A method and system for ingredient optimization for premix

Abstract

This invention discloses a method and system for optimizing the formulation of premixed feed, relating to the field of formulation optimization and process control technology for premixed feed. By integrating multi-source data acquisition, adaptive estimation modeling, and dynamic optimization control, this invention effectively addresses the formulation error problem caused by batch variations of raw materials in premixed feed production. It utilizes multi-sensor fusion technology to acquire raw material characteristic data in real time and updates it online using historical data, significantly improving the accuracy and reliability of nutrient parameter estimation and avoiding the error amplification caused by fixed models in traditional methods. By introducing a chance-constrained optimization model, the nutrient compliance conditions are transformed into probabilistic forms, and statistical uncertainty is handled through safety factors and robustness radii. This allows the formulation ratio to adapt to raw material fluctuations while meeting cost, process, and regulatory constraints, ensuring consistency of nutrient concentration between batches.

Description

Technical Field

[0001] This invention relates to the field of premixed feed formulation optimization and process control technology, and in particular to a formulation optimization method and system for premixed feed. Background Technology

[0002] In modern premixed feed production, the ingredient formulation process needs to achieve a stable and consistent target content under dozens of components, strict nutritional constraints, and batch-to-batch raw material fluctuations. The industry usually uses the coefficient of variation (CV) to measure uniformity. Regulatory and guidelines require the CV of premixed / supplementary feeds to be ≤10%, while for drug-containing diluted premixes, the requirement is even ≤5%. The measurement / mixing error of trace amounts of highly active additives will significantly amplify the final deviation, directly affecting product consistency and compliance.

[0003] Existing processes are mostly based on static formulations using ingredient lists and minimum cost or multi-objective planning. Although SPC and near-infrared (NIR) rapid detection have been introduced to identify raw material differences, the nutritional content of raw materials fluctuates significantly depending on the origin, batch, storage, and processing conditions. Furthermore, the closed loop from online data to formulation decisions is insufficient, making it difficult to perform intra-batch or near-real-time ratio corrections based on the actual characteristics of specific batches. As a result, when multiple raw materials fluctuate simultaneously, inter-batch deviations accumulate, the risk of trace component compliance increases, and further deviations from the target may occur due to nutrient loss during subsequent heat processing. Therefore, there is an urgent need for a technology that can improve the accuracy and consistency of formulation under conditions of uncertain raw material characteristics. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] This invention provides a method and system for optimizing the formulation of premixes to solve the problem that batch-to-batch nutritional variations in raw materials lead to amplified formulation errors, and that existing methods cannot dynamically respond to changes in actual characteristics.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, embodiments of the present invention provide a method for optimizing the formulation of premixes, comprising:

[0008] Step S1: Obtain current batch characteristic data of various raw materials, wherein the current batch characteristic data includes at least one of nutrient content information and moisture content;

[0009] Step S2: Combine the current batch characteristic data with the historical batch characteristic data to obtain updated estimates of the nutritional parameters of each raw material;

[0010] Step S3: Under nutritional constraints, regulatory constraints, and process / measurement constraints, determine the formulation ratio based on the updated estimate;

[0011] Step S4: Generate a batching control command based on the formula ratio and send it to the metering and feeding equipment for execution;

[0012] Step S5: During the execution process, collect weighing or flow feedback, correct the deviation of the batching control command, and write the execution result back to the historical batch characteristic data for adaptive updating;

[0013] The formulation ratio refers to the mass fraction or amount of each raw material in a single batch; the updated estimate includes at least the expected value, standard deviation and / or range of each nutritional parameter.

[0014] As a preferred embodiment of the batching optimization method for premixed materials described in this invention, the current batch characteristic data is obtained by multi-source sensor fusion, wherein the sensors include at least one of near-infrared spectral sensors, and at least one of moisture, temperature, bulk density, particle size or flow rate sensors, and includes steps for removing and calibrating outliers.

[0015] As a preferred embodiment of the ingredient optimization method for premixed materials described in this invention, the adaptive update process employs a multi-scale data processing unit to process macroscopic characteristic data and microscopic characteristic data respectively, and performs feature alignment and weight allocation before the proportion is determined.

[0016] As a preferred embodiment of the ingredient optimization method for premixed materials described in this invention, the updated estimation includes:

[0017] Using historical batch characteristic data as a priori, the current batch characteristic data is statistically updated online to obtain the interval or distribution characterization of each nutrient parameter, and the estimation results of the correlation between raw materials are given.

[0018] As a preferred embodiment of the ingredient optimization method for premixed materials described in this invention, the correlation modeling is implemented using a graph neural network.

[0019] Using raw materials as graph nodes and formulation co-occurrence relationships or statistical correlations as edges, the correlations between raw material characteristics are learned to improve the updated estimation.

[0020] As a preferred embodiment of the ingredient optimization method for premixes described in this invention, the determination of the formula ratio includes solving an optimization problem with cost or deviation risk as the objective, and satisfying the nutritional compliance conditions under opportunity constraints, while simultaneously satisfying the upper and lower limits of raw material usage, ratio, inclusion / exclusion and specified restrictions.

[0021] The nutritional achievement conditions under the opportunity constraint are defined as follows:

[0022] Step C1, in formula optimization, nutrient achievement is written as a probabilistic condition for the total nutrient content of the batch and coupled with the formula decision:

[0023] ,

[0024] in, Represents a probability measure. Indicators of nutritional value The random variable representing the total content in a single batch. As an indicator The lower limit of compliance, As an indicator The upper limit of the standard, and These represent the lower and upper limits of the permissible default probability, respectively. The set of nutritional indicators to be included in the constraints;

[0025] Step C2: When the updated estimate provides the mean and variance of each indicator, the chance constraint is rewritten into a computable form using a normal approximation:

[0026] ,

[0027] in, for Expected estimate for Standard deviation estimate This is the lower limit safety factor. For the upper limit safety factor, The quantile function is the standard normal distribution function;

[0028] Step C3, when there is only a finite sample or when drift exists, absorb statistical uncertainty by replacing it with a distribution-independent conservatism:

[0029] ,

[0030] in, This refers to the risk coefficient on one side obtained by mapping according to the target probability level. The robust radius is measured on a standard deviation scale to cover mean estimation bias and drift.

[0031] Step C4: When the raw material nutrition is only given in intervals, use interval conservatism (lower limit is the lower bound of the interval, upper limit is the upper bound of the interval) or scenario method to solve the problem.

[0032] Step C5: Combine the deterministic substitutions obtained in step C2 or C3 with the cost / deviation risk objectives, raw material upper and lower limits, ratios, inclusion / exclusion, and process / measurement constraints to form a linear-second-order cone mixing model, and use the interior point method or a commercial solver to solve the formulation ratio.

[0033] As a preferred embodiment of the ingredient optimization method for premixed feed described in this invention, the determination of the formula ratio further includes calculating the nutritional deviation compensation value:

[0034] Based on the difference between the updated estimate and the target nutrient range, as well as the nutrient loss estimate in the process, the set dosage of trace high-activity components is adjusted to compensate for the loss.

[0035] The calculation steps for the nutritional deviation compensation value include:

[0036] Step D1: When the updated estimate falls outside the target interval, calculate the basic deviation and record its direction.

[0037] ,

[0038] in, Indicators of nutritional value The basic deviation relative to the target interval. and This represents the lower / upper limit of the target for this indicator. This is an expected estimate of the total content of this indicator in a single batch;

[0039] Step D2: Convert the basic deviation into the total compensation amount that needs to be added on the formulation side, based on process losses.

[0040] ,

[0041] in, Nutritional indicators Total compensation amount This refers to the overall relative loss rate of this indicator in the preparation-transportation-mixing-metering process;

[0042] Step D3: Based on the intra-batch closed-loop capability and inter-batch adjustment strategy, the total compensation is distributed across the two channels:

[0043] ,

[0044] in, This is the batch compensation amount, applied in real time to the current batch. This is the inter-batch compensation amount, used for setting updates in the next batch. The allocation coefficient is determined by the frequency of measurements within the batch, response lag, and... Comprehensive tuning;

[0045] Step D4: Crop the upper limit of amplitude and the upper limit of channel for the compensation amount:

[0046] ,

[0047] in, This represents the upper limit of the amplitude available in the channels within the batch. This is the upper limit of the inter-batch channel. The safety limit for the total compensation of this indicator is set offline based on process limitations, regulations, and formulation sensitivity.

[0048] Step D5: When it is necessary to transfer nutritional compensation to specific trace highly active components, the allocation is performed using least squares or linear programming according to the nutrient-raw material mapping coefficient, while being constrained by the upper and lower limits of raw material usage, ratio, and inclusion / exclusion restrictions.

[0049] As a preferred embodiment of the batching optimization method for premixed materials described in this invention, the batching control command includes setting the target for loss-in-weight or gain-in-weight metering feeders, the feeding sequence and batch metering threshold, and implementing intra-batch closed-loop correction and inter-batch setting update based on weighing / flow feedback.

[0050] Secondly, the present invention provides a batching optimization system for premixed materials, comprising,

[0051] The data acquisition module is used to collect current batch characteristic data of various raw materials, and to associate, calibrate and store the current batch characteristic data with historical batch characteristic data;

[0052] The estimation and modeling module is used to calculate updated estimates of the nutritional parameters of each raw material based on the current batch characteristic data and historical batch characteristic data, establish the correlation between raw materials, and generate parameters and features for formula optimization.

[0053] The optimization and compensation module is used to solve the formula ratio under nutritional, specification and process / measurement constraints, and calculate the nutritional deviation compensation value based on the updated estimate and process loss to adjust the formula ratio;

[0054] The control module is used to generate batching control commands and send them to the metering and feeding equipment, collect weighing or flow feedback, and implement intra-batch corrections and inter-batch setting updates.

[0055] Thirdly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the batching optimization method for premixes as described in the first aspect of the present invention.

[0056] The beneficial effects of this invention are as follows: By integrating multi-source data acquisition, adaptive estimation modeling, and dynamic optimization control, this invention effectively addresses the batch variation error problem caused by raw material batches in premix production. Utilizing multi-sensor fusion technology to acquire raw material characteristic data in real time and updating it online in conjunction with historical data significantly improves the accuracy and reliability of nutrient parameter estimation, avoiding the error amplification caused by fixed models in traditional methods. Introducing a chance-constrained optimization model transforms nutrient compliance conditions into probabilistic forms, and handling statistical uncertainty through safety factors and robustness radii, allows the formulation ratio to adapt to raw material fluctuations while meeting cost, process, and regulatory constraints, ensuring consistency of nutrient concentration between batches. Furthermore, the nutrient deviation compensation mechanism dynamically allocates compensation to intra-batch and inter-batch channels by analyzing differences in target intervals and process losses, achieving precise adjustment of trace components and reducing the risk of cumulative errors.

[0057] This invention achieves full automation from data acquisition to execution through closed-loop control and feedback correction. This not only reduces the need for manual intervention but also enhances the robustness and flexibility of the production system, providing a reliable guarantee for the stable quality of premixed materials. Attached Figure Description

[0058] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation on the scope of this application.

[0059] Figure 1 This is a schematic flowchart of the ingredient optimization method for premixed materials in the embodiments.

[0060] Figure 2 This is a schematic diagram of the ingredient optimization system for premixed materials in the embodiment. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0062] All terms used in this application (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0063] For example, the terms “first” and “second” used in this application are only used to distinguish and describe similar objects, to differentiate the first object from another object, and are not used to describe a specific order or sequence, nor should they be interpreted as indicating or implying relative importance.

[0064] This application proposes a method for optimizing the formulation of premixes, combining... Figure 1 As shown, the method includes:

[0065] Step S1: Obtain current batch characteristic data for multiple raw materials. The current batch characteristic data includes at least one of nutrient content and moisture content. In this embodiment, the current batch characteristic data refers to the sensor measurement and testing records bound to the timeline of a single batch of ingredients, including timestamps, batch identifiers, and measurement point sources, used to summarize batch-level statistics in subsequent steps. The default sampling period is 2 seconds, adjustable from 1 to 5 seconds. The batch aggregation window covers the entire process from feeding to discharging by default, with a minimum duration of 60 seconds to filter out brief disturbances. On-site estimation of nutrient content is primarily based on near-infrared model output, with periodic verification using test values ​​(once a week or every 200 batches) as necessary. When only moisture content is available without nutrient content, the system retains the batch identifier and proceeds to the subsequent update estimation process; fields that are not missing participate in the calculation normally. Optionally, if multiple sampling points exist on-site, they are first aligned chronologically before batch-level summarization. If the sampling interval exceeds 10 seconds or the batch identifier is missing, the time period will be marked as unavailable and the steady-state statistic of the most recent valid batch will be used as a temporary substitute value until the end of the current batch, at which point the test value will be used to fill the gap.

[0066] Step S2: Combine the current batch characteristic data with the historical batch characteristic data to obtain updated estimates of the nutritional parameters of each raw material;

[0067] Step S3: Determine the formulation proportions based on the updated estimate under nutritional constraints, regulatory constraints, and process / measurement constraints; regulatory constraints may refer to regional or local regulations.

[0068] Step S4: Generate a batching control instruction based on the formula ratio and send it to the metering and feeding equipment for execution;

[0069] Step S5: During the execution process, collect weighing or flow feedback, correct deviations in the batching control commands, and write the execution results back to the historical batch characteristic data for adaptive updates;

[0070] In one embodiment, the current batch characteristic data is obtained by multi-source sensor fusion. The sensors include at least one near-infrared spectral sensor and at least one of moisture, temperature, bulk density, particle size, or flow rate sensors, and include outlier removal and calibration steps. Specifically, multi-source sensor fusion is time-aligned and uses weighted fusion to obtain representative batch values. The weights are automatically adjusted based on sensor stability and calibration freshness. The default alignment tolerance is ±5 seconds. Outlier determination uses a three-fold median absolute deviation criterion; continuous out-of-threshold conditions must be met for at least 30 seconds to be considered an outlier segment. Near-infrared model drift is calibrated using a reference sample, with a recommended calibration cycle of every 7 days. Moisture and flow rate sensors are calibrated using single-point or two-point methods, with a recommended cycle of every 30 days. Optionally, when near-infrared spectroscopy is unavailable, only a downgraded version of the current batch characteristic data is generated using moisture content, bulk density, and particle size, which can still be used for subsequent updates and estimations. If any critical sensor is continuously disconnected for more than 120 seconds, the batch enters conservative mode and automatically increases the conservatism during the optimization phase, while recording an alarm.

[0071] In one embodiment, the adaptive update process employs a multi-scale data processing unit to process macroscopic characteristic data (moisture content, bulk density, particle size, bulk material flowability) and microscopic characteristic data (amino acids, vitamins, minerals, etc.) respectively, and performs feature alignment and weight allocation before the ratio is determined.

[0072] For example, the multi-scale data processing unit performs scale standardization on macroscopic and microscopic characteristics separately and aggregates them at the batch level, then aligns them within a unified feature space. The default time window for the macroscopic channel is 1-5 minutes, while the microscopic channel is updated at batch granularity. The initial weights can be set to 30% for macroscopic and 70% for microscopic, and fine-tuning is allowed within 10% based on the validation set error during runtime. When the microscopic channel is missing data, the system degenerates to only the macroscopic channel and automatically tightens the conservatism of the nutrient constraints during the optimization phase. If the effective data in a batch of the macroscopic channel is less than 70%, that batch does not participate in model updates and is only used for control loop closure.

[0073] In one embodiment, the updated estimate includes:

[0074] Using historical batch characteristic data as prior, the system performs online statistical updates on the current batch characteristic data to obtain interval or distribution representations of each nutrient parameter and provides estimates of the correlation between raw materials. Similarly, online statistical updates are based on a sliding time window or a sliding batch number, with the default window being the most recent 100 batches, adjustable from 50 to 200 batches. The update cycle defaults to once at the end of each batch; if the batch duration exceeds 30 minutes, an additional update is inserted midway. Interval representation uses the quantile method, defaulting to using the 5th and 95th quantiles as interval endpoints. Distribution representation outputs mean and standard deviation parameters when the sample is sufficient. The correlation between raw materials is calculated by default as a parallel estimation of linear correlation and rank correlation, taking the larger of the two absolute values ​​as the strength index, with a threshold of not less than 0.3 for establishing the association. Optionally, when frequent raw material replacement leads to insufficient sample size, only interval representations are output and correlation updates are temporarily suspended. For missing data handling, if a raw material has fewer than 5 valid samples within the window, only the interval is provided for that raw material, without outputting distribution parameters.

[0075] In one embodiment, correlation modeling is implemented using a graph neural network:

[0076] Using raw materials as graph nodes and co-occurrence relationships or statistical correlations as edges, the system learns the associations between raw material characteristics to improve and update estimates. Optionally, the edges of the graph structure are determined by both co-occurrence frequency and correlation strength; edges with a co-occurrence frequency of less than 5 are excluded by default. Edge directions are still treated as undirected to align with statistical correlations. The training refresh cycle defaults to 1-7 days; the early stopping criterion for each training session is based on the validation set error no longer improving, with a tolerance of 10 rounds. Graph node features are constructed by concatenating the current batch and representative statistics within the window, and the output is used to correct correlations and parameter estimates for missing raw materials. If the cumulative number of batches is less than 50 or the main raw material changes, leading to topological instability, the system degenerates into an implementation path that only uses the statistical correlation matrix, maintaining consistency between the input and output interfaces. When training resources are insufficient, the current training cycle is skipped, and the model from the previous cycle is reused and trained in the next cycle.

[0077] In one embodiment, determining the formulation ratio includes solving an optimization problem with cost or deviation risk as the objective, while satisfying the nutritional compliance conditions under opportunity constraints, and simultaneously satisfying the upper and lower limits of raw material usage, ratios, inclusion / exclusion and specified restrictions.

[0078] The definition of opportunity-constrained nutritional achievement conditions is as follows:

[0079] Step C1, in formula optimization, nutrient achievement is written as a probabilistic condition for the total nutrient content of the batch and coupled with the formula decision:

[0080] ,

[0081] in, Represents a probability measure. Indicators of nutritional value The random variable representing the total content in a single batch. As an indicator The lower limit of compliance, As an indicator The upper limit of the standard, and These represent the lower and upper limits of the permissible default probability, respectively. The set of nutritional indicators to be included in the constraints;

[0082] Step C2: When the updated estimate provides the mean and variance of each indicator, the chance constraint can be rewritten into a computable form using a normal approximation.

[0083] ,

[0084] in, for Expected estimate for Standard deviation estimate This is the lower limit safety factor. For the upper limit safety factor, It is the quantile function of the standard normal distribution function; this form corresponds to the second-order cone feasible region, which is convenient for solving in conjunction with cost or deviation risk objectives;

[0085] Step C3, when there is only a finite sample or when drift exists, absorb statistical uncertainty by replacing it with a distribution-independent conservatism:

[0086] ,

[0087] in, The risk coefficient on one side, obtained by mapping to the target probability level, is given by the Cantelli inequality and can be written as follows: , The robust radius is measured on a standard deviation scale to cover mean estimation bias and drift; this form also corresponds to a second-order cone structure.

[0088] Step C4: When only a range of raw material nutrients is given, the solution can be obtained by using range conservatism (using the lower bound of the range as the lower limit and the upper bound of the range as the upper limit) or scenario method (sampling within the range to generate representative scenarios and setting the coverage rate).

[0089] Step C5 combines the deterministic substitutions obtained in Step C2 or C3 with the cost / deviation risk targets, raw material upper and lower limits, ratios, inclusion / exclusion, and process / metering constraints to form a linear-second-order cone hybrid model. The formula proportions are then solved using the interior-point method or a commercial solver, and linked with deviation compensation iterations. Furthermore, the optimization solution defaults to an interior-point algorithm, with a convergence tolerance set to 0.1% to 0.01%, and a maximum solution time of 30 seconds, adjustable from 10 to 60 seconds depending on the production line cycle time. The default target range for the probability of default is 1% to 5%, with the corresponding safety factor obtained from a preset mapping. When the sample size is limited, the robustness radius parameter is taken from the upper quantile of historical residuals and decays slowly over time. The solution results are rounded to the nearest metering resolution before being distributed, with a default resolution of 0.1 kg or 0.1% of the formula percentage to meet the minimum operable granularity of the metering equipment. Optionally, if the solution fails to converge or is deemed infeasible within a given time, the system rolls back to the most recent feasible formula and triggers a conservative mode with deviation compensation until the next update is completed; during the rollback, the upper and lower limits of raw materials and their mutual exclusion relationships remain unchanged. If the solution is deemed infeasible three times consecutively, the ratio of non-critical raw materials is automatically relaxed by no more than 1%, and the reason is recorded for manual review.

[0090] Specifically, this paper presents a path to transform nutritional uncertainty into chance constraints and integrate it with the solver. First, a probabilistic model is used to uniformly characterize the achievement criteria. Then, based on data sufficiency, two implementation methods are adopted: When the distribution is approximately normal, the mean and standard deviation, along with quantiles, form a deterministic equivalence, directly mapping the probability level to a safety factor. This results in a well-structured calculation that is easily coordinated with cost and process constraints. When the distribution is unknown or the sample size is limited, a conservative alternative constructed using a one-sided inequality absorbs statistical bias, and a robust radius is introduced on the standard deviation scale to enhance tolerance to drift and estimation errors. Both forms belong to the second-order cone model and can be used with upper and lower bounds on dosage, ratio relationships, inclusion / exclusion, and measurement constraints. The solution is obtained by combining the two methods. When only interval information is available, interval conservatism provides a fast and feasible deployment solution, while scenario-based methods are more flexible when specific confidence levels need to be covered. In terms of parameter settings, target probability levels can be stratified according to nutrient importance, and safety coefficients can be obtained using quantiles or one-sided mapping. The robust radius can be calibrated by historical bias or cross-validation to achieve a trade-off between risk and cost. In this embodiment, the mean and standard deviation parameters are derived from the aforementioned online statistical update window statistics. The one-sided risk coefficient is obtained from the target probability level through a pre-set mapping table. The robust radius parameter uses the high quantile residual of the historical validation set as its initial value and decays at a fixed rate each week to avoid excessive conservatism. Target probability levels can be grouped and set according to nutrient importance, with core nutrients referencing lower default probabilities and secondary nutrients referencing higher default probabilities. Optionally, in the stage where only interval information is available, the number of representative scenarios for the scenario-based method is defaulted to between 10 and 50 to meet the computation time limit requirements during deployment. When a representative scenario becomes infeasible, similar scenarios are merged and the number is reduced by half before retrying.

[0091] In one embodiment, determining the formulation ratio also includes calculating a nutritional deviation compensation value:

[0092] Based on the difference between the updated estimate and the target nutrient range, as well as the nutrient loss estimate in the process, the set dosage of trace high-activity components is adjusted to compensate for the loss.

[0093] The calculation steps for the nutritional deficit compensation value include:

[0094] Step D1: When the updated estimate falls outside the target interval, calculate the basic deviation and record its direction.

[0095] ,

[0096] in, Indicators of nutritional value The basic deviation relative to the target interval (can be positive or negative). and This represents the lower / upper limit of the target for this indicator. This is an expected estimate of the total content of this indicator in a single batch;

[0097] Step D2: Convert the basic deviation into the total compensation amount that needs to be added on the formulation side, based on process losses.

[0098] ,

[0099] in, Nutritional indicators The total compensation amount (positive value indicates an increase, negative value indicates a decrease). This refers to the overall relative loss rate of this indicator in the preparation-transportation-mixing-metering process;

[0100] Step D3: Based on the intra-batch closed-loop capability and inter-batch adjustment strategy, the total compensation is distributed across the two channels:

[0101] ,

[0102] in, This is a batch-wide (weighing / flow feedback closed loop) compensation amount, applied in real time to the current batch. This is the inter-batch compensation amount, used for setting updates in the next batch. The allocation coefficient is determined by the frequency of measurements within the batch, response lag, and... Comprehensive tuning;

[0103] Step D4: Crop the upper limit of amplitude and the upper limit of channel for the compensation amount to avoid over-adjustment.

[0104] ,

[0105] in, This represents the upper limit of the amplitude available in the channels within the batch. This is the upper limit of the inter-batch channel. The safety limit for the total compensation of this indicator is set offline based on process limitations, regulations, and formulation sensitivity.

[0106] Step D5: When nutrient compensation needs to be transferred to specific trace highly active components, it can be allocated using least squares or linear programming based on the nutrient-raw material mapping coefficient, while being constrained by upper and lower limits of raw material usage, ratios, and inclusion / exclusion restrictions. Specifically, the nutrient-raw material mapping coefficient is preferentially taken from the regression results of supplier specifications and historical test matrices. If the two are inconsistent, the test matrix results shall prevail and be reviewed monthly. The default allocation strategy prioritizes trace components with low fluctuations and high stability. The minimum compensation particle size for a single batch is not less than one ten-thousandth of the formulation percentage or the larger of the minimum metering particle size of the equipment. The initial value of the allocation coefficient can be set to 50%, and then automatically fine-tuned within 10% based on the closed-loop deviation. Optionally, when the mapping coefficient is uncertain or there is inter-batch drift, conservative compensation is concentrated on a single stable component until the coefficient stabilizes before proportional allocation. If the compensation reaches the upper limit of the amplitude, the inter-batch channel is reduced first, and then the intra-batch channel is reduced to avoid over-adjustment to the current batch.

[0107] Specifically, the sources of deviation are decomposed into two parts: first, the difference between the updated estimate and the target interval, which is used to obtain the basic deviation in a segmented form while retaining the direction, avoiding unnecessary actions within the interval; second, the loss caused by the processing, which is uniformly converted with a single loss rate to form a total compensation amount that can be directly issued. Then, the total compensation is split into two paths, intra-batch and inter-batch, through the allocation coefficient. Intra-batch paths converge quickly based on weighing or flow feedback, while inter-batch paths are used to slowly correct drift. The sum of the two corresponds to the same physical meaning. To prevent over-adjustment, amplitude and channel upper limits are introduced for trimming, which can be preset according to regulations and equipment capabilities, or fine-tuned during operation based on statistical deviations. At the raw material level, the compensation amount can be distributed among multiple trace components by combining nutrient-raw material mapping, and optimized together with existing dosages, ratios, and process limitations to form a consistent control and setting closed loop. Similarly, the process loss rate is obtained based on the quality balance of small-sample tests and historical batches, and is updated by default every quarter. If new test data is lacking, the results of the previous period are used, and the conservatism is increased during the optimization stage to cover uncertainty. The response of the intra-batch channel is based on weighing or flow feedback, with a default allowable error range of ±3 percentage points. Exceeding this limit triggers secondary compensation. The inter-batch channel update is performed once per batch by default, and an additional correction can be added when three consecutive batches have the same deviation direction. When the compensated formula triggers a mutual exclusion relationship for a certain raw material, the mutual exclusion relationship is maintained first, and the compensation magnitude is reduced until feasibility is restored.

[0108] In one embodiment, the batching control command includes setting the target setpoint, feeding sequence, and batch metering threshold for the loss-in-weight or gain-in-weight metering feeder, and implementing intra-batch closed-loop correction and inter-batch setting updates based on weighing / flow feedback. Optionally, the batching control command includes the target setpoint, feeding sequence, coarse / fine feeding switching point, and advance amount, all issued on a batch-by-batch basis. The default filtering time constant for the weighing signal is 0.5~2 seconds, with an allowable overshoot range of ±3 percentage points; the minimum opening / closing time for valves or frequency converters is 200 milliseconds, and the minimum stable feeding time is 1 second. The intra-batch closed-loop calculation calculates the correction amount based on the remaining deviation and the target setpoint, and updates it in real time without exceeding the amplitude limit; the inter-batch setting is executed once after the batch ends to slowly correct long-term deviations. If the weighing or flow feedback is interrupted for more than 10 seconds, it automatically switches to a safety mode to maintain the current feeding rate until the batch ends, and records an alarm for manual review.

[0109] This application proposes a batching optimization system for premixed materials, combined with Figure 2 As shown, the system includes:

[0110] The data acquisition module is used to collect current batch characteristic data of various raw materials, and to associate, calibrate and store the current batch characteristic data with historical batch characteristic data;

[0111] The estimation and modeling module is used to calculate updated estimates of the nutritional parameters of each raw material based on the current batch characteristic data and historical batch characteristic data, establish the correlation between raw materials, and generate parameters and features for formula optimization.

[0112] The optimization and compensation module is used to solve the formula ratio under nutritional, specification and process / measurement constraints, and calculate the nutritional deviation compensation value based on the updated estimate and process loss to adjust the formula ratio;

[0113] The control module is used to generate batching control commands and send them to the metering and feeding equipment, collect weighing or flow feedback, and implement intra-batch corrections and inter-batch setting updates.

[0114] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a batching optimization method for premixed materials as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0116] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of this application and form different embodiments. For example, all the embodiments above can be used in any combination. The information disclosed in this background section is intended only to enhance the understanding of the general background of this application and should not be construed as an admission or in any way implying that such information constitutes prior art known to those skilled in the art.

Claims

1. A method for ingredient optimization of a premix, characterized in that, The method comprises the following steps: Step S1: obtaining current batch characteristic data of multiple raw materials, the current batch characteristic data at least including one of nutrient content information and moisture content; Step S2: combining the current batch characteristic data with historical batch characteristic data to obtain an updated estimation of each raw material nutrient parameter; Step S3: determining a formula proportion based on the updated estimation under the constraints of nutrient, regulation and process / metering; Step S4: generating a batching control instruction according to the formula proportion and issuing it to a metering and feeding device for execution; Step S5: collecting weighing or flow feedback during execution, correcting the batching control instruction, and writing the execution result back to the historical batch characteristic data for adaptive updating; Wherein, the formula proportion is the mass fraction or amount of each raw material in a single batch; the updated estimation at least includes the expectation, standard deviation and / or interval of each nutrient parameter; In step S3, the determination of the formula proportion includes solving an optimization problem with cost or deviation risk as the target, and meeting the opportunity constraint of nutrient standard, and at the same time meeting the upper and lower limits, ratio, inclusion / exclusion and regulation restrictions of raw material usage; The definition of the opportunity constraint of nutrient standard is as follows: Step C1: in formula optimization, write the nutrient standard as a probability condition of the total nutrient amount of the batch and couple it with the formula decision: ， wherein, denotes a probability measure, denotes a nutritional indicator a random variable for the total content in a single batch, is the indicator of the lower limit of compliance, is the indicator of the upper limit of compliance, and are the allowed violation probabilities for the lower and upper limits, respectively, is the set of nutritional indicators subject to inclusion constraints; Step C2: when the updated estimation gives the mean and variance of each index, use normal approximation to rewrite the opportunity constraint into a calculable form: ， wherein is a desired estimate of is a standard deviation estimate of is a lower-side safety factor, is an upper-side safety factor, is a quantile function of a standard normal distribution function; Step C3: when there are only limited samples or there is drift, use distribution-independent conservative replacement to absorb statistical uncertainty: ， wherein, is a one-sided risk factor mapped from a target probability level, is a robust radius measured in standard deviation scale to cover the mean estimation bias and drift; Step C4: when the nutrient of the raw material is only given as an interval, use interval conservatism or scenario method to enter the solution; Step C5: the deterministic replacement obtained in step C2 or step C3 is used to form a linear-second-order cone mixed model together with the cost / deviation risk target, raw material upper and lower limits, ratio, inclusion / exclusion and process / metering constraints, and the formula proportion is solved using the interior point method or a commercial solver.

2. A method for ingredient optimization of a premix as claimed in claim 1, wherein, The current batch characteristic data is obtained by multi-source sensor fusion, the sensors at least including one of near-infrared spectrum sensors, and at least one of moisture, temperature, bulk density, particle size or flow sensors, and containing outlier rejection and calibration steps.

3. A method for ingredient optimization of a premix as claimed in claim 1, wherein, The adaptive updating process uses a multi-scale data processing unit to process macroscopic and microscopic characteristic data respectively, and performs feature alignment and weight allocation before proportion determination.

4. A method for ingredient optimization of a premix as claimed in claim 1, wherein, The updated estimation includes: Using the historical batch characteristic data as a priori, the current batch characteristic data is updated online to obtain the interval or distribution representation of each nutrient parameter, and the estimation result of the correlation between raw materials is given.

5. A method for ingredient optimization of a premix as claimed in claim 4, characterized in that, Correlation modeling is realized by using graph neural network: Taking raw materials as graph nodes and formula co-occurrence relationship or statistical correlation as edges, the correlation between raw material characteristics is learned to improve the updated estimation.

6. A method for ingredient optimization of a premix as claimed in claim 1, wherein, The determination of the formula proportion also includes calculating a nutrient deviation compensation value: Based on the difference between the updated estimation and the target nutrient range and the nutrient loss estimation of the process, the set amount of trace high-activity components is compensated and adjusted; The calculation steps of the nutrient deviation compensation value include: Step D1, when the update estimate falls outside the target interval, calculate the base bias and record the direction: ， wherein, represents a nutritional indicator a base bias relative to a target interval, and a target lower / upper limit for the indicator, is a desired estimate of the total content of the indicator for a single batch; Step D2, convert the base bias to total compensation needed on the recipe side by process loss: ， wherein is a nutritional index is the total compensation amount, is the overall relative loss rate of the index in the preparation-delivery-mixing-measuring link; Step D3, distribute the total compensation to two channels according to the in-batch closed loop capability and inter-batch setting adjustment strategy: ， wherein, is the batch within compensation amount, acting on the current batch in real time, is the batch between compensation amount, used for setting update of the next batch, is the distribution coefficient, obtained by batch within measurement frequency, response lag and comprehensive setting; Step D4, implement amplitude upper limit and channel upper limit clipping on the compensation: ， wherein, is the upper limit of the magnitude available for the in-batch channel, is the upper limit for the inter-batch channel, is the safety upper limit for the total compensation of the index, all three of which are set offline according to process limitations, regulations, and formulation sensitivity; Step D5, when the nutritional compensation needs to be transmitted to specific micro-high activity components, perform least squares or linear programming allocation according to the nutrition-raw material mapping coefficient, while being subject to the upper and lower limits of raw material usage, ratio and inclusion / exclusion constraints.

7. A method for ingredient optimization of a premix as claimed in claim 1, wherein, The batching control instructions include target given, feeding sequence and batch metering threshold settings for loss-in-weight or gain-in-weight metering feeders, and implement in-batch closed loop correction and inter-batch setting update based on weighing / flow feedback.

8. A formulation optimization system for premixes based on the formulation optimization method for premixes according to any one of claims 1 to 7, characterized by It comprises: a data acquisition module for collecting current batch characteristic data of multiple raw materials, and associating, calibrating and storing the current batch characteristic data with historical batch characteristic data; an estimation and modeling module for calculating updated estimates of each raw material nutritional parameter based on the current batch characteristic data and historical batch characteristic data, establishing the correlation between raw materials and generating parameters and features for formula optimization; an optimization and compensation module for solving the formula proportion under the constraints of nutrition, specification and process / metering, and adjusting the formula proportion according to the updated estimate and process loss to calculate the nutritional bias compensation value; a control module for generating batching control instructions and issuing them to metering and feeding equipment, collecting weighing or flow feedback, and implementing in-batch correction and inter-batch setting update.

9. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program, when executed by a processor, implements the steps of the batching optimization method for premixes according to any one of claims 1-7. The computer program, when executed by a processor, implements the steps of the batching optimization method for premixes according to any one of claims 1-7.

Images