Biological fermentation monitoring method and system based on multi-source data fusion

Abstract

The invention relates to the technical field of microbial fermentation, in particular to a biological fermentation monitoring method and system based on multi-source data fusion. The method comprises the following steps: acquiring multi-source monitoring data in a fermentation tank and constructing a multi-dimensional state vector; on the basis of data of the multi-dimensional state vector in a preset time window, constructing a metabolic entropy of the current fermentation process; a simulated annealing algorithm is adopted to optimize control parameters, the temperature of the simulated annealing algorithm is positively correlated with the metabolic entropy, and minimization of a second energy function is taken as a target; a catastrophe reignition mechanism is set in the optimization process, fermentation abnormity is diagnosed based on triggering of the catastrophe reignition mechanism, and abnormal state characteristics are matched with a preset abnormity database through iterative search of an optimal weight distribution scheme so as to determine the abnormity type with the highest matching degree. By introducing the metabolic entropy and the catastrophe reignition mechanism, the sensitivity, the accuracy and the robustness of monitoring the biological fermentation process are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of microbial fermentation. More particularly, the present application relates to a biological fermentation monitoring method and system based on multi-source data fusion. BACKGROUND

[0002] Biological fermentation is the core production link of modern industries such as pharmaceutical, food, and chemical industries, and the stability, reliability and accuracy of the process inside directly determine the yield and quality of the final product. However, the biological fermentation process is a typical dynamic system, which has inherent characteristics such as high nonlinearity, time-varying, strong coupling of multiple variables, and complex biochemical mechanism.

[0003] In order to ensure process stability, the industry has adopted various monitoring methods from traditional to modern, but all have obvious limitations. The traditional method relies on single variable threshold alarm of parameters such as temperature and pH. This method is seriously deficient in sensitivity to slow and gradual abnormalities caused by multiple factors due to the neglect of the complex coupling relationship between variables, often leading to delayed discovery of faults. In order to overcome this defect, multivariate statistical methods such as principal component analysis are introduced, but due to their linear assumption, their model accuracy will be greatly discounted when applied to the inherent nonlinear interval of biological processes. Even the latest data-driven methods such as deep learning, due to their high dependence on massive historical data covering various fault types for training, their generalization ability and monitoring robustness are difficult to guarantee when facing sparse data and unknown disturbances in actual production.

[0004] In summary, the existing biological fermentation monitoring technology still faces the severe challenges of insufficient sensitivity, accuracy and robustness in actual application. On the one hand, the traditional method cannot sensitively capture the early dynamic changes of the system state due to its static and one-sided perspective; on the other hand, the existing model needs to improve its diagnostic accuracy and robustness to unknown disturbances when facing the complexity and uncertainty of biological processes. Therefore, developing a new method that can fundamentally improve the sensitivity, diagnostic accuracy and overall robustness of the monitoring system to complex working condition changes has become a key technical problem to be solved in the field. SUMMARY

[0005] The present application aims to provide a biological fermentation monitoring method and system based on multi-source data fusion, which solves the problems of low accuracy and poor robustness of abnormal monitoring and diagnosis in the prior art.

[0006] In a first aspect, the present invention provides a bio-fermentation monitoring method based on multi-source data fusion. The detection method includes: acquiring multi-source monitoring data within a fermenter to construct a multi-dimensional state vector and acquiring control parameters of the fermenter; constructing a metabolic entropy based on the multi-dimensional state vector to quantify the complexity of the current fermentation process; defining a first energy function to evaluate the deviation of the current fermentation state, and defining a second energy function to evaluate the effectiveness of the control parameters on the fermentation state; optimizing the control parameters using a simulated annealing algorithm, with the goal of minimizing the second energy function, and outputting the optimal control parameters, wherein temperature is positively correlated with the metabolic entropy; and iteratively optimizing the simulated annealing algorithm... During the process, a catastrophic re-ignition mechanism is established, which is triggered when the first energy function value and the metabolic entropy value simultaneously exceed a preset anomaly threshold. After the catastrophic re-ignition mechanism is triggered, the multidimensional state vector is used as the anomaly state feature, and the temperature in the simulated annealing algorithm is reset to a preset high temperature value. The simulated annealing algorithm for anomaly matching is started according to the preset high temperature value. Multiple anomaly vectors in the preset anomaly database are weighted and combined, and the distance between the weighted combined anomaly vector and the anomaly state feature is minimized as the optimization objective for iterative search. The anomaly type with the highest matching degree is determined according to the optimal weight allocation scheme, and the monitoring result of the anomaly type is output.

[0007] This invention innovatively combines the metabolic entropy of dynamic quantification uncertainty with the catastrophe re-ignition simulated annealing algorithm with adaptive bounce capability to construct an active, closed-loop monitoring framework, fundamentally solving the inherent defects of traditional passive, static threshold monitoring methods in terms of sensitivity, accuracy, and robustness.

[0008] Optionally, the construction of the multidimensional state vector includes: the multi-source monitoring data includes at least three of temperature, pH, dissolved oxygen, and carbon dioxide concentration; filtering and normalizing the acquired data; and combining the preprocessed data to form the multidimensional state vector.

[0009] Optionally, the process of constructing the metabolic entropy includes: arranging multiple multidimensional state vectors within a preset time window to form a state matrix; performing principal component analysis on the state matrix to obtain a set of principal components; calculating the variance of each principal component and the total variance of the matrix; determining the proportion of the variance of each principal component to the total variance, and calculating the metabolic entropy based on the proportion using the Shannon entropy formula.

[0010] The metabolic entropy construction method transforms high-dimensional, abstract process data into a single scalar value with clear physical meaning. Metabolic entropy can accurately quantify the overall complexity and uncertainty of the current process, providing a core basis for subsequent dynamic adaptive monitoring.

[0011] Optionally, the construction of the first energy function includes: obtaining a standard trajectory vector corresponding to the current fermentation stage from a preset standard database; calculating the weighted Euclidean distance between the multidimensional state vector of the current state and the standard trajectory vector as a trajectory deviation term; determining whether each component in the multidimensional state vector of the current state exceeds a preset process constraint range, summing the squares of the ranges exceeding the process constraint range as a constraint term; and performing a weighted summation of the trajectory deviation term and the constraint term to obtain the first energy function.

[0012] By constructing a first energy function through a weighted summation of deviation terms that assess the degree of deviation from fermentation state and constraint terms that penalize violations of process constraints, the two multi-objective requirements of tracking the trajectory and complying with constraints in process monitoring are cleverly unified into a single-objective optimization problem, which greatly simplifies the solution process.

[0013] Optionally, the construction of the second energy function includes: constructing a prediction model based on historical fermentation data; predicting the multidimensional state vector at the next moment using the prediction model based on the current multidimensional state vector and control parameters; calculating the weighted Euclidean distance between the predicted multidimensional state vector and the standard trajectory vector as a state deviation term; calculating the L2 norm of the difference vector between the control parameters to be evaluated and the currently executed control parameters as a control cost term; and performing a weighted summation of the state deviation term and the control cost term to obtain the second energy function.

[0014] Optionally, the positive correlation between temperature and metabolic entropy in the simulated annealing algorithm includes: the positive correlation is achieved through a preset functional relationship, wherein the temperature is set as a monotonically increasing function of the metabolic entropy.

[0015] Optionally, the catastrophic reignition mechanism includes: when the catastrophic reignition mechanism is triggered, setting the current temperature of the simulated annealing algorithm to a preset initial high temperature value, so as to increase the probability of the algorithm accepting state solutions with higher energy function values ​​in subsequent iterations according to the Metropolis criterion.

[0016] When the catastrophic reignition mechanism is triggered, the simulated annealing algorithm's temperature is forcibly reset to a preset initial high temperature value. Based on the Metropolis criterion, this increases the probability of the algorithm accepting a poor solution. This allows the algorithm to forcibly jump out of the failed local optimum region and restart the global search, ensuring that the method still has strong robustness when the fermentation state undergoes a fundamental change.

[0017] Optionally, the construction of the anomaly database includes: collecting historical fermentation data containing various abnormal operating conditions, and extracting a set of multidimensional state vectors corresponding to each abnormal operating condition; processing the set of multidimensional state vectors using a clustering algorithm to divide vectors with similar characteristics into different clusters; and using the centroid vector of each cluster as a standardized anomaly vector of a specific anomaly type to form the anomaly database.

[0018] By employing clustering algorithms to process historical anomaly data and using the centroid vector of each cluster as a standardized anomaly vector for a specific anomaly type to construct an anomaly database, the core patterns of various anomalies were effectively extracted from the chaotic historical data, laying a solid foundation for subsequent rapid and accurate anomaly type identification.

[0019] Optionally, determining the anomaly type with the highest matching degree according to the optimal weight allocation scheme includes: using the triggering of the catastrophic reignition mechanism as a marker of the anomaly event; identifying the anomaly vector with the largest weight value through the optimal weight allocation scheme obtained by iterative search; and outputting the anomaly type corresponding to the anomaly vector with the largest weight value as the anomaly type with the highest matching degree; the anomaly type includes at least one of the following: bacterial contamination anomaly, nutrient deficiency anomaly, dissolved oxygen deficiency anomaly, pH runaway anomaly, temperature runaway anomaly, and metabolic product accumulation anomaly.

[0020] In the second aspect, the bio-fermentation monitoring system based on multi-source data fusion includes: processor; The memory stores computer instructions for a bio-fermentation monitoring method based on multi-source data fusion, which, when executed by the processor, cause the system to perform the aforementioned bio-fermentation monitoring method based on multi-source data fusion.

[0021] The beneficial effects of this invention are as follows: This invention provides a novel monitoring method by constructing a metabolic entropy index to dynamically quantify uncertainty and combining it with a simulated annealing algorithm that includes a catastrophic reignition mechanism. This method uses metabolic entropy to perceive the evolution of the state in real time, exhibiting sensitivity far exceeding traditional methods. Its catastrophic reignition mechanism employs dual triggering conditions of energy function and metabolic entropy, effectively distinguishing between real anomalies and transient disturbances, thus improving diagnostic accuracy. Once an anomaly is confirmed, the algorithm resets the temperature to force a break from local optima and initiate a new global search, ensuring the method's robustness in the face of unknown operating condition changes. In summary, this invention effectively solves the challenges of low accuracy and poor robustness in anomaly monitoring and diagnosis in existing technologies. Attached Figure Description

[0022] Figure 1 This is a flowchart of a bio-fermentation monitoring method based on multi-source data fusion according to an embodiment of the present invention.

[0023] Figure 2 This is a structural block diagram of a bio-fermentation monitoring system based on multi-source data fusion according to an embodiment of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will now be clearly and completely described with reference to the accompanying drawings. Figure 1 The diagram shown is a flowchart of a bio-fermentation monitoring method based on multi-source data fusion according to an embodiment of the present invention.

[0025] S1: Acquire multi-source monitoring data in the fermenter to construct a multi-dimensional state vector, and construct the metabolic entropy of the current fermentation process based on the multi-dimensional state vector.

[0026] First, multi-source monitoring data is acquired within the fermenter. This multi-source monitoring data refers to a sequence of one or more key physical or chemical parameters that can reflect the physiological state of microorganisms and environmental conditions during fermentation in real time. Specifically, it includes at least three of the following: temperature, pH, dissolved oxygen, and carbon dioxide concentration.

[0027] After acquiring the raw monitoring data, preprocessing is required. This preprocessing refers to a series of data purification and transformation processes performed to improve data quality and model applicability, including filtering and normalization. The filtering process uses a median filtering algorithm. Median filtering replaces the current value with the median of the data within a window, effectively suppressing impulse interference while preserving the edge characteristics of the data. The normalization process aims to map data of different dimensions to a unified numerical range. The normalization preprocessing uses a max-min normalization method to map each indicator data to the range of 0 to 1.

[0028] After preprocessing, a multidimensional state vector is constructed. This vector consists of normalized temperature, normalized pH, normalized dissolved oxygen concentration, and normalized carbon dioxide concentration arranged in sequence, forming a four-dimensional column vector. The multidimensional state vector is updated in real-time at a preset sampling period, continuously recording the dynamic evolution trajectory of the fermentation process and providing a data foundation for subsequent metabolic entropy calculations.

[0029] Next, metabolic entropy is constructed based on the multidimensional state vectors. Metabolic entropy is an indicator used to quantify the dynamic complexity and orderliness of the fermentation process. Its construction process includes: first, constructing a state matrix, which is composed of multiple multidimensional state vectors arranged in chronological order within a preset time window. Second, performing principal component analysis on the state matrix to extract a set of principal components that characterize the main dynamic features of the fermentation process. These principal components are determined by several eigenvectors whose cumulative variance contribution rate exceeds a preset threshold.

[0030] Next, the proportion of the variance of each principal component to the total variance is calculated, and the metabolic entropy is calculated based on the Shannon entropy formula according to the variance proportion. The value of the metabolic entropy is used to quantitatively describe the overall complexity and uncertainty of the fermentation process. The higher the value, the more uniform the distribution of the multidimensional states and the more complex the dynamics of the fermentation process; the lower the value, the more stable the fermentation process. When the value of the metabolic entropy suddenly increases and exceeds a preset abnormality threshold, it indicates that an abnormality has occurred in the fermentation process.

[0031] S2: Define the first energy function and the second energy function, and use the simulated annealing algorithm to optimize the control parameters.

[0032] In this embodiment, to achieve active monitoring and optimization of the fermentation process, two energy functions are defined, and a simulated annealing algorithm is used to optimize the control parameters that directly affect the fermentation state. The control parameters include at least one of temperature, feeding rate, stirring speed, and aeration flow rate.

[0033] The first energy function quantifies the deviation between the current fermentation state and the healthy operating state; the smaller the value, the closer the current state is to the ideal operating state. First, a standard trajectory vector corresponding to the current fermentation stage is obtained from a preset standard database. This standard database stores data on the fermentation process in historical normal production batches, dividing the entire fermentation process into several stages along a time axis. For each stage, the corresponding standard trajectory vector is statistically obtained from historical normal batches and stored. Second, the trajectory deviation term is calculated, specifically the weighted Euclidean distance between the current four-dimensional state vector and the standard trajectory vector. The weighted Euclidean distance is calculated as follows: for each of the four dimensions, the difference between the current state vector and the standard state vector in the current dimension is calculated. Then, each difference is multiplied by a preset weight coefficient and squared. The weighted squared values ​​of the four dimensions are summed, and the square root of the sum is taken.

[0034] Next, the constraint terms are calculated. These constraint terms are designed to penalize violations of process safety boundaries. The process constraint ranges specifically include: upper and lower safety limits for temperature to ensure strain activity; the allowable pH range to ensure the activity of key enzymes; and the lower limit for dissolved oxygen concentration to maintain aerobic metabolism in the microorganisms. The constraint terms are calculated as follows: for each dimension, if the current state value exceeds the upper limit of its constraint range or falls below its lower limit, the square of the excess or deficiency value is calculated as the penalty value for the current dimension; if it is within the constraint range, the penalty value for that dimension is zero. The penalty values ​​for all dimensions are summed to obtain the total constraint term. Finally, the trajectory deviation term and the constraint term are weighted and summed to obtain the first energy function.

[0035] The second energy function is used to evaluate the effectiveness of control parameters on the fermentation state, providing an objective function for control parameter optimization. A prediction model is constructed based on historical fermentation data. This embodiment uses a Kalman filter model as the prediction model. The Kalman filter model is an optimal recursive data processing algorithm that can accurately estimate the fermentation state in noisy environments. Using historical fermentation data, the state transition law of the fermentation process and the measurement noise characteristics are learned to establish a state-space model. This model includes state equations and observation equations, which respectively describe the evolution law of the fermentation state over time and the relationship between the measured value and the actual state. Based on the current multidimensional state vector and control parameters, the Kalman filter model is used to predict the multidimensional state vector at the next moment. Specifically, according to the state equations, the current state vector and control parameters are input into the model to calculate the predicted state value at the next moment; the predicted value is corrected using Kalman gain to obtain the optimal estimated state vector at the next moment. The weighted Euclidean distance between the predicted multidimensional state vector and the standard trajectory vector is calculated as the state deviation term. The state deviation term evaluates the degree to which the future state deviates from the healthy trajectory under the current control parameters. The smaller the deviation, the better the current control parameters can maintain the fermentation process on the standard trajectory.

[0036] The L2 norm of the difference vector between the control parameter to be evaluated and the currently executed control parameter is calculated and used as the control cost term. The control cost term quantifies the severity of the control action. The L2 norm is calculated by squared the differences between the control parameter to be evaluated and the current control parameter in each dimension, summing them, and then taking the square root. A larger control cost term indicates a larger control adjustment amplitude, which may impact the fermentation process. The second energy function is obtained by weighted summation of the state deviation term and the control cost term. By adjusting the weights of the state deviation term and the control cost term, the requirements for control accuracy and control stability can be balanced.

[0037] The control parameters are optimized using a simulated annealing algorithm, with the goal of minimizing the second energy function to output the optimal control parameters. Simulated annealing is a heuristic global optimization method that, by introducing a probabilistic acceptance mechanism, can escape local optima.

[0038] The simulated annealing algorithm shows a positive correlation between temperature and metabolic entropy. Specifically, the correlation is as follows: ; in, The temperature at the k-th iteration. The initial temperature is set to 100 in this embodiment; The cooling coefficient is set to 0.9; The metabolic entropy influence factor is set to 0.3; H is the current metabolic entropy value. By incorporating metabolic entropy into temperature calculation, the algorithm can adaptively adjust its search strategy based on the dynamic complexity of the fermentation process.

[0039] The iterative process of the simulated annealing algorithm includes: First, using the current control parameters as the initial solution. In each iteration, a new candidate control parameter solution is randomly generated within a preset neighborhood of the current solution. Then, the prediction model is used to calculate the second energy function value corresponding to the candidate solution. If the energy function value of the new solution is lower, the new solution is accepted as the current solution; if it is higher, the new solution is accepted with a probability related to the energy difference and the current temperature, according to the Metropolis criterion. The algorithm continues to iterate until a preset termination condition is reached. The termination condition includes: the number of iterations reaches a preset maximum value, or the change in the global optimal solution within a consecutive preset number of iterations is less than a preset convergence threshold. When any termination condition is met, the algorithm outputs the currently searched optimal control parameters.

[0040] S3: Establish a catastrophic reignition mechanism during the optimization process of the simulated annealing algorithm, and diagnose fermentation anomalies based on the triggering of the catastrophic reignition mechanism.

[0041] During the iterative optimization of control parameters using the simulated annealing algorithm, the fermentation process status is monitored simultaneously, and a catastrophic reignition mechanism is established to diagnose anomalies. The trigger condition for this catastrophic reignition mechanism is that both the first energy function value and the metabolic entropy value simultaneously exceed a preset anomaly threshold. This dual trigger condition design ensures the accuracy of anomaly diagnosis. The first energy function value exceeding the preset anomaly threshold indicates a significant deviation from the healthy trajectory, although a single deviation may be caused by a brief disturbance. The metabolic entropy value exceeding the preset threshold indicates that fermentation is in a highly uncertain and chaotic state, reflecting a fundamental change in the process dynamics. Only when both exceed the limits simultaneously can a genuine anomaly be confirmed, rather than a brief fluctuation, thus effectively avoiding false alarms.

[0042] When the catastrophic reignition mechanism is triggered, the following operations are performed: The multidimensional state vector at the time the catastrophic reignition mechanism was triggered is immediately locked as an anomalous state feature. This anomalous state feature records key parameter values ​​such as temperature, pH, dissolved oxygen, and carbon dioxide concentration during the fermentation process at the moment the anomalous event occurs, serving as the basis for subsequent anomalous type identification. The temperature in the simulated annealing algorithm is reset to a preset high temperature value. This preset high temperature value is typically set to 2 to 3 times the initial temperature. Temperature reset puts the algorithm into a high-temperature state. According to the Metropolis criterion, the probability of the algorithm accepting a state solution with a higher energy function value increases significantly at high temperatures. This mechanism forces the algorithm to escape potentially invalid local optima and restart the global search, preventing the algorithm from lingering near anomalous states.

[0043] S4: Match the abnormal state characteristics with the preset abnormal database to determine the abnormal type with the highest matching degree.

[0044] After confirming the anomaly and capturing the anomaly state feature vector through the aforementioned catastrophe re-ignition mechanism, the anomaly state feature is matched with a preset anomaly database to determine the specific anomaly type.

[0045] The anomaly database is pre-constructed using the following method: First, historical fermentation data containing various known anomalous operating conditions is collected, including contamination anomalies, nutrient deficiency anomalies, dissolved oxygen deficiency anomalies, pH runaway anomalies, temperature runaway anomalies, and metabolite accumulation anomalies. Second, the K-means clustering algorithm is used to process the multidimensional state vector sets corresponding to each type of anomalous operating condition. Finally, the centroid vector of each cluster is used as a standardized anomaly vector for a specific anomaly type and stored in the anomaly database along with the corresponding anomaly type identifier. This construction method effectively extracts the core patterns of various anomalies from the chaotic historical data, laying the foundation for subsequent identification.

[0046] A separate iterative search process for anomaly identification is initiated. In this embodiment, simulated annealing is used for iterative search to find a set of optimal weight allocation schemes. The weight values ​​in this scheme reflect the similarity between the current anomaly state features and each standardized anomaly vector. The iterative search process includes: initializing each weight value, and in each iteration, generating a new weight vector in the neighborhood of the current weight vector, while ensuring that the weight sum remains 1 and each component is non-negative; calculating the distance corresponding to the new weight vector, and deciding whether to accept the new solution according to the Metropolis criterion; gradually reducing the temperature according to a preset cooling coefficient until the termination condition is met.

[0047] After obtaining the optimal weight allocation scheme, the standardized anomaly vector with the largest weight value is identified. The anomaly type corresponding to the standardized anomaly vector with the largest weight value is output as the anomaly type with the highest matching degree. Simultaneously, the maximum weight value is defined as the confidence level of the diagnostic result. Finally, a structured monitoring result is output to the operator, including the anomaly type, confidence level, time of anomaly occurrence, and specific parameter values ​​of the anomaly state feature vector, providing a decision-making basis for subsequent troubleshooting operations.

[0048] According to a second aspect of the present invention, the present invention also provides a bio-fermentation monitoring system based on multi-source data fusion. Figure 2 This is a structural block diagram of a bio-fermentation monitoring system based on multi-source data fusion according to an embodiment of the present invention. Figure 2As shown, the system includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the bio-fermentation monitoring method based on multi-source data fusion according to the first aspect of the present invention. The system also includes other components well-known to those skilled in the art, such as a communication bus and a communication interface; their configuration and functions are known in the art and therefore will not be described further here.

[0049] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be within the scope of protection of the present invention.

Claims

1. A bio-fermentation monitoring method based on multi-source data fusion, characterized in that, The monitoring method includes: Multi-source monitoring data from the fermenter is acquired to construct a multi-dimensional state vector and obtain the fermenter's control parameters. Based on the multi-dimensional state vector, a metabolic entropy is constructed to quantify the complexity of the current fermentation process. A first energy function is defined to evaluate the deviation of the current fermentation state, and a second energy function is defined to evaluate the effectiveness of the control parameters on the fermentation state. The control parameters are optimized using a simulated annealing algorithm, with the goal of minimizing the second energy function, outputting the optimal control parameters, where temperature is positively correlated with the metabolic entropy. During the iterative optimization of the simulated annealing algorithm, a catastrophic re-ignition mechanism is established, triggered when both the first energy function value and the metabolic entropy value simultaneously exceed a preset abnormal threshold. After triggering the catastrophic re-ignition mechanism, the multi-dimensional state vector is used as an abnormal state feature, and the temperature in the simulated annealing algorithm is reset to a preset high temperature value. The simulated annealing algorithm for anomaly matching is started based on a preset high temperature value. It iteratively searches by weighting and combining multiple anomaly vectors in the preset anomaly database, with the optimization objective of minimizing the distance between the weighted combined anomaly vector and the anomaly state feature. The anomaly type with the highest matching degree is determined according to the optimal weight allocation scheme, and the monitoring result of the anomaly type is output.

2. The bio-fermentation monitoring method based on multi-source data fusion according to claim 1, characterized in that, The construction of the multidimensional state vector includes: The multi-source monitoring data includes at least three of the following: temperature, pH, dissolved oxygen, and carbon dioxide concentration. The acquired data is preprocessed by filtering and normalization, and the preprocessed data are combined to form the multidimensional state vector.

3. The bio-fermentation monitoring method based on multi-source data fusion according to claim 2, characterized in that, The process of constructing the metabolic entropy includes: Arrange the multiple multidimensional state vectors within the preset time window to form a state matrix; Principal component analysis was performed on the state matrix to obtain a set of principal components; Calculate the variance of each principal component and the total variance of the matrix; The proportion of the variance of each principal component to the total variance is determined, and the metabolic entropy is calculated based on the proportion using the Shannon entropy formula.

4. The bio-fermentation monitoring method based on multi-source data fusion according to claim 1, characterized in that, The construction of the first energy function includes: Retrieve the standard trajectory vector corresponding to the current fermentation stage from the preset standard database; Calculate the weighted Euclidean distance between the current state's multidimensional state vector and the standard trajectory vector, and use it as the trajectory deviation term; Determine whether each component in the multidimensional state vector of the current state exceeds the preset process constraint range, and sum the squares of the ranges that exceed the process constraint range as the constraint terms; The first energy function is obtained by weighted summation of the trajectory deviation term and the constraint term.

5. The bio-fermentation monitoring method based on multi-source data fusion according to claim 4, characterized in that, The construction of the second energy function includes: A prediction model is constructed based on historical fermentation data. Based on the current multidimensional state vector and control parameters, the prediction model is used to predict the multidimensional state vector at the next moment. The weighted Euclidean distance between the predicted multidimensional state vector and the standard trajectory vector is calculated and used as the state deviation term. Calculate the L2 norm of the difference vector between the control parameter to be evaluated and the currently implemented control parameter, and use it as the control cost term; The second energy function is obtained by weighted summation of the state deviation term and the control cost term.

6. The bio-fermentation monitoring method based on multi-source data fusion according to claim 5, characterized in that, The positive correlation between temperature and metabolic entropy in the simulated annealing algorithm includes: The positive correlation is achieved through a preset functional relationship, wherein the temperature is set as a monotonically increasing function of the metabolic entropy.

7. The bio-fermentation monitoring method based on multi-source data fusion according to claim 6, characterized in that, The catastrophe reignition mechanism includes: When the catastrophic reignition mechanism is triggered, the current temperature of the simulated annealing algorithm is set to a preset initial high temperature value, so as to increase the probability of the algorithm accepting state solutions with higher energy function values ​​in subsequent iterations, based on the Metropolis criterion.

8. The bio-fermentation monitoring method based on multi-source data fusion according to claim 7, characterized in that, The construction of the anomaly database includes: Collect historical fermentation data containing various abnormal operating conditions, and extract a set of multidimensional state vectors corresponding to each abnormal operating condition; Clustering algorithms are used to process the multidimensional state vector set, dividing vectors with similar characteristics into different clusters; The centroid vector of each cluster is used as a standardized anomaly vector for a specific anomaly type to form an anomaly database.

9. The bio-fermentation monitoring method based on multi-source data fusion according to claim 8, characterized in that, The method of determining the anomaly type with the highest matching degree according to the optimal weight allocation scheme includes: The triggering of the aforementioned catastrophe reignition mechanism is used as a marker of an abnormal event; The optimal weight allocation scheme obtained through iterative search is used to identify the outlier vector with the largest weight value. The anomaly type corresponding to the anomaly vector with the largest weight value is output as the anomaly type with the highest matching degree; The abnormality types include at least one of the following: bacterial contamination abnormality, nutrient deficiency abnormality, dissolved oxygen deficiency abnormality, pH dysregulation abnormality, temperature dysregulation abnormality, and metabolic product accumulation abnormality.

10. A bio-fermentation monitoring system based on multi-source data fusion, characterized in that, include: processor; A memory, wherein a computer program is stored; When the processor is configured to execute the computer program, it implements the bio-fermentation monitoring method based on multi-source data fusion as described in any one of claims 1 to 9.

Images