Preparation method of whole-component enzyme-added fermented meal replacement powder
By constructing an integral component feature matrix and real-time state deviation correction, the problem of incompatible enzyme addition during the fermentation process of meal replacement powder was solved, thereby achieving stability of the fermentation process and improving the conversion rate of dietary fiber, ensuring the consistency of product quality.
Patent Information
- Application Number
- CN202611451400.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-18
- Publication Date
- 2026-11-17
AI Technical Summary
During the preparation of meal replacement powder, batch differences in raw materials can make it difficult to adapt the fixed enzyme addition, fixed ratio, and fixed termination to the actual fermentation state. This can lead to enzyme addition being too early or too late, or termination being delayed, resulting in insufficient or excessive degradation of certain types of substrates.
By constructing a feature matrix of the whole component, enzyme addition time nodes and fermentation termination time thresholds are generated. Combining near-infrared spectral parameters and online viscosity parameters, a material measured state sequence is formed. The state deviation is compared and the enzyme addition time nodes and termination time thresholds are corrected. Single enzyme dynamic reconstruction ratio and network structure collapse index correction are adopted to ensure that the enzyme addition ratio and termination time are adapted to the actual fermentation state.
It achieves stability in the fermentation process of meal replacement powder and improves the conversion rate of soluble dietary fiber, maintains the target gel-state sensory range, enhances the process's resistance to interference, and avoids uneven substrate degradation caused by adding enzymes at a fixed ratio.
Smart Images

Figure CN123100759A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of online detection and computer data control technology for food fermentation, specifically relating to a method for preparing whole-component enzyme-fermented meal replacement powder. Background Technology
[0002] Meal replacement powders are typically composed of a blend of grains, legumes, fruits and vegetables, and medicinal and edible ingredients. There are batch-to-batch variations in starch content, protein content, dietary fiber content, polysaccharide content, water activity, degree of gelatinization, and particle size distribution among grain powders, legume powders, fruit and vegetable powders, and medicinal and edible powders. Conventional meal replacement powder preparation processes generally involve raw material pretreatment, mixing, sterilization, and packaging. Some processes also incorporate enzymes at the powder or slurry stage to induce enzymatic degradation of some macromolecules, thereby improving reconstitution properties, rheological characteristics, and post-fermentation material transformation.
[0003] When the entire set of raw materials enters the fermentation or enzymatic fermentation process together, the indicators of a single raw material are difficult to stably represent the state of the mixture. If enzymes are added or fermentation is terminated at a fixed time, batch-to-batch variations in the carbon-to-nitrogen ratio, water activity, starch gelatinization degree, and initial macromolecular content will cause deviations between the enzymatic hydrolysis rate and the fermentation process, leading to problems such as premature or late enzyme addition, delayed or premature termination. Furthermore, the amylase, cellulase, and protease in the entire enzyme group do not target the same substances. If the entire enzyme group is pumped at a fixed mixing ratio, when the real-time degradation deviations of starch, protein, and polysaccharide substrates are inconsistent, adding enzymes at a fixed ratio can easily result in insufficient degradation of one type of substrate and excessive degradation of another.
[0004] Near-infrared spectroscopy can reflect the absorption characteristics of starch, protein, polysaccharides, and water in materials, while online viscosity can reflect the rheological state formed after starch gelatinization, protein degradation, polysaccharide hydration, and changes in the dietary fiber network. However, relying solely on near-infrared spectral parameters and online viscosity parameters for display or alarm purposes cannot resolve the discrepancy between model-predicted trajectories and actual fermentation states. Especially in high-fiber formulations, the spatial network formed by dietary fiber and amylose can lead to sudden viscosity changes. Judging solely by apparent viscosity or a fixed termination time makes it difficult to promptly identify network collapse caused by excessive cellulase activity. Therefore, in the fermentation process of whole-component meal replacement powder materials, it is crucial to establish a closed loop that integrates raw material physicochemical indicators, substrate degradation trajectories, near-infrared band deviations, online viscosity shear response, and actuator commands, ensuring that enzyme addition timing, single-enzyme ratios, and termination actions are all driven by the material state. Summary of the Invention
[0005] The purpose of this invention is to provide a method for preparing whole-component enzyme-added fermented meal replacement powder, which solves the technical problem that the fixed enzyme addition, fixed ratio and fixed termination are difficult to adapt to the actual fermentation state due to batch fluctuations of raw materials in whole-component meal replacement powder.
[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:
[0007] The method for preparing whole-component enzyme-fermented meal replacement powder includes:
[0008] Physicochemical indicators of grain powder, bean powder, fruit and vegetable powder and medicinal and edible powder were collected before and before the mixed powder entered fermentation. The whole component feature matrix was constructed and the core feature vector of the whole component was extracted.
[0009] The core feature vector of the whole group is input into the fermentation state prediction model, and combined with the substrate degradation kinetic trajectory, the whole group of enzyme addition time nodes and fermentation termination time threshold are generated.
[0010] Near-infrared spectral parameters and online viscosity parameters were collected during fermentation to generate a material measured state sequence.
[0011] The measured state sequence of materials is compared with the degradation kinetic trajectory of the substrate to form a state deviation. The enzyme addition time nodes and fermentation termination time threshold of the whole set are corrected based on the state deviation. The enzyme addition time nodes of the whole set include cellulase addition time nodes that can be corrected independently.
[0012] The enzyme addition valve is triggered based on the revised enzyme addition time node, and the cooling and enzyme inactivation command is triggered based on the revised fermentation termination time threshold, thus obtaining the whole-group enzyme-added fermented meal replacement powder.
[0013] As a preferred technical solution of the present invention, the component feature matrix includes carbon-nitrogen ratio, water activity, starch gelatinization degree, initial starch content, initial protein content, initial dietary fiber content, initial polysaccharide content, and raw material powder particle size distribution. The aforementioned indicators are respectively obtained from raw material acceptance testing, testing after cooking and drying, testing before mixing, and testing before fermentation.
[0014] As a preferred technical solution of the present invention, the core feature vector of the whole component is obtained by standardization and principal component load mapping of the whole component feature matrix, and the principal components whose cumulative contribution rate reaches the preset contribution condition are retained.
[0015] As a preferred technical solution of the present invention, the process of obtaining the whole group core feature vector includes: using the historical qualified batch index mean vector and the historical qualified batch index standard deviation vector to perform element-wise standardization processing on the whole group original index vector of the production batch, and then multiplying the standardization processing result with the transpose of the principal component loading matrix to obtain the whole group core feature vector of the corresponding production batch.
[0016] As a preferred technical solution of the present invention, the fermentation state prediction model includes a feature mapping sub-model, a substrate degradation sub-model, and an instruction timing sub-model. The feature mapping sub-model receives the whole set of core feature vectors and outputs the initial state parameters of the substrate. The substrate degradation sub-model generates the substrate degradation kinetic trajectory based on the initial state parameters of the substrate. The instruction timing model outputs the whole set of enzyme addition time nodes and fermentation termination time thresholds based on the substrate degradation kinetic trajectory.
[0017] As a preferred technical solution of the present invention, the substrate degradation kinetic trajectory includes starch degradation trajectory, protein degradation trajectory and dietary fiber degradation trajectory, the enzyme addition time node of the whole set is the time point corresponding to when at least one of the aforementioned trajectories enters the restriction interval, and the fermentation termination time threshold is the time point corresponding to when the material conversion completion index meets the termination condition.
[0018] As a preferred technical solution of the present invention, the material conversion completion index is obtained by weighting and summing the conversion completion ratios of various substrates included in the evaluation according to their corresponding substrate weights, and then subtracting the viscosity deviation penalty term. The viscosity deviation penalty term is determined by the viscosity deviation penalty coefficient and the difference between the measured online viscosity value and the predicted theoretical viscosity value.
[0019] As a preferred technical solution of the present invention, the near-infrared spectral parameters include absorption characteristics related to starch, protein, polysaccharide and moisture, and the online viscosity parameters include apparent viscosity, viscosity change rate and shear response change. The near-infrared spectral parameters and the online viscosity parameters are time-aligned to form a material measured state sequence.
[0020] As a preferred technical solution of the present invention, the state deviation includes spectral state deviation, viscosity state deviation and fusion state deviation. The fusion state deviation is used to advance, maintain or delay the enzyme addition time node of the whole group, and to advance, maintain or delay the fermentation termination time threshold.
[0021] As a preferred technical solution of the present invention, the whole enzyme group includes amylase, cellulase and protease. After the enzyme addition time node triggers the enzyme addition valve action command, the enzyme addition valve adds the whole enzyme group to the fermenter in a segmented manner. The amount of enzyme added in a single segment is determined by the predicted amount of remaining substrate and the real-time state deviation.
[0022] As a preferred embodiment of the present invention, the cereal powder is obtained by selecting, washing, steaming, drying and grinding at least one of oats, quinoa, millet and brown rice; the bean powder is obtained by selecting, washing, steaming, drying and grinding at least one of black beans, red beans and mung beans; the fruit and vegetable powder is obtained by selecting, washing, dicing, steaming, drying and grinding at least one of dehydrated pumpkin granules and dehydrated sweet potato granules; and the medicinal and edible powder is obtained by selecting, washing, dicing, steaming, drying and grinding at least one of yam, wolfberry and polygonatum.
[0023] As a preferred technical solution of the present invention, after the cooling and enzyme inactivation command is triggered, the fermenter temperature control unit adjusts the material temperature to the enzyme inactivation process temperature range and maintains it until the enzyme activity is lower than the termination criterion. Subsequently, the material is dried, crushed, microwave sterilized and packaged to form a whole-component enzyme-added fermented meal replacement powder product.
[0024] This invention also discloses a whole-component enzyme-added fermented meal replacement powder preparation system, used to execute the whole-component enzyme-added fermented meal replacement powder preparation method described above. The system includes a whole-component index acquisition module, a feature matrix construction module, a fermentation state prediction module, an online state detection module, an error compensation control module, an enzyme addition execution module, a cooling and enzyme inactivation module, and a finished product processing module. The whole-component index acquisition module receives physicochemical indicators from the raw material pretreatment production line and the mixed fermentation tank and outputs them to the feature matrix construction module. The feature matrix construction module generates a whole-component core feature vector and outputs it to the fermentation state prediction module. The fermentation state prediction module outputs the whole-component enzyme addition time node and the fermentation termination time threshold. The data is sent to the error compensation control module; the online status detection module receives real-time detection data from the near-infrared probe and the online viscometer and outputs the material's measured status sequence; the error compensation control module corrects the enzyme addition and termination commands based on the material's measured status sequence and outputs them to the enzyme addition execution module and the cooling enzyme inactivation module, respectively; the enzyme addition execution module includes metering pumps and enzyme addition valves connected to the amylase storage tank, cellulase storage tank, and protease storage tank, respectively, and receives the single-enzyme independent pulse injection command output by the error compensation control module; the cooling enzyme inactivation module receives the cooling enzyme inactivation command and adjusts the fermenter temperature; the finished product processing module receives the material after cooling enzyme inactivation and completes drying, pulverization, microwave sterilization, and packaging.
[0025] As a preferred embodiment of the present invention, the amount of enzyme added in a single stage is determined by the predicted amount of remaining substrate and the real-time state deviation, including the independent dynamic reconstruction of the addition ratios of amylase, cellulase, and protease in the entire enzyme group; the spectral state deviation is decomposed into starch characteristic band deviation, protein characteristic band deviation, and polysaccharide characteristic band deviation; for the single-stage enzyme addition process, an enzyme addition ratio vector matrix is set, and the weights of amylase ratio, cellulase ratio, and protease ratio in the enzyme addition ratio vector matrix are dynamically reconstructed according to the following formula:
[0026] ;
[0027] in, Indicates the fermentation time Time target enzyme Dynamic reconfiguration of allocation weights, Indicates the fermentation time. Indicates the target enzyme type and is selected from amylase, cellulase, and protease. Indicates the target enzyme at the fermentation initiation point The initial proportion weights, This represents the natural exponential function. This represents the substrate sensitivity scaling factor. Indicates the fermentation time Time target enzyme Spectral state deviation of the corresponding substrate characteristic band, Indicates the fermentation time Time target enzyme The corresponding predicted amount of remaining substrate; the single-stage enzyme addition process is based on the target enzyme. The dynamic reconstructed ratio weights issue independent pulse injection commands to each single enzyme in the fermenter; the spectral state deviation and the predicted amount of remaining substrate entering the formula are both normalized dimensionless quantities, the substrate sensitivity scaling factor is a dimensionless coefficient, so the exponential term is a dimensionless quantity, and the dynamic reconstructed ratio weights are still ratio weights.
[0028] As a preferred embodiment of the present invention, the change in shear response in the online viscosity parameter is calculated from the transient shear stress decay characteristics under a double-step shear rate in the fermenter. The change in shear response in the measured material state sequence is converted into the network structure collapse index according to the following formula, which is used to specifically correct the cellulase addition time node in the entire set of enzyme addition time nodes:
[0029] ;
[0030] in, Indicates the fermentation time Network structure collapse index at time, Indicates the fermentation time. This represents the stress relaxation time constant. This indicates the time window for the double step shear test. This represents the integral time variable within the time window of the double-step shear test. Indicates the high step shear rate. Indicates the low step shear rate. Indicates the variable during integration time. High step shear rate Transient shear stress fed back by the material. Indicates the variable during integration time. Low step shear rate Transient shear stress fed back by the material. Represents the integral time variable Find the rate of change. Represents an integral infinitesimal element; when When the stress exceeds the network collapse threshold, a delay instruction is triggered for the cellulase addition time node; the transient shear stress entering this formula is first normalized according to the initial peak value within the same test cycle, and the unit of the network structure collapse index is seconds. -1 The network collapse threshold uses the same unit.
[0031] Compared with the prior art, the present invention has the following beneficial effects:
[0032] This invention employs a whole-group feature matrix construction and fermentation state prediction model to transform batch differences in cereal powder, bean powder, fruit and vegetable powder, and medicinal and edible powder into calculable inputs. The fermentation state prediction model, combined with substrate degradation kinetic trajectories, generates a set of enzyme addition time nodes and fermentation termination time thresholds, avoiding the influence of raw material batch fluctuations on fixed-sequence control. During fermentation, near-infrared spectral parameters and online viscosity parameters form a measured material state sequence, which is compared with the substrate degradation kinetic trajectory to obtain state deviations. The error compensation control module corrects the whole-group enzyme addition time nodes and fermentation termination time thresholds based on the state deviations, ensuring that the addition and termination times are adaptable to the actual fermentation state. Dynamic reconstructed ratios of single enzymes allow amylase, cellulase, and protease to adjust their addition ratios according to the spectral state deviations and predicted remaining substrate amounts, avoiding insufficient or excessive degradation of certain substrates due to fixed enzyme ratios. The network structure collapse index, calculated from the transient shear stress decay characteristics under double-step shear rates, is used to independently correct the cellulase addition time nodes, limiting excessive cellulase activity before irreversible damage to the dietary fiber network. The construction of the whole component feature matrix, the correction of closed-loop state deviation, the dynamic reconstruction ratio of single enzymes and the correction of network structure collapse index jointly improve the stability of meal replacement powder preparation and the conversion rate of soluble dietary fiber, maintain the target gel state sensory range, and enhance the process's anti-interference ability. Attached Figure Description
[0033] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.
[0034] Figure 1 This is the main flowchart of the method of the present invention.
[0035] Figure 2 This is a flowchart of the single-enzyme dynamic reconstruction ratio subprocess of the present invention.
[0036] Figure 3 This is a flowchart of the network structure collapse index correction sub-process of the present invention.
[0037] Figure 4 This is a simplified flowchart of the overall process of the present invention. Detailed Implementation
[0038] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the embodiments of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.
[0039] The following is in conjunction with the appendix Figures 1-4 The embodiments of the present invention will be described in detail below.
[0040] This invention provides a method for preparing whole-component enzyme-fermented meal replacement powder. In specific implementation, after raw materials are inspected, they enter their respective production lines. Grain raw materials such as oats, quinoa, millet, and brown rice are selected, washed, steamed, dried, and ground. Legume raw materials such as black beans, red beans, and mung beans are selected, washed, steamed, dried, and ground. Fruit and vegetable raw materials such as dehydrated pumpkin and sweet potato granules are selected, washed, diced, steamed, dried, and ground. Medicinal and edible raw materials such as yam, goji berries, and polygonatum are selected, washed, diced, steamed, dried, and ground. Goji berries, when meeting the feed particle size requirements, directly enter the washing and steaming processes. After processing, the various raw material powders are mixed, and the mixture enters the enzyme-fermentation control process. After fermentation, the mixture is dried, pulverized, microwave-sterilized, and packaged.
[0041] The specific steps are as follows:
[0042] Step S1: Collect the physicochemical indicators of the whole component and construct the feature matrix of the whole component;
[0043] In practice, the physicochemical indicators include the carbon-to-nitrogen ratio, water activity, starch gelatinization degree, initial starch content, initial protein content, initial dietary fiber content, initial polysaccharide content, and raw material powder particle size distribution. The carbon-to-nitrogen ratio indicates the availability of carbon and nitrogen sources, water activity indicates the availability of water during fermentation and enzymatic reactions, starch gelatinization degree indicates the initial degree to which starch can be acted upon by amylases, and initial macromolecular content indicates the target group for enzyme action. After collection, the indicators are arranged into a whole-group feature matrix according to batch, raw material category, and indicator category, so that the detection data from different sources and different stages are incorporated into a unified data structure.
[0044] Step S2: Formation of core feature vectors for the entire group;
[0045] In practice, the integral group feature matrix is standardized and principal component loadings are mapped to obtain the integral group core feature vector. A mean vector and standard deviation vector are constructed using historical qualified batches. Principal components are selected and retained according to a preset cumulative contribution rate, for example, a preset cumulative contribution rate of 85%-95%, discarding the remaining principal components so that the cumulative contribution rate corresponding to the retained principal components meets the preset contribution condition. If a certain indicator is affected by seasonal or geographical fluctuations in raw materials, the mean and standard deviation can be updated when the batch is archived. The integral group core feature vector is calculated using the following formula:
[0046] ;
[0047] in, Indicates the production batch number. Indicates the first The core feature vector of each production batch. Represents the principal component loading matrix. This represents the matrix transpose operation. Indicates the first The original index vector of the whole group of each production batch, This represents the vector of historical average indicators for qualified batches. This represents the standard deviation vector of historical qualified batch indicators. This indicates element-wise division. The formula first standardizes the original index vector, and then maps it to a low-dimensional feature space using the principal component loading matrix, thus avoiding the direct inclusion of indices with different dimensions into the fermentation state prediction model.
[0048] Step S3: Establishment of fermentation state prediction model;
[0049] In practical implementation, the fermentation state prediction model includes a feature mapping sub-model, a substrate degradation sub-model, and a command timing sub-model. The feature mapping sub-model receives the entire set of core feature vectors and outputs the initial substrate state parameters, namely the initial relative degradability of starch, protein, and dietary fiber. The substrate degradation sub-model generates the substrate degradation kinetic trajectory based on the initial substrate state parameters and outputs the predicted theoretical viscosity value. The command timing model reads the substrate degradation kinetic trajectory and outputs the entire set of enzyme addition time points and fermentation termination time thresholds.
[0050] In its implementation, the feature mapping sub-model includes an input normalization layer, a fully connected mapping layer, a nonlinear activation layer, and a substrate initial state output layer. The input normalization layer receives the entire set of core feature vectors; the fully connected mapping layer maps these core feature vectors to intermediate substrate state vectors; the nonlinear activation layer performs a nonlinear transformation on the intermediate substrate state vectors; and the substrate initial state output layer outputs the initial relative degradable states of starch, protein, and dietary fiber. The substrate degradation sub-model includes a substrate state input layer, a trajectory recursion layer, a trajectory output layer, and a viscosity prediction output layer. The substrate state input layer receives the initial substrate state parameters; the trajectory recursion layer updates the predicted remaining quantities of various substrates according to the fermentation time step; the trajectory output layer outputs the starch degradation trajectory, protein degradation trajectory, and dietary fiber degradation trajectory; and the viscosity prediction output layer outputs the theoretical viscosity prediction value. The instruction time point sub-model includes a restriction interval judgment layer, a termination condition judgment layer, and an instruction output layer. The restriction interval judgment layer reads whether the degradation trajectory of various substrates has entered the enzyme addition restriction interval, the termination condition judgment layer reads whether the substance conversion completion index meets the termination condition, and the instruction output layer outputs the enzyme addition time node and fermentation termination time threshold of the whole set of enzymes.
[0051] In some preferred embodiments, historical qualified batch data includes pre-fermentation physicochemical indicators, near-infrared spectral parameters during fermentation, online viscosity parameters, enzyme addition execution records, enzyme inactivation execution records, and finished product qualification judgment results. This historical qualified batch data is used to calibrate the principal component loading matrix, substrate weights, viscosity bias penalty coefficient, substrate sensitivity scaling factor, network breakdown threshold, near-infrared channel bias weights, and viscosity channel bias weights. The calibration process aims at improving the finished product's reconstitution stability, soluble dietary fiber conversion rate, and material conversion completion index, using methods such as grid search or regression to determine the aforementioned coefficients and thresholds. In offline detection labels, starch can be measured using enzymatic hydrolysis, protein using the Kjeldahl method, and dietary fiber using enzymatic gravimetric analysis. Abnormal batches, such as those with raw material indicators exceeding the historical mean ± 3 standard deviations or those with continuous sensor anomalies, are removed. When the number of newly added qualified batches reaches a preset number, such as 20 batches, the mean vector, standard deviation vector, and principal component loading matrix are updated. For example, training data can use 80 consecutive qualified production batches, and validation data can use 20 consecutive qualified production batches. The input for each batch includes the whole-component core feature vector, spectral state deviation during fermentation, online viscosity parameters, and actuator action records. The supervision labels for each batch include the remaining starch percentage, remaining protein percentage, remaining dietary fiber percentage, enzyme addition execution time, fermentation termination time, and finished product reconstitution stability index obtained from offline detection. During training, the standardized input parameters are first fixed, then the feature mapping sub-model and substrate degradation sub-model are trained. Finally, the constraint range and termination conditions of the instruction timing sub-model are adjusted based on the validation data. After model training is complete, the model parameters are stored in the industrial control computer, and only forward calculations and online deviation corrections are performed during runtime.
[0052] Step S4: Formation of the material's measured state sequence and state deviation;
[0053] In practice, near-infrared probes collect absorption characteristics related to starch, protein, polysaccharides, and moisture, while online viscometers collect apparent viscosity, viscosity change rate, and shear response change. Near-infrared spectral parameters and online viscosity parameters are time-aligned to form a measured material state sequence. The measured material state sequence is compared with the substrate degradation kinetic trajectory to obtain the state deviation. The state deviation is then fused according to the following formula:
[0054] ;
[0055] in, Indicates the fermentation time The deviation of the fusion state at that time Indicates the fermentation time. Indicates the near-infrared channel bias weight. Indicates the fermentation time Spectral state deviation at that time Indicates the viscosity channel deviation weight. Indicates the fermentation time Viscosity state deviation at that time. The near-infrared channel deviation weight and viscosity channel deviation weight are calibrated by the correspondence between spectral deviation, viscosity deviation and termination quality index in historical qualified batches. The aforementioned deviations and weights are all normalized dimensionless quantities to ensure dimensional consistency in the fusion calculation.
[0056] If the spectral status indicates that substrate degradation is slower than the substrate degradation kinetic trajectory and the viscosity decreases slower than the theoretical viscosity prediction, the error compensation control module will trigger subsequent staged enzyme additions earlier and delay the fermentation termination time threshold. If the spectral status indicates that substrate degradation is faster than the substrate degradation kinetic trajectory and the viscosity decreases faster than the theoretical viscosity prediction, the error compensation control module will delay subsequent staged enzyme additions or reduce the amount of enzyme added in subsequent stages, and advance the fermentation termination time threshold. Through this processing, the predicted trajectory is no longer just used as a fixed control plan, but as a benchmark for real-time feedback correction.
[0057] In some preferred embodiments, the sensor jump determination criteria can be set as follows: a single jump in near-infrared absorbance exceeds the mean of the sliding window by ±3 times the standard deviation, or exceeds a preset fluctuation range for 5 consecutive sampling cycles; the channel weight is restored after stabilization for 10 consecutive sampling cycles. This allows the error compensation control module to reduce the weight of abnormal channels when a sensor jump occurs, and restore the weight after stabilization following continuous sampling.
[0058] Step S5: Enzyme addition and cooling / enzyme inactivation commands are triggered;
[0059] In practical implementation, when the corrected time point for adding the entire enzyme group is reached, the error compensation control module outputs an enzyme addition valve action command. The enzyme addition valve adds the entire enzyme group to the fermenter in a segmented manner. The amount of enzyme added in a single segment is determined by the predicted amount of remaining substrate and the real-time state deviation. In some preferred embodiments, the amount of enzyme added in a single segment starts from the base amount of enzyme added, is allocated according to the predicted remaining proportions of starch, protein, and dietary fiber, and is increased, maintained, or decreased according to the fusion state deviation. When the corrected fermentation termination time threshold is reached and the material conversion completion index meets the termination condition, the error compensation control module outputs a cooling and enzyme inactivation command. The material conversion completion index is calculated according to the following formula:
[0060] ;
[0061] in, Indicates the fermentation time. Indicates the fermentation time The material transformation completion index at that time This indicates the total number of substrate categories included in the evaluation. Indicates the substrate category number, Indicates the first Substrate weights Indicates the first Substrate fermentation time The relative content at that time Indicates the first The relative content of the substrate at the fermentation initiation point, This represents the viscosity deviation penalty coefficient. Indicates the fermentation time The measured online viscosity value after normalization. Indicates the fermentation time The theoretical viscosity prediction value after being normalized to the same scale. Because... , , and All calculations are performed using relative or normalized values. This is a dimensionless indicator. The cooling and enzyme inactivation module adjusts the material temperature to the enzyme inactivation process temperature range and maintains it until the enzyme activity is below the termination criterion. The intermediate material after fermentation is dried, pulverized, microwave-sterilized, and packaged to form the finished product of enzyme-added fermented meal replacement powder.
[0062] Step S6: Dynamic reconstruction ratio of single enzymes and inhibition of network structure collapse;
[0063] In specific implementation, the error compensation control module further independently and dynamically reconstructs the addition ratios of amylase, cellulase, and protease within a single enzyme addition process. After preprocessing, the diffuse reflectance spectral data acquired by the near-infrared probe is decomposed into starch characteristic band deviation, protein characteristic band deviation, and polysaccharide characteristic band deviation. In some preferred embodiments, near-infrared spectral preprocessing includes multivariate scattering correction and second derivative. The starch characteristic bands can be 1450 nm and 1940 nm, the protein characteristic bands can be 2050 nm and 2180 nm, and the polysaccharide characteristic bands can be 2280 nm and 2340 nm. The starch characteristic band deviation corresponds to the amylase channel, the protein characteristic band deviation corresponds to the protease channel, and the polysaccharide characteristic band deviation corresponds to the cellulase channel. For the single enzyme addition process, the error compensation control module sets an enzyme addition ratio vector matrix. The weights of the amylase, cellulase, and protease ratios in the enzyme addition ratio vector matrix are dynamically reconstructed according to the following formula:
[0064] ;
[0065] in, Indicates the fermentation time. Indicates the fermentation time Time target enzyme Dynamic reconfiguration of allocation weights, Indicates the target enzyme type and is selected from amylase, cellulase, and protease. Indicates the target enzyme at the fermentation initiation point The initial proportion weights, This represents the natural exponential function. This represents the substrate sensitivity scaling factor. Indicates the fermentation time Time target enzyme Spectral state deviation of the corresponding substrate characteristic band, Indicates the fermentation time Time target enzyme The corresponding predicted residual substrate amount. Both the spectral state bias and the predicted residual substrate amount used in this formula are normalized dimensionless quantities. The substrate sensitivity scaling factor is a dimensionless coefficient; therefore, the exponential term is dimensionless, and the dynamic reconstruction ratio weights remain the same. After dynamic reconstruction is completed, the error compensation control module normalizes the dynamic reconstruction ratio weights of each target enzyme and converts the normalization result into independent pulse injection commands for each single enzyme.
[0066] In practical implementation, the change in shear response in the online viscosity parameters is calculated from the transient shear stress decay characteristics under a double-step shear rate within the fermenter. The online state detection module uses a coaxial cylindrical rotational viscometer installed in the fermenter's bypass pipeline to acquire rheological data. The bypass pipeline is equipped with a cam rotor pump with constant pressure output characteristics, reducing the influence of the shear field from the fermenter's main agitator on the material entering the test chamber. The change in shear response in the measured material state sequence is converted into the network structure collapse index according to the following formula:
[0067] ;
[0068] in, Indicates the fermentation time. Indicates the fermentation time Network structure collapse index at time, This represents the stress relaxation time constant. This indicates the time window for the double step shear test. This represents the integral time variable within the time window of the double-step shear test. Indicates the high step shear rate. Indicates the low step shear rate. Indicates the variable during integration time. High step shear rate Transient shear stress fed back by the material. Indicates the variable during integration time. Low step shear rate Transient shear stress fed back by the material. Represents the integral time variable Find the rate of change. This represents an integral infinitesimal element. The transient shear stress entering this formula is first normalized to the initial peak value within the same test period. The unit of the network structure collapse index is s. -1 The network collapse threshold uses the same unit. The network structure collapse index is used to specifically correct the cellulase addition time node in the entire set of enzyme addition time nodes. When the network structure collapse index is greater than the network collapse threshold, the error compensation control module generates a delay instruction for the cellulase addition time node, thus restricting the cellulase channel in subsequent addition cycles.
[0069] Example 1: In this example, 100 kg of mixed powder was prepared, consisting of grain powder, bean powder, fruit and vegetable powder, and medicinal and edible powder, mixed according to the production formula. The resulting powder had a carbon-to-nitrogen ratio of 18.5, a water activity of 0.82, a starch gelatinization degree of 72%, an initial starch content of 46%, an initial protein content of 18%, an initial dietary fiber content of 12%, an initial polysaccharide content of 6%, and a powder particle size D50 of 90 μm.
[0070] After standardizing the aforementioned indicators, the original indicator vector of the whole group is obtained, and the core feature vector of the whole group is calculated using the principal component loading matrix. After the core feature vector of the whole group is input into the fermentation state prediction model, the fermentation state prediction model generates the enzyme addition time node of 3.6h and the fermentation termination time threshold of 8.2h for the whole group.
[0071] At 3.2 hours of fermentation, near-infrared spectral parameters showed that the decline of starch-related absorption characteristics was slower than the substrate degradation kinetics, and the measured online viscosity was higher than the theoretical viscosity prediction. The error compensation control module corrected the overall enzyme addition time node from 3.6 hours to 3.4 hours and adjusted the single-stage enzyme addition amount from 30% to 35% of the predicted requirement. At 8.0 hours of fermentation, the relative remaining proportions of starch, protein, and dietary fiber were 0.38, 0.44, and 0.52, respectively, with weights of 0.40, 0.35, and 0.25. The viscosity deviation penalty coefficient was 0.20, and the viscosity deviation was 0.03. Based on a starch conversion ratio of 0.62, a protein conversion ratio of 0.56, a dietary fiber conversion ratio of 0.48, and a viscosity deviation penalty of 0.006, the material conversion completion index was calculated to be 0.558. If the termination condition is that the material conversion completion index is not lower than 0.55, the error compensation control module triggers a cooling and enzyme inactivation command.
[0072] Example 2: This example further illustrates the system implementation based on Example 1. In this example, the whole-group index acquisition module is connected to the raw material detection equipment and the fermenter feed detection point. The feature matrix construction module runs on the edge computing node, and the fermentation state prediction module runs on the industrial control computer. Near-infrared probes are installed in the fermenter circulation pipeline or the tank's detection window, and online viscometers are installed in the material circulation branch. The error compensation control module receives the whole-group enzyme addition time nodes and fermentation termination time thresholds output by the fermentation state prediction module, and simultaneously receives the material measured state sequence output by the online state detection module.
[0073] If a sensor reading jumps, the error compensation control module reduces the weight of the abnormal channel and restores the weight after continuous sampling and stabilization. The jump judgment condition can be set to a single jump in near-infrared absorbance exceeding ±3 times the mean of the sliding window, or exceeding the preset fluctuation range for 5 consecutive sampling cycles; the channel weight is restored after 10 consecutive sampling cycles of stabilization. The enzyme addition execution module includes an enzyme storage tank, a metering pump, and an enzyme addition valve, while the cooling enzyme inactivation module includes a heat exchange unit, a temperature detection unit, and an enzyme inactivation maintenance control unit. After receiving the enzyme addition valve action command output by the error compensation control module, the enzyme addition execution module adds the entire group of enzymes to the fermenter in stages through the metering pump and enzyme addition valve; after receiving the cooling enzyme inactivation command, the cooling enzyme inactivation module adjusts the material temperature through the heat exchange unit and maintains the enzyme inactivation process through the temperature detection unit and the enzyme inactivation maintenance control unit.
[0074] Example 3: This example describes a specific product scenario based on Example 1. In this example, the grain raw materials selected are oats, quinoa, millet, and brown rice; the legume raw materials selected are black beans, red beans, and mung beans; the fruit and vegetable raw materials selected are dehydrated pumpkin granules and dehydrated sweet potato granules; and the medicinal and edible raw materials selected are yam, goji berries, and polygonatum. After selection, washing, steaming, drying, and grinding, all raw materials are mixed. The mixed powder undergoes pre-fermentation physicochemical testing before entering the enzyme fermentation process. After fermentation, the material is dried, pulverized, and microwave sterilized before packaging.
[0075] In this embodiment, enzyme addition and termination actions are not based solely on fixed empirical times, but rather on a combination of the whole-component core feature vector, substrate degradation kinetic trajectory, and online status deviation. When the online status deviation indicates that the material conversion process is slower than the substrate degradation kinetic trajectory, the error compensation control module advances or enhances subsequent enzyme addition actions; when the material conversion process is faster than the substrate degradation kinetic trajectory, the error compensation control module delays or weakens subsequent enzyme addition actions and adjusts the fermentation termination time threshold accordingly. This ensures that the fermentation control of a specific batch of meal replacement powder products matches the raw material batch status and the real-time status within the tank.
[0076] Example 4: Based on Example 3, this example addresses the imbalance of multi-enzyme synergistic ratio and spatial network rheological mutation during the fermentation process of high dietary fiber meal replacement powder, and illustrates the specific implementation path of independent dynamic reconstruction ratio and network structure collapse index correction.
[0077] Oats, quinoa, black beans, and dehydrated pumpkin seeds were mixed at a mass ratio of 4:2:2:2. Before the mixture of the above grain powder, bean powder, and fruit and vegetable powder entered the fermentation tank, the whole component index acquisition module acquired the physicochemical properties of the mixed powder. The whole component feature matrix was standardized and principal component loading was applied to output the whole component core feature vector. The fermentation state prediction model generated the substrate degradation kinetic trajectory.
[0078] A near-infrared probe acquires diffuse reflectance spectral data, covering a wavelength range of 1000 nm to 2500 nm. The online state detection module performs multivariate scattering correction and second-derivative preprocessing on the spectral data. Spectral state deviations are decomposed into starch characteristic band deviations, protein characteristic band deviations, and polysaccharide characteristic band deviations. The starch characteristic band deviation is calculated from the absorption peak area deviations at 1450 nm and 1940 nm. The protein characteristic band deviation is calculated from the absorption peak area deviations at 2050 nm and 2180 nm. The polysaccharide characteristic band deviation is calculated from the absorption peak area deviations at 2280 nm and 2340 nm.
[0079] The predicted residual substrate amount is continuously updated as fermentation progresses. The single-stage enzyme addition process no longer uses a fixed mixed enzyme solution for pumping; instead, the amylase, cellulase, and protease tanks in the enzyme addition execution module are each connected to independent high-precision pneumatic diaphragm metering pumps. The enzyme addition ratio vector matrix calculation program is executed by the error compensation control module. Taking a fermentation time of 2.5 hours as an example, the single-stage enzyme addition command is triggered at the overall enzyme addition time node. In the predicted residual substrate amount given by the substrate degradation kinetic trajectory, the predicted residual substrate amount for starch is 0.65, for protein it is 0.78, and for dietary fiber it is 0.82.
[0080] The spectral state deviation output by the online state detection module exhibits discrete characteristics: the calculated deviation value for the starch characteristic band is 0.02, the calculated deviation value for the protein characteristic band is -0.08, and the calculated deviation value for the polysaccharide characteristic band is 0.15. The substrate sensitivity scaling factor is set to 1.2, the initial weighting of amylase is set to 0.40, the initial weighting of cellulase is set to 0.35, and the initial weighting of protease is set to 0.25.
[0081] The error compensation control module calls the calculation model to update the dynamic reconstruction weight ratio. The dynamic reconstruction weight ratio of amylase is calculated according to the weight reconstruction formula, with the exponential term result close to 1.037, and the final amylase reconstruction weight output is 0.415. The spectral state deviation of the protease is negative, indicating that protein degradation is faster than the substrate degradation kinetic trajectory. The exponential term result of the dynamic reconstruction weight ratio of the protease is 0.884, and the final protease reconstruction weight output is 0.221. The characteristic band deviation of polysaccharides shows positive lag, indicating that the encapsulation effect of the dietary fiber network structure hinders substrate degradation. The exponential term result of the dynamic reconstruction weight ratio of cellulase is 1.245, and the final cellulase reconstruction weight output is 0.436.
[0082] After weight calculation, the error compensation control module normalizes the dynamically reconstructed weight ratios of amylase, protease, and cellulase. The normalized amylase, protease, and cellulase ratios are 0.387, 0.206, and 0.407, respectively. Based on the normalization result, the enzyme addition valve action command is decomposed into three independent pulse injection commands: the amylase metering pump frequency is set to 38Hz, the protease metering pump frequency to 20Hz, and the cellulase metering pump frequency to 40Hz. The mechanism for the formation of the material's measured state sequence and state deviation is achieved through a physical closed loop via independently addressed enzyme injection.
[0083] To extract the shear response change from online viscosity parameters, the online state detection module utilizes a coaxial cylindrical rotational viscometer installed in the fermenter's bypass pipeline to acquire rheological data. The bypass pipeline is equipped with a cam rotor pump with constant pressure output characteristics to ensure that the material entering the test chamber is less affected by the shear field of the fermenter's main agitator. The double-step shear test cycle is 2 minutes. Within each test cycle, the rotational speed of the inner cylinder of the rotational viscometer is driven by a programmable servo motor, and the low-step shear rate is set to 15 seconds. -1 The high-step shear rate is set to 120 s. -1 The time window for the double-step shear test was set to 30 seconds. The controller recorded the transient shear stress decay curve triggered at the instant of the step jump, and the stress relaxation time constant was calibrated to 5 seconds.
[0084] Dietary fiber and amylose in the material crosslink to form a complex three-dimensional hydrogen bond network. At high step shear rates, the transient shear stress exhibits a sharp peak followed by rapid decay; at low step shear rates, the transient shear stress decays gradually with partial structural reconstruction. The network collapse index is obtained by integrating the difference in stress decay rates at the two shear rates over time using an integral unit. When cellulase excessively cleaves the hemicellulose backbone, the yield stress of the mixed system decreases rapidly, while the absolute value of the stress decay rate difference increases.
[0085] After 4.2 hours of fermentation, the network structure collapse index calculated from online viscosity parameters reached 8.7 s⁻¹. -1 The network collapse threshold is set to 7.5 seconds. -1 The error compensation control module determines that the macromolecular network framework has entered an irreversible destruction stage. Excessive viscosity loss will lead to stratification and sedimentation when the meal replacement powder is prepared, affecting the stability indicators. At this point, a delayed instruction is generated, and the cellulase addition time node is forcibly blocked by a mask. The cellulase metering pump corresponding to the enzyme addition valve action instruction is locked and does not participate in the action allocation for the subsequent three enzyme addition cycles. The dynamic reconstruction ratio weights of amylase and protease are locally normalized again after removing the cellulase channel and then sent to the enzyme addition execution module.
[0086] The fermenter temperature control unit synchronously reads the material conversion completion index. When the fermentation termination time threshold assessment condition is met, the cooling and enzyme inactivation command is triggered. The jacket cooling water regulating valve opens to 100%. The material temperature decreases at a rate of 2.5℃ / min to the enzyme inactivation process temperature range. The enzyme inactivation holding control unit maintains this temperature range for 15 minutes. Test samples are taken after drying, pulverizing, and microwave sterilization in the finished product processing module. The reconstitution stability is characterized by the transmittance of the supernatant after adding 15g of powder to 150mL of warm water at 60℃ and stirring and letting it stand for 10 minutes. The soluble dietary fiber conversion rate is calibrated by determining the difference in absolute content before and after using the enzyme gravimetric method.
[0087] Table 1 is a comparison table of finished product quality index tests between multiple dynamic correction control and conventional control;
[0088]
[0089] The basic fixed-ratio group did not introduce state bias, the single-spectrum state bias compensation group did not reconstruct the single enzyme ratio, and the dynamic ratio reconstruction combined with collapse index inhibition group used the operating parameters of Example 4. Data comparison reflects the effect of the control strategy on the adjustment of the material microstructure. The fixed-time group had substrate encapsulation residues, low light transmittance, and limited fiber conversion. Dynamic ratio reconstruction reduced the impact of batch-to-batch raw material structure barriers on enzyme addition. The introduction of the network structure collapse index reduced the breakage of the thickening effect caused by excessive enzymatic hydrolysis, maintaining the apparent viscosity of the reconstituted product within the target gel-state sensory range while maintaining the dietary fiber conversion rate.
[0090] The nonlinear mapping formed by the whole component's core feature vector and online rheological feedback overcomes the limitations of traditional meal replacement powder fermentation, which relies on time-series programming. The instruction timing model is no longer a unidirectional output node; the enzyme addition ratio vector matrix transforms the microscopic chemical state into the actuator frequency through a substrate sensitivity scaling factor. The single-stage enzyme addition amount is further extended from total volume control to single-enzyme ratio control targeting specific substrate barriers, enabling near-infrared band deviation, prediction of remaining substrate amount, online shear response, and enzyme addition execution to form a continuous closed loop.
[0091] Comparative Example 1: Comparative Example 1 uses the same raw materials, mixing ratio, fermenter, and enzyme group as Example 1. The only difference is that a whole component feature matrix is not constructed; instead, a fixed enzyme addition time of 3.6 h and a fixed termination time of 8.2 h are used. Because the water activity and starch gelatinization degree in the raw material batch deviate from the historical average, the fixed enzyme addition is later in this batch. Based on the state sequence of Example 1, the material conversion completion index under the fixed time control is 0.51 at 8.2 h, which is lower than the termination condition of 0.55. If termination is still at 8.2 h, the viscosity of the powder after termination will be higher, with a viscosity deviation of 8%. Comparative Example 1 illustrates that the lack of a whole component feature matrix makes it impossible to identify batch differences in fermentation state prediction.
[0092] Comparative Example 2: Comparative Example 2 uses the same raw materials, the same mixing ratio, and the same whole-component feature matrix as Example 1. The only difference is that near-infrared spectral parameters and online viscosity parameters are not collected, and the fermentation state prediction model is run in open loop. The model initially predicts the whole-component enzyme addition time node to be 3.6h and the fermentation termination time threshold to be 8.2h. During fermentation, if the actual viscosity of the material decreases slower than the substrate degradation kinetic trajectory, the open-loop model cannot feed back the difference to the enzyme addition and termination commands. Calculated based on the same batch in Example 1, the material conversion completion index is 0.528 when the open loop is controlled at 8.2h, and 0.558 after closed-loop compensation. Comparative Example 2 illustrates that the lack of dual-channel feedback prevents the predicted trajectory from being corrected according to the actual fermentation state.
[0093] Comparative Example 3: Comparative Example 3 used the same raw materials, fermenter, near-infrared probe, and online viscosity detection conditions as Example 4. The only difference was that the entire enzyme group was still pumped in a fixed proportion in a single stage, without independent dynamic reconstruction of amylase, cellulase, and protease. After 2.5 hours of fermentation, the deviations in the characteristic bands of starch, protein, and polysaccharides were inconsistent. Fixed-proportion pumping could not increase the participation rate of cellulase channels or decrease the participation rate of protease channels. Based on the same batch in Example 4, the soluble dietary fiber conversion rate was 21.4% under fixed-proportion control and 42.8% under dynamic proportion reconstruction control. Comparative Example 3 illustrates that the lack of dynamic reconstruction of single enzyme proportions makes it difficult to adapt the single-stage enzyme dosage to the multi-substrate deviation differentiation state.
[0094] Comparative Example 4: Comparative Example 4 uses the same raw materials, the same single-enzyme dynamic reconstruction ratio algorithm, and the same independent pulse injection structure as Example 4. The only difference is that the network structure collapse index is not calculated, and the cellulase addition time point is not delayed or masked. When the polysaccharide characteristic band deviation is positively hysteretic, the cellulase dynamic reconstruction ratio weight continues to increase, but a large molecular network collapse trend has already appeared in the online shear response. If the cellulase channel continues to be executed, the material is prone to stratification and precipitation after remixing. Calculated based on the same batch in Example 4, the apparent viscosity of the single spectral state deviation compensation group is 1240 mPa·s, and the apparent viscosity of the dynamic ratio reconstruction and collapse index inhibition group is 3100 mPa·s. Comparative Example 4 shows that the lack of a network structure collapse index prevents the cellulase channel from being restricted in a timely manner according to the rheological structure destruction state.
[0095] In the actual production process, after raw material acceptance, grains, legumes, fruits and vegetables, and medicinal and edible raw materials undergo pretreatment respectively. Before the mixed powder enters fermentation, the carbon-to-nitrogen ratio, water activity, degree of gelatinization, starch content, and protein content are measured. The measured values are first standardized according to the mean and standard deviation of historical qualified batches. After standardization, the original index vector of the whole component is multiplied by the principal component loading matrix to obtain the core feature vector of the whole component. The core feature vector of the whole component is not the terminal control command, but the input to the fermentation state prediction model. After reading the core feature vector of the whole component, the fermentation state prediction model provides the initial state parameters of the substrate and generates the substrate degradation kinetic trajectory.
[0096] In the initial stage of fermentation, the error compensation control module does not immediately execute enzyme addition, but waits for the substrate degradation kinetic trajectory to enter the limiting range. Near-infrared spectral parameters and online viscosity parameters are input into the error compensation control module according to the sampling period. If both spectral attenuation and viscosity decrease are slower than the substrate degradation kinetic trajectory, the error compensation control module determines that the substrate transformation is slower than predicted, advances the overall enzyme addition time point, and increases the current segment enzyme addition amount. If the spectral state deviation of each band shows that the lag degree of different substrates is different, the error compensation control module further generates an enzyme addition ratio vector matrix and issues pulse injection commands to amylase, cellulase, and protease respectively. If the double step shear test shows that the network structure collapse index exceeds the network disintegration threshold, the error compensation control module delays or masks the cellulase channel, so that the enzyme addition ratio meets the substrate transformation while reducing rheological structure damage.
[0097] During the mid-fermentation stage, after the entire enzyme group enters the fermenter, the relative remaining proportions of starch, protein, and dietary fiber continue to decrease. The error compensation control module continuously calculates the material conversion completion index. When the material conversion completion index reaches the termination condition and the viscosity deviation does not exceed the allowable range, the error compensation control module generates a cooling and enzyme inactivation command. After the cooling and enzyme inactivation command is executed, the temperature control unit lowers the material temperature and maintains it until the enzyme activity is below the termination criterion. The material after cooling and enzyme inactivation no longer enters the enzyme addition cycle but instead enters the drying, pulverizing, and microwave sterilization processes. The batch number of the packaged finished product is linked to the fermentation process data for subsequent updates to the qualified batch database.
[0098] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.
[0099] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for preparing whole-component enzyme-fermented meal replacement powder, characterized in that, include: Physicochemical indicators of grain powder, bean powder, fruit and vegetable powder and medicinal and edible powder were collected before and before the mixed powder entered fermentation. The whole component feature matrix was constructed and the core feature vector of the whole component was extracted. The core feature vector of the whole group is input into the fermentation state prediction model, and combined with the substrate degradation kinetic trajectory, the whole group of enzyme addition time nodes and fermentation termination time threshold are generated. Near-infrared spectral parameters and online viscosity parameters were collected during fermentation to generate a material measured state sequence. The measured state sequence of materials is compared with the degradation kinetic trajectory of the substrate to form a state deviation. The enzyme addition time nodes and fermentation termination time threshold of the whole set are corrected based on the state deviation. The enzyme addition time nodes of the whole set include cellulase addition time nodes that can be corrected independently. The enzyme addition valve is triggered based on the revised enzyme addition time node, and the cooling and enzyme inactivation command is triggered based on the revised fermentation termination time threshold, thus obtaining the whole-group enzyme-added fermented meal replacement powder.
2. The method for preparing whole-component enzyme-fermented meal replacement powder according to claim 1, characterized in that, The overall component characteristic matrix includes carbon-nitrogen ratio, water activity, starch gelatinization degree, initial starch content, initial protein content, initial dietary fiber content, initial polysaccharide content, and raw material powder particle size distribution. The aforementioned indicators are respectively obtained from raw material acceptance testing, testing after cooking and drying, testing before mixing, and testing before fermentation.
3. The method for preparing whole-component enzyme-fermented meal replacement powder according to claim 1, characterized in that, The core feature vector of the whole component is obtained by standardizing the feature matrix of the whole component and mapping it with principal component loads, and retains the principal components whose cumulative contribution rate reaches the preset contribution condition.
4. The method for preparing whole-component enzyme-fermented meal replacement powder according to claim 3, characterized in that, The process of obtaining the core feature vector of the whole group includes: using the mean vector of historical qualified batch indicators and the standard deviation vector of historical qualified batch indicators to perform element-wise standardization on the original index vector of the whole group of the production batch, and then multiplying the standardization result with the transpose of the principal component loading matrix to obtain the core feature vector of the whole group of the corresponding production batch.
5. The method for preparing whole-component enzyme-fermented meal replacement powder according to claim 1, characterized in that, The fermentation state prediction model includes a feature mapping sub-model, a substrate degradation sub-model, and a command time point sub-model. The feature mapping sub-model receives the whole set of core feature vectors and outputs the initial state parameters of the substrate. The substrate degradation sub-model generates the substrate degradation kinetic trajectory based on the initial state parameters of the substrate. The command time point model outputs the whole set of enzyme addition time points and fermentation termination time thresholds based on the substrate degradation kinetic trajectory.
6. The method for preparing whole-component enzyme-fermented meal replacement powder according to claim 5, characterized in that, The substrate degradation kinetic trajectory includes starch degradation trajectory, protein degradation trajectory and dietary fiber degradation trajectory. The enzyme addition time point of the whole set is the time point corresponding to when at least one of the aforementioned trajectories enters the restriction interval. The fermentation termination time threshold is the time point corresponding to when the material conversion completion index meets the termination condition.
7. The method for preparing whole-component enzyme-fermented meal replacement powder according to claim 6, characterized in that, The material conversion completion index is obtained by weighting and summing the conversion completion rates of various substrates included in the evaluation according to their respective substrate weights, and then subtracting the viscosity deviation penalty term. The viscosity deviation penalty term is determined by the viscosity deviation penalty coefficient and the difference between the measured online viscosity value and the predicted theoretical viscosity value.
8. The method for preparing whole-component enzyme-added fermented meal replacement powder according to claim 1, characterized in that, Near-infrared spectral parameters include absorption characteristics related to starch, protein, polysaccharide, and moisture. Online viscosity parameters include apparent viscosity, viscosity change rate, and shear response change. The near-infrared spectral parameters and online viscosity parameters are time-aligned to form a sequence of measured material states.
9. The method for preparing whole-component enzyme-fermented meal replacement powder according to claim 1, characterized in that, The enzyme set includes amylase, cellulase, and protease. After the enzyme addition time point triggers the enzyme addition valve action command, the enzyme addition valve adds the enzyme set to the fermenter in a segmented manner. The amount of enzyme added in a single segment is determined by the predicted amount of remaining substrate and the real-time state deviation.
10. The method for preparing whole-component enzyme-fermented meal replacement powder according to claim 9, characterized in that, The amount of enzyme added in a single stage is determined by the predicted amount of remaining substrate and the real-time state deviation, including the independent dynamic reconstruction of the addition ratios of amylase, cellulase, and protease in the entire enzyme group; the spectral state deviation is decomposed into starch characteristic band deviation, protein characteristic band deviation, and polysaccharide characteristic band deviation; for the single-stage enzyme addition process, an enzyme addition ratio vector matrix is set, and the weights of amylase ratio, cellulase ratio, and protease ratio in the enzyme addition ratio vector matrix are dynamically reconstructed according to the following formula: ; in, Indicates the fermentation time Time target enzyme Dynamic reconfiguration of allocation weights, Indicates the fermentation time. Indicates the target enzyme type and is selected from amylase, cellulase, and protease. Indicates the target enzyme at the fermentation initiation point The initial proportion weights, This represents the natural exponential function. This represents the substrate sensitivity scaling factor. Indicates the fermentation time Time target enzyme Spectral state deviation of the corresponding substrate characteristic band, Indicates the fermentation time Time target enzyme The corresponding predicted amount of remaining substrate; the single-stage enzyme addition process is based on the target enzyme. The dynamic reconstructed ratio weights issue independent pulse injection commands to each single enzyme in the fermenter; the spectral state deviation and the predicted amount of remaining substrate entering the formula are both normalized dimensionless quantities, the substrate sensitivity scaling factor is a dimensionless coefficient, so the exponential term is a dimensionless quantity, and the dynamic reconstructed ratio weights are still ratio weights.
