Machine learning-based hpp multi-product dwell time adaptive control method

Abstract

The present application relates to the technical field of food high pressure processing process control, and particularly relates to a HPP multi-product pressure holding time self-adaptive regulation and control method based on machine learning, comprising: establishing a product processing unit according to product categories and loading positions; forming a time window and a pressure interval and a medium temperature interval; determining the upper and lower boundaries of the equivalent processing temperature of each product processing unit by using a physical temperature model and a machine learning residual model; accumulating the lower boundary of the effective processing amount according to the minimum processing contribution rate in the interval; determining the limiting processing unit and performing the remaining pressure holding accessibility judgment; and outputting the end pressure holding request when all product processing units continuously meet the standards, the upper boundary of the temperature is not exceeded, the batch is reliable and the shortest pressure holding time is met. The present application provides a HPP multi-product pressure holding time self-adaptive regulation and control method based on machine learning, which is used for reducing the influence of measurement errors, temperature estimation errors and loading differences on the HPP pressure holding termination judgment.

Description

Technical Field

[0001] This invention relates to the field of high-pressure food processing and sterilization process control technology, and in particular to a machine learning-based adaptive control method for holding pressure time of multiple HPP products. Background Technology

[0002] High Pressure Processing (HPP) is a process that applies isostatic pressure to sealed packaged food within a pressure chamber using a pressure medium. This is used for sterilization, microbial control, or shelf-life control of packaged food. Existing HPP equipment typically performs batch processing according to pre-set pressurization procedures, target pressures, nominal holding times, and depressurization procedures. When a batch includes multiple product categories, the longer holding time in the standard process for each product category or a uniform holding time determined by process engineers is usually used.

[0003] Existing HPP (High-Pressure Processing) pressure holding control methods typically determine the end of pressure holding based on the stage status recorded by the equipment controller and the nominal pressure holding time, or calculate the equivalent throughput based on pressure and medium temperature measurements. When using the nominal pressure holding time for control, the actual throughput contribution during pressure drop, pressure recovery, and plateau fluctuations cannot be distinguished according to the corresponding pressure history. When calculating throughput using a single pressure measurement and a single medium temperature measurement, existing methods do not separately account for pressure sensor calibration errors, temperature measurement point differences, product initial temperature differences, packaging heat exchange differences, and loading position differences in the throughput calculation process.

[0004] The pressure within the pressure chamber exhibits isostatic pressure transmission characteristics. Products in different loading positions within the same batch may still have different pre-batch temperatures, medium heat exchange conditions, and pressure adiabatic temperature rise processes. The internal temperature of the product cannot usually be directly measured in each production batch. When the medium temperature is used to replace the product temperature, the estimation error of the product's internal temperature will be included in the calculation of the processing contribution rate. When the predictive model directly outputs the target processing volume or the end result of pressure holding, inputs outside the model training range, underestimation of model residuals, and abnormal sensor data may cause the processing volume to be reached earlier than expected.

[0005] The existing control method neither integrates the uncertainty of pressure measurement, the uncertainty of product temperature estimation, the uncertainty of processing contribution function parameters, and the calculation of the lower bound of the processing capacity based on loading differences, nor does it identify and limit the processing units according to product category and loading location, and determine whether they can reach the target processing capacity within the longest permissible holding time. As a result, a time difference may occur between the nominal holding time and the actual time when each product processing unit reaches the target processing capacity. This time difference may correspond to premature termination of holding time or continued operation beyond the necessary holding time.

[0006] Therefore, it is necessary to establish processing units for products within the same HPP batch according to product category and loading location, determine the pressure range and medium temperature range using process measurement data, determine the upper and lower bounds of the equivalent processing temperature of each product processing unit using physical temperature model and machine learning residual model, and accumulate the lower bound of effective processing volume according to the minimum processing contribution rate within the pressure and equivalent processing temperature range, so that the end of pressure holding judgment is executed based on the lower bound of processing volume that can be confirmed by all product processing units. Summary of the Invention

[0007] To achieve the above-mentioned objectives, this invention provides a machine learning-based adaptive control method for holding pressure time of multiple HPP products, which solves or at least alleviates the problems that fixed nominal holding pressure time cannot reflect the actual pressure-temperature history, the inability to separately determine the status of multiple products in the same batch, and the lack of safety constraints in the end-holding pressure result directly generated by the machine learning model.

[0008] To achieve the above objectives, the present invention provides the following technical solution: an adaptive control method for holding pressure time of multiple HPP products based on machine learning, comprising: acquiring empirically validated process parameters for each product category within the current HPP batch, and establishing product processing units according to product category and loading position; collecting pressure, medium temperature, stage status, and sensor status to form a continuous time window, and determining the pressure range and medium temperature range of each time window; using a physical temperature model and a machine learning residual model to determine the equivalent processing temperature lower bound and equivalent processing temperature upper bound of each product processing unit; accumulating the effective processing volume lower bound based on the minimum processing contribution rate of the processing contribution function within the range, and determining the product processing unit with the lowest lower bound processing volume achievement rate as the restricted processing unit; outputting a request to end holding pressure when all product processing units continuously meet the standards, the upper temperature bound is not exceeded, the batch is reliable, and the shortest holding pressure time is reached; otherwise, outputting a non-end holding pressure command.

[0009] To further realize the present invention, the following technical solutions may be preferred:

[0010] Preferably, the step of obtaining the verified process parameters of each product category within the current HPP batch and establishing product processing units based on product category and loading location includes: obtaining the target pressure range, minimum platform pressure, processing contribution function, target throughput, shortest holding time, longest holding time, upper limit of equivalent processing temperature, and batch processing permission relationship for each product category; determining the common pressure range of the target pressure range for each product category, the maximum value of the shortest holding time, and the minimum value of the longest holding time; and outputting a batch processing prohibition instruction when the common pressure range is empty, the maximum value is greater than the minimum value, or there is no batch processing permission relationship.

[0011] Preferably, establishing product processing units based on product category and loading location includes: forming an association record between products and loading locations based on product category, packaging form, product quality, loading basket number, loading layer, axial position, and radial position; identifying products with the same product category and the same loading location partition as a product processing unit; and associating each product processing unit with corresponding thermophysical parameters, packaging heat transfer parameters, processing contribution function, target processing volume, and loading difference uncertainty.

[0012] Preferably, the acquisition of pressure, medium temperature, stage status, and sensor status forms a continuous time window. Determining the pressure range and medium temperature range for each time window includes: determining the pressure range based on pressure measurement values, pressure sensor calibration errors, measurement differences between pressure sensors, and data continuity; and determining the medium temperature range based on medium temperature measurement values, temperature sensor calibration errors, measurement differences between temperature measuring points, and data continuity. When the data missing ratio, timestamp continuity, sensor status, or stage status consistency does not meet the corresponding reliability conditions, the corresponding time window is set as an unreliable window.

[0013] Preferably, the determination of the equivalent processing temperature lower bound and the equivalent processing temperature upper bound of each product processing unit using the physical temperature model and the machine learning residual model includes: calculating the physical predicted temperature based on the medium temperature, pre-batch product temperature, pressure change rate, product equivalent heat capacity, packaging heat exchange parameters, and pressure adiabatic temperature rise parameters; inputting the product category, packaging form, loading location, pressurization time, pressure change rate, medium temperature change, and pressurization status into the machine learning residual model to obtain the temperature residual correction amount limited by the residual boundary.

[0014] Preferably, determining the equivalent processing temperature lower bound and the equivalent processing temperature upper bound of each product processing unit includes: adding the physical predicted temperature to the temperature residual correction amount to obtain an equivalent processing temperature estimate; determining the total temperature uncertainty based on the temperature sensor measurement uncertainty, machine learning residual uncertainty, and loading difference uncertainty; subtracting the total temperature uncertainty from the equivalent processing temperature estimate to obtain the equivalent processing temperature lower bound, and adding the total temperature uncertainty to the equivalent processing temperature estimate to obtain the equivalent processing temperature upper bound.

[0015] Preferably, the machine learning residual uncertainty is determined by the residual quantile between the measured product temperature and the estimated equivalent processed temperature in the independent calibration datasets separated according to the HPP processing cycle; when the model input approaches the boundary of the validated applicable range, the measurement difference between sensors increases, or the model is in a pressurization or pressure compensation state, the corresponding uncertainty is increased; when the model input exceeds the validated applicable range, the temperature residual correction is set to zero, and shortening the holding time based on the machine learning residual model is prohibited.

[0016] Preferably, the step of accumulating the lower bound of effective processing volume based on the minimum processing contribution rate of the processing contribution function within the interval includes: within the applicable range where the processing contribution function is verified to increase monotonically with pressure and equivalent processing temperature, inputting the lower bound of the pressure interval and the lower bound of the equivalent processing temperature into the processing contribution function using a conservative parameter set to obtain the minimum processing contribution rate; accumulating the lower bound of the effective processing volume of each product processing unit according to the minimum processing contribution rate, time window duration, window confidence flag, and window inclusion coefficient, and performing temperature limit judgment using the upper bound of the equivalent processing temperature.

[0017] Preferably, determining the product processing unit with the lowest lower limit processing volume achievement rate as the restricted processing unit includes: calculating the ratio of the effective lower limit of processing volume to the corresponding target processing volume for each product processing unit; determining achievability based on the remaining processing volume of the current restricted processing unit, the verified maximum allowable processing contribution rate, and the remaining allowable holding time; outputting a process non-compliance instruction when the sum of the effective lower limit of processing volume and the maximum increase in processing volume within the remaining allowable holding time is less than the corresponding target processing volume; outputting a manual review instruction when it is necessary to confirm the reason for non-compliance; and outputting a batch isolation instruction when the processing compliance of the corresponding product processing unit cannot be confirmed.

[0018] Preferably, the step of outputting the end-of-pressure-holding request when all product processing units continuously meet the standards, the upper limit of the temperature is not exceeded, the batch is reliable, and the shortest holding time is reached includes: verifying the lower limit processing volume achievement rate, the upper limit of the equivalent processing temperature, the pressure status, and the batch reliability status of all product processing units within a continuously preset number of reliable time windows; sending a pre-end-of-pressure-holding request to the HPP equipment controller when the end conditions are met; freezing the batch calculation data and outputting the end-of-pressure-holding request after receiving the safety interlock confirmation; when the end conditions are not met, outputting a continue-of-pressure-holding instruction when the batch is reliable and there are no process abnormalities, only the processing volume has not yet met the standards or the shortest holding time has not yet been reached; outputting a standard process instruction when the adaptive judgment condition fails but the equipment and the verified standard process allow continued processing; outputting an abnormality handling request when the temperature exceeds the limit, the pressure abnormality cannot be confirmed, or the process cannot meet the standards; and outputting a manual review instruction or a batch isolation instruction respectively when manual confirmation of the cause of the abnormality is required or the compliance of product processing cannot be confirmed.

[0019] The beneficial effects of this invention are:

[0020] This invention accumulates the lower bound of the effective processing capacity based on product category and loading location, and uses the product processing unit with the lowest lower bound processing capacity achievement rate within the same batch as the limiting processing unit for ending pressure holding. When accumulating the lower bound of the effective processing capacity, pressure measurement error, product temperature estimation error, processing contribution function parameter error, and loading difference are included in the calculation process. This invention uses the lower bound of the equivalent processing temperature to calculate the processing contribution and uses the upper bound of the equivalent processing temperature to determine temperature exceeding the limit. When the target processing capacity cannot be reached within the remaining allowable pressure holding time, an abnormal handling request is output to reduce the situation of prematurely ending pressure holding and continuing to operate beyond the maximum pressure holding time. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the execution environment and data connection relationship of the present invention.

[0022] Figure 2 This is a flowchart of the method of the present invention.

[0023] Figure 3 This is a flowchart of the effective processing capacity lower bound accumulation and limit processing unit determination of the present invention.

[0024] Figure 4 This is a flowchart of the reachability determination and end pressure holding request output of the present invention. Detailed Implementation

[0025] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] The HPP treatment described in this invention includes a process of applying isostatic pressure to sealed packaged food within a pressure chamber using a liquid pressure medium. The treatment time referred to in this invention is a reference equivalent treatment time determined for validated target or indicator microorganisms; this treatment time does not represent complete sterilization of all microorganisms.

[0028] In this embodiment, the processing contribution function outputs a dimensionless relative processing contribution rate relative to the reference pressure and reference temperature. This relative processing contribution rate is multiplied by the time window duration (in seconds) to obtain the processing volume (in reference equivalent seconds). The target processing volume is also expressed in reference equivalent seconds. When minutes are used as the time unit, the time window duration, cumulative processing volume, and target processing volume are converted using the same units.

[0029] The product processing unit referred to in this invention is a computational object determined by both the product category and the loading location. When the same product category is distributed in different loading locations, multiple product processing units can be formed; when different product categories are in the same loading location, corresponding product processing units are formed respectively.

[0030] The lower and upper temperature bounds referred to in this invention are computational boundaries determined based on temperature estimates and corresponding uncertainties. The lower temperature bound is used to calculate the minimum processing contribution that can be confirmed within the corresponding time window, and the upper temperature bound is used to determine whether the product exceeds the verified temperature range. The uncertainty can be represented by the upper limit of calibration error, the residual quantile of independent calibration data, or the error boundary confirmed by the process.

[0031] Example 1: Method Execution Environment and Parameter Management

[0032] See Figure 1 As shown, the method in this embodiment is executed by an industrial computer that is communicatively connected to the HPP device, which includes a pressure-bearing cavity, a loading basket, a pressurizing mechanism, a pressure replenishing mechanism, a pressure relief mechanism, a device controller, a pressure measuring component, a temperature measuring component, and a safety interlock component.

[0033] The pressure measurement assembly includes a main pressure sensor located at the pressure interface of the pressure chamber and a calibration pressure sensor located at the pressure medium pipeline. The main pressure sensor and the calibration pressure sensor output pressure measurement values, sensor status, calibration status, and fault status according to a unified time reference. When the HPP equipment is equipped with only one process pressure sensor, the pressure range is determined based on the calibration error, dynamic response error, and data continuity of the process pressure sensor.

[0034] The temperature measurement component includes temperature sensors located at the pressure medium inlet, the pressure medium reflux position, and the pressure chamber temperature measurement point. These temperature sensors can be PT100 platinum resistance temperature sensors. The temperature sensors transmit the pressure medium inlet temperature, pressure medium reflux temperature, and pressure chamber temperature to an industrial computer. The industrial computer then associates and stores this temperature data with the batch number, equipment cycle number, and a unified timestamp.

[0035] The equipment controller is used to execute pressure boosting, pressure holding, pressure replenishment, and pressure relief control programs, and outputs the current stage status, target pressure, pressure replenishment status, valve status, equipment alarm status, and safety interlock status. The industrial computer is used to execute product processing unit establishment, data window division, temperature upper and lower bound determination, processing volume lower bound accumulation, limit processing unit determination, and pressure holding termination judgment.

[0036] The industrial computer outputs pre-end pressure holding request, end pressure holding request, continue pressure holding command, standard process command, and abnormal handling request to the equipment controller. The end pressure holding request does not bypass the safety interlock component of the equipment controller to control the pressure relief mechanism. After confirming the pressure status, valve status, control mode, communication status, and safety interlock status, the equipment controller performs controlled pressure relief.

[0037] The batch and product information includes batch number, processing cycle number, product category, packaging form, unit weight, product equivalent dimensions, pre-batch product temperature, loading basket number, loading layer, axial position, radial position, and standard process number. This batch and product information can be obtained through a production execution system, barcode reader, RFID device, or operating terminal.

[0038] The verified process parameters include target pressure range, minimum platform pressure, target throughput, shortest holding time, longest holding time, upper limit of equivalent processing temperature, processing contribution function parameters, product equivalent heat capacity, packaging heat exchange parameters, pressure adiabatic temperature rise parameters, applicable range of machine learning model, residual correction boundary, model residual uncertainty, loading difference uncertainty, and batch processing permission relationship.

[0039] The target throughput, processing contribution function, and validated temperature range are determined based on microbial challenge tests for the corresponding product category. The product equivalent heat capacity, packaging heat transfer parameters, and pressure adiabatic temperature rise parameters are determined based on loading temperature distribution tests, product internal temperature measurements, or validated thermophysical property data. The machine learning model residual correction boundary and model residual uncertainty are determined based on calibration data independent of the training data.

[0040] The treatment contribution function parameters include a reference parameter set and a conservative parameter set. The conservative parameter set is determined based on the allowable deviation, confidence boundary, or independent validation results of the challenge experiment fitting parameters. When no conservative parameter set is established, the treatment contribution rate calculated from the reference parameter set is multiplied by a validated treatment contribution correction coefficient not greater than one.

[0041] After obtaining the product information for the current batch, a common pressure range is determined based on the target pressure range for each product category. The lower limit of the common pressure range is the maximum value among the lower limits of the target pressure for each product category, and the upper limit of the common pressure range is the minimum value among the upper limits of the target pressure for each product category. When the lower limit of the common pressure range is greater than the upper limit of the common pressure range, the current batch is set to a state where batch processing is prohibited.

[0042] Accordingly, the shortest holding time for a batch is the maximum value among the shortest holding times for each product category, and the longest holding time for a batch is the minimum value among the longest holding times for each product category. When the shortest holding time for a batch is greater than the longest holding time for a batch, the current batch is set to a state where batch processing is prohibited.

[0043] When product categories have batch processing permits and the common pressure range, minimum batch holding time, and maximum batch holding time meet the batch processing conditions, product processing units are established based on product category and loading location. The loading basket can be divided axially into inlet-side, intermediate, and return-side zones, and radially into center and outer zones; the number and boundaries of these zones are determined based on the pressure chamber structure and loading temperature distribution test results.

[0044] The current batch includes product category A and product category B. Product category A is loaded into the central partition of the first loading basket and the outer partition of the second loading basket, respectively. When product category B is loaded into the central partition of the second loading basket, three product processing units are established. Each product processing unit is associated with the target throughput and processing contribution function of the corresponding product category, and is also associated with the heat transfer correction parameter and loading difference uncertainty of the corresponding loading location.

[0045] The physical temperature model, machine learning residual model, and corresponding uncertainty parameters are associated with the HPP equipment number. Before applying the model to another HPP equipment, the model parameters are calibrated using the temperature distribution verification data of the other HPP equipment. If changes occur to the pressure sensor, temperature sensor, pressure medium circulation structure, packaging form, or loading method that exceed the approved range, the corresponding error boundaries and the applicable scope of the model are reconfirmed.

[0046] Example 2: Adaptive Pressure Holding Control Method

[0047] See Figure 2 As shown in the figure, this embodiment provides a machine learning-based adaptive control method for holding pressure time of HPP multi-products, including the following steps.

[0048] Step S100: Obtain verified process parameters and establish product processing units.

[0049] Obtain verified process parameters for each product category within the current HPP batch, and establish product processing units based on product category and loading location.

[0050] In this embodiment, before the HPP equipment begins pressurization, the batch number, processing cycle number, product category, packaging form, product quality, pre-batch product temperature, loading basket number, and loading position are read. Based on the standard process number, the target pressure range, minimum platform pressure, target throughput, shortest holding time, longest holding time, upper limit of equivalent processing temperature, and processing contribution function are retrieved.

[0051] Each product category's process parameters have an activation status, version number, effective date, and applicable equipment number. When a process parameter is not in an activated state, the product category is inconsistent with the standard process number, the process parameter has exceeded its approved validity period, or the applicable equipment number is inconsistent with the current HPP equipment, the batch process parameter status will be set to invalid, and the output of the end-of-pressure holding request based on the adaptive calculation results will be prohibited.

[0052] For multiple product categories, check the batch processing permission relationship, common pressure range, minimum batch holding time, and maximum batch holding time separately. If any batch condition is not met, output a batch processing instruction or a manual review instruction, and do not enter the adaptive holding time calculation process.

[0053] When establishing a product processing unit, associated records are created according to product category, loading basket number, layer, axial partition, and radial partition. Each product processing unit has a unit number, product category, product quantity, total mass, packaging form, location partition, physical temperature model parameters, residual model version, target throughput, and upper temperature limit.

[0054] Step S200: Form a continuous time window and generate a credibility state.

[0055] Data such as pressure, medium temperature, stage status, and sensor status are collected to form continuous time windows, and the pressure range and medium temperature range of each time window are determined.

[0056] The sampling period for pressure data can be set from 10 milliseconds to 1 second, the sampling period for temperature data can be set from 100 milliseconds to 2 seconds, and the time window length can be set from 0.5 seconds to 5 seconds. In this embodiment, the sampling period for pressure data is set to 200 milliseconds, the sampling period for temperature data is set to 500 milliseconds, and the time window length is set to 1 second.

[0057] The pressure holding stage is identified based on the stage status output by the equipment controller. After the equipment enters the pressure holding stage, the starting time when the lower boundary of the pressure range continuously falls within the common pressure range for a preset confirmation duration is determined as the effective pressure holding candidate starting point. If the equipment enters the pressure holding stage but the lower boundary of the pressure range has not yet entered the common pressure range, the effective processing volume of the pressure holding stage will not be accumulated.

[0058] For each time window, calculate the average measurement value of the main pressure sensor, verify the average measurement value of the pressure sensor, pressure standard deviation, pressure change rate, average medium inlet temperature, average medium return temperature, temperature change rate, pressure compensation status, and consistency of stage status.

[0059] When calculating the pressure range, both pressure measurements and pressure uncertainty are used. When multiple reliable pressure measurement signals are available, the lower bound of the pressure range is the minimum value after subtracting the corresponding error boundary from each pressure measurement, and the upper bound is the maximum value after adding the corresponding error boundary to each pressure measurement. If the difference between pressure sensors exceeds the verified consistency range, the pressure uncertainty is increased, and a pressure sensor deviation flag is generated.

[0060] When calculating the medium temperature range, the medium inlet temperature, medium return temperature, pressure chamber temperature, and corresponding temperature measurement errors are comprehensively considered. When there are multiple reliable temperature measurement points, the medium temperature range covers the measured values ​​and error boundaries of each reliable temperature measurement point; when the difference between temperature measurement points exceeds the verified consistency range, the medium temperature range is expanded and a temperature distribution deviation indicator is generated.

[0061] A window confidence flag is generated based on sampling integrity, timestamp continuity, sensor status, and stage status consistency. The window confidence flag is set to a confident state when the data missing ratio within the time window does not exceed the allowable ratio, the timestamps are continuous, at least one process pressure signal is valid, the medium temperature signal meets the temperature model input conditions, and the stage status is consistent.

[0062] A single unreliable time window is not included in the effective processing volume lower bound for adaptive termination judgment, and the corresponding data anomaly type is recorded. When the number of consecutive unreliable time windows reaches the preset number or the cumulative unreliable duration reaches the allowed duration, the adaptive termination judgment state is exited; if the equipment status and verified standard process still allow continued processing, a standard process instruction is output; if the cause of the data anomaly needs to be confirmed, a manual review instruction is also output; if the equipment failure or safety status does not allow continued processing, a controlled termination request is output.

[0063] After recovery from an unreliable state, the recovery time window is calculated when the pressure, medium temperature, stage status, and sensor status continuously meet the reliable conditions for a preset recovery time. Batches that have exited the adaptive termination judgment state will not have the early termination pressure holding function reactivated due to data recovery, unless the recovery rules are already included in the empirical process documentation for the corresponding product and equipment.

[0064] Step S300: Calculate the physical predicted temperature

[0065] The equivalent processing temperature lower limit and equivalent processing temperature upper limit of each product processing unit are determined using a physical temperature model and a machine learning residual model.

[0066] For each product processing unit, the physical predicted temperature is calculated based on the medium temperature, pre-batch product temperature, product equivalent heat capacity, packaging heat exchange parameters, pressure change rate, and pressure adiabatic temperature rise parameters.

[0067] In one specific implementation, the physical temperature model employs a first-order lumped thermal parameter model:

[0068]

[0069] in, Number the product processing unit. For the physical prediction temperature of the product processing unit, The medium represents temperature. For equivalent heat capacity, The equivalent heat transfer parameters for packaging and products. For pressure adiabatic temperature rise parameters, For pressure.

[0070] The medium temperature is determined based on the medium temperature range and the loading location of the product processing unit. The inlet-side product processing unit uses a weighted average of the inlet temperature and the pressure chamber temperature; the intermediate-position product processing unit uses the pressure chamber temperature; and the return-side product processing unit uses a weighted average of the return temperature and the pressure chamber temperature. The weighting coefficient is determined based on loading temperature distribution tests.

[0071] The pressure change rate during the pressurization and repressurization phases is determined by the pressure measurements within a continuous pressure window. When the pressure data is reliable, the pressure adiabatic temperature rise term is calculated based on the pressure change rate; when the pressure data is unreliable, the temperature state used for adaptive termination judgment is not updated using the aforementioned time window, and a data anomaly flag is output.

[0072] Step S400: Correct the physically predicted temperature using a machine learning residual model.

[0073] The residual model input vector is formed by taking into account product category, packaging form, product quality, loading location, pre-batch product temperature, pressurization time, pressure change rate, medium temperature change, pressurization status, and physical prediction temperature.

[0074] The machine learning residual model can employ a gradient boosting tree regression model, a random forest regression model, or a support vector regression model. The training label for the machine learning residual model is the difference between the measured internal temperature of the product in the validation batch and the physically predicted temperature at the same time. The measured internal temperature of the product is obtained by a temperature probe placed inside the simulated packaging or the validation packaging.

[0075] When using a gradient boosting tree regression model, the number of trees can be set from 100 to 500, the maximum tree depth can be set from 2 to 8, the learning rate can be set from 0.01 to 0.1, and the loss function can be either the squared error loss function or the absolute error loss function. The model parameters are determined based on the validation set error, and the number of trees is stopped from increasing when the validation set error has not decreased for a preset number of consecutive epochs.

[0076] The data is divided into training, validation, and independent calibration sets according to the HPP processing cycle, and all time windows within the same processing cycle are assigned to only one of these datasets. Data partitioning can also be performed by device number, processing date, or product batch to reduce residual underestimation caused by consecutive time windows within the same processing cycle entering both the training and calibration sets simultaneously.

[0077] The machine learning residual model does not directly output the target processing volume, holding time, or end of holding time conclusion. Instead, it outputs a temperature residual correction. This temperature residual correction is tailored according to the empirical residual boundaries of the corresponding product category, packaging form, and loading zone.

[0078] The estimated equivalent treatment temperature is determined according to the following formula:

[0079]

[0080] in, This is an estimated value for the equivalent treatment temperature. For the output of the machine learning residual model, This indicates a clipping operation performed according to the verified residual boundary.

[0081] If a model input field is missing, the model version is invalid, or the input exceeds the verified applicable range, the temperature residual correction is set to zero, and the upper and lower bounds of the temperature are determined using the physical predicted temperature and the expanded model uncertainty. In this case, the holding time is not shortened based on the output of the machine learning residual model.

[0082] Step S500: Determine the upper and lower limits of the equivalent treatment temperature.

[0083] The total temperature uncertainty is determined based on the temperature sensor measurement uncertainty, machine learning residual uncertainty, and loading difference uncertainty.

[0084] The temperature sensor measurement uncertainty is determined based on the temperature sensor calibration certificate, the conversion error of the acquisition module, and the consistency of the temperature measurement points. The machine learning residual uncertainty is taken as the residual quantile between the measured product temperature and the estimated equivalent processed temperature in the independent calibration dataset. The loading difference uncertainty is determined based on the temperature distribution verification results among different packaged products within the same location zone.

[0085] In one specific implementation, the total temperature uncertainty is determined according to the conservatism and determination of each uncertainty component:

[0086]

[0087] in, The total temperature uncertainty is... For temperature measurement uncertainty, For model residual uncertainty, This represents the loading difference uncertainty.

[0088] When using the conservative sum of each uncertainty component, each uncertainty component is determined separately according to its corresponding calibration range, product range, and equipment range. The conservative sum is used to reduce the situation where the lower limit of the temperature rises above the corresponding estimated boundary due to mutual cancellation of error components. When the uncertainty components have statistical correlation and corresponding joint distribution verification results have been obtained, a verified joint uncertainty calculation method can also be used.

[0089] The lower limit and upper limit of the equivalent treatment temperature are determined according to the following formula:

[0090]

[0091]

[0092] in, The lower limit of the equivalent treatment temperature, The upper limit of the equivalent treatment temperature.

[0093] When the residual model input is within the applicable range, the model residual uncertainty determined by the corresponding calibration data is used; when the residual model input is close to the boundary of the applicable range, the model residual uncertainty is increased according to the preset magnification factor; when the residual model input exceeds the applicable range, the uncertainty corresponding to the model failure state is used.

[0094] When the sensor is in a rapid pressurization or repressurization state, the dynamic temperature uncertainty is increased according to the pressure change rate. When the difference between different temperature measurement points exceeds the consistency threshold, the temperature measurement uncertainty is increased according to the maximum difference between the measurement points. After the sensor returns to normal, the corresponding uncertainty is restored to the normal range when the consistency condition is met within a consecutive preset number of confidence windows.

[0095] Step S600, Lower bound of cumulative effective processing volume

[0096] The lower bound of the cumulative effective processing volume is determined by the minimum processing contribution rate of the processing contribution function within the specified interval.

[0097] See Figure 3 As shown, the treatment contribution function for each product category is determined by challenge tests against the target or indicator microorganisms, defining the corresponding applicable pressure range, applicable temperature range, and monotonicity range. Within these applicable ranges, when the treatment contribution function monotonically increases with pressure and equivalent treatment temperature, the minimum treatment contribution rate is calculated using the lower bound of the pressure range and the lower bound of the equivalent treatment temperature.

[0098] In one specific implementation, the contribution function is processed using the following formula:

[0099]

[0100] in, For product categories, The contribution rate for dimensionless relative processing. The treatment contribution coefficient under the reference condition. For reference pressure, It is a pressure-sensitive parameter. For reference temperature, This is a temperature-sensitive parameter.

[0101] The treatment contribution function uses a conservative parameter set determined by challenge experiments. When the reference parameter set has fitting errors but a conservative parameter set has not yet been established, independent validation data is used to determine the treatment contribution correction coefficient, and this correction coefficient is used to correct the relative treatment contribution rate. The treatment contribution correction coefficient, along with its applicable products, pressure range, and temperature range, is stored in association with the process parameter version.

[0102] For the One time window, product processing unit The minimum treatment contribution rate is determined based on the minimum value of the treatment contribution function within the pressure and temperature ranges:

[0103]

[0104] When the processing contribution function monotonically increases with pressure and temperature within a verified range, the minimum processing contribution rate is determined by the lower bound of the pressure interval and the lower bound of the equivalent processing temperature. If the processing contribution function does not satisfy the monotonicity condition, a minimum value search is performed within the corresponding interval based on the verified function, or the corresponding time window is set to a window not used for adaptive termination judgment.

[0105] The lower bound of the effective processing capacity of the product processing unit is accumulated according to the following formula:

[0106]

[0107] in, This represents the lower bound of the effective processing volume up to the current moment, in reference equivalent seconds. As a confidence indicator for the window, A coefficient is included for the window. The minimum processing contribution rate for dimensionless data. The duration of the time window is expressed in seconds.

[0108] For time windows that are reliable and fall within the validated process range, the window reliability flag is set to included. For time windows that are reliable and fall within the validated compensable low-pressure range, the minimum processing contribution rate is calculated based on the lower bound of the actual pressure range. Time windows with unreliable data or where the pressure or temperature exceeds the applicable range of the processing contribution function are not used to form an adaptive end-of-pressure conclusion and trigger standard process or manual review status.

[0109] When calculating the processing contribution using the equivalent processing temperature lower bound, the equivalent processing temperature upper bound is also used to perform temperature limit judgment. When the equivalent processing temperature upper bound of any product processing unit exceeds the corresponding verified temperature upper limit, a temperature limit violation event is generated, the adaptive end pressure holding request is stopped, and an abnormal handling request is output; if the equipment safety status does not allow continued pressure holding, a controlled termination request is output; if the reason for the limit violation needs to be confirmed, a manual review instruction is also output; if the temperature limit violation may affect the product processing compliance, a batch isolation instruction is also output.

[0110] Step S700: Determine the restriction processing unit

[0111] The ratio of the effective processing capacity lower bound of each product processing unit to the corresponding target processing capacity is determined as the lower bound processing capacity achievement rate, and the product processing unit with the lowest lower bound processing capacity achievement rate is determined as the restricted processing unit.

[0112] The lower limit processing volume achievement rate and the limit processing unit are determined according to the following formula:

[0113]

[0114]

[0115] in, This represents the lower limit of the processing capacity achieved by the product processing unit. The target processing volume for the corresponding product category is expressed in reference equivalent seconds. This is the current restricted processing unit.

[0116] The target throughput is a validated fixed target throughput, which is not reduced by the machine learning residual model. When the lower bound throughput achievement rate of multiple product processing units is the same, the product processing unit with the shorter remaining allowable holding time is preferentially identified as the limiting processing unit; when the remaining allowable holding time is the same, the product processing unit whose equivalent processing temperature upper bound is close to the corresponding temperature upper limit is identified as the limiting processing unit.

[0117] After each time window ends, record the unit number, product category, loading location, effective throughput lower limit, target throughput, lower limit throughput achievement rate, temperature upper and lower limits, and determination time of the current restricted processing unit. When the restricted processing unit changes, retain the restricted processing unit records before and after the change.

[0118] Step S800: Perform a residual pressure reachability check.

[0119] See Figure 4 As shown, based on the remaining target processing capacity of the current limited processing unit, the verified maximum allowable processing contribution rate, and the remaining allowable holding time, it is determined whether the corresponding product processing unit can still reach the target processing capacity before the longest holding time of the batch.

[0120] The verified maximum permissible processing contribution rate is determined based on the upper limit of the common pressure range, the upper limit of the permissible temperature for the corresponding product category, and the processing contribution function. Processing contribution rates exceeding the verified pressure or temperature range are not used. The remaining permissible holding time is determined by subtracting the current actual holding time from the longest holding time of the batch.

[0121] When the sum of the current effective processing capacity lower limit and the maximum increase processing capacity within the remaining allowable holding time is still less than the target processing capacity, the corresponding product processing unit is set to an unacceptable state and a process unacceptable instruction is output; if the reason for unacceptability needs to be confirmed, a manual review instruction is also output; if the processing compliance of the corresponding product processing unit cannot be confirmed, a batch isolation instruction is also output; if equipment failure or safety conditions do not allow continued holding pressure, a controlled termination request is output.

[0122] If the processing volume target can still be achieved within the remaining allowable holding time, the estimated remaining holding time is calculated based on the current minimum processing contribution rate and the remaining processing volume. This estimated remaining holding time is used for human-machine interface display and production scheduling, and does not replace the determination of whether the effective processing volume lower limit has been met.

[0123] Step S900: Perform stress anomaly and data anomaly handling.

[0124] During the pressure holding phase, pressure drop events are detected, and the following information is recorded: the start time of the pressure drop, the lowest value of the lower limit of the pressure range, the recovery time, the duration, the pressure replenishment status, the equipment alarm status, and the data continuity.

[0125] If the data is continuous during the pressure drop, the pressure range can be determined, the processing contribution function covers the corresponding pressure range, and the equipment does not experience an unrecoverable failure, the pressure drop event is determined as a computable pressure drop event, and the effective processing volume continues to accumulate according to the lower bound of the pressure range.

[0126] If data interruption, pressure sensor failure, pressure falling below the applicable range of the processing contribution function, or inability to confirm the equipment alarm status occurs during a pressure drop, the pressure drop event is classified as an unconfirmable pressure anomaly, and the batch reliability status is set to invalid. After the occurrence of such an unconfirmable pressure anomaly, no end-of-pressure-holding request is output based on the adaptive calculation results; if the equipment status and verified standard process still allow continued processing, a standard process instruction is output; if the cause of the anomaly needs to be confirmed, a manual review instruction is also output; if product processing compliance cannot be confirmed through subsequent processing, a batch isolation instruction is also output; if equipment failure or safety conditions prevent continued processing, a controlled termination request is output.

[0127] If temperature data is temporarily missing but there are verified redundant temperature measurement points, the medium temperature range is expanded based on the remaining reliable temperature measurement points; if there are no temperature measurement points that meet the input conditions of the temperature model, the corresponding window is set to unreliable and a standard process or manual verification instruction is output.

[0128] If the continuous unreliable state exceeds the allowed duration, the adaptive termination judgment will stop; if the equipment status and verified standard process still allow continued processing, the current batch will be switched to the standard process state, and subsequent processing will be completed according to the pre-approved fixed holding time; if equipment failure or safety conditions prevent continued processing, a controlled termination request will be output; if the cause of the unreliable state needs to be confirmed, a manual review instruction will also be output. The lower limit of the processing volume accumulated before switching to the standard process state is only used for batch recording and is not used to shorten the fixed holding time.

[0129] Step S1000: Output a request to end pressure holding or a command to not end pressure holding.

[0130] The system outputs a request to end the pressure holding process when all product processing units continuously meet the standards, the upper temperature limit is not exceeded, the batch is reliable, and the shortest pressure holding time is reached; otherwise, it outputs a command to not end the pressure holding process.

[0131] In this embodiment, the number of continuous compliance confirmation windows can be set to 3 to 10, and can also be adjusted according to the time window length and the equipment control response cycle. During the continuous confirmation period, a pre-end pressure holding request is generated when the lower limit processing volume achievement rate of all product processing units is not less than 1, the upper limit of all equivalent processing temperatures does not exceed the corresponding upper temperature limit, the pressure status meets the termination conditions, the batch credibility status is valid, and the actual pressure holding time is not less than the shortest pressure holding time of the batch.

[0132] The pre-end pressure holding request includes batch number, processing cycle number, request generation time, limited processing unit, lower limit processing volume achievement rate of each product processing unit, upper temperature limit status, batch reliability status, shortest pressure holding time status, and data verification code.

[0133] Upon receiving the pre-end pressure holding request, the equipment controller verifies the equipment control mode, pressure status, valve status, pressurization mechanism status, pressure replenishment mechanism status, pressure relief mechanism status, communication status, and safety interlock status. If the verification is successful, the equipment controller returns a confirmation signal allowing the end of the pressure holding process.

[0134] Upon receiving the confirmation signal to end the process, the industrial computer freezes the raw process data index, time window data, model version, model input, temperature upper and lower limits, throughput lower limit, limiting processing units, and control decision records for the current batch, and outputs an end-of-pressure request to the equipment controller. The equipment controller then executes the pressure relief procedure according to the verified pressure relief protocol.

[0135] If any termination condition is not met, the corresponding non-termination pressure holding instruction or handling request is output according to the unmet condition or anomaly type. If the batch is reliable and there are no process anomalies, but the lower bound processing volume achievement rate of any product processing unit is less than one or the actual pressure holding time has not reached the minimum pressure holding time of the batch, a continue pressure holding instruction is output. If the batch reliability is invalid, the model input exceeds the verified applicable range, or the process parameters required for adaptive calculation are invalid, but the equipment status and verified standard process still allow continued processing, a standard process instruction is output. If a temperature exceeding the limit, an unconfirmed pressure anomaly, or a process failure occurs, an anomaly handling request is output. If equipment failure or a safety condition does not allow continued pressure holding, a controlled termination request is output. When it is necessary to confirm process parameters, data status, model status, or the cause of the anomaly, a manual review instruction is output in addition to the corresponding main instruction; if product processing compliance cannot be confirmed, a batch isolation instruction is output in addition to the corresponding main instruction. The main instruction and additional instructions can be output simultaneously based on the same anomaly event.

[0136] Example 3: Model Calibration and Validation

[0137] This embodiment illustrates the implementation methods for determining physical temperature model parameters, training the machine learning residual model, calibrating uncertainty, and verifying control logic. Model calibration data and control logic test data are separated according to a processing loop; the control logic test does not use the training data as independent verification data.

[0138] 1. Determination of physical temperature model parameters

[0139] Select validation packaging that matches the production packaging form, product quality, and loading method. Place temperature probes at the product center, the midpoint between the product's geometric center and the packaging surface, and near the packaging surface. Record the temperature probe measurements along with the pressure medium inlet temperature, reflux temperature, cavity temperature, and pressure data using a unified time reference.

[0140] Validation cycles were performed under multiple pre-batch product temperatures, loading masses, and pressurization procedures. The equivalent heat capacity, packaging heat transfer parameters, and pressure adiabatic temperature rise parameters were determined based on the error between the measured product temperature and the physical temperature model output. When training the machine learning residual model, the physical temperature model parameters were kept constant, and the training labels were obtained by subtracting the physically predicted temperature from the temperature probe measurements.

[0141] 2. Training and Calibration of Machine Learning Residual Models

[0142] The dataset is divided into training, validation, and independent calibration sets according to the HPP processing loop, and the entire time window of the same processing loop is allocated to only one of the datasets. When there are multiple HPP devices, calibration is performed separately according to the device number, or the device number feature is set in the training data, and cross-device validation is performed using the processing loop of the device that is not included in the training set.

[0143] The machine learning residual model employs a gradient boosting tree regression model. Inputs include product category code, packaging form code, product quality, loading location code, pre-batch product temperature, current physical prediction temperature, pressurization time, pressure change rate, medium temperature change rate, pressurization status, and cumulative time since self-pressurization began. The model output is the temperature residual correction for the current window.

[0144] After model training is complete, the temperature residual distributions corresponding to product category, packaging form, and loading location are calculated in an independent calibration set, and the model residual uncertainty is determined based on the selected coverage level. The coverage level, residual quantiles, maximum uncovered bias, and applicable input range are saved in association with the model version.

[0145] 3. Temperature range coverage verification

[0146] Temperature range coverage verification is performed using a processing loop that does not include the training set, validation set, or residual boundary determination process. In each processing loop, internal temperature probes are set at each predetermined loading location, and the probe measurements and their calibration error boundaries are compared with the equivalent processing temperature lower bound and equivalent processing temperature upper bound at the same time.

[0147] The validation log includes the number of independent validation cycles, product category, packaging form, loading location, pre-batch product temperature range, pressure range, medium temperature range, number of temperature range coverage windows, number of uncovered windows, coverage ratio, and maximum uncovered deviation. When uncovered windows exist, the corresponding model residual uncertainty is increased or the model's applicable range is narrowed based on the uncovered location, stage status, and model input range.

[0148] 4. Processing capacity lower bound verification

[0149] Using the internal temperature probe measurements and pressure measurements from the independent verification cycle, the probe reference processing capacity is calculated according to the same processing contribution function, and the probe reference processing capacity is compared with the lower bound of the effective processing capacity of the corresponding product processing unit.

[0150] Within the applicable range of the processing contribution function, record the lower limit of the processing capacity of each product processing unit, the probe reference processing capacity, the difference between the two, the limiting processing unit number, and the time of achievement. When the lower limit of the processing capacity is higher than the probe reference processing capacity, check the temperature range, pressure range, processing contribution function parameters, and time synchronization status, and adjust the uncertainty or processing contribution correction factor according to the check results.

[0151] 5. Verification of normal pressure holding conditions for multiple products

[0152] Within the same batch, at least two product categories should be set up, and the same product category should be distributed across at least two loading locations. During the pressure holding phase, the pressure range, upper and lower temperature limits, lower limit of effective throughput, target throughput, and lower limit throughput achievement rate of each product processing unit should be recorded.

[0153] When some product processing units reach their target throughput while the limiting processing unit has not yet reached its target throughput, the system continues to maintain pressure. When all product processing units continuously reach their target throughput, the upper temperature limit is not exceeded, the batch is reliable, and the minimum pressure holding time is met, the system generates a pre-end pressure holding request.

[0154] 6. Pressure sensor offset verification condition

[0155] In the calibration test environment, a slow offset within a preset range is added to the verification pressure signal, or historical verification data with known pressure deviations is replayed. When the measurement difference between the main pressure sensor and the verification pressure sensor exceeds the normal consistency range, the system expands the pressure range and uses the lower bound of the expanded pressure range for cumulative processing.

[0156] When the measurement difference exceeds the unconfirmable threshold, the system sets the batch reliability status to invalid, stops outputting the adaptive end pressure holding request, and outputs a standard process instruction or a manual review instruction.

[0157] 7. Short-term pressure drop verification condition

[0158] The pressure replenishment event is executed in the process verification mode permitted by the equipment, or it can be replayed using historical processing data that includes the pressure replenishment event. Pressure and temperature data remain continuous during the pressure drop, and when the processing contribution function covers the corresponding pressure range, the system accumulates the processing volume according to the lower limit of the pressure range and the upper limit of the temperature range.

[0159] When pressure data is interrupted, pressure is below the applicable range of the processing contribution function, or the equipment status cannot be confirmed, the system will set the corresponding window to untrusted, and exit the adaptive termination judgment state when the continuous untrusted state exceeds the allowed duration.

[0160] 8. Model input out of range verification condition

[0161] If the pre-batch product temperature, product quality, pressurization time, or loading location are set to input states that are outside the scope of the residual model's verified applicability, the system will set the machine learning residual correction to zero and use the physical predicted temperature and the expanded model residual uncertainty to determine the upper and lower bounds of the temperature.

[0162] Under the aforementioned operating conditions, the system records the out-of-range fields, out-of-range magnitudes, rollback model versions, and expanded uncertainties of the model input, and does not shorten the pressure holding time based on the output of the machine learning residual model.

[0163] 9. Remaining time cannot meet the verification conditions

[0164] The control logic test is performed using data that shows the pressure is consistently below the common pressure range and includes multiple calculable pressure drop events, so that the limited processing unit still has remaining processing capacity when it approaches the maximum holding time of the batch.

[0165] When the sum of the current effective processing capacity lower limit and the maximum increase processing capacity within the remaining allowable holding time is less than the target processing capacity, the system will set the restricted processing unit to an unacceptable state and output a process unacceptable instruction, a manual review instruction, or a batch isolation instruction.

[0166] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A machine learning-based adaptive control method for holding pressure time of multiple HPP products, characterized in that, include: The system acquires verified process parameters for each product category within the current HPP batch and establishes product processing units based on product category and loading location. It collects pressure, medium temperature, stage status, and sensor status data to form continuous time windows, determining the pressure and medium temperature ranges for each time window. Using a physical temperature model and a machine learning residual model, it determines the lower and upper bounds of the equivalent processing temperature for each product processing unit. Based on the minimum processing contribution rate of the processing contribution function within the specified range, it accumulates the lower bound of the effective processing volume and identifies the product processing unit with the lowest lower bound processing volume achievement rate as the restricted processing unit. The system outputs a request to end the pressure holding process when all product processing units continuously meet the standards, the upper temperature limit is not exceeded, the batch is reliable, and the shortest pressure holding time is reached; otherwise, it outputs a command to not end the pressure holding process.

2. The method according to claim 1, characterized in that, The process of obtaining empirical process parameters for each product category within the current HPP batch and establishing product processing units based on product category and loading location includes: obtaining the target pressure range, minimum platform pressure, processing contribution function, target throughput, minimum holding time, maximum holding time, upper limit of equivalent processing temperature, and batch processing permission relationship for each product category; determining the common pressure range of the target pressure range for each product category, the maximum value of the minimum holding time, and the minimum value of the maximum holding time; and outputting a batch processing prohibition instruction when the common pressure range is empty, the maximum value is greater than the minimum value, or there is no batch processing permission relationship.

3. The method according to claim 2, characterized in that, The process of establishing product processing units based on product category and loading location includes: forming an association record between products and loading locations based on product category, packaging form, product quality, loading basket number, loading layer, axial position, and radial position; identifying products with the same product category and the same loading location partition as a product processing unit; and associating each product processing unit with corresponding thermophysical parameters, packaging heat transfer parameters, processing contribution function, target processing volume, and loading difference uncertainty.

4. The method according to claim 1, characterized in that, The collected pressure, medium temperature, stage status, and sensor status form a continuous time window. Determining the pressure range and medium temperature range for each time window includes: determining the pressure range based on pressure measurements, pressure sensor calibration errors, measurement differences between pressure sensors, and data continuity; and determining the medium temperature range based on medium temperature measurements, temperature sensor calibration errors, measurement differences between temperature measuring points, and data continuity. When the data missing ratio, timestamp continuity, sensor status, or stage status consistency does not meet the corresponding reliability conditions, the corresponding time window is set as an unreliable window.

5. The method according to claim 1, characterized in that, The process of determining the equivalent processing temperature lower bound and equivalent processing temperature upper bound of each product processing unit using a physical temperature model and a machine learning residual model includes: calculating the physical predicted temperature based on the medium temperature, pre-batch product temperature, pressure change rate, product equivalent heat capacity, packaging heat exchange parameters, and pressure adiabatic temperature rise parameters; and inputting the product category, packaging form, loading location, pressurization time, pressure change rate, medium temperature change, and pressurization status into the machine learning residual model to obtain the temperature residual correction amount constrained by the residual boundary.

6. The method according to claim 5, characterized in that, The determination of the equivalent processing temperature lower bound and the equivalent processing temperature upper bound of each product processing unit includes: adding the physical predicted temperature to the temperature residual correction amount to obtain an estimated equivalent processing temperature; determining the total temperature uncertainty based on the temperature sensor measurement uncertainty, machine learning residual uncertainty, and loading difference uncertainty; subtracting the total temperature uncertainty from the estimated equivalent processing temperature to obtain the equivalent processing temperature lower bound, and adding the total temperature uncertainty to the estimated equivalent processing temperature to obtain the equivalent processing temperature upper bound.

7. The method according to claim 6, characterized in that, The machine learning residual uncertainty is determined by the residual quantile between the measured product temperature and the estimated equivalent processed temperature in the independent calibration datasets separated according to the HPP processing cycle. When the model input approaches the boundary of the validated applicable range, the measurement difference between sensors increases, or the model is in a pressurization or pressure compensation state, the corresponding uncertainty is increased. When the model input exceeds the validated applicable range, the temperature residual correction is set to zero, and shortening the holding time based on the machine learning residual model is prohibited.

8. The method according to claim 1, characterized in that, The step of accumulating the lower bound of effective processing volume based on the minimum processing contribution rate of the processing contribution function within the interval includes: within the applicable range where the processing contribution function is verified to increase monotonically with pressure and equivalent processing temperature, inputting the lower bound of the pressure interval and the lower bound of the equivalent processing temperature into the processing contribution function using a conservative parameter set to obtain the minimum processing contribution rate; accumulating the lower bound of the effective processing volume of each product processing unit according to the minimum processing contribution rate, time window duration, window confidence flag, and window inclusion coefficient, and performing temperature limit judgment using the upper bound of the equivalent processing temperature.

9. The method according to claim 1, characterized in that, The step of identifying the product processing unit with the lowest lower limit processing volume achievement rate as the restricted processing unit includes: calculating the ratio of the effective processing volume lower limit of each product processing unit to the corresponding target processing volume; determining the achievability based on the remaining processing volume of the current restricted processing unit, the verified maximum allowable processing contribution rate, and the remaining allowable holding time; outputting a process non-compliance instruction when the sum of the effective processing volume lower limit and the maximum increase in processing volume within the remaining allowable holding time is less than the corresponding target processing volume; outputting a manual review instruction when it is necessary to confirm the reason for non-compliance; and outputting a batch isolation instruction when the processing compliance of the corresponding product processing unit cannot be confirmed.

10. The method according to claim 1, characterized in that, The step of outputting an end-of-pressure-holding request when all product processing units continuously meet the standards, the upper limit of the temperature is not exceeded, the batch is reliable, and the minimum holding time is reached includes: verifying the lower limit processing volume achievement rate, equivalent processing temperature upper limit, pressure status, and batch reliability status of all product processing units within a continuously preset number of reliable time windows; sending a pre-end-of-pressure-holding request to the HPP equipment controller when the end conditions are met; freezing the batch calculation data and outputting an end-of-pressure-holding request after receiving confirmation from the safety interlock; outputting a continue-of-pressure-holding instruction when the end conditions are not met, provided that the batch is reliable and there are no process abnormalities, only the processing volume has not yet met the standards or the minimum holding time has not yet been reached; outputting a standard process instruction when the adaptive judgment condition fails but the equipment and the verified standard process allow continued processing; outputting an abnormality handling request when the temperature exceeds the limit, the pressure abnormality cannot be confirmed, or the process cannot meet the standards; and outputting a manual review instruction or a batch isolation instruction when manual confirmation of the cause of the abnormality is required or the compliance of product processing cannot be confirmed.

Images