate by 20%, strengthening feeding in the area near coordinates (x,y,z), or issuing audible and visual alarms, and pushing the suggested instructions to the fermentation process control system.
[0047] Example 1
[0048] A 10m³ fermenter was selected as the monitoring object. The fermenter was divided into three monitoring layers along its height: an upper layer (7-10m), a middle layer (3-7m), and a lower layer (0-3m). Three fiber-optic chemical substance concentration detection probes were evenly distributed in a ring around each layer. Probes in adjacent layers were staggered at 60°, for a total of nine probes. Each probe's sampling frequency was set to 5Hz, and glucose concentration data was collected simultaneously. The three-dimensional coordinates of each probe, the fermenter diameter (4m), and the height (10m) were also collected, along with process parameters such as a stirring rate of 200r / min, a temperature of 37℃, and a pressure of 0.1MPa. The 3σ criterion was used to initially screen 100 consecutive sets of data from a single probe, removing discrete data that were less than or equal to 3 times the standard deviation of the mean. The concentration gradient values between adjacent time points were calculated, and data exceeding 5 times the historical average gradient were identified as gradient abrupt changes and removed. For probe data with a missing duration of 8 seconds, weighted interpolation was performed using data from three neighboring probes (weighted at 0.7) and historical data from the same probe (weighted at 0.3) to complete the missing data. Simultaneously, probe response time was monitored, and the weighting coefficient for probe data with response times exceeding the standard value by 20% was reduced by 50%. Geometric and process parameters were standardized in terms of dimensions. A three-dimensional coordinate system was established with the tank center as the origin, and the actual installation coordinates of each probe were mapped to this coordinate system, with the mapping error controlled within 0.01m. Time alignment of all data was achieved based on 1ms timestamps, with an alignment deviation ≤5ms, forming a multi-source dataset in the format {timestamp, probe number, three-dimensional coordinates, glucose concentration, stirring rate, temperature, pressure}. Data layer fusion: Based on the probe's historical detection accuracy (values 0.8-0.95) and real-time working status score (values 0.75-0.98), dynamic weighting coefficients are calculated with weight ratios of 0.6 and 0.4 to complete the weighted average initial fusion of multi-probe data; Feature layer fusion: Features such as radial / axial concentration gradients, mean and variance of concentrations in each layer, and the correlation between stirring rate and concentration are extracted from the initial fused data, input into an improved Bayesian network model to complete feature fusion, and output a concentration feature matrix; Decision layer fusion: The velocity field data of the fermenter stirring flow field is obtained through CFD simulation, and it is incorporated as a covariate into the Kriging interpolation semivariogram function. After correction, the glucose concentration prediction of the entire grid (10cm×10cm×10cm) is completed, with a prediction error ≤0.05mol / L.The system generates a three-dimensional distribution map of glucose concentration in the fermenter and two-dimensional cross-sectional distribution maps at heights of 0.3H, 0.5H, and 0.7H, supporting dynamic updates at 10s / frame. With a concentration variation coefficient threshold of 0.2 and a spatial distribution entropy threshold of 0.8, it identifies uneven mixing in the bottom edge region of the fermenter. This region has a concentration variation coefficient of 0.25, an entropy value of 0.72, and a maximum concentration gradient of 0.12 mol / L·cm. The output three-dimensional coordinate range of the abnormal region is (1.8-2.0m, -0.2-0.2m, 0-0.5m).
[0049] Example 2
[0050] Includes: a multi-probe data acquisition unit: composed of 9 IP68-rated fiber optic concentration probes, a coordinate positioning module with a positioning accuracy of ±0.005m, a stirring rate sensor with a range of 0-500r / min, a temperature sensor with a range of 0-100℃, and a pressure sensor with a range of 0-0.5MPa, transmitting data via Modbus TCP protocol at a transmission rate of 100Mbps; a data preprocessing unit: equipped with algorithms for outlier removal, missing value completion, and data standardization, triggering a fault warning when the outlier rate of a single probe exceeds 10%; a spatiotemporal registration unit: supporting real-time probe coordinate calibration, automatically updating coordinates when displacement exceeds 0.01m; a data fusion and calculation unit: equipped with a GPU acceleration module, with a calculation latency ≤1s, and a built-in model self-optimization module, completing parameter correction every 10 batches, improving prediction accuracy by ≥5%; a visualization and anomaly identification unit: supporting distribution map rotation and zoom operations, and automatically generating diagnostic reports including suggestions for the causes of anomalies; and a storage and output unit: using an industrial database to store 1000... The batch data is equipped with an audible and visual alarm, which automatically triggers an alarm when a severe anomaly is detected.
[0051] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.
[0052] 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 foregoing embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit 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 present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.
Claims
1. A method for constructing a spatial distribution map of concentration in a fermenter using multi-probe data fusion, characterized in that, Includes the following steps: Step S1, Multi-probe data acquisition: Deploy chemical substance concentration detection probes at multiple preset spatial monitoring points in the fermenter. The number of detection probes is no less than 3 and they are distributed non-collinearly. Each probe synchronously collects real-time chemical substance concentration data at the corresponding point according to a preset sampling frequency. At the same time, it also collects the installation coordinate information of each probe, the geometric parameters of the fermenter, and the auxiliary process parameters such as stirring rate, temperature, and pressure during the fermentation process. Step S2, Raw Data Preprocessing: Outlier identification and removal, and missing value completion processing are performed on the collected concentration data from each probe; and the geometric parameters of the tank and auxiliary process parameters are standardized to unify the data units. Step S3, Spatiotemporal Registration of Multi-Source Data: Establish a three-dimensional spatial coordinate system with the center of the fermenter as the origin, map the installation coordinate information of each probe to this coordinate system, and align the time dimension of concentration data and auxiliary process parameters of different probes based on timestamps to form a spatiotemporally unified multi-source dataset. Step S4, Multi-probe data fusion calculation: Construct a fusion algorithm model, which includes a data layer fusion module, a feature layer fusion module, and a decision layer fusion module connected in sequence; The data layer fusion module is used to perform weighted average initial fusion of the concentration data of each probe after spatiotemporal registration to obtain initial fused concentration data. Its weighting coefficient is dynamically adjusted according to the historical detection accuracy and real-time working status of the probe. The feature layer fusion module is used to extract the quantitative distribution features of concentration gradient and spatial autocorrelation from the initial fused concentration data, and calculate the correlation features with concentration changes in combination with the auxiliary process parameters. The quantitative distribution features and correlation features are fused through an improved Bayesian network model to output the fused concentration feature matrix. The decision layer fusion module is used to input the concentration feature matrix into a Kriging space interpolation model that combines the flow field characteristics in the fermenter to calculate the estimated chemical substance concentration data in the entire space of the fermenter. Step S5: Concentration Spatial Distribution Map Generation and Anomaly Identification: The estimated chemical concentration data of the entire space is visualized to generate spatial distribution maps of chemical concentration in fermenters in different dimensions; the spatial distribution maps are analyzed by using preset gradient thresholds and distribution uniformity evaluation indicators to identify areas of uneven mixing of chemical substances in the tank or abnormal areas with significant concentration gradients, and the corresponding abnormal locations and gradient quantification data are output.
2. The method for constructing a multi-probe data fusion map of concentration spatial distribution in a fermenter according to claim 1, characterized in that: In step S1, the detection probes are deployed in a layered manner. The fermenter is divided into at least three monitoring layers in the height direction: upper layer, middle layer, and lower layer. The probes in each monitoring layer are evenly distributed in a ring, and the probes in adjacent monitoring layers are staggered.
3. The method for constructing a multi-probe data fusion model of concentration spatial distribution in a fermenter according to claim 1, characterized in that: In step S4, when the data layer fusion module dynamically adjusts the weighting coefficients, the indicators for determining the real-time working status of the probe include the probe's response time, data fluctuation amplitude, and calibration deviation value. When any of the indicators exceeds its corresponding preset threshold, the weighting coefficient of the corresponding probe data is automatically reduced.
4. The method for constructing a multi-probe data fusion model of concentration spatial distribution in a fermenter according to claim 3, characterized in that: In step S4, the construction process of the improved Bayesian network model is as follows: the concentration data and auxiliary process parameters of each probe monitoring point are used as observation nodes, the concentration of unmonitored points are used as hidden variable nodes, and constraints describing the physical mechanism of mass transfer in the fermenter are introduced as priors for the network structure to optimize the Bayesian network.
5. The method for constructing a multi-probe data fusion model of concentration spatial distribution in a fermenter according to claim 1, characterized in that: In step S4, the construction process of the improved Kriging spatial interpolation model based on the flow field characteristics in the fermenter is as follows: the velocity field data of the fermenter stirring flow field associated with the stirring rate is introduced as a covariate, the influence factor of the flow field on the diffusion and transport of chemical substances is incorporated into the semivariance function of the Kriging interpolation, and the spatial correlation calculation results are corrected.
6. The method for constructing a multi-probe data fusion model of concentration spatial distribution in a fermenter according to claim 1, characterized in that: In step S5, the concentration spatial distribution map includes at least two of the following: a three-dimensional distribution map, a two-dimensional distribution map of cross-sections at different heights, a radial concentration gradient curve, and an axial concentration gradient curve, and supports real-time dynamic updating of the distribution map.
7. The method for constructing a multi-probe data fusion map of concentration spatial distribution in a fermenter according to claim 1, characterized in that: In step S5, the distribution uniformity evaluation index is a combination of concentration variation coefficient and spatial distribution entropy value. When the concentration variation coefficient exceeds a preset threshold and the spatial distribution entropy value is lower than a preset threshold, it is determined that there is a problem of uneven mixing.
8. A system for constructing a multi-probe data fusion method for a fermenter concentration spatial distribution map according to any one of claims 1-7, characterized in that: include: The multi-probe data acquisition unit consists of multiple chemical substance concentration detection probes, a coordinate positioning module, and process parameter sensors. It is used to collect concentration data, probe coordinate data, and auxiliary process parameters at various points in the fermenter. The data preprocessing unit is communicatively connected to the multi-probe data acquisition unit and is used to perform outlier removal, missing value completion, and data standardization processing of the raw data. The spatiotemporal registration unit is used to establish a three-dimensional spatial coordinate system for the fermenter, enabling spatiotemporal dimension alignment of multi-source data. The data fusion computing unit is equipped with the fusion algorithm model as described in claim 1 and the improved Kriging space interpolation model combined with the flow field characteristics in the fermenter, and is used to complete the multi-level fusion of multi-probe data and the prediction of global concentration data. The visualization and anomaly detection unit is used to generate spatial distribution maps of chemical substance concentrations and to perform the identification and quantification of mixing inhomogeneities and concentration gradient anomalies. The storage and output unit is used to store all collected data, intermediate calculation data, distribution map data and final results, and output anomaly identification result report.
9. The system for constructing a multi-probe data fusion method for a fermenter concentration spatial distribution map according to claim 8, characterized in that: The data fusion computing unit also includes a model self-optimization module, which can periodically correct the weight coefficients, the improved Bayesian network model, and the parameters of the improved Kriging space interpolation model in the fusion algorithm model based on historical monitoring data and batch data of the fermentation process, so as to improve the accuracy of concentration prediction.
10. The system for constructing a multi-probe data fusion method for a fermenter concentration spatial distribution map according to claim 8 or 9, characterized in that: It also includes a control and early warning unit, which is connected to the visualization and anomaly identification unit. The control and early warning unit is used to generate stirring rate adjustment instructions, feeding strategy suggestions or early warning information based on the identified uneven mixing areas or significant concentration gradient anomalies, and send them to the fermentation process control system or user interface.