Peptide milk powder production line monitoring method and system based on multi-source data fusion

Abstract

The invention relates to the technical field of data processing, in particular to a peptide milk powder production line monitoring method and system based on multi-source data fusion. The method comprises the following steps: acquiring real-time working condition data and off-line detection data of a peptide milk powder production line; calculating a relative error of the offline data and the extended Kalman filtering output, and mapping the relative error into a state drift divergence; calculating a blind area heat cumulative gradient based on the raw milk density, the real-time temperature and the pipeline flow velocity; constructing a time confidence coefficient smooth surface in combination with fluid residence duration to generate a backtracking migration gain coefficient; and extracting a historical deviation vector, and calculating a dynamic compensation item by using a gain coefficient to correct a filtering posteriori state updating equation so as to output a corrected real-time state of the peptide milk powder production line. The method can more accurately evaluate and predict the deviation, scientifically analyze the heat exchange accumulative transfer effect, ensure time-space reasonable attenuation of the historical deviation, improve the monitoring accuracy and real-time performance, effectively balance the sterilization effect and peptide activity protection, and reduce the production loss.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for monitoring peptide milk powder production lines based on multi-source data fusion. Background Technology

[0002] As a functional dairy product rich in high-value bioactive peptides, peptide milk powder's production process covers multiple continuous fluid dynamics and thermodynamics processing steps, including milk storage, milk purification, sterilization, drying, and packaging. In the core sterilization process, it is necessary to simultaneously meet the dual requirements of killing pathogenic microorganisms and preserving the activity of heat-sensitive peptides to the greatest extent.

[0003] Currently, for the condition monitoring of peptide milk powder production lines, the extended Kalman filter algorithm is usually introduced for online dynamic state optimization estimation. The extended Kalman filter algorithm uses the thermodynamic mechanism model of peptide degradation and microbial lethality as the state transition equation, and combines the real-time observation data of the single-point temperature sensor of the sterilizer to linearize the nonlinear system by performing a first-order Taylor expansion, thereby continuously predicting and correcting the instantaneous state inside the sterilization pipeline.

[0004] However, when traditional extended Kalman filtering algorithms are applied to peptide milk powder production lines, significant physical time lag exists between the microbial survival data obtained from autoclaves and incubators and the peptide activity retention rate results from biochemical tests during the quality inspection stage. Offline detection data typically lags by 24 to 48 hours. Furthermore, since the sterilization process must simultaneously achieve the dual goals of killing microorganisms and preserving peptide activity, traditional algorithms cannot achieve the optimal balance under dynamic fluctuations. This often results in either overheating leading to peptide degradation and inactivation, or underheating causing excessive microbial levels. The traditional algorithm structure lacks time-series backtracking and delay compensation mechanisms, relying solely on continuous, noisy temperature sensor data for real-time correction. It cannot effectively integrate high-precision but time-delayed offline detection data to correct historical operating conditions or iteratively optimize current model prediction biases. This wastes valuable multi-source heterogeneous detection data and, when sterilization temperatures fluctuate dynamically, the traditional algorithm cannot accurately balance the contradiction between overheating causing peptide degradation and inactivation and underheating causing excessive microbial levels, ultimately affecting the quality of the finished peptide milk powder. Summary of the Invention

[0005] To address the problem that traditional extended Kalman filter algorithms cannot effectively integrate offline detection data with significant delays, and cannot accurately balance the contradiction between excessive heating causing peptide degradation and inactivation and insufficient heating causing excessive microbial levels, leading to distorted model predictions and thus affecting the quality of finished peptide milk powder, this invention provides a peptide milk powder production line monitoring method and system based on multi-source data fusion.

[0006] In a first aspect, the present invention provides a method for monitoring a peptide milk powder production line based on multi-source data fusion, employing the following technical solution: A monitoring method for a peptide milk powder production line based on multi-source data fusion includes: acquiring real-time operating condition data and corresponding historical offline detection data of the peptide milk powder production line; the real-time operating condition data includes the real-time temperature of the sterilizer and the instantaneous flow rate of the sterilization pipeline; the offline detection data includes the microbial survival rate and peptide activity retention rate; calculating the relative error between the offline detection data and the state prediction data output by the extended Kalman filter at the same historical moment, and performing nonlinear amplification mapping on the relative error to determine the state drift divergence; acquiring the real-time density of raw milk, and based on the state drift divergence, combined with the real-time temperature of the sterilizer and the instantaneous flow rate of the sterilization pipeline, calculating the non-offline detected data through continuous-time integration. The cumulative transfer effect of fluid heat exchange on historical deviations during the blind zone period covered by the detection data is used to extract the blind zone heat accumulation gradient. The spatial residence time of the fluid is determined based on the instantaneous flow velocity and total length of the sterilization pipeline. A time confidence smoothing surface is constructed by combining the total operating time of the peptide milk powder production line with the blind zone period to impose spatiotemporal attenuation constraints on the blind zone heat accumulation gradient and generate a backtracking migration gain coefficient. The historical true deviation vector of the offline detection data is extracted. The product of the backtracking migration gain coefficient and the historical true deviation vector is used as a dynamic compensation term to correct the posterior state update equation of the extended Kalman filter algorithm, so as to output the corrected real-time state of the peptide milk powder production line.

[0007] This invention assesses the prediction bias of the Extended Kalman Filter (EKF) by calculating state drift divergence, more accurately reflecting the relative error between offline detection data and filter output, and providing a reliable basis for subsequent blind zone thermal accumulation gradient extraction. By extracting the blind zone thermal accumulation gradient, it analyzes the cumulative transfer effect in periods not covered by offline detection data, more accurately reflecting the continuous impact of fluid heat exchange on historical deviations, and providing a key cumulative effect indicator for retrospective correction of historical states. By constructing a time confidence smoothing surface and generating a retrospective migration gain coefficient, it achieves spatiotemporal attenuation constraints on historical deviations, dynamically adjusting based on fluid residence time and blind zone periods, ensuring that the impact of historical deviations on the current state attenuates reasonably with time and spatial distance. Through dynamic compensation terms to correct the EKF posterior state update equation, it achieves effective fusion of multi-source heterogeneous data, improving the accuracy and real-time performance of peptide milk powder production line monitoring, and effectively balancing the contradiction between sterilization effect and peptide activity protection for finished peptide milk powder products.

[0008] Furthermore, the method for obtaining the offline detection data corresponding to the historical moment is as follows: obtain the timestamp of the test report corresponding to the offline detection data issued by the laboratory, subtract the timestamp from the actual production timestamp of the historical production batch of the peptide milk powder production line, and dynamically obtain the lag time of the offline detection data; trace back by using the batch identifier recorded by the manufacturing execution system of the peptide milk powder production line, and extract the state prediction data output by the extended Kalman filter algorithm at the corresponding historical moment within the same production batch, so as to complete the timestamp interpolation alignment of multi-source heterogeneous data. The state prediction data includes microbial survival rate and peptide activity retention rate.

[0009] This invention calculates the lag time of offline detection data by using timestamp differences, achieving accurate assessment of data latency and ensuring the accurate determination of historical moments. This provides a reliable time alignment basis for subsequent state drift divergence calculation. By tracing and extracting state prediction data corresponding to historical moments through batch identifiers, it achieves spatiotemporal alignment of multi-source heterogeneous data, ensuring accurate matching between real-time operating data and offline detection data. This provides a time-consistent data basis for calculating state drift divergence and improves the accuracy of multi-source data fusion.

[0010] Furthermore, the state drift divergence satisfies: In the formula, For the current moment The state drift divergence corresponding to historical moments, For the current moment, This refers to the lag time in offline testing data for historical production batches of peptide milk powder production lines. and Each represents the current time. The microbial survival rate and peptide activity retention rate in the offline detection data corresponding to historical moments. and Each represents the current time. Microbial survival rate and peptide activity retention rate in the extended Kalman filter state prediction data corresponding to historical moments. To prevent extremely small constants with a denominator of 0, It is a natural exponential function.

[0011] This invention achieves a scientific assessment of state drift divergence by constructing an exponential amplification model that incorporates the squared relative error between microbial survival rate and peptide activity retention rate. This model more accurately reflects the overall deviation between offline detection data and filtered prediction data. The exponential function nonlinearly amplifies the relative error, ensuring that even small deviations can be effectively identified, thereby accurately assessing the accuracy of extended Kalman filter predictions.

[0012] Furthermore, the thermal accumulation gradient in the blind zone satisfies: In the formula, For the current moment The thermal accumulation gradient in the blind zone For the current moment The state drift divergence corresponding to historical moments, For the current moment, This refers to the lag time in offline testing data for historical production batches of peptide milk powder production lines. For continuous time integral variables within the blind zone period, This is a function representing the real-time temperature of the sterilizer during the blind zone period. Let be the instantaneous flow rate function of the sterilization pipeline during the blind zone period. This refers to the real-time density of the raw milk. This is a preset heat exchange coefficient used for dimensional normalization and calibration of the cumulative effect of thermal kinetic energy. To prevent extremely small constants with a denominator of 0, It is the natural logarithm function.

[0013] This invention achieves a scientific assessment of the thermal accumulation gradient in the blind zone by constructing a product model that includes state drift divergence and continuous-time integral. It accurately reflects the cumulative transfer effect of fluid heat exchange on historical deviation states during periods not covered by offline detection data. The logarithmic function reflects the nonlinear characteristics of the heat exchange effect, thus effectively quantifying the thermal accumulation impact during blind zone periods.

[0014] Furthermore, the backtracking migration gain coefficient satisfies: In the formula, For the current moment The backtracking migration gain coefficient, For the current moment The thermal accumulation gradient in the blind zone The total length of the sterilization pipeline. For the current moment The instantaneous flow rate of the sterilization pipeline, For the current moment, This refers to the lag time in offline testing data for historical production batches of peptide milk powder production lines. To prevent extremely small constants with a denominator of 0, It is a natural exponential function.

[0015] This invention achieves a scientific evaluation of the backtracking migration gain coefficient by constructing a fractional function that includes the blind zone thermal accumulation gradient and the spatiotemporal decay term. This more accurately reflects the ratio between the thermal accumulation effect and the fluid residence time. The spatiotemporal decay term ensures that the influence of historical deviations decays reasonably over time, thereby effectively balancing the contribution of historical deviations to the correction of the current state.

[0016] Furthermore, the generation of the backtracking migration gain coefficient also includes boundary condition processing: in response to the current moment used to characterize the current absolute running time of the peptide milk powder production line being less than the offline detection data lag time, the backtracking migration gain coefficient is set to 0.

[0017] Furthermore, the historical true deviation vector is obtained by subtracting the offline detection data from the state prediction data.

[0018] Furthermore, the modification of the posterior state update equation of the extended Kalman filter algorithm includes: subtracting the dynamic compensation term from the posterior state update equation to obtain the modified posterior state update equation, wherein the historical true deviation vector is the difference vector between the offline detection value and the historical prediction value.

[0019] Furthermore, after outputting the corrected real-time status of the peptide milk powder production line, the system also includes: in response to the microbial survival rate or peptide activity retention rate in the corrected real-time status of the peptide milk powder production line exceeding the corresponding threshold, sending a PID fine-tuning interlock control command to adjust the sterilizer exhaust valve or steam control valve, thereby completing the monitoring of the peptide milk powder production line based on multi-source data fusion.

[0020] Secondly, the present invention provides a peptide milk powder production line monitoring system based on multi-source data fusion, which adopts the following technical solution: A peptide milk powder production line monitoring system based on multi-source data fusion includes a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the aforementioned peptide milk powder production line monitoring method based on multi-source data fusion is implemented.

[0021] By adopting the above technical solution, a computer program is generated from the above-mentioned method for monitoring peptide milk powder production lines based on multi-source data fusion, and stored in a memory so that it can be loaded and executed by a processor. In this way, a terminal device can be made based on the memory and the processor for convenient use.

[0022] The present invention has the following technical effects: (1) In view of the problem that traditional algorithms can only utilize real-time temperature data with strong noise and cannot integrate high-precision but significantly time-delayed offline detection data, resulting in the waste of high-value multi-source heterogeneous detection data, this invention evaluates the model state drift divergence by calculating the relative error between offline detection data and state prediction data at the same historical moment, and then constructs a dynamic compensation term through backtracking migration gain coefficient to correct the posterior state update equation of extended Kalman filter. This breaks the structural limitations of traditional algorithms that lack time-series backtracking and delay compensation, and effectively integrates high-precision quality data such as microbial survival rate and peptide activity retention rate obtained from offline testing into the real-time state estimation process. This not only leverages the continuous monitoring advantages of real-time working condition data, but also utilizes the high-precision characteristics of offline detection data, realizing the complementary fusion of multi-source heterogeneous data, fundamentally correcting the historical prediction bias of the model, and improving the accuracy of real-time state estimation of the production line.

[0023] (2) In response to the monitoring blind spots caused by offline detection lag, traditional algorithms cannot quantify the cumulative transmission effect of fluid heat exchange on historical deviations during the blind spot period, which leads to the problem of inaccurate prediction of the current state. This invention combines the real-time density of raw milk, the real-time temperature of the sterilizer and the instantaneous flow velocity of the pipeline, and calculates the heat accumulation gradient of the blind spot through continuous time integration to accurately evaluate the historical cumulative effect of fluid heat exchange in the sterilization pipeline during the blind spot period. At the same time, based on the fluid space residence time, the total operating time of the production line and the blind spot period, a time confidence smooth surface is constructed to constrain the heat accumulation gradient of the blind spot in a time and space, and generate a reasonable backtracking migration gain coefficient. This fills the state monitoring gap of the offline detection blind spot, and avoids overcompensation of historical deviations through time and space decay constraints, ensuring the continuity and reliability of the state estimation of the sterilization process throughout the entire time and process, and effectively solving the problem of model prediction distortion caused by the accumulation of blind spot deviations.

[0024] (3) In view of the contradiction that traditional algorithms cannot balance the killing of pathogenic microorganisms and the maximum preservation of the activity of thermosensitive peptides under dynamic operating conditions, this invention uses a modified extended Kalman filter algorithm to output real-time estimation results that fit the real thermodynamic state inside the sterilization pipeline. This can more accurately grasp the dynamic balance between sterilization temperature, fluid residence time and peptide degradation and microbial lethality. Based on accurate real-time status monitoring, the sterilization process parameters can be dynamically optimized, which avoids the degradation and inactivation of high-value bioactive peptides caused by overheating and eliminates the problem of microbial over-standard caused by insufficient heating, ensuring the safety of dairy products and improving the quality stability and batch consistency of peptide milk powder products.

[0025] (4) Through precise real-time status monitoring and dynamic process optimization, this invention can predict the deviation trend of the sterilization process in advance, adjust the operating parameters in a timely manner, avoid the whole batch of finished products being unqualified due to process runaway, and reduce the loss of high-value raw milk, bioactive peptides and other raw materials; at the same time, the automated status estimation of multi-source data fusion replaces the traditional mode of relying on manual test results to adjust the process in a lag, reduces the workload and operational errors of manual intervention, and improves the automation and intelligent control level of the production line; stable process control can maximize the retention of peptide activity, enhance the competitiveness of the product, and at the same time reduce the risk of food safety control, taking into account the multiple needs of production efficiency, product quality and safety control. Attached Figure Description

[0026] Figure 1 This is a flowchart of a peptide milk powder production line monitoring method based on multi-source data fusion according to an embodiment of the present invention.

[0027] Figure 2 This is a schematic diagram comparing the dynamic monitoring of microbial survival rate in a peptide milk powder production line monitoring method based on multi-source data fusion according to an embodiment of the present invention.

[0028] Figure 3 This is a schematic diagram comparing the dynamic monitoring of peptide activity retention rate in a peptide milk powder production line monitoring method based on multi-source data fusion according to an embodiment of the present invention. Detailed Implementation

[0029] 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, not all, of the embodiments of the present invention. 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.

[0030] This invention discloses a method for monitoring peptide milk powder production lines based on multi-source data fusion, referring to... Figure 1 This includes steps S001-S005: S001: Obtain real-time operating data and corresponding historical offline detection data of the peptide milk powder production line. The real-time operating data includes the real-time temperature of the sterilizer and the instantaneous flow rate of the sterilization pipeline. The offline detection data includes the microbial survival rate and peptide activity retention rate.

[0031] Specifically, the current moment, which characterizes the absolute time, is acquired in real time through industrial sensors deployed on the sterilization equipment. The continuous operating parameters include the real-time temperature of the sterilizer and the instantaneous flow rate of the sterilization pipeline. In this embodiment, the sampling period is 3 seconds. Simultaneously, the system is connected to a laboratory testing information system to obtain data on the time elapsed since the last time. The offline testing data of the most recent batch of historical production batches using the same process is used to dynamically obtain the lag time of the offline testing data by subtracting the timestamp of the test report corresponding to the offline testing data issued by the laboratory from the actual production timestamp of the historical production batch. The offline detection data is a vector, including microbial survival rate and peptide activity retention rate. The system traces the batch ID recorded by the production line MES and extracts the state prediction value output by the extended Kalman filter algorithm at the corresponding historical moment within the same historical production batch to complete the timestamp interpolation alignment of multi-source heterogeneous data.

[0032] S002: Calculate the relative error between the offline detection data and the state prediction data output by the extended Kalman filter at the same historical moment, and then perform nonlinear amplification mapping on the relative error to determine the state drift divergence.

[0033] It should be noted that in peptide milk powder production, both microbial lethality and peptide activity degradation are typical nonlinear thermodynamic processes. Offline detection data represents the true product quality at historical production moments, accurately reflecting the actual implementation effect of the sterilization process at those moments. The extended Kalman filter's state prediction data, however, is based on predictions from a thermodynamic mechanism model and real-time sensor data output. The deviation between these two directly characterizes the degree of deviation between the model prediction and the actual production state, serving as the core benchmark for model correction. Therefore, this step calculates the state drift divergence by constructing a nonlinear amplification model of the relative error between microbial survival rate and peptide activity retention rate, enabling accurate assessment of the distortion degree in historical model predictions.

[0034] Specifically, the state drift divergence satisfies: ; In the formula, For the current moment The state drift divergence corresponding to historical moments, For the current moment, This refers to the lag time in offline testing data for historical production batches of peptide milk powder production lines. and Each represents the current time. The microbial survival rate and peptide activity retention rate in the offline detection data corresponding to historical moments. and Each represents the current time. Microbial survival rate and peptide activity retention rate in the extended Kalman filter state prediction data corresponding to historical time points. To prevent extremely small constants with a denominator of 0, in this embodiment, ; It is a natural exponential function.

[0035] The relative difference between the theoretically predicted model value and the actual offline test value in the laboratory is considered as the fundamental error of the system state deviation. This error is nonlinearly amplified and mapped using an exponential function with the base of the natural logarithm. As the fundamental error increases... It will exhibit a dramatic exponential growth trend, thus characterizing the initial state distortion energy of the system at a specific historical moment at the algorithmic level.

[0036] S003: Obtain the real-time density of raw milk. Based on the state drift divergence, combined with the real-time temperature of the sterilizer and the instantaneous flow rate of the sterilization pipeline, calculate the cumulative transfer effect of fluid heat exchange on historical deviation state during the blind zone period not covered by offline detection data through continuous time integration, and extract the blind zone thermal accumulation gradient.

[0037] It should be noted that the sterilization process of peptide milk powder is a continuous fluid heat exchange process. Deviations in state at historical moments accumulate and are transmitted directionally as the fluid flows continuously and heats exchange within the pipeline. Fluctuations in operating conditions such as temperature and flow rate during blind periods not covered by offline monitoring data directly affect the transmission and evolution of historical deviations, thus directly impacting the accuracy of current state predictions. Therefore, this step, based on state drift divergence and combining real-time operating data and real-time density properties of raw milk during the blind period, assesses the cumulative thermal gradient in the blind period through continuous-time integration. This allows for a more accurate reconstruction of the cumulative transmission effect of fluid heat exchange on historical deviations during the blind period.

[0038] Specifically, the thermal accumulation gradient in the blind zone satisfies: ; In the formula, For the current moment The thermal accumulation gradient in the blind zone, For the current moment The state drift divergence corresponding to historical moments, For the current moment, This refers to the lag time in offline testing data for historical production batches of peptide milk powder production lines. For continuous time integral variables within the blind zone period, This is a function representing the real-time temperature of the sterilizer during the blind zone period. Let be the instantaneous flow rate function of the sterilization pipeline during the blind zone period. This refers to the real-time density of the raw milk. The preset heat exchange coefficient for dimensional normalization and thermo-kinetic energy accumulation effect calibration was obtained through historical batch heat exchange calibration experiments and regression analysis of real-time sensor data. Using data from the past 30 days of the same process batches stored in the MES system, combined with offline detection results, the heat exchange model was fitted using the least squares method to determine the coefficient. The initial value is automatically updated every 24 hours during normal operation. In order to adapt to fluctuations in the composition of raw milk, in this embodiment, The range of values ​​is The unit is This range of values ​​can cover the conventional impacts of raw milk composition, equipment heat exchange performance, and fluctuations in production conditions on heat exchange efficiency, ensuring that the gradient calculation results are accurately matched with the actual cumulative heat effect of sterilization. To prevent extremely small constants with a denominator of 0, It is the natural logarithm function.

[0039] Among them, the reciprocal of the product of continuous real-time temperature and flow rate reflects the heat retention intensity per unit mass of the sterilized fluid inside the pipe. When the pipe temperature rises and the flow rate decreases, the heat energy absorbed by the fluid inside the sterilizer increases sharply, and the thermodynamic cumulative characteristic value formed by integration rises. This cumulative characteristic acts as a multiplier factor directly on the state drift energy divergence, truly reflecting the driving effect of the continuous operating environment on the evolution of hysteresis error. In addition, in the actual deployment of industrial control systems, the above continuous integration is based on the sampling period and is realized through discrete numerical accumulation calculation.

[0040] S004: Determine the spatial residence time of the fluid based on the instantaneous flow rate and total length of the sterilization pipeline. Combine the total operating time of the peptide milk powder production line with the blind zone period to construct a time confidence smooth surface. Apply spatiotemporal attenuation constraints to the thermal accumulation gradient in the blind zone to generate the backtracking migration gain coefficient.

[0041] It should be noted that the impact of historical deviations on the current production state is not constant; it naturally decays as the fluid flows and diffuses within the sterilization pipeline and as production operation time progresses. Directly using historical deviations to indiscriminately correct the current state can easily lead to overcompensation, negatively impacting the stability of state estimation and the robustness of the algorithm for continuous production processes. Therefore, this step constructs a dual spatiotemporal decay constraint combining fluid residence time and operating time dimensions to generate a backtracking migration gain coefficient. This dynamically adapts to the efficiency of historical deviation migration to the current moment under different operating conditions, ensuring both the effective contribution of historical deviations to the current state correction and avoiding overcompensation through reasonable decay constraints.

[0042] Specifically, the backtracking migration gain coefficient satisfies: ; In the formula, For the current moment The backtracking migration gain coefficient, For the current moment The thermal accumulation gradient in the blind zone, The total length of the sterilization pipeline. For the current moment The instantaneous flow rate of the sterilization pipeline, For the current moment, This refers to the lag time in offline testing data for historical production batches of peptide milk powder production lines. To prevent extremely small constants with a denominator of 0, It is a natural exponential function.

[0043] Among them, the ratio of the physical length of the pipe to the current flow velocity represents the spatial penetration residence time of the fluid under the current operating conditions. As the residence time increases, the micro-diffusion effect inside the fluid is enhanced, which reduces the efficiency of the migration and transmission of historical errors to the current moment. At the same time, the negative exponential term in the post-processing uses the ratio of the total running time to the lag time to construct a time confidence smooth surface, which dynamically reduces the historical discrete deviation based on spatiotemporal constraints and transforms it into a dynamic correction weight that must be borne at the current moment.

[0044] Specifically, the generation of backtracking migration gain coefficients also includes boundary condition processing: In response to the fact that the current moment of the current absolute running time used to characterize the peptide milk powder production line is less than the offline detection data lag time, the backtracking migration gain coefficient is set to 0.

[0045] S005: Extract the historical true deviation vector of the offline detection data, use the product of the backtracking migration gain coefficient and the historical true deviation vector as a dynamic compensation term to correct the posterior state update equation of the extended Kalman filter algorithm, and output the corrected real-time status of the peptide milk powder production line.

[0046] It should be noted that the Extended Kalman Filter (EKF) algorithm is the core framework for state estimation of the nonlinear thermodynamic system in a peptide milk powder production line. Its posterior state update process directly determines the accuracy of the real-time state output. The historical true deviation vector extracted from offline detection data is the most reliable basis for correcting model system errors and improving prediction accuracy. Therefore, this step constructs a dynamic compensation term by backtracking the migration gain coefficient and the historical true deviation vector to correct the posterior state update equation of the EKF. This allows for the deep integration of high-precision offline quality detection data into the real-time state estimation process, breaking the limitation of traditional algorithms that can only utilize real-time sensor data and achieving efficient complementary fusion of multi-source heterogeneous data.

[0047] Specifically, the historical true deviation vector is obtained by subtracting the offline detection data from the state prediction data.

[0048] Specifically, the modification of the posterior state update equation of the extended Kalman filter algorithm includes: Based on the posterior state update equation, the dynamic compensation term is subtracted to obtain the modified posterior state update equation, where the historical true deviation vector is the difference vector between the offline detection value and the historical estimated value.

[0049] Specifically, after outputting the corrected real-time status of the peptide milk powder production line, it also includes: In response to the microbial survival rate or peptide activity retention rate in the real-time status of the corrected peptide milk powder production line exceeding the corresponding threshold, a PID fine-tuning interlock control command is sent to adjust the sterilizer exhaust valve or steam control valve, thus completing the monitoring of the peptide milk powder production line based on multi-source data fusion. In this embodiment, the corresponding thresholds are the microbial survival rate... Peptide activity retention rate This can be obtained through relevant industry standards.

[0050] like Figure 2 and Figure 3 As shown, compared with the state prediction data of traditional extended Kalman filtering, the curve fluctuation amplitude of the state prediction data after compensation and correction in this invention is reduced, effectively suppressing the estimation bias and sensor noise interference of traditional algorithms. It can more smoothly and accurately track the real change trend of microbial survival rate and peptide activity retention rate during sterilization. This verifies that the present invention effectively solves the problem of inaccurate monitoring of traditional algorithms by multi-source fusion and dynamic compensation correction of offline high-precision detection data and real-time operating data, improves the accuracy and stability of core state monitoring of peptide milk powder production line, and provides reliable real-time data basis for precise interlocking control of sterilization process parameters.

[0051] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0052] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A monitoring method for peptide milk powder production lines based on multi-source data fusion, characterized in that, include: Acquire real-time operating data and corresponding historical offline detection data of the peptide milk powder production line. The real-time operating data includes the real-time temperature of the sterilizer and the instantaneous flow rate of the sterilization pipeline, while the offline detection data includes the microbial survival rate and peptide activity retention rate. The relative error between the offline detection data and the state prediction data output by the extended Kalman filter at the same historical moment is calculated, and the relative error is nonlinearly amplified and mapped to determine the state drift divergence. The real-time density of raw milk is obtained. Based on the state drift divergence, combined with the real-time temperature of the sterilizer and the instantaneous flow rate of the sterilization pipeline, the cumulative transfer effect of fluid heat exchange on historical deviation state in the blind zone period not covered by offline detection data is calculated by continuous time integration, and the blind zone heat accumulation gradient is extracted. The spatial residence time of the fluid is determined based on the instantaneous flow rate and total length of the sterilization pipeline. A time confidence smooth surface is constructed by combining the total operating time of the peptide milk powder production line with the blind zone period. The spatiotemporal attenuation constraint of the thermal accumulation gradient in the blind zone is applied to generate the backtracking migration gain coefficient. The historical true deviation vector of the offline detection data is extracted, and the product of the backtracking migration gain coefficient and the historical true deviation vector is used as a dynamic compensation term to correct the posterior state update equation of the extended Kalman filter algorithm, so as to output the corrected real-time status of the peptide milk powder production line.

2. The method for monitoring a peptide milk powder production line based on multi-source data fusion according to claim 1, characterized in that, The offline detection data for the corresponding historical moment is obtained as follows: The timestamp of the test report corresponding to the offline test data issued by the laboratory is obtained, and the difference between the timestamp and the actual production timestamp of the historical production batch of the peptide milk powder production line is calculated to dynamically obtain the lag time of the offline test data. By tracing the batch identifiers recorded by the manufacturing execution system of the peptide milk powder production line, the state prediction data output by the extended Kalman filter algorithm at the corresponding historical moment within the same production batch is extracted to complete the timestamp interpolation alignment of multi-source heterogeneous data. The state prediction data includes microbial survival rate and peptide activity retention rate.

3. The method for monitoring a peptide milk powder production line based on multi-source data fusion according to claim 1, characterized in that, The state drift divergence satisfies: ； In the formula, For the current moment The state drift divergence corresponding to historical moments, For the current moment, This refers to the lag time in offline testing data for historical production batches of peptide milk powder production lines. and Each represents the current time. The microbial survival rate and peptide activity retention rate in the offline detection data corresponding to historical moments. and Each represents the current time. Microbial survival rate and peptide activity retention rate in the extended Kalman filter state prediction data corresponding to historical moments. To prevent extremely small constants with a denominator of 0, It is a natural exponential function.

4. The method for monitoring a peptide milk powder production line based on multi-source data fusion according to claim 1, characterized in that, The thermal accumulation gradient in the blind zone satisfies: ； In the formula, For the current moment The thermal accumulation gradient in the blind zone For the current moment The state drift divergence corresponding to historical moments, For the current moment, This refers to the lag time in offline testing data for historical production batches of peptide milk powder production lines. For continuous time integral variables within the blind zone period, This is a function representing the real-time temperature of the sterilizer during the blind zone period. Let be the instantaneous flow rate function of the sterilization pipeline during the blind zone period. This refers to the real-time density of the raw milk. This is a preset heat exchange coefficient used for dimensional normalization and calibration of the cumulative effect of thermal kinetic energy. To prevent extremely small constants with a denominator of 0, It is the natural logarithm function.

5. The method for monitoring a peptide milk powder production line based on multi-source data fusion according to claim 1, characterized in that, The backtracking migration gain coefficient satisfies: ； In the formula, For the current moment The backtracking migration gain coefficient, For the current moment The thermal accumulation gradient in the blind zone The total length of the sterilization pipeline. For the current moment The instantaneous flow rate of the sterilization pipeline, For the current moment, This refers to the lag time in offline testing data for historical production batches of peptide milk powder production lines. To prevent extremely small constants with a denominator of 0, It is a natural exponential function.

6. The method for monitoring a peptide milk powder production line based on multi-source data fusion according to claim 5, characterized in that, The generation of backtracking migration gain coefficients also includes boundary condition processing: In response to the fact that the current moment of the current absolute running time used to characterize the peptide milk powder production line is less than the offline detection data lag time, the backtracking migration gain coefficient is set to 0.

7. The method for monitoring a peptide milk powder production line based on multi-source data fusion according to claim 1, characterized in that, The historical true deviation vector is obtained by subtracting the offline detection data from the state prediction data.

8. The method for monitoring a peptide milk powder production line based on multi-source data fusion according to claim 1, characterized in that, The modification of the posterior state update equation of the extended Kalman filter algorithm includes: Based on the posterior state update equation, the dynamic compensation term is subtracted to obtain the modified posterior state update equation, where the historical true deviation vector is the difference vector between the offline detection value and the historical estimated value.

9. The method for monitoring a peptide milk powder production line based on multi-source data fusion according to claim 1, characterized in that, After outputting the corrected real-time status of the peptide milk powder production line, it also includes: In response to the microbial survival rate or peptide activity retention rate in the real-time status of the corrected peptide milk powder production line exceeding the corresponding threshold, a PID fine-tuning interlock control command is sent to adjust the sterilizer exhaust valve or steam control valve, thereby completing the monitoring of the peptide milk powder production line based on multi-source data fusion.

10. A monitoring system for a peptide milk powder production line based on multi-source data fusion, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement a peptide milk powder production line monitoring method based on multi-source data fusion according to any one of claims 1-9.

Images