A bio-enzyme preparation production abnormality positioning method based on process model constraints

Abstract

The application discloses a kind of based on process model constraint biological enzyme preparation production abnormal positioning method, it is related to production supervision technical field.The application extracts feature data segment from historical biological enzyme preparation production record by multidimensional feature extraction index, establishes multidimensional feature vector according to feature data segment, establishes biological enzyme process flow, sets multiple constraint terms to biological enzyme process flow, establishes standard process constraint model based on multidimensional feature vector with constraint term, generates corresponding real-time multidimensional feature vector based on real-time production data, maps in standard process constraint model according to corresponding process space-time sequence with real-time multidimensional feature vector, carries out abnormal discrimination, labels multiple abnormal positions in standard process constraint model according to abnormal discrimination result, and outputs abnormal sharing degree according to the real-time production data of abnormal position, and then locates the fault source point of biological enzyme process flow by residual propagation and standard process constraint model.

Description

Technical Field

[0001] This invention relates to the field of production monitoring technology, specifically to a method for locating anomalies in the production of biological enzyme preparations based on process model constraints. Background Technology

[0002] The production process of bio-enzyme preparations is highly complex and dynamic, involving multivariate coupling, multi-process linkage, and nonlinear reaction mechanisms. Traditional bio-enzyme production anomaly monitoring systems primarily rely on univariate threshold alarms, post-event statistical analysis, or simple rule-based judgments, making it difficult to capture the deep-seated spatiotemporal correlations within the production process. Conventional methods, such as statistical process control (SPC) or machine learning-based outlier detection, often sever the intrinsic connections between process steps, resulting in significant lags in response to anomalies and a high likelihood of false alarms or missed alarms. More importantly, existing technologies generally lack systematic modeling of the inherent physicochemical constraints of the process, leading to insufficient anomaly localization accuracy, difficulties in fault tracing, and challenges in guiding rapid intervention and process optimization on the production floor.

[0003] With the development of industrial big data and digital twin technologies, it has become possible to construct process models that reflect the dynamic characteristics of the entire bio-enzyme production process. However, existing models mostly focus on the simulation of a single stage or data-driven prediction, failing to organically integrate the spatiotemporal patterns implicit in historical production data with the structural constraints of the process flow. This results in poor interpretability of anomaly identification results, making it difficult to clearly reveal the propagation path of anomalies between processes, and thus difficult to accurately trace the root cause of the failure. In addition, alignment deviations between real-time production data and historical benchmark data, as well as the fusion processing of multi-source heterogeneous data, also pose technical challenges to accurate anomaly localization. To address these challenges, a method for anomaly localization in bio-enzyme preparation production based on process model constraints is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a method for locating anomalies in the production of biological enzyme preparations based on process model constraints, so as to solve the problems in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for locating anomalies in the production of biological enzyme preparations based on process model constraints, comprising the following steps: Step S1: Obtain historical bio-enzyme production records, set multi-dimensional feature extraction indicators, extract several feature data fragments from historical bio-enzyme preparation production records through multi-dimensional feature extraction indicators, perform spatiotemporal alignment on each feature data fragment, and then establish a multi-dimensional feature vector based on the spatiotemporally aligned feature data fragments. Step S2: Establish the bio-enzyme process flow, set multiple constraints for the bio-enzyme process flow, and then establish a standard process constraint model based on multi-dimensional feature vectors and multiple constraints. Step S3: Collect real-time production data of bio-enzymes, and generate corresponding real-time multidimensional feature vectors based on the real-time production data. Map the real-time multidimensional feature vectors to the standard process constraint model according to the corresponding process time and space order for anomaly detection. Step S4: Based on the anomaly identification results, mark multiple anomaly locations in the standard process constraint model, and output the anomaly sharing degree based on the real-time production data of the anomaly locations. Then, through residual propagation and the standard process constraint model, locate the fault source point of the bio-enzyme process.

[0006] Furthermore, the process of obtaining historical bio-enzyme production records, setting multi-dimensional feature extraction indicators, and extracting several feature data fragments from historical bio-enzyme preparation production records using these indicators includes: The historical bio-enzyme production records cover the entire production chain and the entire life cycle, ensuring the integrity, normal status, and abnormal status of the data; Set up multi-dimensional feature extraction indicators, which include analog quantity indicators, equipment status indicators, and process setpoint indicators; A sliding time window is used to extract historical bioenzyme production records. The entire historical bioenzyme production record is traversed through the sliding time window, and a basic data segment is generated for each step. The generated basic data segments are then filtered to obtain several feature data segments.

[0007] Furthermore, the process of spatiotemporally aligning each feature data segment and then constructing a multidimensional feature vector based on the spatiotemporally aligned feature data segments includes: The process steps include fermentation → enzymatic hydrolysis → extraction; A unified timeline is set according to the process steps. A reference timestamp is set on the process steps on the unified timeline. Then, the original timestamps of all data in each feature data segment are extracted, mapped to the unified timeline and compared with the corresponding reference timestamps. The absolute value of the time deviation is obtained based on the comparison results. The feature data segments are time-aligned based on the absolute value of the time deviation. Each feature data fragment is labeled with the corresponding process step number and equipment number. Then, based on the spatial location of the equipment corresponding to each process step, each feature data fragment is associated with the spatial location of the equipment in the actual production process. After the feature data segments are spatiotemporally aligned, the basic feature values ​​of the feature data segments under each indicator are obtained. All basic feature values ​​are then subjected to dimensionality reduction and standardization to obtain multidimensional feature vectors.

[0008] Furthermore, the process of establishing a bio-enzyme process flow and setting multiple constraints for the bio-enzyme process flow includes: A bio-enzyme process flow is established based on the process steps, and the bio-enzyme process flow includes a fermentation process stage, an enzymatic hydrolysis process stage, and an extraction process stage. Based on historical bio-enzyme production records, multiple production devices are bound to each industrial stage. At the same time, multiple constraints are set for the bio-enzyme process flow, including material balance constraints, energy balance constraints, timing constraints, and safety interlock logic constraints.

[0009] Furthermore, the process of establishing a standard process constraint model based on multidimensional feature vectors and multiple constraint terms includes: The standard process constraint model includes a physical process layer, a constraint rule layer, and a feature mapping layer. The physical process layer contains a 3D model of the equipment. The constraint rule layer transforms four constraint items—material balance, energy balance, timing constraints, and safety interlock logic—into mathematical constraint equations to construct a constraint rule library. The feature mapping layer classifies multidimensional feature vectors corresponding to normal states by process stage and equipment number, calculates the mean and standard deviation of multidimensional feature vectors for each process link and each production equipment in different time intervals, and determines the normal range of core feature values ​​for each dimension.

[0010] Furthermore, the process of collecting real-time production data of biological enzymes and generating corresponding real-time multidimensional feature vectors based on this data includes: Various sensors and data acquisition devices are deployed at each stage of the bio-enzyme production process to form a data acquisition network that monitors each stage of the entire bio-enzyme production process. The data acquisition network covers multi-dimensional feature extraction indicators of the entire bio-enzyme production process. It adopts the process of converting historical bio-enzyme production records into multi-dimensional feature vectors and converting real-time production data into real-time multi-dimensional feature vectors of the same dimension.

[0011] Furthermore, the real-time multidimensional feature vectors are mapped onto the standard process constraint model according to the corresponding process spatiotemporal order for anomaly detection. The specific process includes: Based on the timestamp of the real-time multidimensional feature vector, the corresponding time interval is located in the standard process constraint model, and the constraint rules within the time interval are matched from the constraint rule library. Based on the process stage number and equipment number of the real-time multidimensional feature vector, the corresponding process link and equipment position are located in the physical process layer of the standard process constraint model, and the constraint rules of the corresponding process link and equipment are matched. Obtain the Euclidean distance between the real-time multidimensional feature vector and the multidimensional feature vector under normal conditions in the corresponding process link and time interval in the standard process constraint model. Set a distance threshold. When the Euclidean distance is greater than or equal to the distance threshold, it is determined to be an abnormal feature deviation. Otherwise, it is determined to be an abnormal feature deviation. The real-time multidimensional core feature values ​​of the real-time multidimensional feature vector are reverse-mapped to the statistical values ​​of the corresponding indicators and substituted into the corresponding mathematical constraint equations of the constraint rule layer of the standard process constraint model. If any constraint equation is not satisfied, it is judged as a constraint violation anomaly; otherwise, it is judged as an unconstrained violation anomaly. If either the feature deviation or constraint violation is deemed abnormal, the entire system is judged as an abnormal production system, and the abnormality judgment result is output; otherwise, the entire system is judged as a normal production system.

[0012] Furthermore, the process of marking multiple anomaly locations in the standard process constraint model based on the anomaly identification results, and outputting the anomaly sharing degree based on the real-time production data of the anomaly locations, includes: In the physical flow layer of the standard process constraint model, the corresponding 3D model of the equipment is located by matching spatial coordinates based on the process stage number and equipment number of the anomaly judgment result. If multiple abnormal locations occur at the same or adjacent timestamps, and the abnormality involves related devices, then the relevant abnormal markers are connected by dashed lines in the device's 3D model, labeled with the words "related abnormality," and numbered according to the order in which the abnormalities occurred; otherwise, no number is marked. The anomaly sharing degree is used to quantify the degree of deviation between real-time production data at anomaly locations and the standard process constraint model, as well as the impact range of the corresponding anomaly on surrounding process links and equipment, providing a quantitative basis for prioritizing fault sources. Based on the above process, the anomaly sharing degree of each anomaly location is obtained, and the fault source analysis is performed sequentially by sorting the anomaly sharing degree values ​​from largest to smallest.

[0013] Furthermore, the process of locating the fault source points in the bio-enzyme process through residual propagation and standard process constraint models includes: A residual propagation model is established, which is based on the physical process layer of the standard process constraint model and constructs a residual propagation network based on the material flow direction, energy transfer path, and equipment correlation of bio-enzyme production. The residual is defined as the difference between each real-time production data and the median value of the corresponding dimension feature value in the feature mapping layer of the standard process constraint model within the normal range of the real-time production data. The anomaly sharing degree values ​​are sorted from largest to smallest. Fault source analysis is performed on each anomaly location in the residual propagation network, and several residual propagation paths are traversed. Starting from the abnormal location, trace the source of the residual along the residual propagation path of the residual propagation network, obtain the residual propagation intensity of each residual propagation path, perform partial ordering on each residual propagation path according to the magnitude of the residual propagation intensity, and trace back to the initial position with no upstream path in sequence, which is recorded as the candidate fault source point. For the candidate fault source locations obtained from the traced residuals, forward simulation is performed using the standard process constraint model. The parameters of the candidate fault source locations are set to outliers, and the simulation results are used to verify that the corresponding candidate fault source locations are indeed fault source locations.

[0014] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This invention constructs a multidimensional feature vector with spatiotemporal alignment capabilities to capture the dynamic correlation characteristics in the bio-enzyme production process. Combined with multiple constraints set by prior process knowledge, the established standard process constraint model can accurately characterize the spatial and temporal boundaries of normal production. Compared with traditional methods, it effectively improves the sensitivity and accuracy of anomaly detection and achieves near real-time monitoring response.

[0015] 2. This invention achieves precise location of fault sources in the bio-enzyme process by identifying abnormal occurrences, marking abnormal locations, calculating abnormal sharing degree, and implementing residual propagation mechanisms. It effectively reveals the propagation path and impact range of abnormalities between processes, and enhances the interpretability of diagnostic results. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0017] Figure 1 This is a flowchart of a method for locating anomalies in the production of biological enzyme preparations based on process model constraints, as described in this invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0019] Please see Figure 1As shown, a method for locating anomalies in the production of biological enzyme preparations based on process model constraints includes the following steps: Step S1: Obtain historical bio-enzyme production records, set multi-dimensional feature extraction indicators, extract several feature data fragments from historical bio-enzyme preparation production records through multi-dimensional feature extraction indicators, perform spatiotemporal alignment on each feature data fragment, and then establish a multi-dimensional feature vector based on the spatiotemporally aligned feature data fragments. Step S2: Establish the bio-enzyme process flow, set multiple constraints for the bio-enzyme process flow, and then establish a standard process constraint model based on multi-dimensional feature vectors and multiple constraints. Step S3: Collect real-time production data of bio-enzymes, and generate corresponding real-time multidimensional feature vectors based on the real-time production data. Map the real-time multidimensional feature vectors to the standard process constraint model according to the corresponding process time and space order for anomaly detection. Step S4: Based on the anomaly identification results, mark multiple anomaly locations in the standard process constraint model, and output the anomaly sharing degree based on the real-time production data of the anomaly locations. Then, through residual propagation and the standard process constraint model, locate the fault source point of the bio-enzyme process.

[0020] Furthermore, step S1 is implemented through the following process: Step S101: Obtain historical bio-enzyme production records, set multi-dimensional feature extraction indicators, and extract several feature data fragments from the historical bio-enzyme preparation production records using these indicators. The specific process includes: The historical bioenzyme production records cover the entire production chain and life cycle, ensuring the integrity, normal status, and abnormal status of the data. The data sources mainly include enterprise production management devices, equipment operation and maintenance logs, and quality inspection databases. Specifically, it covers complete records of more than 100 production batches in the past 3 years, including production data of different product models (such as neutral protease, cellulase, and amylase) and different production scales. The historical bioenzyme production records include: original monitoring data (temperature, pH, dissolved oxygen concentration, stirring speed, and fermenter pressure during fermentation; substrate concentration, enzyme dosage, reaction temperature, and degree of hydrolysis during enzymatic hydrolysis; filtration pressure, purification column flow rate, and eluent concentration during extraction); process operation data (raw material input and ratio, inoculum size, reaction time setting, stirring intensity adjustment records, and acid-base regulator dosage); equipment status data (cumulative runtime, operating current, vibration frequency, noise level, and maintenance records for production equipment such as fermenters, enzymatic hydrolysis reactors, filters, and purification columns); fault history (past fault types, frequency of occurrence, and repair effects); and quality testing data (enzyme activity, cell concentration, substrate residue, finished product purity, yield, and moisture content at each process stage). Set up multi-dimensional feature extraction indicators, which include analog quantity indicators, equipment status indicators, and process setpoint indicators; Among them, analog indicators are used to focus on the continuously changing physical and chemical parameters during the production process, directly reflecting the state of the process reaction, such as the fermentation stage: temperature (25-35℃), pH value (4.5-6.5), and dissolved oxygen concentration; Equipment status indicators are used to reflect the health of production equipment and are key indicators for judging equipment failure and process setpoints. Examples of basic operating parameters include: cumulative operating time (h), continuous operating time of a single run (h), operating current (A), voltage fluctuation value (V), and vibration amplitude (mm / s). The process setpoints are used to represent the benchmark for production operation and reflect the parameter constraint range for normal production. For example, the parameter settings for each stage are: fermentation temperature setpoint ±2℃, fermentation time setpoint ±2h, enzymatic hydrolysis temperature setpoint ±1℃, enzymatic hydrolysis time setpoint ±1h, and extraction pressure setpoint ±0.02MPa. Since the production of bio-enzymes is continuous, a sliding time window is used to extract historical bio-enzyme production records. The window length is set to 5 minutes (to ensure coverage of the dynamic change cycle of process parameters), and the step size is set to 30 seconds (to ensure a balance between data continuity and redundancy). The sliding time window iterates through all historical bio-enzyme production records, and a basic data segment is generated for each step. Each segment contains all the original data of analog quantity indicators, equipment status indicators, and process setpoint indicators within 5 minutes. The generated basic data fragments are screened based on the following criteria: data integrity (no missing values ​​in the fragments, and the missing data rate of a single indicator is ≤1%) and validity (the fragments must contain clear process status information). At least 50 valid data fragments are retained for each indicator, and the ratio of normal state fragments to abnormal state fragments is controlled at 3:1 to ensure a balanced data distribution. Finally, several feature data fragments are obtained through screening.

[0021] Step S102: Perform spatiotemporal alignment on each feature data segment, and then establish a multidimensional feature vector based on the spatiotemporally aligned feature data segments. The specific process includes: The purpose of spatiotemporal alignment of feature data segments is to eliminate interference caused by time differences in data acquisition from different sensors and misalignment of data from different process steps, to ensure the spatiotemporal consistency and comparability of feature data segments, and to provide a standardized data foundation for subsequent feature vector construction. The process steps include fermentation → enzymatic hydrolysis → extraction. A unified timeline is set according to the process steps. A reference timestamp is set on the unified timeline for each process step. Then, the original timestamps of all data in each feature data segment are extracted, mapped to the unified timeline and compared with the corresponding reference timestamps. The absolute value of the time deviation is obtained based on the comparison results. The original timestamps of feature data segments with an absolute time deviation of more than 1 second and less than or equal to 5 seconds are corrected. Feature data segments with an absolute time deviation of more than 5 seconds are directly removed. No operation is performed on feature data segments with an absolute time deviation of less than or equal to 1 second. After correction, the data of various indicators at the same time point are accurately matched. Once the feature data segments are time-aligned, each feature data segment is labeled with a corresponding process step number (e.g., F01 = fermentation stage, E01 = enzymatic hydrolysis stage, E02 = extraction stage) and equipment number (e.g., F-T01 = fermenter temperature sensor, E-R01 = enzymatic hydrolysis reactor). Then, based on the spatial location of the equipment corresponding to each process step, each feature data segment is associated with the spatial location of the equipment in the actual production process to ensure that the spatial attributes of the feature data segments are consistent with the actual production. After the feature data segments are spatiotemporally aligned, the statistical values ​​of the feature data segments under each indicator are obtained, including the mean, maximum, minimum and standard values, and then the basic feature values ​​of each feature data segment are obtained. Dimensionality reduction is performed on all basic feature values. By obtaining core feature values ​​and variance contribution rates, the top m principal components with a cumulative variance contribution rate of ≥85% are selected as the final feature dimensions. The basic feature values ​​are reduced to m-dimensional core feature values, which ensures that core information is not lost and greatly improves the efficiency of subsequent model processing. Here, m is a natural number less than 10 and greater than 1. The m-dimensional core feature values ​​are standardized using Z-score to eliminate the influence of different metrics, thereby obtaining a standardized multi-dimensional feature vector.

[0022] Furthermore, step S2 is implemented through the following process: Step S201: Establish the bio-enzyme process flow and set multiple constraints for the bio-enzyme process flow. The specific process includes: A bio-enzyme process flow is established based on the process steps, and the bio-enzyme process flow includes a fermentation process stage, an enzymatic hydrolysis process stage, and an extraction process stage. Based on historical bio-enzyme production records, multiple production equipment are linked to each industrial stage. For example, the production equipment for the fermentation process stage includes stainless steel fermenters (equipped with temperature control devices, pH adjustment devices, dissolved oxygen detection devices, stirring devices, and feeding devices); the production equipment for the enzymatic hydrolysis process stage includes enzymatic hydrolysis reactors (equipped with temperature control devices, stirring devices, pH adjustment devices, and sampling ports); and the production equipment for the extraction process stage includes plate and frame filters, ion exchange chromatography columns, gel filtration chromatography columns, vacuum concentrators, and freeze dryers. Meanwhile, multiple constraints are set for the bio-enzyme process, including material balance constraints, energy balance constraints, timing constraints, and safety interlock logic constraints. Based on the law of conservation of mass, material balance constraints are established for each process stage, clarifying the quantitative relationship between input materials, output materials, and process losses. The formula for calculating the constraint is: m in =m out +m loss , where m in To input the total mass of the materials, m out To output the total mass of the material, m loss For process loss quality; Based on the law of conservation of energy, and considering heat transfer, mass transfer, and energy loss during the production process, energy balance constraints are established for each process stage. The formula for calculating the constraint is: Q in =Q out +Q loss Q in For the input energy (including thermal energy from the heating device, mechanical energy from the stirring motor, etc.), Q out For output energy (including energy absorbed by material heating, energy dissipated by evaporation, etc.), Q loss Energy loss (including equipment heat dissipation, pipeline heat dissipation, etc.); The timing constraints clearly define the sequence and time interval requirements of each process stage and operation step in the bio-enzyme process to ensure the orderliness and rationality of the production process. Strictly follow the process flow of fermentation → enzymatic hydrolysis → extraction, and do not reverse the order of key operations; if the fermentation stage is not completed (enzyme activity does not meet the standard), the enzymatic hydrolysis stage must not be entered; if the enzymatic hydrolysis stage is not terminated, the extraction stage must not be entered. Safety interlock logic constraints are based on equipment safety thresholds and process safety standards to establish a linkage between equipment operating status, process parameters, and safety actions, ensuring the safety of the production process. For example, temperature interlock: when the fermenter temperature is ≥35℃, the cooling device is automatically started; when the temperature is ≥38℃, an over-temperature alarm is triggered and the machine is shut down; when the enzymatic hydrolysis reactor temperature is ≥42℃, heating is immediately stopped and cooling is started, and enzyme addition is stopped at the same time. Pressure interlock: When the fermenter pressure is ≥0.35MPa, the pressure relief valve will automatically open; when the pressure is ≥0.4MPa, an overpressure alarm will be triggered and the machine will be shut down immediately; when the filtration pressure in the extraction stage is ≥0.2MPa, the feed flow rate will be automatically reduced; when the pressure is ≥0.25MPa, the machine will be shut down and an alarm will be triggered. Level interlock: When the liquid level in the fermenter or enzymatic hydrolysis reactor is ≥90%, feeding is stopped; when the liquid level is ≤10%, stirring is stopped and an alarm is triggered. Component interlock: When the pH value of the fermentation broth is <4.0 or >7.0, the feeding will be automatically stopped and acid-base adjustment will be initiated; when the adjustment is ineffective (pH value <3.5 or >7.5), the machine will be shut down and an alarm will be triggered; when the degree of hydrolysis of the enzymatic hydrolysate is <10% and the reaction time is ≥48h, an abnormal reaction alarm will be triggered, etc.

[0023] Step S202: Establish a standard process constraint model based on multi-dimensional feature vectors and multiple constraint terms. The specific process includes: The standard process constraint model includes a physical process layer, a constraint rule layer, and a feature mapping layer. It integrates the process flow, constraint rules, and multi-dimensional feature vectors to form a benchmark model that can reflect normal production patterns. The physical process layer is based on the bio-enzyme process flow. It uses 3D modeling technology to construct a 1:1 matching 3D model of the equipment to the actual production scenario. The 3D model of the equipment restores the production equipment, pipeline connections, sensor installation positions and material flow of each process stage, and marks the equipment number, process stage number and key operation nodes. The production equipment parameters (such as fermenter volume and stirring power) and process parameters (such as normal operating temperature range) are associated with the equipment 3D model to realize the visualization of the process flow. It supports clicking on the equipment model with the mouse to query the historical operating data, constraint rules and feature vector range of the corresponding production equipment, providing a spatial reference for subsequent spatiotemporal mapping. The constraint rule layer transforms the four constraint terms of material balance, energy balance, timing constraints, and safety interlock logic into mathematical constraint equations, and constructs a constraint rule library. Each constraint equation is associated with the corresponding process stage, production equipment, and indicators. The feature mapping layer classifies multidimensional feature vectors corresponding to normal states by process stage and equipment number, calculates the mean and standard deviation of multidimensional feature vectors for each process link and each production equipment in different time intervals, and determines the normal range of core feature values ​​for each dimension (mean ± 2 times standard deviation).

[0024] Furthermore, step S3 is implemented through the following process: Step S301: Collect real-time production data of the bio-enzyme and generate corresponding real-time multidimensional feature vectors based on the real-time production data. The specific process includes: Various sensors and data acquisition devices are deployed in each process stage of bio-enzyme production to form a data acquisition network that monitors each process stage of the entire bio-enzyme production process. The data acquisition network consists of process parameter sensors (such as temperature sensors, pH sensors, dissolved oxygen sensors, etc.), equipment status sensors (such as vibration sensors, current sensors, voltage sensors, etc.), and data transmission devices (such as industrial Internet of Things gateways). The data acquisition network covers multi-dimensional feature extraction indicators of the entire bio-enzyme production process, including process parameters, equipment status parameters and process operation records of each process stage of fermentation, enzymatic hydrolysis and extraction. After time synchronization between various sensors, real-time production data of each process link in the bio-enzyme production process are acquired at the same acquisition frequency of 5 seconds / time. The process involves converting historical bio-enzyme production records into multi-dimensional feature vectors, and then converting real-time production data into real-time multi-dimensional feature vectors of the same dimension.

[0025] Step S302: Map the real-time multidimensional feature vectors to the standard process constraint model according to the corresponding process spatiotemporal order for anomaly detection. The specific process includes: Based on the timestamp of the real-time multidimensional feature vector, the corresponding time interval is located in the standard process constraint model, and the constraint rules within the time interval are matched from the constraint rule library (such as material balance constraints and time sequence constraints in the 24th hour of the fermentation stage). For real-time multidimensional feature vectors that span time intervals (such as the vector transitioning from the 48th hour of the fermentation stage to the enzymatic hydrolysis stage), the constraint rules of both time intervals are matched simultaneously to avoid omissions. Based on the process stage number and equipment number of the real-time multidimensional feature vector, the corresponding process link and equipment position are located in the physical flow layer of the standard process constraint model, and the constraint rules of the corresponding process link and equipment are matched (such as the safety interlock constraint of fermenter temperature and the energy balance constraint of enzymatic reactor); for real-time multidimensional feature vectors involving multi-equipment collaboration (such as the linkage data between feed pump and fermenter), the constraint rules of associated equipment are matched at the same time to ensure the integrity of the mapping. Calculate the Euclidean distance between the real-time multidimensional feature vector and the corresponding multidimensional feature vector under normal conditions in the standard process constraint model at the corresponding process stage and time interval. The formula for calculating the Euclidean distance is: d=√[(x1-y1)²+...+(x i -y i )²+...+(x m -y m )²], where x i Let y be the i-th eigenvalue of the real-time multidimensional feature vector. iLet i be the i-th eigenvalue of a normal vector. Based on the Euclidean distance statistics of multidimensional eigenvectors, a distance threshold is set. When the Euclidean distance is greater than or equal to the distance threshold, it is determined to be an abnormal feature deviation. Otherwise, it is determined to be an abnormal feature deviation. i is an integer less than or equal to m and greater than 1. The real-time multidimensional core feature values ​​of the real-time multidimensional feature vector are reverse-mapped to the statistical values ​​of the corresponding indicators, and then substituted into the corresponding mathematical constraint equations of the constraint rule layer of the standard process constraint model. The compliance of material balance, energy balance, timing constraints, and safety interlock logic is checked one by one. If any constraint equation is not satisfied, it is judged as a constraint violation anomaly; otherwise, it is judged as an unconstrained violation anomaly. If either feature deviation or constraint violation is judged as abnormal, the whole is judged as a production abnormality and the abnormality judgment result is output. If both are judged as normal, the whole is judged as a normal production. The anomaly detection results include: anomaly marker (0 = normal, 1 = abnormal), anomaly type (feature deviation anomaly, constraint violation anomaly, or mixed anomaly), the process stage number, equipment number, anomaly timestamp, violation constraint item number (constraint violation anomaly), and Euclidean distance value (feature deviation anomaly).

[0026] Furthermore, step S4 is implemented through the following process: Step S401: Based on the anomaly identification results, mark multiple anomaly locations in the standard process constraint model, and output the anomaly sharing degree based on the real-time production data of the anomaly locations. The specific process includes: In the physical process layer of the standard process constraint model, based on the process stage number and equipment number of the anomaly judgment result, the corresponding three-dimensional model of the equipment (such as the installation position of the temperature sensor of fermenter F-T01) is located by spatial coordinate matching. If multiple abnormal locations occur at the same timestamp or adjacent timestamps (≤10 minutes), and the abnormality involves related equipment (such as fermenter and feed pump, enzymatic reactor and temperature control device), then the relevant abnormal markers are connected by dashed lines in the equipment 3D model, labeled with the words "related abnormality", and numbered according to the order in which the abnormality occurred (1, 2, 3, ...). Otherwise, no number is marked. The anomaly sharing degree is used to quantify the degree of deviation between real-time production data at anomaly locations and the standard process constraint model, as well as the impact range of the corresponding anomaly on surrounding process links and equipment, providing a quantitative basis for prioritizing fault sources. The higher the anomaly sharing degree A, the greater the severity and scope of the anomaly. The abnormal sharing degree A is obtained by summing the parameter deviation degree A1, constraint violation degree A2, and correlation impact degree A3. The calculation formula is: A = (A1 + A2 + A3). The process of obtaining the parameter deviation A1 includes: Parameter deviation is used to quantify the degree of deviation between real-time production data and the corresponding normal parameter range in the standard process constraint model: For real-time production data corresponding to a single abnormal location (such as fermenter temperature), obtain the normal range of the corresponding indicator data in the model [μ-2σ, μ+2σ] (μ is the historical mean, σ is the historical standard deviation). Calculate the absolute deviation Δj = |x - μ| between the real-time production data j and the center value μ of the normal range. Let the absolute deviation Δj be the residual e. The half-width of the normal range is W = μ + 2σ - μ = 2σ. If Δj ≥ W, then A1 = 1. If Δj is less than W, then A1 = 0. For example, the normal temperature range of the fermenter is [28℃, 32℃] (μ=30℃, σ=1℃), and the real-time temperature j=34℃. Then Δj=4℃, W=2℃. Since Δj>W, A1=1 in the end. The process of obtaining the constraint violation degree A2 includes: Constraint violation severity is used to quantify the severity of constraint violations at out-of-place locations: The number of constraints n violated at abnormal locations is counted, and the weight ω of each constraint is determined. For each violation constraint, a violation coefficient v is assigned according to the degree of deviation. The constraint violation degree A2 = Σ(ω×v) / n. If no constraint is violated, then A2 = 0. The process of obtaining the correlation influence A3 includes: The correlation impact level is used to quantify the range of impact of anomalies on surrounding processes and equipment. Centered on the equipment corresponding to the abnormal location, determine the associated equipment and process steps based on the material flow direction and energy transfer path of the process flow (e.g., associated equipment of the fermenter includes feed pump, cooling device, pH adjustment device, and associated process steps include the enzymatic hydrolysis stage after fermentation). The total number of related equipment and process links k, and the number r of those with data fluctuations (deviation from the normal range ±5%), are counted. The degree of influence of the correlation is A3 = k / r. If there are no related equipment, then A3 = 0. Based on the above process, the anomaly sharing degree of each anomaly location is obtained, and the fault source analysis is performed sequentially by sorting the anomaly sharing degree values ​​from largest to smallest.

[0027] Step S402: Locate the fault source in the bio-enzyme process using residual propagation and the standard process constraint model. The specific process includes: A residual propagation model is established, which is based on the physical process layer of the standard process constraint model. Based on the material flow, energy transfer path, and equipment correlation of bio-enzyme production, a residual propagation network is constructed to clarify the propagation direction and path of the residuals. The residual e is defined as the difference between each real-time production data and the median value of the corresponding dimension feature value in the feature mapping layer of the standard process constraint model within the normal range of the real-time production data. Then, based on the abnormality sharing degree value, the abnormality locations are sorted from largest to smallest and the fault source analysis is performed on each abnormality location in the residual propagation network. Based on the fault source analysis results, several residual propagation paths are traversed. The residual propagation paths follow the principle of material / energy upstream → downstream and equipment input → output. For example, the residual in the fermentation stage can propagate to the enzymatic hydrolysis stage (downstream process), feed pump (associated input device), and cooling device (associated control device). A propagation weight p is assigned to each residual propagation path, with a value ranging from 0 to 1. p=1 indicates that the residual will definitely propagate, and p=0 indicates that the residual cannot propagate. The propagation weight p is determined by statistically analyzing the probability of the corresponding residual propagating from position A to position B in historical biological enzyme preparation production records. For example, if the probability of the fermenter temperature residual propagating to the feed pump flow rate in historical data is 75%, then the propagation weight p for this path is 0.75. The residual will attenuate during propagation due to factors such as process buffering and equipment adjustment. An attenuation coefficient α is set, where α = 0.8-1.0. The farther the propagation distance (such as across process stages), the smaller the attenuation coefficient. Starting from the abnormal location, trace the source of the residual along the residual propagation path of the residual propagation network, calculate the residual propagation intensity I of each residual propagation path, I=Σ(p*α*e), perform partial ordering on each residual propagation path according to the magnitude of the residual propagation intensity, and trace back to the initial position without upstream path in sequence, and record the initial position as the candidate fault source point. For candidate fault source locations obtained from tracing residuals, forward simulation is performed using the standard process constraint model. The parameters of the candidate fault source locations are set as outliers to simulate their impact on downstream process links and equipment. If the simulation results are consistent with the residuals and outlier types of the actual outlier locations, the corresponding candidate fault source location is verified as a fault source location. If they are inconsistent, the next residual propagation path is traced according to the ranking results of residual propagation intensity until a matching fault source location is found, and the fault source location is mapped to the physical process layer of the standard process constraint model.

[0028] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for locating anomalies in the production of biological enzyme preparations based on process model constraints, characterized in that, Includes the following steps: Step S1: Obtain historical bio-enzyme production records, set multi-dimensional feature extraction indicators, extract several feature data fragments from historical bio-enzyme preparation production records through multi-dimensional feature extraction indicators, perform spatiotemporal alignment on each feature data fragment, and then establish a multi-dimensional feature vector based on the spatiotemporally aligned feature data fragments. Step S2: Establish the bio-enzyme process flow, set multiple constraints for the bio-enzyme process flow, and then establish a standard process constraint model based on multi-dimensional feature vectors and multiple constraints. Step S3: Collect real-time production data of bio-enzymes, and generate corresponding real-time multidimensional feature vectors based on the real-time production data. Map the real-time multidimensional feature vectors to the standard process constraint model according to the corresponding process time and space order for anomaly detection. Step S4: Based on the anomaly identification results, mark multiple anomaly locations in the standard process constraint model, and output the anomaly sharing degree based on the real-time production data of the anomaly locations. Then, through residual propagation and the standard process constraint model, locate the fault source point of the bio-enzyme process.

2. The method for locating anomalies in the production of biological enzyme preparations based on process model constraints according to claim 1, characterized in that, The process of acquiring historical bio-enzyme production records, setting multidimensional feature extraction indicators, and extracting several feature data fragments from historical bio-enzyme preparation production records using these indicators includes: The historical bio-enzyme production records cover the entire production chain and the entire life cycle, ensuring the integrity, normal status, and abnormal status of the data; Set up multi-dimensional feature extraction indicators, which include analog quantity indicators, equipment status indicators, and process setpoint indicators; A sliding time window is used to extract historical bioenzyme production records. The entire historical bioenzyme production record is traversed through the sliding time window, and a basic data segment is generated for each step. The generated basic data segments are then filtered to obtain several feature data segments.

3. The method for locating anomalies in the production of biological enzyme preparations based on process model constraints according to claim 2, characterized in that, The process of spatiotemporally aligning each feature data segment and then constructing a multidimensional feature vector based on the spatiotemporally aligned feature data segments includes: The process steps include fermentation → enzymatic hydrolysis → extraction; A unified timeline is set according to the process steps. A reference timestamp is set on the process steps on the unified timeline. Then, the original timestamps of all data in each feature data segment are extracted, mapped to the unified timeline and compared with the corresponding reference timestamps. The absolute value of the time deviation is obtained based on the comparison results. The feature data segments are time-aligned based on the absolute value of the time deviation. Each feature data fragment is labeled with the corresponding process step number and equipment number. Then, based on the spatial location of the equipment corresponding to each process step, each feature data fragment is associated with the spatial location of the equipment in the actual production process. After the feature data segments are spatiotemporally aligned, the basic feature values ​​of the feature data segments under each indicator are obtained. All basic feature values ​​are then subjected to dimensionality reduction and standardization to obtain multidimensional feature vectors.

4. The method for locating anomalies in the production of biological enzyme preparations based on process model constraints according to claim 3, characterized in that, The process of establishing a bio-enzyme process flow and setting multiple constraints for the bio-enzyme process flow includes: A bio-enzyme process flow is established based on the process steps, and the bio-enzyme process flow includes a fermentation process stage, an enzymatic hydrolysis process stage, and an extraction process stage. Based on historical bio-enzyme production records, multiple production devices are bound to each industrial stage. At the same time, multiple constraints are set for the bio-enzyme process flow, including material balance constraints, energy balance constraints, timing constraints, and safety interlock logic constraints.

5. The method for locating anomalies in the production of biological enzyme preparations based on process model constraints according to claim 4, characterized in that, The process of establishing a standard process constraint model based on multidimensional eigenvectors and multiple constraint terms includes: The standard process constraint model includes a physical process layer, a constraint rule layer, and a feature mapping layer. The physical process layer contains a 3D model of the equipment. The constraint rule layer transforms four constraint items—material balance, energy balance, timing constraints, and safety interlock logic—into mathematical constraint equations to construct a constraint rule library. The feature mapping layer classifies multidimensional feature vectors corresponding to normal states by process stage and equipment number, calculates the mean and standard deviation of multidimensional feature vectors for each process link and each production equipment in different time intervals, and determines the normal range of core feature values ​​for each dimension.

6. The method for locating anomalies in the production of biological enzyme preparations based on process model constraints according to claim 5, characterized in that, The process of collecting real-time production data of biological enzymes and generating corresponding real-time multidimensional feature vectors based on this data includes: Various sensors and data acquisition devices are deployed at each stage of the bio-enzyme production process to form a data acquisition network that monitors each stage of the entire bio-enzyme production process. The data acquisition network covers multi-dimensional feature extraction indicators of the entire bio-enzyme production process. It adopts the process of converting historical bio-enzyme production records into multi-dimensional feature vectors and converting real-time production data into real-time multi-dimensional feature vectors of the same dimension.

7. The method for locating anomalies in the production of biological enzyme preparations based on process model constraints according to claim 6, characterized in that, The real-time multidimensional feature vectors are mapped to the standard process constraint model according to the corresponding process spatiotemporal order for anomaly detection. The specific process includes: Based on the timestamp of the real-time multidimensional feature vector, the corresponding time interval is located in the standard process constraint model, and the constraint rules within the time interval are matched from the constraint rule library. Based on the process stage number and equipment number of the real-time multidimensional feature vector, the corresponding process link and equipment position are located in the physical process layer of the standard process constraint model, and the constraint rules of the corresponding process link and equipment are matched. Obtain the Euclidean distance between the real-time multidimensional feature vector and the multidimensional feature vector under normal conditions in the corresponding process link and time interval in the standard process constraint model. Set a distance threshold. When the Euclidean distance is greater than or equal to the distance threshold, it is determined to be an abnormal feature deviation. Otherwise, it is determined to be an abnormal feature deviation. The real-time multidimensional core feature values ​​of the real-time multidimensional feature vector are reverse-mapped to the statistical values ​​of the corresponding indicators and substituted into the corresponding mathematical constraint equations of the constraint rule layer of the standard process constraint model. If any constraint equation is not satisfied, it is judged as a constraint violation anomaly; otherwise, it is judged as an unconstrained violation anomaly. If either the feature deviation or constraint violation is deemed abnormal, the entire system is judged as an abnormal production system, and the abnormality judgment result is output; otherwise, the entire system is judged as a normal production system.

8. The method for locating anomalies in the production of biological enzyme preparations based on process model constraints according to claim 7, characterized in that, The process of marking multiple anomaly locations in the standard process constraint model based on the anomaly identification results, and outputting the anomaly sharing degree based on the real-time production data of the anomaly locations, includes: In the physical flow layer of the standard process constraint model, the corresponding 3D model of the equipment is located by matching spatial coordinates based on the process stage number and equipment number of the anomaly judgment result. If multiple abnormal locations occur at the same or adjacent timestamps, and the abnormality involves related devices, then the relevant abnormal markers are connected by dashed lines in the device's 3D model, labeled with the words "related abnormality," and numbered according to the order in which the abnormalities occurred; otherwise, no number is marked. The anomaly sharing degree is used to quantify the degree of deviation between the real-time production data of the anomaly location and the standard process constraint model, as well as the impact range of the corresponding anomaly on the surrounding process links and equipment. This provides a quantitative basis for prioritizing fault sources, thereby obtaining the anomaly sharing degree of each anomaly location, and sorting the fault sources in descending order of the anomaly sharing degree value.

9. The method for locating anomalies in the production of biological enzyme preparations based on process model constraints according to claim 8, characterized in that, The process of locating the fault source points in the bio-enzyme process through residual propagation and standard process constraint models includes: A residual propagation model is established, which is based on the physical process layer of the standard process constraint model and constructs a residual propagation network based on the material flow direction, energy transfer path, and equipment correlation of bio-enzyme production. The residual is defined as the difference between each real-time production data and the median value of the corresponding dimension feature value in the feature mapping layer of the standard process constraint model within the normal range of the real-time production data. The anomaly sharing degree values ​​are sorted from largest to smallest. Fault source analysis is performed on each anomaly location in the residual propagation network, and several residual propagation paths are traversed. Starting from the abnormal location, trace the source of the residual along the residual propagation path of the residual propagation network, obtain the residual propagation intensity of each residual propagation path, perform partial ordering on each residual propagation path according to the magnitude of the residual propagation intensity, and trace back to the initial position with no upstream path in sequence, which is recorded as the candidate fault source point. For the candidate fault source locations obtained from the traced residuals, forward simulation is performed using the standard process constraint model. The parameters of the candidate fault source locations are set to outliers, and the simulation results are used to verify that the corresponding candidate fault source locations are indeed fault source locations.

Images