Management system for bioenzymatic-based processing of agricultural products

Abstract

This invention belongs to the technical field of agricultural product processing production management, process data processing, and scheduling control, and discloses a management system for agricultural product processing based on bio-enzymatic hydrolysis. The system includes a batch feature builder, an enzymatic hydrolysis progress predictor, a scheduling coupling processor, and an execution deviation write-back unit. By integrating raw material batch attributes, enzyme batch potency, formulation version, stage inspection results, and transfer records, it generates batch-level enzymatic hydrolysis characteristics, predicts a reliable time window for the target hydrolysis state, and maps it to a reaction resource occupancy window. This window, along with post-processing capacity, inspection waiting time, and delivery constraints, forms a batch node plan. The system decomposes raw material enzymatic hydrolysis deviation, enzyme activity deviation, and scheduling waiting time deviation based on actual inspection values ​​and transfer times. It writes back and corrects the prediction inputs and scheduling boundaries for subsequent batches, reducing premature transfers, target waiting times, and node conflicts caused by mismatches between fixed process durations and actual enzymatic hydrolysis progress.

Description

Technical Field

[0001] This invention belongs to the field of agricultural product processing production management, process data processing and scheduling control technology, and discloses an agricultural product processing management system based on bio-enzymatic hydrolysis. Background Technology

[0002] In agricultural product processing, the management of bio-enzymatic hydrolysis typically relies on a production management system to complete batch filing, formula issuance, material input records, reaction process records, inspection records, and transfer records. Existing systems, when establishing production tasks, generally use the product's process specifications as a basis, entering raw material batches, enzyme batches, reaction temperatures, pH levels, stirring times, holding times, enzyme inactivation times, and post-processing steps as fixed fields, and generating batch processing orders according to a preset process route. During production execution, operators requisition agricultural raw materials and enzymes according to the processing orders, entering the material input time, reaction start time, sampling and inspection time, enzyme inactivation time, and transfer time into the system. Inspection personnel enter indicators such as the degree of hydrolysis, viscosity, and soluble solids after stage or final inspections. Based on these records, the system generates batch ledgers, material consumption tables, process flow tables, and quality traceability tables to indicate when, with what formula, and using what raw materials and enzymes were processed for a particular batch. In terms of data organization, raw material information, enzyme preparation information, process parameters, and test results are usually scattered across different forms such as materials, production, and quality, and are linked to each other through batch numbers or work order numbers. Such systems can support post-event traceability and standardized on-site records, but the relevant fields mostly exist as static record bases and do not distinguish the differences in enzymatic hydrolysis progress between different batches during the planning and generation stage.

[0003] Current scheduling methods typically treat enzymatic hydrolysis as a fixed-duration production process. When arranging reaction vessels, post-processing steps, and inspection tasks, the system first determines the feeding sequence based on order quantity, planned date, and equipment availability. Then, it calculates the estimated enzyme inactivation time and estimated transfer time according to the enzymatic hydrolysis duration recorded in the formula. Subsequent processes such as filtration, concentration, drying, blending, or packaging also use this estimated time as a prerequisite. If actual inspection results show that the batch has not yet reached the target hydrolysis state, common handling methods include extending the insulation period on-site, adding samples, or temporarily adjusting the post-processing queue. If the batch reaches the target state ahead of schedule, it is often necessary to wait for the original transfer time or for post-processing resources to be released. Some management systems set buffer times in the production plan or adjust the batch sequence according to equipment idle status. However, buffer times are usually based on experience, and equipment sequence adjustments are mainly based on resource availability, without considering the reaction target attainment time as a calculable node boundary. In continuous production scenarios, the release time of the reaction tank, the sampling time of the inspectors, and the material receiving time of the post-processing steps in the production schedule are usually calculated in conjunction with the same standard duration. If the actual compliance time of the preceding batch deviates, subsequent nodes can only maintain operation through manual reassignment. The planning boundaries usually do not change automatically with the results of stage inspections and the status of enzyme batches.

[0004] Existing technologies also utilize historical batch data for process analysis. A common approach is to calculate average reaction times based on product category, raw material category, formulation version, or enzyme type, generating abnormal batch labels, quality deviation records, or process review prompts. The system then correlates stage inspection results with endpoint inspection results to determine whether a batch requires extended reaction time, additional enzyme preparation, or manual review. This data processing typically occurs after batch completion, with output primarily used for quality traceability, process review, or subsequent manual adjustments. However, there is a lack of unified data representation regarding raw material hydrolysis solubility, enzyme potency changes, stage inspection trends, and actual sequence delays. This data cannot directly alter reaction resource occupancy boundaries or create movable node constraints during post-processing queueing. Even if the system saves raw material moisture content, pretreatment methods, enzyme potency upon warehousing, opening status, and stage inspection sequences, these data are often categorized into quality analysis fields and not converted into target hydrolysis state attainment time ranges. While historical data enters the management system, it remains at the recording and review level, not participating in the constraint generation process for production planning.

[0005] The main technical problem lies in the fact that existing bio-enzymatic hydrolysis agricultural product processing management systems fail to translate the uncertainty of the actual batch enzymatic hydrolysis target time into scheduling constraints, leading to deviations between planned nodes formed by fixed process durations and the actual reaction progress. Differences in the variety, maturity, moisture content, pretreatment state, and key components of agricultural raw materials can alter substrate release rates; batch potency of enzyme preparations, opening status, and storage processes can change catalytic capabilities; and stage test results reflect real-time changes in reaction rates. These factors collectively determine the time range in which the target hydrolysis state occurs. While existing systems store some of this data, they lack a computational chain between batch-level enzymatic hydrolysis characteristics, reliable time windows, and resource occupancy windows, and they do not correct the target window for subsequent batches after writing back actual test values, enzyme inactivation time, and transfer time. This results in production plans operating based on average working hours, while the enzymatic hydrolysis process depends on batch reaction state changes. When reaction delays, early target achievement, or transfer waiting occur, deviations can only be recorded as anomalies for individual batches and cannot be carried back to the scheduling generation process for the next batch of similar raw materials or the same enzyme preparation. The planned nodes lack an update mechanism driven by the reaction status, making it difficult to form a closed-loop coupling between reaction resource release time, inspection waiting time, and post-processing takeover time. Summary of the Invention

[0006] The purpose of this invention is to provide a management system for agricultural product processing based on bio-enzymatic hydrolysis, which can effectively solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A management system for agricultural product processing based on bio-enzymatic hydrolysis includes: a batch feature builder, which is used to generate batch-level enzymatic hydrolysis features based on the batch attributes of agricultural raw materials, batch potency of enzyme preparations, process formula version, historical enzymatic hydrolysis test results and actual transfer records;

[0009] An enzymatic hydrolysis process predictor is used to output a reliable time window for reaching the target hydrolysis state and an abnormal risk indicator based on the batch-level enzymatic hydrolysis characteristics.

[0010] A scheduling coupling processor is used to map the trusted time window to a reactive resource occupancy window and generate a batch node plan together with post-processing processes, inspection waits, and order delivery constraints.

[0011] The execution bias writeback is used to adjust the reliable time window for subsequent batches based on the actual inspection value and the actual turnaround time.

[0012] Preferably, the batch feature builder establishes a three-dimensional index according to raw material batch, enzyme preparation batch, and formulation version. It aligns the fields of raw material moisture content, pretreatment method, key component content, enzyme preparation warehousing potency, opening status, feeding ratio, reaction temperature range, and pH range, and configures source identifiers and valid ranges for the aligned fields. It marks historical test results as time segments according to the reaction start-up stage, main reaction stage, and plateau stage, and generates batch-level enzymatic hydrolysis features containing batch source, reaction conditions, and stage test sequences.

[0013] Preferably, the enzymatic hydrolysis process predictor inputs the batch-level enzymatic hydrolysis characteristics into the enzymatic hydrolysis stratification model. The enzymatic hydrolysis stratification model determines the reaction rate stable type, delayed release type, rapid over-hydrolysis type, or fluctuation sensitive type based on the historical target achievement time residual, stage inspection slope, raw material composition offset, and enzyme potency offset. It also calls the corresponding time window prediction model for different types. The time window prediction model outputs the upper bound time, lower bound time, window confidence, and early enzyme inactivation risk indicator of the target hydrolysis state.

[0014] Preferably, the scheduling coupling processor generates reaction resource occupancy boundaries based on the upper and lower bound times, sets feeding nodes, enzymatic hydrolysis target achievement nodes, inspection nodes, enzyme inactivation nodes, and sequence transfer nodes as constrainable nodes in the directed scheduling graph, and sets reaction tank occupancy, post-processing capacity, inspection waiting time, batch switching on the same line, and order delivery time as constraint edges, and limits the movable range of the constrainable nodes according to the window confidence level, generating a batch node plan that includes the triggering order of each node.

[0015] Preferably, the batch feature builder sets batch comparability labels for the time series segments, and the batch comparability labels are jointly determined by the similarity of raw material sources, the consistency of pretreatment, the batch correlation of enzyme preparations, and the consistency of process formulation versions.

[0016] When generating the batch-level enzymatic hydrolysis characteristics, the historical batch set is first screened based on the batch comparability label, then the actual transsequence records in the historical batch set are aligned with the stage test sequences, and the missing test fragments are configured with fragment confidence markers that are jointly defined by adjacent stages and similar batches.

[0017] Preferably, the time window prediction model includes a category benchmark sub-model and a deviation correction sub-model. The category benchmark sub-model outputs a benchmark achievement curve according to the reaction rate stable type, delayed release type, rapid over-decomposition type, or fluctuation sensitive type category. The deviation correction sub-model corrects the benchmark achievement curve on the time axis according to the raw material composition offset, enzyme potency offset, and stage test slope of the current batch, and determines the time period when the corrected curve intersects the target hydrolysis state interval as the reliable time window.

[0018] Preferably, the scheduling coupling processor sets window constraint nodes and fixed constraint nodes for each batch in the directed scheduling diagram. The window constraint nodes include enzymatic hydrolysis target achievement nodes and inspection nodes, and the fixed constraint nodes include feeding nodes, enzyme inactivation nodes and transfer nodes.

[0019] When the resource occupancy boundaries of different batches overlap with the capacity of post-processing steps, a node adjustment sequence is generated according to window reliability, order delivery time, batch switching correlation and early enzyme inactivation risk marker, and the batch node plan is updated based on the node adjustment sequence.

[0020] Preferably, the execution deviation writer receives the planned inspection time, planned enzyme inactivation time, and planned transsequence time from the batch node plan, and generates a node deviation sequence with the actual inspection time, actual enzyme inactivation time, and actual transsequence time;

[0021] The execution deviation write-back device, according to the category of the enzymatic solubility stratification model, enzyme potency offset, stage inspection sequence change, and post-processing process waiting record, splits the node deviation sequence into raw material enzymatic solubility deviation, enzyme activity deviation, and scheduling waiting deviation, and writes them into the feature fields of the corresponding batches.

[0022] Preferably, when the execution deviation writer writes the raw material enzymatic hydrolysis deviation, enzyme activity deviation, and scheduling waiting deviation, it establishes an association index between the deviation source and the batch comparability label and the fragment confidence marker.

[0023] When the same type of deviation occurs consecutively from the same raw material source or the same batch of enzyme preparation, the weight of the corresponding feature field in the deviation correction sub-model is increased according to the consecutive number of deviation sources and the corresponding node positions, and the weight of historical batch samples formed by low-confidence test fragments is reduced, generating an updated feature set for predicting the time window of subsequent batches.

[0024] Preferably, after obtaining the updated feature set, the scheduling coupling processor recalculates the reliable time window for batches that have not yet been fed, and updates only the movable range of the enzymatic hydrolysis target node and the inspection node for batches that have been fed but have not been inactivated.

[0025] The batches that have undergone enzyme inactivation are kept locked at the transfer node, and the resource usage changes generated by the recalculation are written into the constraint edges of the directed scheduling graph. The capacity constraints of the unlocked post-processing steps are reordered, and a rolling batch node plan is generated according to the node adjustment sequence.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0027] 1. By constructing batch-level enzymatic hydrolysis characteristics based on agricultural raw material batch attributes, enzyme batch potency, formulation version, historical test results, and actual transfer records, and outputting reliable time windows for the target hydrolysis state based on enzymatic hydrolysis stratification, production planning no longer determines enzyme inactivation and transfer nodes based on a single fixed process duration. After the reliable time window is mapped to a reaction resource occupancy window, it can form a batch node plan together with post-processing capacity, inspection waiting time, and order delivery constraints, enabling the enzymatic hydrolysis target node, inspection node, enzyme inactivation node, and transfer node to be updated in a linked manner within the same scheduling logic. This approach incorporates the target achievement time offset caused by raw material differences and enzyme activity differences into the planning calculation process, reducing batch state deviations caused by premature transfer before reaching the target or long waiting times after reaching the target.

[0028] 2. By associating node deviation sequences, enzymatic solubility categories, enzyme potency offsets, stage test sequence changes, and post-processing waiting records, the actual test time, enzyme inactivation time, and transsequence time can be broken down into raw material enzymatic solubility deviation, enzyme activity deviation, and scheduling waiting deviation, and written into the feature fields of subsequent batches. For the same type of deviation occurring consecutively from the same raw material source or the same enzyme batch, the system can update the feature weights of the deviation correction sub-model and reduce the sample weights formed by low-confidence test fragments. For batches not yet fed, the confidence time window is recalculated; for batches fed but not yet inactivated, the movable node range is updated; and for batches inactivated, the transsequence node remains locked, ensuring that the rolling batch node plan remains consistent with the actual execution status. Attached Figure Description

[0029] Figure 1 The flowchart of the overall closed-loop management system for agricultural product processing based on bio-enzymatic hydrolysis of the present invention is shown below.

[0030] Figure 2 This is a flowchart illustrating the batch-level enzymatic hydrolysis feature construction process of the present invention.

[0031] Figure 3 This is a flowchart of the enzymatic hydrolysis process prediction and reliable time window generation of the present invention;

[0032] Figure 4 This is a flowchart of the scheduling coupling, deviation write-back, and rolling update process of the present invention. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] Please refer to Figure 1This embodiment provides a management system for agricultural product processing based on bio-enzymatic hydrolysis, deployed within the production management platform of an agricultural product processing enterprise. It integrates data processing for batch archiving, hydrolysis progress prediction, reaction resource usage, inspection waiting, enzyme inactivation and transfer, and post-processing in the bio-enzymatic hydrolysis process. The system includes a batch feature builder, a hydrolysis progress predictor, a scheduling coupling processor, and an execution deviation write-back unit. The batch feature builder extracts calculable fields from production work orders, raw material batch files, enzyme preparation batch files, process formula versions, historical hydrolysis inspection records, and actual transfer records. It aligns the batch attributes of agricultural raw materials, enzyme preparation batch potency, feed ratio, reaction temperature range, pH range, stage inspection values, actual enzyme inactivation time, and actual transfer time of the scattered records within the same batch using batch numbers to form batch-level hydrolysis features. After reading the batch-level hydrolysis features, the hydrolysis progress predictor does not directly determine the completion time based on a fixed process duration, but... It combines the enzymatic hydrolysis capability of raw materials, the potency of enzyme preparations, and historical test residual outputs to reach the target hydrolysis state within a reliable time window. Simultaneously, it generates abnormal risk indicators such as the risk of premature enzyme inactivation, the risk of delayed achievement of targets, or the risk of waiting for sequence transition. The scheduling coupling processor maps the reliable time window to a reaction resource occupancy window, and organizes the release of the reaction tank, the capacity of the post-processing process, the test waiting time, and the order delivery time into a unified batch node plan. The execution deviation write-back device reads the actual test value and the actual sequence transition time after the batch execution is completed or the stage test is completed, writes the deviation between the planned value and the actual value back to the corresponding batch feature field, and drives the reliable time window of subsequent batches to be recalculated. This allows the prediction result of the bio-enzymatic hydrolysis process to directly participate in the generation of the scheduling boundary, and allows the scheduling execution deviation to correct the prediction input of the next batch in reverse. Through the above data closed loop, this embodiment converts the deviation between the fixed-duration scheduling and the actual enzymatic hydrolysis process into a calculable, write-back, and updateable management constraint.

[0035] In this embodiment, reference Figure 2When the batch feature builder standardizes the data entering the system, it uses the production batch number as the primary index and the raw material batch number, enzyme batch number, and process formulation version number as related indexes. Raw material batch attributes include raw material variety, origin batch, moisture content, pretreatment method, key component content, and warehousing time. Enzyme batch potency includes warehousing inspection potency, opening status, number of uses, remaining usable records, and historical batches. Process formulation version includes target hydrolysis state, feed ratio range, stage sampling nodes, and post-processing acceptance conditions. Historical enzymatic hydrolysis test results include the degree of hydrolysis, viscosity, or solubility indicators corresponding to the reaction initiation stage, main reaction stage, and plateau stage. The actual transsequence record includes the actual enzyme inactivation time, actual transsequence time, post-processing wait start time, and post-processing receiving time. The batch feature builder converts the above fields into a unified batch-level enzymatic hydrolysis feature vector and retains the source identifier, collection time, and availability status of each field to avoid setting missing fields directly to zero, which could lead to model misjudgment. If a test value is missing at a certain stage, a fragment confidence marker is generated from the test values ​​of adjacent stages and the historical curves of the same batch. The fragment confidence marker is only used as the model confidence input and does not replace the actual test data. This embodiment aligns the same batch of dispersed production data and quality data to ensure that the input data used for subsequent prediction and scheduling has a consistent reference basis.

[0036] In this embodiment, reference Figure 3 When calculating the reliable time window for the target hydrolysis state, the enzymatic hydrolysis process predictor first calls the enzymatic hydrolysis stratification model to classify batch-level enzymatic hydrolysis characteristics. These categories include stable reaction rate, delayed release, rapid over-hydrolysis, and fluctuation-sensitive types. Category identification is not based on a single raw material name or fixed process formulation, but rather on a combination of historical target achievement time residuals, stage test slopes, raw material composition offsets, enzyme potency offsets, and fragment confidence markers. Stable reaction rate corresponds to batches with concentrated historical batch target achievement time residuals and stable stage test slopes. The release type corresponds to batches with slow changes in the initiation stage and a delayed main reaction stage; the rapid over-decomposition type corresponds to batches with a large slope in the initiation stage and an early plateau stage; and the fluctuation-sensitive type corresponds to batches with dispersed test curves and large variations in potency under the same formulation. The enzymatic hydrolysis process predictor calls the corresponding time window prediction model based on the identified category and outputs the lower bound time, upper bound time, window confidence, and abnormal risk indicator of the target hydrolysis state. This embodiment enables different raw material batches and different enzyme preparation batches to form distinguishable reaction process expressions, avoiding the use of a single average working time to cover batch-to-batch differences.

[0037] ;

[0038] in, Indicates the first Batch-level enzymatic hydrolysis feature vectors for each production batch Indicates the batch attribute code of the raw materials. This indicates the enzyme preparation's potency status code. Indicates the process formula version code, This indicates the stage test sequence encoding. Indicates the actual transposition record encoding. This represents the fragment confidence tag encoding. For example, if a batch has the following encodings: raw material attribute code 0.62, enzyme potency status code 0.81, process formulation version code 0.35, stage test sequence code 0.58, actual transsequence record code 0.44, and fragment confidence tag code 0.90, then the input vector for that batch is... Its physical meaning is to compress the raw materials, enzyme preparations, formulations, testing, sequencing, and data reliability into a calculable state description under the same batch.

[0039] In this embodiment, when the scheduling coupling processor maps the reliable time window to the reaction resource occupancy window, it constructs a directed scheduling graph starting from the feeding node, with the enzymatic hydrolysis target completion node, inspection node, enzyme inactivation node, and transfer node as intermediate constraints, and the subsequent processing step receiving node as the endpoint. The edges in the directed scheduling graph do not simply represent the sequence of processes, but also represent the executable interval between adjacent nodes that is jointly restricted by resource occupancy, inspection waiting, on-line switching, and delivery constraints. The lower bound of the reliable time window is used to limit the earliest time that inspection or enzyme inactivation can be triggered, and the upper bound of the reliable time window is used to limit the risk of delayed target completion if it is exceeded. The identified time and window confidence level are used to limit the movable range of nodes. If the confidence level is low, the scheduling coupling processor retains a verification buffer before the inspection node and reduces the locking degree of the post-processing acceptance schedule. If the confidence level is high, the movable range between the inspection node and the enzyme inactivation node is narrowed. The reaction resource occupation window can be used for queuing calculation of subsequent batches entering the reaction resources. The batch node plan includes the batch feeding sequence, expected inspection time, expected enzyme inactivation time, expected transfer time and post-processing acceptance sequence. In this embodiment, the reaction prediction results are incorporated into the production plan generation process by converting the enzymatic hydrolysis target time range into the resource occupation boundary.

[0040] In this embodiment, after receiving the actual test value and the actual transfer time, the execution deviation write-back device compares the planned test time with the actual test time, the planned enzyme inactivation time with the actual enzyme inactivation time, and the planned transfer time with the actual transfer time to form a node deviation sequence. If the stage test value shows that the degree of hydrolysis is lower than the target range and the actual enzyme inactivation time is delayed, the execution deviation write-back device will preferentially mark the deviation as a candidate source of raw material enzymatic solubility deviation or enzyme activity deviation, and make a judgment based on the historical deviations of the same raw material source or the same enzyme batch. If the target hydrolysis state has been reached but the post-processing receiving time is delayed, the deviation will be marked as a scheduling waiting deviation. If the stage test curve deviates from the historical similar raw material and the same enzyme batch performs normally in other raw materials, the deviation will be classified into the raw material enzymatic solubility deviation field. If the same enzyme batch shows reaction delay in multiple raw material batches, the deviation will be classified into the enzyme activity deviation field. The deviation field after writing back will be used as the input of the subsequent batch time window prediction model. In this embodiment, through deviation splitting and source writing back, the deviation of a production execution will no longer stay at the abnormal record level, but will enter the next round of scheduling calculation. The batch-level enzymatic hydrolysis feature fields and data processing methods are shown in Table 1.

[0041] Table 1. Batch-level enzymatic hydrolysis characteristic fields and data processing methods

[0042] Raw material batch attributes Variety, source, moisture content, pretreatment method, content of key components Unified coding and generation of raw material composition offset Enzymatic stratification model Enzyme preparation status Records of valence upon entry, opening status, batch used, and valence offset. Formation of enzyme titer offset Time window prediction model Recipe version Target hydrolysis state, feed ratio, and stage sampling settings As a basis for batch alignment and target interval Category Benchmark Submodel Test sequence Test values ​​for the initiation phase, main reaction phase, and plateau phase Formation stage test slope and fragment confidence marker Deviation Correction Submodel Transorder record Actual enzyme inactivation time, actual transfection time, and post-processing waiting time records. Forming a node deviation sequence Execution bias writeback

[0043] In a preferred embodiment, the batch feature builder organizes batch data according to the fields shown in Table 1. The field categories are not required to come from the same database table. Instead, a joint index is established using the production batch number, raw material batch number, enzyme preparation batch number, and process formulation version number. The raw material batch attribute field is used to characterize the substrate release conditions, the enzyme preparation state field is used to characterize the catalytic ability, the formulation version field is used to determine the target hydrolysis state and stage sampling rules, the test sequence field is used to form the reaction process curve, and the transsequence record field is used to determine the offset position between the planned node and the actual node. The position of entering the model is determined by the calculation purpose of the field. When generating batch-level enzymatic hydrolysis features, the batch feature builder sets non-mixable rules according to the field categories. For example, missing fragments in the test sequence can only generate fragment confidence markers and do not participate in replacing the true test value. The post-processing waiting time in the transsequence record can only participate in the scheduling waiting deviation judgment and does not participate in the direct judgment of the raw material enzymatic hydrolysis category. This embodiment ensures that the input source used for enzymatic hydrolysis prediction, scheduling coupling, and deviation write-back is clear through the correspondence between field categories and calculation positions.

[0044] Preferably, the batch feature builder establishes a three-dimensional index based on raw material batches, enzyme batches, and formulation versions. The first dimension of the three-dimensional index is used to aggregate batches from the same raw material source and with the same pretreatment method. The second dimension is used to aggregate batches from the same enzyme preparation or with the same potency state. The third dimension is used to aggregate the target hydrolysis state and stage inspection rules under the same process formulation version. When aligning fields, the moisture content of raw materials, pretreatment method, and content of key components are entered into the raw material attribute sub-vector; the potency upon entry, opening status, and feeding ratio are entered into the enzyme preparation status sub-vector; the reaction temperature range and pH range are entered into the process condition sub-vector; the stage inspection values ​​are sorted according to the offset of the sampling time relative to the feeding time; and the actual transfer records are sorted according to the time sequence of enzyme inactivation, transfer, and post-treatment receiving. After the field alignment is completed, the batch feature builder writes the source identifier into the appendix of each field to distinguish between system automatic import, inspection entry, and production execution write-back. In this embodiment, the three-dimensional index avoids masking reaction differences caused by aggregating historical batches only by product name.

[0045] Furthermore, the batch feature builder sets time segment labels for the reaction initiation stage, main reaction stage, and plateau stage of the stage test sequence. The time segment labels are determined based on the relative position of the sampling time after feeding and the target process time. The initiation stage corresponds to the segment where the reaction curve has not yet formed a stable slope after the enzyme preparation comes into contact with the substrate. The main reaction stage corresponds to the segment where the degree of hydrolysis or solubility index continues to change. The plateau stage corresponds to the segment where the test value tends to the target hydrolysis state and enters a slow change state. If a batch lacks an initiation stage test value, the batch feature builder uses the initiation stage curve of an adjacent batch of the same type as a confidence reference and sets the confidence flag of the initiation stage segment of the current batch to low confidence. If the main reaction stage test value of a batch is complete, the confidence flag of the main reaction stage segment is set to high confidence. If the plateau stage test value comes from a retest after delayed transsequence, the confidence flag of the plateau stage segment is also written into the transsequence waiting source to avoid misjudging the numerical change caused by waiting as a change in the actual reaction rate. This embodiment enables the test record to enter the reaction category identification and time window prediction through stage segmentation processing.

[0046] ;

[0047] in, Indicates the first The batch in The phase test slope of each test segment, Indicates the first The test value of each sampling point This represents the test value of the previous sampling point. Indicates the first The time offset of each sampling point from the feeding time This indicates the time offset between the previous sampling point and the feeding time. For example, if the degree of hydrolysis of a batch is 18 two hours after feeding and 30 four hours after feeding, then... Its physical meaning is that the degree of hydrolysis within the test segment changes by 6 units per hour.

[0048] In a preferred embodiment, when the batch feature builder filters the historical batch set, it first generates batch comparability labels based on the similarity of raw material sources, consistency of pretreatment, batch correlation of enzyme preparations, and consistency of process formulation versions. The similarity of raw material sources is determined by the differences in raw material varieties, origin batches, and content of key components. The consistency of pretreatment is determined by the degree of consistency of washing, crushing, soaking, heat treatment, or other existing pretreatment records. The batch correlation of enzyme preparations is determined by the enzyme preparation batch number, warehousing potency, opening status, and number of uses. The consistency of process formulation versions is determined by the target hydrolysis state, feeding ratio, and stage sampling rules. The batch comparability labels are divided into three states: directly comparable, requiring correction for comparison, and not comparable. Only historical batches that are directly comparable or require correction for comparison enter the time window prediction sample pool. Uncomparable batches are only retained as traceability records. This embodiment uses batch comparability screening to maintain homology and process consistency between historical samples and current batches.

[0049] In this embodiment, when processing historical batches requiring correction and comparison, the batch feature builder aligns the actual transsequence records of the historical batches with the stage test sequences. The actual transsequence records are used to determine whether the test values ​​were obtained during the normal reaction process or during the waiting post-processing period. If a test value occurs within the waiting interval where the target hydrolysis state has been reached but transsequence has not yet occurred, the batch feature builder marks it as a waiting impact test segment. The waiting impact test segment does not participate in the basic discrimination of the reaction rate category, but only in the source analysis of scheduling waiting bias. If a historical batch lacks the main reaction stage test value but has complete start-up and plateau stage test values, the batch feature builder generates a missing segment confidence label based on the main reaction stage curve of the same type of directly comparable batches, and uses the label as a confidence field in the model input. If there are too many missing segments, the historical batch is only used for scheduling waiting statistics and not for training with a confidence time window. This embodiment reduces the interference of historical batch samples on reaction process prediction by jointly aligning the actual transsequence records and test sequences.

[0050] Preferably, after reading the batch comparability labels, the stratification model of enzymatic hydrolysis process predictor uses full weights for directly comparable samples, assigns weights to comparative samples requiring correction based on fragment confidence markers, and does not include incomparable samples in the stratification process. The stratification model calculates the weighted distances between the current batch and historical batches in terms of raw material composition offset, enzyme potency offset, and stage test slope, and selects historical batches with smaller distances and higher fragment confidence as the nearest neighbor batch set. The distribution of the time to reach the target in the nearest neighbor batch set is used to determine whether the current batch belongs to the reaction rate stable type. The reaction can be classified as either delayed-release, rapid over-reaction, or fluctuation-sensitive. If the slopes of the start-up phase and the main reaction phase of neighboring batches are concentrated and the historical residual distribution of the time to reach the target is narrow, it is determined to be a stable reaction rate type. If the slope of the start-up phase is low and the slope of the main reaction phase is delayed, it is determined to be a delayed-release type. If the slope of the start-up phase is high and the plateau phase appears too early, it is determined to be a rapid over-reaction type. If neighboring batches still have discrete residuals under the same formulation, it is determined to be a fluctuation-sensitive type. This embodiment uses the comparability of samples and the characteristics of the reaction phase to jointly determine the category, so that the prediction model does not depend on a single process name.

[0051] Furthermore, the time window prediction model of the enzymatic hydrolysis process predictor includes a category benchmark sub-model and a deviation correction sub-model. The category benchmark sub-model saves a benchmark achievement curve for each enzymatic hydrolysis category. The benchmark achievement curve is obtained by fitting the stage test sequence of the high-confidence segment in the nearest batch set. The deviation correction sub-model reads the raw material composition offset, enzyme potency offset, and stage test slope of the current batch and performs time axis correction and curve amplitude correction on the benchmark achievement curve. The time axis correction is used to describe whether the reaction is advanced or delayed, and the curve amplitude correction is used to describe the degree to which the endpoint test value is close to the target interval. The continuous time period when the corrected curve intersects with the target hydrolysis state interval is determined as the confidence time window. The left end of the confidence time window is the lower bound time, and the right end is the upper bound time. If the corrected curve quickly crosses the target interval in a shorter interval, the risk of premature enzyme inactivation is written into the current batch. If the corrected curve is still not close to the target interval at the upper bound time, the risk of delayed achievement is written into the current batch. In this embodiment, the combination of category benchmark and deviation correction enables the time window to change with batch differences.

[0052] ;

[0053] in, Indicates the first Each batch at time offset The corrected predicted test value at the location, Indicates the first Category of each batch The benchmark compliance curve This represents the time axis correction amount formed by the offset of raw material composition and the offset of enzyme potency. This represents the correction amount for the test value formed by the slope of the stage test. For example, if the category baseline curve reaches the test value of 40 after 6 hours, and the time axis correction amount for the current batch is 1 hour and the test value correction amount is 2, then in... Hourly Its physical meaning is that the current batch is delayed by 1 hour compared to the category baseline curve and the test value level is shifted up by 2 units of measurement.

[0054] In a preferred embodiment, the calculation of the reliable time window uses the target hydrolysis state interval as the judgment boundary. The target hydrolysis state interval is given by the process formulation version. Let the lower limit of the target hydrolysis state be L and the upper limit be U. The enzymatic hydrolysis process predictor searches for the desired hydrolysis state on the corrected prediction curve. The continuous time interval is used, and the interval with high consistency with the confidence mark of the stage test segment is used as the output window. If there are multiple candidate intervals, the interval with high overlap with the actual compliance time distribution of the nearest batch is selected. If there are rapid changes in test values ​​before and after the candidate interval, the window confidence is reduced. If the slope of the stage test near the candidate interval gradually slows down and the segment confidence mark is high, the window confidence is increased. The abnormal risk indicator is generated by the position of the candidate interval relative to the original planned process time. When the candidate interval is earlier than the original planned process time and the slope is still high, a rapid over-solution risk is formed. When the candidate interval is later than the original planned process time, a delayed compliance risk is formed. In this embodiment, the target hydrolysis state is no longer represented by a single expected completion point, but by a time range with confidence and risk indicators.

[0055] ;

[0056] in, Indicates the first The reliable time window for each batch This indicates the time offset from the time of material feeding. Indicates the lower limit of the target hydrolysis state. This indicates the upper limit of the target hydrolysis state. Indicates time offset The corresponding window credibility, This indicates the minimum confidence level requirement. For example, if the target hydrolysis state has a lower limit of 38 and an upper limit of 45, and the minimum confidence level requirement is 0.70, then if the predicted test value at 6 hours is 39 and the confidence level is 0.76, the predicted test value at 7 hours is 43 and the confidence level is 0.82, and the predicted test value at 8 hours is 46 and the confidence level is 0.79, then the confidence time window includes the interval between 6 and 7 hours. Its physical meaning is that the batch enters the target hydrolysis state within this interval and the prediction confidence level meets the requirements.

[0057] In this embodiment, reference Figure 4 The scheduling coupling processor distinguishes between window-constrained nodes and fixed-constrained nodes in the directed scheduling graph. Window-constrained nodes include enzymatic hydrolysis target achievement nodes and inspection nodes, while fixed-constrained nodes include feeding nodes, enzyme inactivation nodes, and transfer nodes. Feeding nodes are restricted by raw material preparation and reaction resource idle status, enzyme inactivation nodes are restricted by reliable time windows and inspection nodes, and transfer nodes are restricted by post-processing capacity and material receiving sequence. The movable range of enzymatic hydrolysis target achievement nodes is determined by the reliable time window, and the movable range of inspection nodes is jointly determined by the reliable time window, inspection waiting time, and stage sampling rules. When the reaction resource occupancy boundaries of different batches overlap with the post-processing capacity, the scheduling coupling processor generates a node adjustment sequence based on window reliability, order delivery time, batch switching correlation, and early enzyme inactivation risk indicator. The node adjustment sequence does not change the locked nodes, but only rearranges the nodes of batches that have not been fed, batches that have been fed but not inactivated, and post-processing nodes that have not been received. This embodiment distinguishes between node locking and node movable range, enabling rolling scheduling to maintain the stability of executed records.

[0058] Furthermore, when the scheduling coupling processor generates the node adjustment sequence, it sorts batches on the same reaction resource according to the resource occupancy window and batches on the same post-processing step according to the expected transfer node. It also locates the conflict positions between the two sorting methods. If the reliable time window of a batch overlaps with the available window for post-processing material receipt, the enzyme inactivation node and transfer node of that batch remain continuous. If the reliable time window is earlier than the available window for post-processing material receipt, a waiting flag is set after the check node and the waiting reason is written into the scheduling waiting field. If the reliable time window is later than the available window for post-processing material receipt, the post-processing step capacity is released to other unlocked batches. If both batches can enter the post-processing step, the batch with high window reliability, high risk of premature enzyme inactivation, and high batch switching correlation is selected first. The batch switching correlation is determined by whether adjacent batches use similar formulas, the same raw material category, or compatible post-processing conditions. This embodiment reduces the chain queuing conflict caused by a single node change by bidirectional matching of the reaction resource occupancy window and the post-processing capacity window.

[0059] In a preferred embodiment, when the deviation write-back device decomposes the planned and actual nodes, it does not directly treat all time differences as process anomalies. Instead, it generates different deviation source fields based on the location of the deviation. If the planned inspection time is close to the actual inspection time but the inspection value is lower than the target range and the actual enzyme inactivation time is delayed, the deviation source candidate is raw material enzymatic solubility deviation or enzyme activity deviation. If the planned inspection time has already met the standard but the actual transfer time is delayed, the deviation source candidate is scheduling waiting deviation. If the same raw material source shows a low slope in the start-up phase in different enzyme batches, the weight of raw material enzymatic solubility deviation is increased. If the same enzyme batch shows a delayed time to meet the standard in different raw material sources, the weight of enzyme activity deviation is increased. If multiple batches are waiting for the same post-processing step after meeting the inspection standard, the weight of scheduling waiting deviation is increased. The deviation write-back device writes the identified deviation source into the batch feature field and sets the occurrence node, deviation direction, associated batch, and confidence source for the deviation field. This embodiment decomposes the deviation source so that subsequent time window corrections can distinguish the different causes of raw materials, enzyme activity, and scheduling waiting. The correspondence between the node deviation sequence and the deviation source field is shown in Table 2.

[0060] Table 2. Correspondence between node deviation sequences and deviation source fields

[0061] The test value is below the target range and the enzyme inactivation is shifted. Stage inspection slope, raw material composition offset Raw material enzymatic hydrolysis deviation Deviation Correction Submodel Multiple batches of raw materials showed delayed compliance times. Enzyme potency offset and batch usage records of the same enzyme Enzyme activity deviation Enzyme preparation state vector Delay in sequence transfer after inspection meets standards Post-processing waiting record, material receiving time Scheduling wait bias Directed scheduling graph constraint edges Deviation occurs when low-confidence segments are included in the prediction. Fragment confidence markers and missing data test records Sample confidence bias Historical Sample Weights

[0062] In this embodiment, the deviation writeback execution unit categorizes the node deviation sequence according to Table 2. The node deviation sequence is formed by the difference between the planned inspection time, planned enzyme inactivation time, planned transfer time and the actual inspection time, actual enzyme inactivation time, and actual transfer time. The main correlation data is used to limit the scope of deviation source discrimination. The deviation source field is used to determine the writeback position. The subsequent processing position is used to determine the impact on the prediction model or scheduling chart. The raw material enzymatic solubility deviation is written into the raw material attribute sub-vector and deviation correction sub-model. The enzyme activity deviation is written into the enzyme preparation state sub-vector and potency offset record. The scheduling waiting deviation is written into the directed scheduling chart constraint edge. The sample confidence deviation is written into the historical sample weight. If a node deviation meets multiple performance criteria at the same time, the deviation writeback execution unit retains multiple candidate sources and determines the main source based on the continuous deviation distribution of the same raw material source, the same enzyme preparation batch and the same post-processing procedure. In this embodiment, through field-based writeback, the prediction input and scheduling constraints of subsequent batches can be updated synchronously.

[0063] ;

[0064] in, Indicates the first The node deviation sequence of each batch Indicates the deviation at the inspection node. Indicates the deviation at the enzyme inactivation node. Indicates the deviation of the transition node. , , These represent the actual testing time, actual enzyme inactivation time, and actual transfection time, respectively. , , These represent the planned testing time, planned enzyme inactivation time, and planned transsequencing time, respectively. For example, if a batch has a planned testing time of 5 hours and an actual testing time of 6 hours, a planned enzyme inactivation time of 7 hours and an actual enzyme inactivation time of 8 hours, and a planned transsequencing time of 8 hours and an actual transsequencing time of 10 hours, then... The physical meaning of this is that the test node is moved back by 1 hour, the enzyme inactivation node is moved back by 1 hour, and the transsequence node is moved back by 2 hours.

[0065] Preferably, when the deviation writer writes the raw material enzymaticity deviation, enzyme activity deviation, and scheduling waiting deviation, it establishes an association index between the deviation source and the batch comparability label and fragment confidence marker. The association index includes the deviation source field, associated raw material source, associated enzyme batch, associated process formulation version, occurrence node position, and sample confidence status. If multiple consecutive batches from the same raw material source show a low slope in the start-up phase and the enzyme batches are different, the association index sets the raw material source as the main association item. If the same enzyme batch shows a decrease in the slope of the main reaction phase under multiple raw material sources, the association index sets the enzyme batch as the main association item. If the same post-processing step shifts after multiple batches meet the standard, the association index sets the post-processing step capacity constraint edge as the main association item. If the deviation comes from a low-confidence test fragment, the association index reduces the weight of the sample on the deviation correction sub-model. This embodiment prevents a single abnormal record from being directly amplified into a global model correction through the joint index of deviation source and sample confidence.

[0066] Furthermore, the deviation writeback unit generates an updated feature set based on the associated index. The updated feature set includes the confirmed raw material solubility deviation, enzyme activity deviation, scheduling waiting deviation, sample confidence weight, and consecutive deviation count for the current batch. The consecutive deviation count is accumulated separately according to the same raw material source, the same enzyme batch, or the same post-processing process. If deviations from the same source occur consecutively in adjacent batches, the weight of the corresponding feature field in the deviation correction sub-model is increased. If the deviation comes from a low-confidence test segment, the weight of the corresponding historical sample is decreased. If the deviation only occurs in the waiting interval after the target has been met, the time window prediction model does not change the raw material solubility category, but updates the post-processing capacity constraint edge in the directed scheduling graph. If the deviation occurs in the non-target stage and is accompanied by a change in the stage test slope, the time window prediction model recalculates the reliable time window for the same type of unfeeded batches. In this embodiment, the updated feature set controls the model correction range, so that the prediction adjustment and scheduling adjustment act on the corresponding data links respectively.

[0067] In a preferred embodiment, after the scheduling coupling processor obtains the updated feature set, it recalculates the reliable time window for batches that have not yet been fed, and reconstructs the reaction resource occupancy window based on the new reliable time window. For batches that have been fed but have not yet undergone enzyme inactivation, it only updates the movable range of the enzymatic hydrolysis target node and the inspection node to avoid changing the feeding time that has already occurred and the recorded reaction start time. For batches that have undergone enzyme inactivation, it keeps the transfer node locked, and adjusts the order of unlocked batches only when the post-processing process has not received materials and there is a waiting conflict. After the resource occupancy change is written into the constraint edge of the directed scheduling graph, the scheduling coupling processor recalculates the executable order between the unlocked nodes and forms a rolling batch node plan. The rolling batch node plan retains the trigger source, the window before update, the window after update, and the node locking status for each rescheduling, which facilitates the subsequent execution deviation write-back device to determine whether the scheduling change comes from the reaction process change or from the post-processing capacity change. In this embodiment, by setting different update ranges for batches with different execution states, the plan recalculation is kept consistent with the on-site execution record.

[0068] ;

[0069] in, Indicates the first The plan for selecting nodes in the candidate node sequence in each batch. Indicates the first A sequence of candidate nodes in a batch This represents the set of candidate node sequences that satisfy the node locking state and the trusted time window. This represents the deviation of the candidate node sequence from the reliable time window. This represents the post-processing wait time after the candidate node sequence is generated. This represents the amount of batch switching inconsistency generated by the candidate node sequence. This represents the risk of premature enzyme inactivation corresponding to the candidate node sequence. , , , These represent the weights of each constraint. For example, if there are two candidate node sequences in a batch, sequence A has... , , , Sequence B , , , ,Pick , , , At that time, the comprehensive constraint value of sequence A is 2×1+1×3+1×1+3×0=6, and the comprehensive constraint value of sequence B is 2×2+1×1+1×1+3×1=9. Therefore, sequence A is selected as the node plan, which means that the node combination with less scheduling conflict is selected under the conditions of locking nodes and time window.

[0070] In this embodiment, after the rolling batch node plan is generated, the scheduling coupling processor divides each batch into four execution states: no feeding, feeding but enzyme inactivation, enzyme inactivation but not transferred, and transferred. For batches that have not been fed, the feeding node, enzyme hydrolysis target completion node, inspection node, enzyme inactivation node, and transfer node can be recalculated. For batches that have been fed but enzyme inactivation, the feeding node remains locked, and the enzyme hydrolysis target completion node, inspection node, and enzyme inactivation node can be updated. For batches that have been inactivated but not transferred, the enzyme inactivation node remains locked, and the transfer waiting order can only be adjusted under post-processing receiving constraints. For batches that have been transferred... All nodes of the sequence batch are locked and entered into the historical sample library. If the updated feature set shows that the reliable time window has moved forward, the feeding order of the batches that have not been fed can be rearranged, and the inspection nodes of the batches that have been fed but have not yet been inactivated can enter the executable range in advance. If the reliable time window has moved backward, the post-processing capacity can be released to other unlocked batches. If the update only comes from scheduling waiting deviation, the time window is not recalculated, and only the post-processing acceptance order is adjusted. This embodiment, through the layering of execution status, ensures that the system will not rewrite the real production records that have already been formed when it is rolling updated.

[0071] In a preferred embodiment, when the system runs in parallel with multiple batches, the batch feature builder continuously receives new batch documentation data, the enzymatic hydrolysis process predictor generates a reliable time window for the new batch, the scheduling coupling processor inserts the new batch into the existing directed scheduling graph, and the execution deviation writeback builder updates the feature set based on the inspection and transordering results of the batch being executed. When inserting a new batch, if it shares the same enzyme preparation batch with an existing batch, the new batch inherits the potency offset record in the enzyme preparation state subvector. If it shares the same raw material source with an existing batch, the new batch inherits the association index of the raw material enzymatic hydrolysis deviation. If it shares the same post-processing step with an existing batch, the new batch enters the same capacity constraint edge sorting. If the formulation version of the new batch is inconsistent with the historical sample, it only inherits the comparable information at the raw material and enzyme preparation levels, and does not inherit the target hydrolysis state interval. This embodiment enables the new batch to utilize historical writeback information without confusing different process objectives through hierarchical inheritance of raw materials, enzyme preparations, formulations, and post-processing constraints.

[0072] Preferably, when the system records abnormal risk identifiers, the abnormal risk identifiers are written into the batch node plan instead of a separate alarm table. The early enzyme inactivation risk identifier is bound to the enzyme hydrolysis target completion node and the enzyme inactivation node; the delayed target completion risk identifier is bound to the enzyme hydrolysis target completion node and the post-processing receiving node; the sequence transition waiting risk identifier is bound to the sequence transition node and the post-processing capacity constraint edge. When the risk identifier enters the scheduling coupling processor, it is used to limit the candidate node sequence set; when it enters the execution deviation writeback unit, it is used to assist in identifying the source of deviation. For example, if the batch corresponding to the early enzyme inactivation risk identifier has a final test value that exceeds the target range, the deviation correction sub-model increases the weight of the fast over-solution related features. If the batch corresponding to the delayed target completion risk identifier has a delayed final enzyme inactivation and the same trend occurs multiple times with the same enzyme preparation batch, the enzyme activity deviation field is given a higher weight. If the batch corresponding to the sequence transition waiting risk identifier still experiences waiting after passing the test, the scheduling waiting deviation field is written into the post-processing capacity constraint edge. In this embodiment, by embedding the risk identifiers into the node plan, the predicted risk, scheduling node, and execution deviation are placed in the same data link.

[0073] In this embodiment, the system's data update cycle is triggered by production execution events, which include new batch filing, material feeding confirmation, stage inspection entry, enzyme inactivation confirmation, sequence conversion confirmation, and post-processing material receiving confirmation. The batch feature builder updates the batch-level enzymatic hydrolysis features during new batch filing and stage inspection entry. The enzymatic hydrolysis process predictor recalculates the reliable time window after material feeding confirmation and stage inspection entry. The scheduling coupling processor updates the directed scheduling graph after changes in the reliable time window, post-processing material receiving confirmation, and sequence conversion confirmation. The execution deviation write-back device generates a node deviation sequence after stage inspection entry, enzyme inactivation confirmation, and sequence conversion confirmation. If the event only changes the traceability field without affecting the raw materials, enzyme preparations, inspection, or sequence conversion constraints, the time window recalculation is not triggered. If the event changes the target hydrolysis state range or formulation version, the corresponding batch formulation index is re-established and the old version samples are isolated. This embodiment uses an event triggering mechanism to focus data processing on fields that affect the enzymatic hydrolysis process and scheduling boundaries.

[0074] In a preferred embodiment, to avoid the long-term accumulation of historical samples causing expired data to affect current predictions, the batch feature builder sets sample lifecycle states for historical batches. Sample lifecycle states include currently available, traceable only, and pending review. Currently available samples have complete batch comparability labels, high fragment confidence markers, and clear sources of deviation, and can be included in the analyzability stratification model and time window prediction model. Traceable only samples are retained in the batch archive but do not participate in model correction. Samples pending review usually come from batches with missing test fragments, inconsistent formula versions, or conflicting transfer records. They can only be restored to currently available samples after subsequent supplementation of test or transfer records. When the deviation writeback device updates sample weights, it will synchronously update the sample lifecycle state. If the same historical sample continuously causes prediction deviations, it will change from currently available to pending review. If the sample pending review completes the key fields and is consistent with the nearest batch, it will re-enter the currently available state. This embodiment, through sample lifecycle management, ensures that historical data remains reviewable and traceable when used for subsequent batch predictions.

[0075] Furthermore, when outputting the batch node plan, the system writes the plan content into the management system record in a form that is usable for production execution. This includes the batch number, feeding node, lower bound of predicted target time, upper bound of predicted target time, inspection node, enzyme inactivation node, transfer node, post-processing takeover node, node lock status, abnormal risk identifier, and plan version number. The plan version number is updated as the reliable time window or the constraint edge of the directed scheduling diagram changes. Multiple plan versions for the same batch are retained. When generating the node deviation sequence, the execution deviation writer selects the plan version that is effective during actual execution as the comparison object and does not use subsequent updated but unexecuted plan versions. If the batch undergoes multiple reschedulings during execution, the trigger source and affected nodes are recorded for each rescheduling. Unaffected nodes inherit the lock status of the previous version. This embodiment uses plan version management to make the calculation basis of execution deviation clear and avoids interpreting the actual execution deviation with an incorrect plan version.

[0076] In this embodiment, the management system for agricultural product processing based on bio-enzymatic hydrolysis forms a closed-loop processing chain during operation, consisting of batch-level enzymatic hydrolysis characteristics, enzymatic stratification, reliable time windows, reaction resource occupancy windows, directed scheduling graphs, node deviation sequences, and updated feature sets. Batch-level enzymatic hydrolysis characteristics provide unified input, enzymatic stratification determines the reaction process category, reliable time windows express the time range in which the target hydrolysis state occurs, reaction resource occupancy windows convert the time range into scheduling boundaries, directed scheduling graphs organize reaction tanks, inspection waiting areas, and post-processing capacities into constrained nodes, node deviation sequences describe the offset between the plan and the actual execution, and updated feature sets write back the source of deviation to the raw materials, enzyme preparations, scheduling, and sample confidence fields. The above processing chain revolves around the main problem of mismatch between fixed process duration and actual enzymatic hydrolysis progress, without introducing hardware structures or external scenarios unrelated to bio-enzymatic hydrolysis management. This embodiment maintains the same calculation relationship between batch reaction states, production plan nodes, and actual execution deviations through end-to-end data referencing and progressive correction.

Claims

1. A management system for agricultural product processing based on bio-enzymatic hydrolysis, characterized in that, include: Batch feature builder is used to generate batch-level enzymatic hydrolysis features based on agricultural raw material batch attributes, enzyme batch potency, process formulation version, historical enzymatic hydrolysis test results, and actual transsequencing records. An enzymatic hydrolysis process predictor is used to output a reliable time window for reaching the target hydrolysis state and an abnormal risk indicator based on the batch-level enzymatic hydrolysis characteristics. A scheduling coupling processor is used to map the trusted time window to a reactive resource occupancy window and generate a batch node plan together with post-processing processes, inspection waits, and order delivery constraints. The execution bias writeback is used to adjust the reliable time window for subsequent batches based on the actual inspection value and the actual turnaround time.

2. The management system for agricultural product processing based on bio-enzymatic hydrolysis according to claim 1, characterized in that, The batch feature builder establishes a three-dimensional index based on raw material batches, enzyme preparation batches, and formulation versions. It aligns the fields for raw material moisture content, pretreatment method, key component content, enzyme preparation potency upon warehousing, opening status, feeding ratio, reaction temperature range, and pH range. It also configures source identifiers and valid ranges for the aligned fields and marks historical test results as time-series segments according to the reaction initiation stage, main reaction stage, and plateau stage, generating batch-level enzymatic hydrolysis features that include batch source, reaction conditions, and stage test sequences.

3. The management system for agricultural product processing based on bio-enzymatic hydrolysis according to claim 2, characterized in that, The enzymatic hydrolysis process predictor inputs the batch-level enzymatic hydrolysis characteristics into the enzymatic hydrolysis stratification model. The enzymatic hydrolysis stratification model determines the reaction rate stable type, delayed release type, rapid over-hydrolysis type, or fluctuation sensitive type based on the historical target achievement time residual, stage inspection slope, raw material composition offset, and enzyme potency offset. It also calls the corresponding time window prediction model for different types. The time window prediction model outputs the upper bound time, lower bound time, window confidence, and early enzyme inactivation risk indicator of the target hydrolysis state.

4. The management system for agricultural product processing based on bio-enzymatic hydrolysis according to claim 3, characterized in that, The scheduling coupling processor generates reaction resource occupancy boundaries based on the upper and lower bound times, sets feeding nodes, enzymatic hydrolysis standard achievement nodes, inspection nodes, enzyme inactivation nodes, and transfer nodes as constrainable nodes in the directed scheduling graph, and sets reaction tank occupancy, post-processing capacity, inspection waiting time, batch switching on the same line, and order delivery time as constraint edges. The movable range of the constrainable nodes is limited according to the window confidence level, and a batch node plan containing the triggering order of each node is generated.

5. The management system for agricultural product processing based on bio-enzymatic hydrolysis according to claim 4, characterized in that, The batch feature builder sets batch comparability labels for the time series segments. The batch comparability labels are jointly determined by the similarity of raw material sources, the consistency of pretreatment, the batch correlation of enzyme preparations, and the consistency of process formulation versions. When generating the batch-level enzymatic hydrolysis characteristics, the historical batch set is first screened based on the batch comparability label, then the actual transsequence records in the historical batch set are aligned with the stage test sequences, and the missing test fragments are configured with fragment confidence markers that are jointly defined by adjacent stages and similar batches.

6. The management system for agricultural product processing based on bio-enzymatic hydrolysis according to claim 5, characterized in that, The time window prediction model includes a category benchmark sub-model and a deviation correction sub-model. The category benchmark sub-model outputs a benchmark achievement curve based on the reaction rate stable type, delayed release type, rapid over-decomposition type, or fluctuation sensitive type. The deviation correction sub-model corrects the benchmark achievement curve on the time axis based on the raw material composition offset, enzyme potency offset, and stage test slope of the current batch, and determines the time period when the corrected curve intersects the target hydrolysis state interval as the reliable time window.

7. The management system for agricultural product processing based on bio-enzymatic hydrolysis according to claim 6, characterized in that, The scheduling coupling processor sets window constraint nodes and fixed constraint nodes for each batch in the directed scheduling graph. The window constraint nodes include the enzymatic hydrolysis target achievement node and the inspection node. The fixed constraint nodes include the feeding node, the enzyme inactivation node and the transfer node. When the resource occupancy boundaries of different batches overlap with the capacity of post-processing steps, a node adjustment sequence is generated according to window reliability, order delivery time, batch switching correlation and early enzyme inactivation risk marker, and the batch node plan is updated based on the node adjustment sequence.

8. The management system for agricultural product processing based on bio-enzymatic hydrolysis according to claim 7, characterized in that, The execution deviation write-back device receives the planned inspection time, planned enzyme inactivation time, and planned transsequence time from the batch node plan, and generates a node deviation sequence with the actual inspection time, actual enzyme inactivation time, and actual transsequence time. The execution deviation write-back device, according to the category of the enzymatic solubility stratification model, enzyme potency offset, stage inspection sequence change, and post-processing process waiting record, splits the node deviation sequence into raw material enzymatic solubility deviation, enzyme activity deviation, and scheduling waiting deviation, and writes them into the feature fields of the corresponding batches.

9. The management system for agricultural product processing based on bio-enzymatic hydrolysis according to claim 8, characterized in that, When writing the raw material enzymatic hydrolysis deviation, enzyme activity deviation, and scheduling waiting deviation, the execution deviation writer establishes an association index between the deviation source and the batch comparability label and the fragment confidence marker. When the same type of deviation occurs consecutively from the same raw material source or the same batch of enzyme preparation, the weights of the corresponding feature fields in the deviation correction sub-model are increased according to the consecutive number of deviation sources and the corresponding node positions, thereby generating an updated feature set for predicting the time window of subsequent batches.

10. The management system for agricultural product processing based on bio-enzymatic hydrolysis according to claim 9, characterized in that, After obtaining the updated feature set, the scheduling coupling processor recalculates the reliable time window for batches that have not yet been fed, and updates only the movable range of the enzymatic hydrolysis target node and the inspection node for batches that have been fed but have not been inactivated. The batches that have undergone enzyme inactivation are kept locked at the transfer node, and the resource usage changes generated by the recalculation are written into the constraint edges of the directed scheduling graph. The capacity constraints of the unlocked post-processing steps are reordered, and a rolling batch node plan is generated according to the node adjustment sequence.

Images