Intelligent production control system for high-activity selenium yeast based on gradient feeding monitoring

Abstract

The invention relates to the technical field of program control, and discloses a high-activity selenium yeast intelligent production control system based on gradient feeding monitoring. The system comprises a data acquisition module, a control parameter acquisition module, a gradient calculation module, an intelligent control module and a material supplementing control module. The system collects process production data, extracts key control parameters and weights, calculates real-time gradient values of the parameters, forms six-dimensional gradient vectors, generates control gradient indexes in combination with the parameter weights and time sequence attenuation weights, and then forms an addition control decision. And according to the decision and the process state dynamic data, the material supplementing opportunity is identified, an active adding instruction is generated, programmed control and adjustment of the raw materials and the additives are achieved, and the control precision and the production automation level are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of program control, in particular to an intelligent production control system for high-activity selenium yeast based on gradient feeding monitoring. Background Art

[0002] Modern industrial batch production processes are complex, multivariable, nonlinear dynamic systems that require precise control and coordinated management of multiple key process parameters, including temperature, pressure, pH, flow rate, and concentration. Batch production is characterized by intermittent and cyclical processes, and the patterns of process parameter changes and control requirements vary significantly within each production cycle.

[0003] Currently, batch production processes primarily rely on traditional process control systems based on PID controllers. These control systems use preset control programs and fixed control algorithms to adjust various process parameters throughout the production process. Traditional process control systems typically employ a single feedback control strategy. Based on the deviation between real-time sensor data and setpoints, they calculate control output signals and adjust the operating status of corresponding actuators, such as heaters, agitators, and dosing devices. Control programs typically implement segmented control based on time series or process stages, with each stage utilizing a preset combination of control parameters.

[0004] However, batch production processes are characterized by significant nonlinearity, time-varying characteristics, and strong coupling. In actual production, processes between different batches often exhibit significant variability, primarily due to factors such as slight fluctuations in raw material quality, changes in environmental conditions, variations in equipment status, and inconsistent operating conditions. Even with identical process parameters, reaction kinetics, conversion rates, and product formation patterns can vary significantly between batches, making it difficult to ensure the stability of final product quality and production efficiency.

[0005] Furthermore, traditional process control systems lack a deep understanding of process mechanisms and predictive capabilities. Feeding control strategies are typically based on empirical formulas or preset time schedules, and are unable to dynamically adjust to real-time process conditions and material requirements. This open-loop feeding control approach often results in waste of raw materials and can cause localized uneven concentration distribution or imbalanced ratios, impacting process stability and product quality. Improper control of feeding timing and feeding rate can cause fluctuations in process parameters, further exacerbating batch-to-batch variability.

[0006] Existing monitoring systems rely primarily on single-point measurement and discrete sampling, lacking the ability to continuously monitor process gradients. Traditional monitoring methods are unable to promptly capture transient changes in process parameters and local anomalies, resulting in a time lag in the control system's response. Gradient monitoring technology enables real-time tracking of process parameter trends, providing richer and more accurate information feedback for intelligent control.

[0007] Therefore, an intelligent production control system for high-activity selenium yeast based on gradient feeding monitoring was proposed. Summary of the Invention

[0008] The object of the present invention is to provide a high-activity selenium yeast intelligent production control system based on gradient feeding monitoring, comprising a data acquisition module for collecting process production data; a control parameter acquisition module for extracting key control parameters and weights of the production data; a gradient calculation module for obtaining real-time gradient values ​​of various parameters in the process production data and obtaining a six-dimensional gradient vector; analyzing the six-dimensional gradient vector to obtain process state dynamic data; an intelligent control module for fusing the six-dimensional gradient vector based on parameter importance weights and time series attenuation weights of key control parameters to generate a control gradient index; determining the type and amount of additives based on the control gradient index and generating an addition control decision; a feeding control module for identifying the optimal addition timing window based on the addition control decision and process state dynamic data and generating an active addition control instruction; and adding reaction raw materials and functional additives according to the active addition control instruction.

[0009] To achieve the above object, the present invention provides the following technical solutions:

[0010] An intelligent production control system for high-activity selenium yeast based on gradient feeding monitoring, comprising:

[0011] Data acquisition module, used to collect process production data in real time;

[0012] The control parameter acquisition module extracts key control parameters of process production data through the control parameter identification model;

[0013] The gradient calculation module is used to obtain the real-time gradient values ​​of various parameters in the process production data and obtain the six-dimensional gradient vector; a multi-dimensional gradient vector field model is constructed to analyze the six-dimensional gradient vector and obtain the dynamic data of the process status;

[0014] The intelligent control module performs weighted fusion of the six-dimensional gradient vector based on the parameter importance weights and time-series attenuation weights of key control parameters to generate a control gradient index. The module also determines the type and amount of additives based on the control gradient index and generates an additive control decision.

[0015] The feeding control module identifies the optimal addition timing window based on the addition control decision and process status dynamic data, and generates active addition control instructions; the reaction raw materials and functional additives are added according to the active addition control instructions.

[0016] Preferably, the process production data includes solid concentration, reactant concentration, target product concentration, additive concentration, gas solubility and pH value;

[0017] The control parameter identification model includes: a feature extraction layer, which preprocesses and quantizes the process production data and extracts time domain feature parameters and frequency domain feature parameters; a weight calculation layer, which uses an attention mechanism to evaluate the importance of time domain feature parameters and frequency domain feature parameters and calculates the influence weight coefficient of each parameter; a parameter screening layer, which obtains key control parameters based on weight thresholds and correlation analysis; and a dynamic adjustment layer, which updates the influence weight coefficient based on historical key control parameter feedback.

[0018] Preferably, the six-dimensional gradient vector acquisition process includes: performing continuous time series sampling on parameters in the process production data, calculating the rate of change of each parameter in the continuous time window, obtaining the temperature change gradient, pressure change gradient, flow change gradient, concentration change gradient, pH change gradient and dissolved oxygen change gradient, and forming a six-dimensional gradient vector;

[0019] The multidimensional gradient vector field model includes: a vector decomposition layer, which decomposes the six-dimensional gradient vector and extracts gradient components, including heat transfer state component, fluid dynamic state component, material transportation state component, chemical reaction state component, ion balance state component and redox state component; a field strength calculation layer, which calculates the corresponding field strength distribution based on the gradient components to obtain multi-dimensional field strength, including temperature field strength, pressure field strength, flow field strength, concentration field strength, pH field strength and dissolved oxygen field strength; a state fusion layer, which uses tensor operations to perform weighted fusion of multi-dimensional field strengths to obtain fused field strength characteristics; and a prediction output layer, which predicts process state dynamic data based on the fused field strength characteristics, including stability assessment, trend prediction and abnormality warning.

[0020] Preferably, the parameter importance weight acquisition process includes:

[0021] Obtain historical process production data, calculate the contribution of each parameter to quality, and assign parameter importance weights based on the contribution. The temporal decay weight acquisition process includes: setting a time decay strategy and assigning weights based on time distance;

[0022] The control gradient index acquisition process includes: weighted calculation, normalizing the parameter importance weight and the time series attenuation weight, combining the real-time gradient value, and obtaining the weighted gradient control component; multi-dimensional fusion operation, analyzing the weighted gradient control component, and obtaining the control gradient value; index standardization, using the hyperbolic tangent function to process the control gradient value and generate a standardized control gradient index.

[0023] Preferably, the generation process of adding a control decision includes:

[0024] Type mapping judgment, establishing a correspondence between the control gradient index and the additive type, selecting a buffer additive when the index value is less than a first threshold, selecting a nutrient additive when the index value is greater than or equal to the first threshold and less than a second threshold, selecting a regulating additive when the index value is greater than or equal to the second threshold and less than a third threshold, selecting a catalytic additive when the index value is greater than or equal to the third threshold and less than a fourth threshold, and selecting an inhibitory additive when the index value is greater than or equal to the fourth threshold;

[0025] Precise dosage calculation, based on the control gradient index, combined with the proportional integral differential control algorithm to optimize the dosage of various additives and generate additive control decisions, including additive type code, precise dosage value, execution priority sorting and estimated cost information.

[0026] Preferably, the optimal adding opportunity window identification process includes:

[0027] Calculate the stability index based on the dynamic data of process status and analyze the fluctuation degree of each parameter to determine the stability level;

[0028] Predict the dosing time window based on the action mechanism and response characteristics of different additives;

[0029] Establish a time conflict detection mechanism between multiple additives and obtain conflict detection results; when the expected time interval between the addition of two and / or more additives is less than the preset time, it is determined to be a time conflict, and a priority queue management mechanism is used to resolve the conflict according to the priority order in the addition control decision, with high-priority additives being executed first and low-priority additives being delayed;

[0030] Taking into account the stability index, stability level, dosing time window and conflict detection results, a dynamic programming algorithm is used to obtain the optimal addition time window for each additive.

[0031] Preferably, the dosing control program process includes:

[0032] Parse the active addition control instructions to extract the dosing parameters and execution sequence; dispatch the corresponding metering pumps, valves and piping systems according to the dosing requirements; use closed-loop control to monitor the dosing flow and cumulative amount in real time to ensure dosing accuracy; continuously monitor changes in process parameters during the dosing process and adjust the dosing rate and method in a timely manner; reset the system and confirm the status after the dosing is completed, and record the dosing execution log.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] 1. The present invention constructs a multi-dimensional gradient vector field model, comprehensively analyzes the gradient trends of the six-dimensional parameters (temperature, pressure, flow, concentration, pH and dissolved oxygen), extracts various state information such as heat transfer, fluid dynamics, chemical reaction, etc. of the process, and thus fully perceives the dynamic evolution of the process state. By predicting process stability, trend changes and abnormal fluctuations, real-time perception and trend prediction of key process states are achieved. By integrating tensor calculation and state field strength modeling, not only the ability to understand complex biological reaction processes is improved, but also the timeliness and accuracy of control decisions are significantly improved, so that the entire high-activity selenium yeast production process is transformed from "static control" to "dynamic intelligent regulation", thereby optimizing reaction efficiency and reducing fluctuation risks while ensuring product quality stability.

[0035] 2. The present invention introduces a dual fusion based on "parameter importance weight" and "time series decay weight" to perform weighted processing on the six-dimensional gradient information monitored in real time to form a control gradient index. This control gradient index not only reflects the actual impact of each parameter on the process, but also takes into account the decay trend of the parameter in the time dimension, thereby realizing a scientific evaluation and regulation basis for parameter changes. On this basis, the system can intelligently match the additive type and accurately calculate the additive dosage according to the standardized control gradient index, realizing closed-loop control of the entire process from "quantitative perception" to "precise execution", significantly improving material utilization efficiency and product consistency, and providing strong control support for high-standard yeast production.

[0036] 3. The present invention constructs a dynamic programming algorithm based on stability index and time window prediction, and integrates the mechanism of action and response characteristics of additives to dynamically identify the optimal addition time window. At the same time, a multi-additive time conflict detection mechanism is established, and a priority queue management strategy is adopted to effectively avoid addition conflicts within a short period of time. This approach not only ensures the maximum effect of various additives, but also avoids resource waste and product quality degradation caused by addition order conflicts, thereby improving the synergy of the overall production process. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1A schematic diagram of the structure of a high-activity selenium yeast intelligent production control system based on gradient feeding monitoring provided by the present invention;

[0038] Figure 2 Schematic diagram of the intelligent production control process of high-activity selenium yeast based on gradient feeding monitoring provided by the present invention;

[0039] Figure 3 A schematic diagram of the intelligent production control process provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0041] Example 1:

[0042] The present invention provides a high-activity selenium yeast intelligent production control system based on gradient feeding monitoring. The technical solution is as follows. Figure 1 and Figure 2 :

[0043] Data acquisition module, used to collect process production data in real time;

[0044] The control parameter acquisition module extracts key control parameters of process production data through the control parameter identification model;

[0045] The gradient calculation module is used to obtain the real-time gradient values ​​of various parameters in the process production data and obtain the six-dimensional gradient vector; a multi-dimensional gradient vector field model is constructed to analyze the six-dimensional gradient vector and obtain the dynamic data of the process status;

[0046] The intelligent control module performs weighted fusion of the six-dimensional gradient vector based on the parameter importance weights and time-series attenuation weights of key control parameters to generate a control gradient index. The module also determines the type and amount of additives based on the control gradient index and generates an additive control decision.

[0047] The feeding control module identifies the optimal addition timing window based on the addition control decision and process status dynamic data, and generates active addition control instructions; the reaction raw materials and functional additives are added according to the active addition control instructions.

[0048] Furthermore, the process production data includes solid concentration, reactant concentration, target product concentration, additive concentration, gas solubility and pH value;

[0049] The control parameter identification model includes: a feature extraction layer, which preprocesses and quantizes the process production data and extracts time domain feature parameters and frequency domain feature parameters; a weight calculation layer, which uses an attention mechanism to evaluate the importance of time domain feature parameters and frequency domain feature parameters and calculates the influence weight coefficient of each parameter; a parameter screening layer, which obtains key control parameters based on weight thresholds and correlation analysis; and a dynamic adjustment layer, which updates the influence weight coefficient based on historical key control parameter feedback.

[0050] In this embodiment, the present invention comprehensively reflects the dynamic change state in the production process of high-activity selenium yeast by introducing multi-dimensional process production data. By constructing a control parameter identification model including a feature extraction layer, a weight calculation layer, a parameter screening layer and a dynamic adjustment layer, deep mining and feature extraction of production data are achieved. Through adaptive adjustment capabilities, parameter weights and screening strategies can be optimized based on historical feedback information, thereby improving the entire control system's perception and response speed to changes in production status, ensuring that feeding decisions are more accurate and timely, effectively improving the yield and quality of high-activity selenium yeast, reducing resource waste and production fluctuation risks, and having significant industrial application value and intelligent control advantages.

[0051] The six-dimensional gradient vector acquisition process includes: collecting six core process parameters in the reactor, namely temperature, pressure, feed flow rate, substance concentration, pH value and dissolved oxygen concentration, in real time through a sensor array, sampling each parameter in a continuous time series at a sampling frequency of ten times per second, calculating the rate of change of each parameter in a continuous time window, forming a temperature change gradient, a pressure change gradient, a flow change gradient, a concentration change gradient, a pH change gradient and a dissolved oxygen change gradient, and combining them to form a six-dimensional gradient vector;

[0052] The multidimensional gradient vector field model includes: a vector decomposition layer, which decomposes the six-dimensional gradient vector and extracts the temperature gradient component to reflect the heat transfer state, the pressure gradient component to reflect the fluid dynamic state, the flow gradient component to reflect the material transportation state, the concentration gradient component to reflect the chemical reaction state, the pH gradient component to reflect the ion balance state, and the dissolved oxygen gradient component to reflect the redox state; a field strength calculation layer, which calculates the corresponding field strength distribution based on each gradient component, wherein the temperature field strength is obtained by multiplying the temperature gradient by the heat conduction coefficient, the pressure field strength is obtained by multiplying the pressure gradient by the flow resistance coefficient, and the flow field strength is obtained by multiplying the flow The six field strength components are weighted and fused by tensor operations. The weight distribution is determined according to the degree of influence of each parameter on the process, and the comprehensive dynamic change characteristics of the process state are extracted. The prediction output layer predicts the process state at the next moment through the state space model based on the fused field strength and historical state transition mode, and generates stability assessment, trend prediction and abnormal warning.

[0053] In this embodiment, by constructing a six-dimensional gradient vector and introducing a multidimensional gradient vector field model, high-frequency, continuous dynamic monitoring of changes in multiple key process parameters during the production process of high-activity selenium yeast is achieved, and the changing trends of core variables such as temperature, pressure, flow, concentration, pH, and dissolved oxygen in the production process can be accurately captured. By functionally decomposing the six-dimensional gradient vector according to process attributes, not only is the physical analysis capability of the significance of each parameter change improved, but also a systematic description of the heat transfer state, fluid dynamics state, material transport state, chemical reaction state, ion balance state, and redox state is achieved. The field strength calculation layer further converts each gradient component into a corresponding process field strength value, and combines tensor weighted fusion technology to extract comprehensive dynamic change characteristics that reflect the overall picture of the process operation. Finally, by predicting the state space model in the output layer, the evolution trend of the process state can be perceived in advance, and stability assessment, trend prediction, and abnormal warning information can be output, thereby providing strong data support for intelligent control and precise feeding. This deep dynamic perception and prediction capability effectively improves the controllability and stability of the process, enhances the system's adaptability to complex and changing working conditions, and reduces the frequency of manual intervention and abnormal response delays.

[0054] Furthermore, the parameter importance weight acquisition process includes:

[0055] Obtain historical process production data, collect historical process data to establish a parameter effect association database, use the principal component analysis method to calculate the contribution of each parameter to the final product quality, and assign importance weights according to the contribution; for example, the temperature parameter weight accounts for 25%, the pressure parameter weight accounts for 15%, the flow parameter weight accounts for 20%, the concentration parameter weight accounts for 25%, the pH parameter weight accounts for 10%, and the dissolved oxygen parameter weight accounts for 5%; the time series decay weight acquisition process includes: setting a time decay strategy, and assigning weights to historical data according to time distance; for example, the current moment weight is full, the weight of the previous minute data is 95% of the full score, the weight of the previous five minutes data is 80% of the full score, the weight of the previous ten minutes data is 60% of the full score, and the weight of the previous thirty minutes data is 30% of the full score, to ensure that recent data plays a dominant role in control decisions;

[0056] The control gradient index acquisition process includes:

[0057] Weight normalization processing: multiply the parameter importance weight and the time series attenuation weight to obtain the comprehensive weight of each parameter at different times;

[0058] Gradient value weighted calculation, multiplying the real-time gradient value of each parameter with the corresponding comprehensive weight to obtain six weighted gradient components;

[0059] Multi-dimensional fusion operation, which linearly combines and sums the six weighted gradient components to form a single control gradient value;

[0060] Indicator standardization processing uses the hyperbolic tangent function to map the control gradient value to a standard range of zero to one, generating a standardized control gradient indicator for subsequent decision-making.

[0061] In this embodiment, by introducing a dual mechanism of parameter importance weights and time-series decay weights, accurate dynamic weighted processing of process control data is achieved. By constructing a parameter effect association database between historical process data and product quality, and combining it with the principal component analysis method, the actual contribution of each core parameter to the final product quality can be scientifically quantified, thereby allocating parameter importance weights with basis, making the control strategy more targeted and scientific. At the same time, the time-series decay mechanism effectively reflects the importance of data timeliness, ensuring that current and recent operating condition changes play a dominant role in control judgment, significantly improving the timeliness and sensitivity of control response. On this basis, by weighted fusion of real-time gradient values ​​and comprehensive weights, and adopting multi-dimensional linear combination, a unified control gradient index is formed, and then mapped to a unified evaluation interval through standardization processing, not only the stability and comparability of the gradient index are enhanced, but also a clear and intuitive reference basis for subsequent intelligent control and feeding decisions is provided. This method overall improves the system's comprehensive perception and intelligent judgment capabilities of complex process states, helping to achieve higher quality, higher efficiency, and lower volatility intelligent production control.

[0062] Furthermore, the generation process of the added control decision includes:

[0063] Type mapping judgment, establishing a correspondence between the control gradient index and the additive type, selecting a buffer additive when the index value is less than a first threshold, selecting a nutrient additive when the index value is greater than or equal to the first threshold and less than a second threshold, selecting a regulating additive when the index value is greater than or equal to the second threshold and less than a third threshold, selecting a catalytic additive when the index value is greater than or equal to the third threshold and less than a fourth threshold, and selecting an inhibitory additive when the index value is greater than or equal to the fourth threshold;

[0064] Precise dosage calculation: The dosage of various additives is calculated based on the proportional-integral-differential control algorithm. The proportional control part determines the basic dosage according to the current deviation, the integral control part makes compensation adjustments based on the accumulated deviation, and the differential control part makes predictive adjustments based on the deviation change trend. The total dosage is obtained by weighted summation of the three parts.

[0065] Constraint optimization: set upper and lower limits for the amount of addition to avoid excessive addition, set addition rate constraints to prevent system shock, set cost constraints to control economic efficiency, and use multi-objective optimization algorithms to find the optimal addition plan while meeting all constraints.

[0066] Decision information output generates a complete additive control decision including additive type code, accurate dosage value, execution priority ranking and estimated cost information.

[0067] In the present embodiment, by accurately parsing and mapping the control gradient index, a corresponding relationship between the index interval and different types of additives is established, and intelligent identification and automatic classification of additive types are realized, effectively improving the pertinence and scientificity of the feeding decision. By introducing the proportional integral differential (PID) control algorithm to dynamically calculate the dosage, it is not only possible to respond to the current process state deviation in a timely manner, but also to take into account historical trends and future change trends, thereby achieving accurate control of the additive dosage. At the same time, the system sets multiple constraints including upper and lower limits of the dosage, the dosage acceleration rate and the cost, and adopts a multi-objective optimization algorithm to ensure that the feeding scheme achieves the best balance between process stability, economy and execution feasibility, effectively preventing the system fluctuations and resource waste caused by excessive or too fast addition. The final generated addition control decision contains multidimensional information such as type code, specific dosage, execution priority and cost estimation, which facilitates the system to respond quickly and automatically execute, greatly improving the decision-making efficiency, dosage accuracy and overall economic benefits of the intelligent control system, and ensuring the stability and high-quality output of the high-activity selenium yeast production process.

[0068] Furthermore, the optimal adding time window identification process includes:

[0069] The stability index is calculated based on the dynamic data of the process status, and the stability level is determined by analyzing the fluctuation degree of each parameter. The stability level of the system is determined by analyzing the fluctuation degree of each parameter. When the stability index is greater than 0.8, it is determined to be a high stability state and suitable for immediate dosing. When the index is between 0.6 and 0.8, it is determined to be a medium stability state and requires cautious dosing. When the index is less than 0.6, it is determined to be a low stability state and dosing should be postponed.

[0070] Predict the dosing time window based on the action mechanism and response characteristics of different additives; for example, a 30-minute prediction window is set for buffering additives to take into account their sustained stabilization effect, a 60-minute prediction window is set for nutrient additives to take into account their metabolic absorption process, a 15-minute prediction window is set for regulating additives to take into account their rapid regulation effect, a 45-minute prediction window is set for catalytic additives to take into account their reaction promotion cycle, and a 20-minute prediction window is set for inhibitory additives to take into account their inhibitory response time;

[0071] Establish a time conflict detection mechanism between multiple additives and obtain conflict detection results; when the expected time interval between the addition of two and / or more additives is less than the preset time, it is determined to be a time conflict, and a priority queue management mechanism is used to resolve the conflict according to the priority order in the addition control decision, with high-priority additives being executed first and low-priority additives being delayed;

[0072] Taking into account the stability index, stability level, dosing time window and conflict detection results, a dynamic programming algorithm is used to obtain the optimal addition time window for each additive.

[0073] In this embodiment, by introducing stability index calculation and multi-dimensional process parameter fluctuation analysis, the stability level of the current state of the system can be dynamically evaluated, thereby providing a scientific risk assessment basis for the addition of additives. The action mechanism and response characteristics of different types of additives are fully considered. By setting differentiated prediction time windows, the use of each type of additive is more in line with its actual effect cycle, which improves the accuracy and timeliness of the feeding strategy. At the same time, the system constructs an additive time conflict detection and priority management mechanism to avoid interference or resource competition between different types of additives, and effectively ensures the continuity and coordination of the process system operation. Finally, through the dynamic programming algorithm, multiple factors such as stability level, prediction window and time conflict are comprehensively considered to achieve intelligent identification and reasonable arrangement of the best time to add various additives, significantly improving the automation level and execution reliability of the additive control, and providing strong support for the continuous optimization and quality assurance of the high-activity selenium yeast production process.

[0074] Furthermore, the dosing control program process includes:

[0075] Parse the active addition control instructions to extract the dosing parameters and execution sequence; dispatch the corresponding metering pumps, valves and piping systems according to the dosing requirements; use closed-loop control to monitor the dosing flow and cumulative amount in real time to ensure dosing accuracy; continuously monitor changes in process parameters during the dosing process and adjust the dosing rate and method in a timely manner; reset the system and confirm the status after the dosing is completed, and record the dosing execution log.

[0076] Specifically, the active addition control instruction is structured analyzed, the additive type identification code, target addition amount value, setting addition acceleration rate parameters and execution time window range are extracted, the instruction integrity and parameter rationality are verified, and a standardized execution parameter set is generated; according to the additive type identification code, the equipment configuration database is queried, the corresponding metering pump number, control valve position and delivery pipeline path are automatically selected, a complete addition channel connection is established, the equipment self-test program is executed to confirm that the equipment status is normal, and the pipeline is pre-flushed and the equipment is preheated; the closed-loop feedback control system is started to achieve precise addition, the target flow value of the flow controller is set, the actual flow data in the pipeline is monitored in real time, the deviation between the target value and the actual value is calculated, and the speed of the metering pump is adjusted through the control algorithm to achieve Flow rate is tracked and the added amount is calculated cumulatively. When the cumulative amount reaches 98% of the target addition amount, the addition is stopped to ensure accuracy. During the entire addition process, the real-time changes of the six-dimensional process parameters in the reactor are monitored once per second, and the instantaneous change rate of each parameter is calculated. When the change rate of any parameter exceeds the preset safety threshold, an abnormal alarm is immediately triggered, and the adaptive control strategy is started to dynamically adjust the addition acceleration rate to reduce the impact on the system. After the addition task is completed, a standardized equipment reset operation sequence is executed, the relevant control valves are closed to stop the material flow, the metering pump is stopped to prevent further transportation, and a detailed execution log is recorded including timestamp, additive code, actual addition amount, execution time and completion status, and the equipment operation record and maintenance plan are updated.

[0077] In this embodiment, structured parsing for proactively adding control instructions enables standardized extraction of additive task parameters and rational scheduling of execution sequences, significantly improving the clarity of control logic and execution consistency. Combined with the automatic matching function of the equipment configuration database, the system intelligently selects and links metering pumps, valves, and delivery pipelines, automatically constructing a complete dosing channel and performing pre-processing preparations, reducing human intervention and improving automation. A closed-loop feedback control mechanism precisely adjusts the dosing flow rate and cumulative amount in real time, effectively ensuring the accuracy of the additive dosage and proactively stopping dosing when the target value is approached to prevent overdosing. Simultaneously, the system monitors the dynamic changes of multidimensional process parameters within the reactor at high frequency and in real time, enabling timely identification of abnormal fluctuations and initiating adaptive control strategies to dynamically adjust the dosing rate and reduce the impact of the addition operation on system stability. A standardized reset process and detailed execution log after task completion not only facilitate process traceability and operational audits, but also provide a decision-making basis for subsequent equipment maintenance and operational optimization, significantly improving the accuracy, safety, and intelligence of dosing control.

[0078] The present invention provides a high-activity selenium yeast intelligent production control system based on gradient feeding monitoring, specifically referring to Figure 3Through multi-dimensional data acquisition, deep feature extraction, and dynamic perception mechanisms, the system achieves precise dynamic control and intelligent management of the production process. The system collects real-time process data, including solid concentration, reactant concentration, target product concentration, additive concentration, gas solubility, and pH. Using a control parameter identification model, it extracts key parameters and their weights. Combining a six-dimensional gradient vector and a multi-dimensional gradient vector field model, it performs high-frequency dynamic monitoring and comprehensive prediction of process status. This forward-looking perception capability not only identifies changing trends in the production process but also generates stability assessments and anomaly warnings, providing a reliable basis for intelligent feeding. The intelligent control module integrates parameter importance with time-decrease weights to generate scientifically sound control gradient indicators, accurately determine additive type and dosage, and optimize the optimal addition timing window, ensuring efficient and targeted feeding decisions. Furthermore, closed-loop feedback and adaptive adjustment mechanisms further enhance feeding accuracy and system stability. Overall, the system significantly improves the yield and quality of high-activity selenium yeast, reduces resource waste and the risk of production fluctuations, demonstrating significant industrial application value and technological potential, and providing an innovative solution for intelligent production.

[0079] Example 2:

[0080] See also Figure 2 , the present invention provides a specific application embodiment of an intelligent production control system for high-activity selenium yeast based on gradient feeding monitoring in an industrial production environment. This embodiment takes the production line of a biopharmaceutical company with an annual output of 500 tons of high-activity selenium yeast as the application scenario, and elaborates on the collaborative working mechanism of each technical module. The specific scheme is: real-time collection of process production data; extraction of key control parameters of process production data through a control parameter identification model; acquisition of real-time gradient values ​​of each parameter in the process production data, and acquisition of a six-dimensional gradient vector; construction of a multi-dimensional gradient vector field model to analyze the six-dimensional gradient vector and obtain dynamic data of the process state; weighted fusion of the six-dimensional gradient vector based on the parameter importance weight and time series attenuation weight of the key control parameters to generate a control gradient index; determination of the additive type and addition amount based on the control gradient index, and generation of an addition control decision; identification of the optimal addition time window based on the addition control decision and the dynamic data of the process state, and generation of an active addition control instruction; addition of reaction raw materials and functional additives based on the active addition control instruction.

[0081] The 10-cubic-meter reactor in this high-activity selenium yeast production line is fully equipped with automated control equipment. A data acquisition module uses a distributed sensor network to monitor changes in key process parameters during the fermentation process in real time. The entire production cycle lasts 72 hours and is divided into three phases: the culture phase (0-24 hours), the rapid proliferation phase (24-48 hours), and the selenium enrichment phase (48-72 hours). Each phase has significantly different process parameter control requirements.

[0082] The data acquisition module uses a high-precision sensor array, including temperature sensors, pressure sensors, flow sensors, an online concentration detector, a pH meter, and a dissolved oxygen analyzer, to synchronously collect process data at a rate of 10 times per second. In this application scenario, process data specifically includes yeast cell concentration, glucose concentration, selenomethionine concentration, nutrient additive concentration, oxygen solubility, and pH value. This data forms a continuous time series data stream, providing the data foundation for subsequent intelligent analysis.

[0083] The control parameter acquisition module constructs a four-layer control parameter identification model, specifically optimized for the complex characteristics of the high-activity selenium yeast fermentation process. The feature extraction layer uses a hybrid frequency domain analysis method combining wavelet transform and fast Fourier transform to preprocess six types of process production data. It extracts time-domain feature parameters, including statistical features such as mean, variance, skewness, and kurtosis, as well as frequency-domain feature parameters, including dynamic features such as the dominant frequency component, spectral energy distribution, and frequency-domain correlation, generating a total of 48-dimensional feature vectors.

[0084] The weight calculation layer introduces a self-attention mechanism, evaluating the importance of the 48-dimensional feature vector through a multi-head attention network. Specifically, eight attention heads are set up, each focusing on feature extraction at a different time scale and frequency range. The influence weight coefficient of each parameter is calculated using the attention weight matrix. During the culture period, the temperature and pH parameters have higher weights (0.35 and 0.28, respectively); during the rapid proliferation period, the weights of dissolved oxygen and glucose concentrations are dominant (0.42 and 0.31, respectively); during the selenium enrichment period, the weights of selenomethionine concentration and nutrient salt concentration are the most critical (0.48 and 0.25, respectively).

[0085] The parameter screening layer dynamically selects a set of key control parameters for each production stage based on a set weight threshold (0.15) and Pearson correlation analysis (threshold 0.7). The dynamic adjustment layer updates parameter weights using an exponential moving average algorithm based on key control parameter feedback from 30 historical batches. The learning rate is set to 0.1, enabling the system to achieve adaptive optimization capabilities.

[0086] The gradient calculation module implements groundbreaking multi-scale gradient vector field intelligent perception technology. This technology utilizes a sliding sampling mechanism with a 30-second time window to calculate the real-time rate of change of each process parameter, generating a six-dimensional gradient vector. Compared to traditional single-point monitoring methods, this technology can capture subtle trends in parameter changes, improving prediction accuracy by 35%.

[0087] The vector decomposition layer of the multidimensional gradient vector field model uses a combination of principal component analysis and independent component analysis to intelligently decompose the six-dimensional gradient vector according to bioprocess characteristics. The heat transfer state component reflects the balance between metabolic heat generation and heat dissipation during fermentation, the fluid dynamics state component reflects the effects of agitation and aeration, the material transport state component characterizes the mass transfer of nutrients, the chemical reaction state component indicates the growth and metabolic state of the yeast, and the redox state component reflects cellular respiration and selenium conversion.

[0088] The field intensity calculation layer incorporates a bioprocess physics model to calculate the corresponding field intensity distribution based on each gradient component. The temperature field intensity is calculated using a heat conduction model modified by the Arrhenius equation, accounting for the temperature sensitivity of the biological reaction. The flow field intensity is combined with a rheological model to describe the non-Newtonian fluid properties of the fermentation broth. The concentration field intensity utilizes Fick's diffusion law combined with the Monod kinetic model to accurately reflect the mass transfer and consumption of nutrients. The dissolved oxygen field intensity is calculated using a dynamic model of the kLa mass transfer coefficient to assess oxygen mass transfer efficiency in real time.

[0089] The state fusion layer uses tensor decomposition technology to perform an adaptive weighted fusion of the six field strength components, with the weight coefficients automatically adjusted based on the current fermentation stage and historical data. The prediction output layer builds a state prediction model based on a long short-term memory (LSTM) network, capable of predicting process state changes 15-30 minutes in advance, generating a stability assessment index (ranging from 0 to 1), a trend prediction vector, and anomaly warning levels (normal / caution / warning / dangerous).

[0090] The intelligent control module implements an innovative multi-weight fusion adaptive control algorithm that simultaneously considers parameter importance weights and time series decay weights to achieve scientific weighted fusion of real-time gradient values.

[0091] To determine parameter importance weights, the system collected process data from 500 historical batches and established a database linking 18 process parameters to product quality indicators (selenium content, protein content, and activity). A random forest regression model was used to calculate the contribution of each parameter to final product quality. The specific weights assigned were: yeast cell concentration (weighted at 0.28), glucose concentration (weighted at 0.22), selenomethionine concentration (weighted at 0.25), pH (weighted at 0.12), dissolved oxygen concentration (weighted at 0.08), and nutrient salt concentration (weighted at 0.05).

[0092] The time series decay weight uses an exponential decay function with a decay coefficient of 0.05 / min. The specific weight distribution is: 1.0 for the current moment, 0.95 for the previous minute, 0.78 for the previous five minutes, 0.61 for the previous ten minutes, and 0.22 for the previous thirty minutes, ensuring that recent data plays a dominant role in control decisions.

[0093] The control gradient index acquisition process adopts a weighted combination algorithm: first, the parameter importance weights and time-series attenuation weights are normalized, and then a weighted product operation is performed with the real-time gradient value to obtain six weighted gradient control components; secondly, a linear combination method with adaptive weights is used to fuse the six components to form a control gradient value; finally, an improved hyperbolic tangent function is used for normalization, mapping the value to the interval [0,1] to generate a standardized control gradient index.

[0094] The additive control decision-making process established a precise five-level mapping relationship: when the control gradient index is less than 0.2, phosphate buffer additives are selected to maintain pH stability; when the index is between [0.2, 0.4), yeast extract nutrient additives are selected to promote bacterial growth; when the index is between [0.4, 0.6), selenomethionine regulating additives are selected to control selenium content; when the index is between [0.6, 0.8), glucose catalytic additives are selected to improve metabolic activity; and when the index is greater than or equal to 0.8, ethanol inhibitory additives are selected to prevent excessive fermentation.

[0095] The PID control algorithm is used for precise dosage calculation, with output dosage ranges of: 0.1-0.5 kg / h for buffering, 0.5-2.0 kg / h for nutrient, 0.2-1.0 kg / h for regulating, 1.0-3.0 kg / h for catalytic, and 0.1-0.8 kg / h for inhibitor. Safety constraints are also set: the maximum single dosage does not exceed 0.5% of the reactor volume, and the dosage acceleration rate does not exceed 100 kg / h.

[0096] The feeding control module implements dynamic timing window intelligent optimization scheduling technology, which can accurately identify the optimal feeding time through multi-dimensional analysis.

[0097] The stability index is calculated using a comprehensive evaluation model, which performs a weighted assessment based on the fluctuation degree of six parameters in the process status dynamic data.

[0098] The dosing time window prediction is differentiated based on the biological action mechanism of different additives. A 45-minute prediction window is set for buffering additives (phosphates) to account for the time it takes to establish pH buffering; a 90-minute prediction window is set for nutritional additives (yeast extract) to account for their metabolic decomposition and absorption and utilization processes; a 60-minute prediction window is set for regulatory additives (selenomethionine) to account for their cellular uptake and conversion cycle; a 30-minute prediction window is set for catalytic additives (glucose) to account for their rapid metabolic characteristics; and a 20-minute prediction window is set for inhibitory additives (ethanol) to account for the rapid response of their inhibitory effects.

[0099] The time conflict detection mechanism establishes a precise conflict matrix, identifying a time conflict when the planned addition intervals for multiple additives are less than 15 minutes. Priority is ranked as follows: inhibitors > buffers > regulators > catalysts > nutrients. Dynamic queue management ensures that high-priority additives are immediately added, while lower-priority additives are automatically delayed by 15 minutes, ensuring maximum additive effectiveness.

[0100] The dynamic programming algorithm uses state transition equations for optimization, with the objective function being a weighted combination of maximizing production efficiency and minimizing cost. The state space encompasses four dimensions: system stability, additive type, dosing time, and dosage. The Bellman optimization principle is used to determine the optimal dosing window for each additive, achieving global optimal control.

[0101] The dosing control program utilizes a distributed execution architecture, parsing active dosing control instructions in real time to extract key parameters such as additive code, target dosing amount, set dosing acceleration rate, and execution priority. The system automatically dispatches the corresponding peristaltic pump, electric control valve, and stainless steel piping system to establish a complete dosing channel.

[0102] The closed-loop control system utilizes mass flow controllers for precise dosing, maintaining a ±2% deviation between setpoint and actual value. Process parameters are monitored at 1Hz during dosing. Adaptive control is initiated when the rate of change of any parameter exceeds a preset threshold, dynamically adjusting the dosing rate. After dosing, a standardized reset procedure is executed, maintaining a detailed log with timestamps, additive type, actual dosage, execution duration, and system response.

[0103] Through systematic technological innovation, the present invention realizes intelligent, precise and efficient control of the production process of high-activity selenium yeast. Multidimensional gradient sensing technology significantly improves the system's understanding and prediction capabilities of complex biological processes, making the control strategy more scientific and reasonable. The multi-weight fusion algorithm ensures the accuracy and timeliness of control decisions, and effectively responds to various changes and interferences in the production process. Dynamic scheduling optimization technology realizes the coordinated management of multiple additives, maximizing the effects of various additives. Overall, the system not only significantly improves product quality and production efficiency, but also greatly reduces production costs and resource consumption, providing advanced technical solutions for the industrial production of high-activity selenium yeast. Please refer to Table 1 for details. It has important industrial application value and promotion prospects.

[0104] Table 1 Performance comparison between the intelligent control system of the present invention and traditional control technology

[0105]

[0106] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A high-activity selenium yeast intelligent production control system based on gradient feeding monitoring, characterized in that: include: Data acquisition module, used to collect process production data in real time; The control parameter acquisition module extracts key control parameters of process production data through the control parameter identification model; Gradient calculation module, used to obtain the real-time gradient value of each parameter in the process production data and obtain the six-dimensional gradient vector; Construct a multi-dimensional gradient vector field model to analyze the six-dimensional gradient vector and obtain dynamic data of process status; The intelligent control module performs weighted fusion of the six-dimensional gradient vector based on the parameter importance weights and time-series attenuation weights of key control parameters to generate a control gradient index. The module also determines the type and amount of additives based on the control gradient index and generates an additive control decision. The feeding control module identifies the optimal feeding window based on the feeding control decision and process status dynamic data, and generates active feeding control instructions; The reaction raw materials and functional additives are added according to the active addition control instructions.

2. The intelligent production control system for high-activity selenium yeast based on gradient feeding monitoring according to claim 1, characterized in that: The process production data include solid concentration, reactant concentration, target product concentration, additive concentration, gas solubility and pH value; The control parameter identification model It includes: feature extraction layer, which pre-processes and quantizes the process production data, and extracts time domain feature parameters and frequency domain feature parameters; The weight calculation layer uses the attention mechanism to evaluate the importance of time domain feature parameters and frequency domain feature parameters, and calculates the influence weight coefficient of each parameter; the parameter screening layer obtains key control parameters based on weight thresholds and correlation analysis; the dynamic adjustment layer updates the influence weight coefficient based on historical key control parameter feedback.

3. The intelligent production control system for high-activity selenium yeast based on gradient feeding monitoring according to claim 1, characterized in that: The six-dimensional gradient vector acquisition process includes: Continuous time series sampling of parameters in process production data is performed, and the rate of change of each parameter in the continuous time window is calculated to obtain the temperature change gradient, pressure change gradient, flow change gradient, concentration change gradient, pH change gradient and dissolved oxygen change gradient, forming a six-dimensional gradient vector; The multidimensional gradient vector field model includes: a vector decomposition layer, which decomposes the six-dimensional gradient vector and extracts gradient components, including heat transfer state component, fluid dynamic state component, material transportation state component, chemical reaction state component, ion balance state component and redox state component; a field strength calculation layer, which calculates the corresponding field strength distribution based on the gradient components to obtain multi-dimensional field strength, including temperature field strength, pressure field strength, flow field strength, concentration field strength, pH field strength and dissolved oxygen field strength; a state fusion layer, which uses tensor operations to perform weighted fusion of multi-dimensional field strengths to obtain fused field strength characteristics; and a prediction output layer, which predicts process state dynamic data based on the fused field strength characteristics, including stability assessment, trend prediction and abnormality warning.

4. The intelligent production control system for high-activity selenium yeast based on gradient feeding monitoring according to claim 1, characterized in that: The parameter importance weight acquisition process includes: Obtain historical process production data, calculate the contribution of each parameter to quality, and assign parameter importance weights based on the contribution. The temporal decay weight acquisition process includes: setting a time decay strategy and assigning weights based on time distance; The control gradient index acquisition process includes: weighted calculation, normalizing the parameter importance weight and the time series attenuation weight, combining the real-time gradient value, and obtaining the weighted gradient control component; multi-dimensional fusion operation, analyzing the weighted gradient control component, and obtaining the control gradient value; index standardization, using the hyperbolic tangent function to process the control gradient value and generate a standardized control gradient index.

5. The intelligent production control system for high-activity selenium yeast based on gradient feeding monitoring according to claim 1, characterized in that: The generation process of adding a control decision includes: Type mapping judgment, establishing a correspondence between the control gradient index and the additive type, selecting a buffer additive when the index value is less than a first threshold, selecting a nutrient additive when the index value is greater than or equal to the first threshold and less than a second threshold, selecting a regulating additive when the index value is greater than or equal to the second threshold and less than a third threshold, selecting a catalytic additive when the index value is greater than or equal to the third threshold and less than a fourth threshold, and selecting an inhibitory additive when the index value is greater than or equal to the fourth threshold; Precise dosage calculation, based on the control gradient index, combined with the proportional integral differential control algorithm to optimize the dosage of various additives and generate additive control decisions, including additive type code, precise dosage value, execution priority sorting and estimated cost information.

6. The intelligent production control system for high-activity selenium yeast based on gradient feeding monitoring according to claim 1, characterized in that: The optimal adding time window identification process includes: Calculate the stability index based on the dynamic data of process status and analyze the fluctuation degree of each parameter to determine the stability level; Predict the dosing time window based on the action mechanism and response characteristics of different additives; Establish a time conflict detection mechanism between multiple additives and obtain conflict detection results; when the expected time interval between the addition of two and / or more additives is less than the preset time, it is determined to be a time conflict, and a priority queue management mechanism is used to resolve the conflict according to the priority order in the addition control decision, with high-priority additives being executed first and low-priority additives being delayed; Taking into account the stability index, stability level, dosing time window and conflict detection results, a dynamic programming algorithm is used to obtain the optimal addition time window for each additive.

7. The intelligent production control system for high-activity selenium yeast based on gradient feeding monitoring according to claim 1, characterized in that: The dosing process comprises: Parse the active addition control instructions to extract the dosing parameters and execution sequence; dispatch the corresponding metering pumps, valves and piping systems according to the dosing requirements; use closed-loop control to monitor the dosing flow and cumulative amount in real time to ensure dosing accuracy; continuously monitor changes in process parameters during the dosing process and adjust the dosing rate and method in a timely manner; reset the system and confirm the status after the dosing is completed, and record the dosing execution log.

Images