A quality management control method for low gi instant rice production

Abstract

The application discloses a quality management control method for low-GI instant rice production and belongs to the technical field of production management, and specifically comprises the following steps: constructing a quality-energy consumption correlation track library based on historical production data; analyzing process nodes and resource flow transfer chains to establish a production resource flow transfer network and a preset resource topology thereof; comparing differences between actual production and the preset topology to screen different batches, extracting performance parameters of key process nodes to form a unit performance parameter set; deconstructing a network model to identify a change path of resource flow transfer relationships between processes in the different batches and combining the unit performance parameter set to generate a level joint optimization parameter set; establishing a quantitative mapping relationship between quality and energy consumption and constructing a cross-unit parameter collaborative decision-making model; and generating an optimized process instruction set according to real-time material characteristics to drive each processing unit to perform collaborative production. The application realizes adaptive regulation and control of raw material fluctuations, significantly reduces production energy consumption while stabilizing product quality.

Description

Technical Field

[0001] This invention relates to the field of production management technology, and specifically to a quality management and control method for the production of low-GI instant rice. Background Technology

[0002] In the current production process of low-GI instant rice, firstly, the production process largely relies on pre-set fixed parameter templates. However, the key initial characteristics of raw materials (such as rice) (such as amylose content, moisture content, and protein content) naturally fluctuate from batch to batch. Fixed process parameters cannot dynamically adapt to this upstream "input disturbance," resulting in significant fluctuations in key quality indicators such as GI value, texture, and rehydration properties of the final product between different batches, making it difficult to consistently guarantee the product qualification rate. Secondly, to ensure that even the worst batch of raw materials meets the lower quality limit, process parameters are often set based on conservative experience, which easily leads to "over-processing" of most regular batches, resulting in high energy consumption (such as steam and electricity) and increased production costs.

[0003] Currently, industry improvements to these issues largely focus on optimizing local parameters of single processes (such as cooking and drying) or introducing offline monitoring feedback. There is a lack of methods to achieve closed-loop correlation and dynamic optimization of raw material characteristics, multi-process collaborative technology, final product quality, and overall production energy consumption from a system-wide perspective. Existing methods struggle to quantify and utilize the process patterns implicit in historical "high-quality, low-consumption" batches, and cannot predict and proactively configure the globally optimal process chain based on real-time raw material characteristics. Therefore, developing a quality management and control method capable of sensing raw material changes, self-learning historical optimal patterns, and dynamically coordinating process parameters across the entire process to achieve the dual goals of quality stability and energy consumption optimization has become an urgent need to improve the technological level and economic benefits of low-GI convenience rice production. Summary of the Invention

[0004] The purpose of this invention is to provide a quality management and control method for the production of low-GI instant rice, and to solve the following technical problems:

[0005] Existing methods are unable to quantify and utilize the process patterns implied in "high-quality and low-consumption" batches in historical production, nor can they predict and proactively configure the globally optimal process chain based on real-time raw material characteristics.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A quality management and control method for the production of low-GI instant rice includes the following steps:

[0008] S1. Based on historical production data, extract the initial characteristic parameters, full-process process parameters, quality output parameters and production energy consumption data of each batch of materials, and establish a production quality-energy consumption correlation trajectory library.

[0009] S2, perform process node analysis and resource flow chain reconstruction on the production quality-energy consumption correlation trajectory library, construct a production resource flow network containing multi-level processing units based on the resource flow chain, and determine the preset resource topology of the production resource flow network;

[0010] S3, based on the preset resource topology, filter out the difference batches that deviate from the preset resource topology from historical production data, extract the quality output parameters and production energy consumption parameters of key process nodes in the difference batches and integrate them to obtain a set of unit performance parameters;

[0011] S4, deconstruct the production resource flow network model, identify changes in the inter-process resource flow relationship corresponding to different batches, and establish a joint optimization parameter set in combination with the unit performance parameter set;

[0012] S5. Based on the joint optimization parameter set, determine the quantitative mapping relationship between quality output parameters and production energy consumption data, and construct a cross-unit parameter collaborative decision-making model based on the quantitative mapping relationship.

[0013] S6. Obtain the set of joint optimization parameters constructed based on real-time material initial characteristic data and input it into the cross-unit parameter collaborative decision-making model to generate the process parameter instruction set of each processing unit and drive each processing unit to perform collaborative production.

[0014] As a further aspect of the present invention: the specific process of establishing the production quality-energy consumption correlation trajectory library in S1 is as follows:

[0015] Extract the initial material characteristic parameters of the corresponding batch from the raw material inspection records, extract the full process parameters of the corresponding batch through each process from the distributed control system log, extract the quality output parameters of the corresponding batch from the laboratory information management system and finished product inspection report, and extract the total production energy consumption data of the corresponding batch from the energy management system.

[0016] The initial characteristic parameters, full-process process parameters, quality output parameters, and total production energy consumption data of the materials under the same batch identifier are associated and merged to form a batch production trajectory with a unified timestamp sequence; the batch production trajectories of all batches are aggregated and stored to obtain the production quality-energy consumption associated trajectory library.

[0017] As a further aspect of the present invention: the specific construction process of the production resource circulation network in S2 is as follows:

[0018] Based on the production quality-energy consumption correlation trajectory library, each independent process recorded in chronological order is extracted as a network node; the upstream and downstream connection relationship between nodes is determined according to the fixed process execution sequence in the production order and batch record.

[0019] Analyze the data fields related to each pair of upstream and downstream processes in the production quality-energy consumption correlation trajectory library: if the data field represents the quantity or composition change of physical raw materials or semi-finished products, then define the connection between these nodes as a material flow edge; if the data field represents the consumption and measurement of electricity, steam or fuel, then define the connection between these nodes as an energy flow edge; if the data field represents process state indicators such as temperature, viscosity or pH value, then define the connection between these nodes as a state flow edge.

[0020] Traverse all process sequences and associated data fields in the database, create a directed edge with a clear type identifier for each pair of upstream and downstream nodes with actual flow relationship, and then generate a production resource flow network composed of process nodes and typed directed edges.

[0021] As a further aspect of the present invention: the specific process of determining the preset resource topology of the production resource circulation network in step S2 is as follows:

[0022] The batch with the smallest total production energy consumption data value is selected from the production quality-energy consumption correlation trajectory library as the benchmark batch. The complete production data of the benchmark batch is parsed. For each directed edge in the production resource circulation network, the circulation volume data of the resource type corresponding to the edge in the batch is extracted, and the proportion of the circulation volume data to the total consumption data of the same type of resource in the batch is calculated. The calculated proportion value is assigned as the preset weight parameter of the directed edge. After the preset weight parameter is assigned to all directed edges in the production resource circulation network, the weighted directed network structure generated is determined as the preset resource topology.

[0023] As a further aspect of the present invention: in S3, the specific screening process for the difference batches is as follows:

[0024] Traverse each historical production batch in the production quality-energy consumption correlation trajectory library, and for each batch, calculate the actual circulation ratio of that batch on each resource circulation edge in the production resource circulation network based on its production data;

[0025] The actual flow ratio of the batch on each side is compared with the flow ratio parameter of the corresponding side in the preset resource topology, and the overall deviation is calculated; the batch with the overall deviation greater than the preset deviation threshold is marked as the difference batch.

[0026] As a further aspect of the present invention: in S3, the specific process for generating the unit performance parameter set is as follows:

[0027] For each selected batch with discrepancies, the actual circulation ratio of the batch on each resource circulation edge in the production resource circulation network is compared with the circulation ratio parameter of the corresponding edge in the preset resource topology, and the ratio deviation value is calculated.

[0028] The process nodes connected to edges whose proportional deviation values ​​exceed the preset node screening threshold are marked as key process nodes of the difference batch. The quality output parameters and production energy consumption parameters corresponding to all marked key process nodes in the difference batch are extracted from the production quality-energy consumption correlation trajectory library.

[0029] Using the unique identifier of the key process node as an index, the extracted node parameters are collected and stored in a structured manner to generate the unit performance parameter set corresponding to the batch with the difference.

[0030] As a further aspect of the present invention: the specific process of establishing the joint optimization parameter set in S4 is as follows:

[0031] For each differential batch, all direct upstream and direct downstream nodes corresponding to its key process nodes are determined based on the connection relationship of the production resource flow network; for the directed edge formed by the key process node and each of the direct upstream or direct downstream nodes, the actual flow data represented by the differential batch on the directed edge is extracted.

[0032] The actual flow data is compared with the flow ratio parameter on the same directed edge in the preset resource topology to obtain the relationship deviation value. When the relationship deviation value is greater than zero, it is recorded as a positive deviation, and when the relationship deviation value is less than zero, it is recorded as a negative deviation. If the absolute value of the relationship deviation value is greater than the preset judgment threshold, it is determined that the inter-process resource flow relationship corresponding to the directed edge has changed, and the node pair with the changed directed edge and its corresponding relationship deviation value and deviation direction are recorded.

[0033] For each changed node, the directed edge and its relationship deviation value and direction are associated with the unit performance parameter set corresponding to the source key process node of the directed edge; according to the topology of the production resource flow network, the quality output parameters and production energy consumption parameters contained in the unit performance parameter set are combined with all the change relationship deviation values ​​issued by the source key process node; the combined parameter and deviation value data are integrated to generate a joint optimization parameter set.

[0034] As a further aspect of the present invention: in step S5, the specific process for determining the quantization mapping relationship is as follows:

[0035] A training dataset is constructed by collecting the joint optimization parameter sets corresponding to all differential batches. The quality output parameters and relationship deviation values ​​recorded in the training dataset are used as input features, and the corresponding production energy consumption data are used as target variables. The model is trained using regression analysis. The trained model is determined as the quantitative mapping relationship between the quality output parameters and the production energy consumption data.

[0036] As a further aspect of the present invention: the specific construction process of the cross-unit parameter collaborative decision-making model in S5 is as follows:

[0037] Using the aforementioned joint optimization parameter set as model training samples, the quality output parameters and production energy consumption parameters contained in the unit performance parameter set corresponding to multiple different batches recorded in the samples are used as basic features, and the directed edges, relationship deviation values ​​and deviation directions of the node pairs corresponding to the changes in inter-process resource flow relationships identified in each batch are used as association features.

[0038] Using the gradient boosting tree algorithm, a regression model is obtained by supervising learning with the basic features and related features as input feature vectors and the production energy consumption parameters as target variables. The model hyperparameters are adjusted through cross-validation, and the error in the model's prediction of production energy consumption is verified to be within a preset tolerance range using independent historical batch data. The validated regression model is then serialized and stored to obtain a cross-unit parameter collaborative decision-making model.

[0039] The beneficial effects of this invention are:

[0040] 1) It is understandable that any batch of production follows a complex mapping relationship determined by equipment physics and reaction mechanisms, encompassing raw material characteristics, process execution, and resource consumption. However, this relationship can manifest as different data paths due to inherent fluctuations in raw materials. This invention first defines the quantitative boundaries of input, output, and resource consumption for each process. Then, it extracts the complete parameter sequence of each batch flowing through all processes from historical production data and calculates the resource allocation ratio that should be followed among the processes when the system operates at optimal energy efficiency. This ratio structure is the "energy efficiency fingerprint" of the production system, which can be understood as the inherent mass-energy conversion law under conditions without abnormal raw material interference. During real-time control, the resource demand calculated in real-time based on the current raw material characteristics is compared with this long-term benchmark. Since the benchmark has defined a cooperative rhythm, the optimization instruction set for the current raw material will be achieved by re-coordinating the resource allocation of upstream and downstream processes, rather than adjusting isolated single-point parameters. This approach of "first establishing a cooperative benchmark, then generating dynamic compensation instructions" fundamentally improves the synergy and reliability of achieving stable quality and optimal energy efficiency under varying raw material conditions.

[0041] 2) This invention abstracts the physically continuous production process into a mathematical resource flow network. Specifically, each process unit and its performance parameters are mapped to a node and its attributes in the network, while the actual flow of materials, energy, and states constitutes the directed edges between nodes. This mapping constructs a topological and causal model of the production process. By analyzing the transmission of node attributes in the network structure, the cross-process constraint law of "how the process execution and resource consumption of the upstream process should affect and limit the operating window of the downstream process under optimal conditions" can be quantified. It can be understood that when the characteristics of raw materials deviate from the standard, it will disrupt the original balance, causing the actual operating state of a certain node to be unable to be smoothly connected by the states of its upstream and downstream nodes through the learned benchmark law, thus producing a significant transmission deviation. This invention internalizes adjustment rules to cope with disturbances by identifying and learning the changes in the transmission paths in these deviation samples. It can be understood that the learning objective of this model is directly derived from the inherent physical and chemical connections in the production process, rather than the imitation of a fixed formula. Therefore, its core logic has the ability to extrapolate unknown raw material conditions, can adapt to production systems of different varieties and production lines, and exhibits excellent adaptability and generalization ability. Furthermore, its decision-making process of "prediction-coordinated control based on network transmission" has clear causal explanatory power.

[0042] 3) This invention achieves precise diagnosis of raw material fluctuations and customized optimization of energy consumption guidance by deconstructing the transmission path and amplitude of deviation samples in the production resource flow network. The transmission path not only reveals the root cause of the fluctuations, but also, more importantly, locates the starting point and diffusion direction of energy consumption anomalies. For example, abnormally high moisture content will first manifest as a significant increase in hot air energy consumption at the drying node, with a short path and concentrated energy consumption impact; while abnormally high amylose content starts from the gelatinization node, and its impact is transmitted to the aging and drying nodes, meaning that steam energy consumption and electrical energy consumption have undergone a complex redistribution among multiple processes. The transmission amplitude further quantifies the scale of this energy consumption redistribution and the inherent energy efficiency trade-off relationship of the system. For example, for high amylose samples, the system quantifies a unique set of "energy efficiency trade-off fingerprints" such as "gelatinization steam energy consumption needs to increase by 5% (amplitude A), but this can reduce the heat preservation energy consumption in the aging stage (time shortened by 10%, amplitude B), and ultimately significantly reduce the drying hot air energy consumption by 15% (amplitude C)". By accumulating such "path-amplitude" combinations to construct an anomaly-energy efficiency pattern library, when a new batch of raw materials is identified as approaching a certain pattern, the matching energy efficiency optimization logic can be directly invoked. The generated collaborative instructions (such as moderately increasing pressure and delaying the gelatinization unit to achieve low-temperature energy-saving operation of the drying unit) have the core objective and ultimate effect of proactively seeking and minimizing the total production energy consumption of the system while ensuring quality. Thus, this analysis directly transforms diagnostic capabilities into dynamic energy efficiency optimization prescription generation capabilities, providing a precise decision-making core for continuously reducing production energy consumption. It realizes the evolution from executing static "prescriptions" to autonomously generating dynamic "prescriptions," providing core decision support for stable production and cost reduction and efficiency improvement. Attached Figure Description

[0043] The invention will now be further described with reference to the accompanying drawings.

[0044] Figure 1 This is a schematic diagram of a quality management and control method for the production of low-GI instant rice according to the present invention. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Please see Figure 1 As shown, this invention is a quality management and control method for the production of low-GI instant rice, comprising the following steps:

[0047] S1. Based on historical production data, extract the initial characteristic parameters, full-process process parameters, quality output parameters and production energy consumption data of each batch of materials, and establish a production quality-energy consumption correlation trajectory library.

[0048] The historical production data is obtained from multiple production information systems: initial material characteristic parameters (such as amylose content, moisture, and protein content) are derived from the raw material incoming inspection database or batch records of online near-infrared analyzers; full-process parameters (such as cooking temperature and time, aging temperature and humidity, and drying temperature curves) are extracted from the historical logs of the production line's distributed control system or programmable logic controller, with each parameter bearing a precise timestamp and batch identifier; quality output parameters (such as finished product GI value, rehydration time, and hardness) are obtained from inspection reports in the laboratory information management system; and production energy consumption data (such as cumulative steam flow and electricity consumption in each process segment) are aggregated from the factory's energy management system by batch and time period. All data are cleaned, aligned, and merged using a unified production order number or batch number as a unique association key, forming complete "batch trajectories." The collection of all batch trajectories constitutes the production quality-energy consumption correlation trajectory library, which is stored in the form of a time-series database or data tables, supporting multi-dimensional queries by batch, process, and time range.

[0049] S2, perform process node analysis and resource flow chain reconstruction on the production quality-energy consumption correlation trajectory library, construct a production resource flow network containing multi-level processing units based on the resource flow chain, and determine the preset resource topology of the production resource flow network;

[0050] The process node analysis first divides the continuous production line into multiple logically independent functional units based on the process flow diagram and equipment layout, such as "gelatinization reactor," "crystal grower," and "dehydration and forming unit," with each unit serving as a network node. The resource flow chain is also determined by analyzing the temporal sequence and logical dependencies of data in the trajectory database: if there is a significant statistical correlation and temporal lag between the output material indicators (such as gelatinization degree) of upstream unit A and the input state or energy consumption of downstream unit B, then a flow edge from A to B is identified. Flow edges are classified according to the nature of the transmitted entities: transmitting physical materials is defined as "material flow," transmitting steam, electricity, etc., is defined as "energy flow," and transmitting process state indicators such as temperature and viscosity is defined as "state flow." Thus, nodes and directed edges constitute a directed network, i.e., the production resource flow network. When determining its preset resource topology, the batch with the lowest total production energy consumption and meeting quality standards is selected from the trajectory database as the "golden batch." The proportion of resources flowing through this batch on each edge of the network relative to the total amount of similar resources in the system is calculated, and this proportion is used as the "standard flow weight" of that edge. After assigning weights to all edges in the network, the resulting weighted directed network is the preset resource topology.

[0051] S3, based on the preset resource topology, filter out the difference batches that deviate from the preset resource topology from historical production data, extract the quality output parameters and production energy consumption parameters of key process nodes in the difference batches and integrate them to obtain a set of unit performance parameters;

[0052] The batch discrimination screening process involves traversing all batches in the trajectory database, calculating the actual resource flow ratio on each edge during actual production for each batch, and comparing it with the standard flow weight of the corresponding edge in the preset resource topology. The sum of squares or average absolute deviation of the proportional deviations on all edges is calculated as the "overall deviation" of the batch. A deviation threshold is set, and batches with an overall deviation greater than the threshold are marked as "discrepant batches". For each discrepant batch, the edges with the largest deviations are identified, and the process nodes connected to these edges are determined as the "critical process nodes" of the batch. Subsequently, the specific quality output parameters (such as the gelatinization degree of the intermediate product produced by the node) and production energy consumption parameters (such as the amount of steam or electricity consumed by the node) recorded by these critical nodes during the production process are extracted from the trajectory database for the discrepant batch. These parameters are packaged by node to form the "unit performance parameter subset" of each critical node in the batch. The subsets of all critical nodes in the batch are collected to form the unit performance parameter set of the discrepant batch.

[0053] S4, deconstruct the production resource flow network model, identify changes in the inter-process resource flow relationship corresponding to different batches, and establish a joint optimization parameter set in combination with the unit performance parameter set;

[0054] For a differential batch and its key process node (denoted as node K), based on the connection structure of the production resource flow network, locate all direct upstream and downstream nodes of node K. For each directed edge between node K and each directly associated node, extract the actual flow data of the differential batch on that edge and calculate its "relationship deviation value" compared to the standard flow weight of that edge in the preset resource topology. If the deviation value of an edge exceeds a set threshold, it is determined that the resource flow relationship between the two processes represented by that edge has "changed". Record the changed edges (i.e., node pairs) and their deviation values ​​and directions (positive or negative). Then, associate and combine this "changed relationship" data with the unit performance parameter set of the differential batch and the performance parameter subset corresponding to node K (including the quality and energy consumption data of node K itself). Repeat this process for all different batches in the network, and summarize and store the resulting combined data of "(key node performance data) + (changes in surrounding relationships caused by it)" to form a system-level joint optimization parameter set. It can be understood that this parameter set reveals "how a node changes its resource interaction relationship with upstream and downstream processes when a certain node experiences specific performance changes due to raw material fluctuations".

[0055] S5. Based on the joint optimization parameter set, determine the quantitative mapping relationship between quality output parameters and production energy consumption data, and construct a cross-unit parameter collaborative decision-making model based on the quantitative mapping relationship.

[0056] The system-level joint optimization parameter set is used as the training dataset. Each sample in this dataset contains two features: 1) state features: the quality output parameters of key process nodes; 2) relationship features: the deviation values ​​of each relationship change edge triggered by that node. The training objective variable is the total production energy consumption data, and machine learning algorithms such as gradient boosting decision trees are used. Supervised learning is performed using the state features and relationship features as input and the energy consumption data as output to train a regression model. In essence, the model learns the quantitative mapping relationship between the quality output parameters and the production energy consumption data, enabling it to predict the energy consumption performance of each process under given node states and specific network relationship perturbations.

[0057] Based on this mapping relationship, with the optimization objective of "minimizing predicted energy consumption while satisfying the target quality constraint", the decision variables are the adjustable process parameters of each process node, and a decision model is constructed. It is understandable that production itself is a deterministic system governed by physical and chemical laws. The same raw material characteristics, under the same process logic, will approximately lead to similar system states and resource consumption results. These laws are hidden in past production records. After clarifying the above quantitative mapping relationship, constructing a cross-unit parameter collaborative decision-making model is essentially transforming the discovered laws into a computable decision rule base. For example, transforming "If we want to control the GI value of the final product within a certain target range, and the current amylose content of the raw material is known, then what numerical combination relationship must be satisfied between the steam input of the 'gelatinization reactor,' the aging time of the 'crystal grower,' and the drying temperature of the 'dehydration and forming device' to minimize the total energy consumption" into a search and optimization calculation within the model, based on the known mapping relationship, will simulate thousands of different parameter combination schemes, and quickly evaluate the predicted product quality and total energy consumption under each scheme, and finally select the optimal solution with the lowest energy consumption among all schemes that ensure quality compliance.

[0058] S6. Obtain the set of joint optimization parameters constructed based on real-time material initial characteristic data and input it into the cross-unit parameter collaborative decision-making model to generate the process parameter instruction set of each processing unit and drive each processing unit to perform collaborative production.

[0059] When a new batch of materials is put into production, the online monitoring system acquires its initial characteristic parameters (such as moisture and amylose content) in real time. First, based on historical similarity patterns, it quickly matches or performs lightweight simulations to deduce the nodes that this batch may become in critical processes, and predicts its unit performance parameters (state characteristics) and potential relationship changes (relationship characteristics), thereby dynamically generating a real-time "system-level joint optimization parameter set" instance for the current batch. This real-time parameter set is input into a pre-trained cross-unit parameter collaborative decision-making model. Based on the input state and relationship characteristics, the decision-making model calculates the optimal process parameter values ​​that minimize the predicted total energy consumption and cover all processing units, i.e., the "process parameter instruction set for each processing unit". This instruction set is sent to the controllers (such as PLCs and DCSs) of each unit on the production line through industrial communication protocols (such as OPCUA), driving the cooking, aging, and drying equipment to automatically adjust parameters according to the instructions and execute collaborative production, thereby achieving dynamic and optimal control of the current specific raw material batch.

[0060] In a preferred embodiment of the present invention, the specific process of establishing the production quality-energy consumption correlation trajectory library in step S1 is as follows:

[0061] Extract the initial material characteristic parameters of the corresponding batch from the raw material inspection records, extract the full process parameters of the corresponding batch through each process from the distributed control system log, extract the quality output parameters of the corresponding batch from the laboratory information management system and finished product inspection report, and extract the total production energy consumption data of the corresponding batch from the energy management system.

[0062] The initial characteristic parameters, full-process process parameters, quality output parameters, and total production energy consumption data of the materials under the same batch identifier are associated and merged to form a batch production trajectory with a unified timestamp sequence; the batch production trajectories of all batches are aggregated and stored to obtain the production quality-energy consumption associated trajectory library.

[0063] In another preferred embodiment of the present invention, the specific construction process of the production resource circulation network in step S2 is as follows:

[0064] Based on the production quality-energy consumption correlation trajectory library, each independent process recorded in chronological order is extracted as a network node; the upstream and downstream connection relationship between nodes is determined according to the fixed process execution sequence in the production order and batch record.

[0065] Analyze the data fields related to each pair of upstream and downstream processes in the production quality-energy consumption correlation trajectory library: if the data field represents the quantity or composition change of physical raw materials or semi-finished products, then define the connection between these nodes as a material flow edge; if the data field represents the consumption and measurement of electricity, steam or fuel, then define the connection between these nodes as an energy flow edge; if the data field represents process state indicators such as temperature, viscosity or pH value, then define the connection between these nodes as a state flow edge.

[0066] Traverse all process sequences and associated data fields in the database, create a directed edge with a clear type identifier for each pair of upstream and downstream nodes with actual flow relationship, and then generate a production resource flow network composed of process nodes and typed directed edges.

[0067] By analyzing the timestamp sequence recorded in the trajectory database, logically independent yet continuous functional stages on the production line are identified. For example, in the production of low-GI instant rice, three core stages are identified: "cooking," "aging," and "drying," each completing a specific physical or chemical transformation. Each such stage is defined as a network node. The reason nodes can be extracted through time-series records is that the trajectory database faithfully records when the material enters which equipment or section and begins recording the process parameters for that stage. This switching point between time and space naturally delineates the boundaries of the processes. The connections between nodes are then determined. Because production orders and batch records mandate fixed and irreversible process flows, based on this fixed execution sequence followed by all batches, the "cooking" node is automatically determined to be upstream of the "aging" node, and the "aging" node is upstream of the "drying" node, thus establishing a connection from "cooking" to "aging," and then from "aging" to "drying." This is feasible because industrial production processes are pre-designed, stable pipelines with predetermined material flow directions. Then, we move into a more refined analysis phase, defining the essence of the connection—resource flow edges. Each established upstream and downstream connection is given a specific meaning. To do this, all data related to the "cooking" and "aging" processes in the trajectory database is scanned. It can be found that from "cooking" to "aging," more than just the material itself is transferred. Three flows need to be distinguished: First, data records the weight or flow rate of semi-finished products transported from the cooking tank to the aging chamber, clearly representing the transfer of physical raw materials; therefore, a material flow edge is established between these two nodes. Second, the trajectory database shows that the "aging" process is associated with power consumption data used to drive the air cooler and humidity controller, indicating that energy is input to the "aging" node to maintain the environment; therefore, an energy flow edge is established from the abstract "energy supply end" to the "aging" node. Furthermore, it was noted that the "cooking" process in the trajectory database outputs a "gelatinization degree" index, and the efficiency and results of the "aging" process are clearly closely related to the gelatinization degree of the input material. This "gelatinization degree" is a key process state index, generated from the "cooking" node and influencing the "aging" node as a key condition. Therefore, a state transition edge is established between these two nodes. This classification is accomplished by parsing the physical meaning of the data fields. This parsing is possible because these data were given clear engineering meaning and units during collection. All identified process node pairs in the entire trajectory database are traversed. For each pair of nodes with actual physical dependencies or resource transfer relationships, one or more directed edges labeled with "material," "energy," or "state" types are created. Once all nodes and edges are defined and connected, an abstract but precise production resource flow network reflecting how resources flow and transform between different functional units in actual production is constructed.

[0068] In another preferred embodiment of the present invention, the specific process of determining the preset resource topology of the production resource circulation network in step S2 is as follows:

[0069] The batch with the smallest total production energy consumption data value is selected from the production quality-energy consumption correlation trajectory library as the benchmark batch. The complete production data of the benchmark batch is parsed. For each directed edge in the production resource circulation network, the circulation volume data of the resource type corresponding to the edge in the batch is extracted, and the proportion of the circulation volume data to the total consumption data of the same type of resource in the batch is calculated. The calculated proportion value is assigned as the preset weight parameter of the directed edge. After the preset weight parameter is assigned to all directed edges in the production resource circulation network, the weighted directed network structure generated is determined as the preset resource topology.

[0070] First, all completed historical production batches in the production quality-energy consumption correlation trajectory library are scanned to identify the batch with the lowest recorded total production energy consumption. For example, among hundreds of batch records for producing low-GI instant rice, a comparison reveals that "batch number P-1023," assuming it produces qualified products, has the lowest total steam and electricity consumption among all batches. This batch is therefore selected as the benchmark batch. The reason for choosing the batch with the lowest total energy consumption is that, while ensuring quality, this batch achieves the highest energy efficiency, and its resource allocation pattern in the production process is closest to the optimal performance of the current production system under ideal conditions. It can serve as the gold standard for measuring the energy efficiency of other batches. Parsing the complete production data of the benchmark batch means extracting all information corresponding to "batch P-1023" from the trajectory library, including the initial characteristics of its raw materials, the specific process parameters of each process unit (such as "cooking," "aging," and "drying"), the output status of each process, and detailed energy consumption details. Operations are then performed on each directed edge in the production resource flow network. Taking an energy flow edge connecting the "cooking" node and the "aging" node in the network as an example, this edge represents the flow of steam energy from the supply end to the "cooking" unit. It is necessary to extract the flow data of the resource type corresponding to this edge within the batch, that is, to find the specific value of steam consumed in the "cooking" process of batch "P-1023" from the detailed energy consumption data (e.g., 850 kg of steam). Next, the proportion of this flow data to the total consumption data of the same type of resource in the batch needs to be calculated. This requires first calculating the total steam consumed by "P-1023" throughout the entire production process (e.g., total steam consumption of 1000 kg), and then dividing the steam consumption of the "cooking" process (850 kg) by the total steam consumption (1000 kg) to obtain a proportion of 0.85 (i.e., 85%). The principle behind this calculation is that it quantifies the distribution ratio of a specific type of resource (such as steam) among various process nodes under optimal energy efficiency conditions.

[0071] The calculated proportion is assigned as the preset weight parameter to the directed edge. Thus, the energy flow edge representing the steam input in the "cooking" process is assigned a weight of 0.85. This process is repeated for every directed edge in the network (including other energy edges, material edges, and state edges). For example, the proportion of hot air energy flowing to the "drying" node relative to the total hot air consumption is calculated, or the relative intensity proportion of the "gelatinization" state transfer from "cooking" to "aging" is calculated. After assigning the preset weight parameter to all directed edges in the production resource flow network, the initially abstract network representing only connections is transformed into a weighted directed network structure where each edge has a specific numerical weight. This structured network is ultimately determined as the preset resource topology. Essentially, it is a "resource allocation map under optimal energy efficiency," precisely describing how various resources flow quantitatively along different paths in the network in historical best practices.

[0072] It transforms the vague concept of "efficient production" into a clearly comparable blueprint composed of concrete figures. This pre-defined resource topology defines the reasonable proportions of resources such as steam, electricity, materials, and process conditions that should occupy at each stage of the process to achieve optimal energy efficiency under existing equipment and process conditions. It provides an absolute and scientific benchmark for all subsequent steps. For example, when analyzing any new batch, its actual resource flow ratio can be compared with this blueprint. Any significant deviation immediately indicates potential energy efficiency losses or process mismatches, making problem detection and anomaly localization quick and accurate.

[0073] In another preferred embodiment of the present invention, the specific screening process for the difference batches in step S3 is as follows:

[0074] Traverse each historical production batch in the production quality-energy consumption correlation trajectory library, and for each batch, calculate the actual circulation ratio of that batch on each resource circulation edge in the production resource circulation network based on its production data;

[0075] The actual flow ratio of the batch on each side is compared with the flow ratio parameter of the corresponding side in the preset resource topology, and the overall deviation is calculated; the batch with the overall deviation greater than the preset deviation threshold is marked as the difference batch.

[0076] In another preferred embodiment of the present invention, the specific process for generating the unit performance parameter set in step S3 is as follows:

[0077] For each selected batch with discrepancies, the actual circulation ratio of the batch on each resource circulation edge in the production resource circulation network is compared with the circulation ratio parameter of the corresponding edge in the preset resource topology, and the ratio deviation value is calculated.

[0078] The process nodes connected to edges whose proportional deviation values ​​exceed the preset node screening threshold are marked as key process nodes of the difference batch. The quality output parameters and production energy consumption parameters corresponding to all marked key process nodes in the difference batch are extracted from the production quality-energy consumption correlation trajectory library.

[0079] Using the unique identifier of the key process node as an index, the extracted node parameters are collected and stored in a structured manner to generate the unit performance parameter set corresponding to the batch with the difference.

[0080] In another preferred embodiment of the present invention, the specific process of establishing the joint optimization parameter set in step S4 is as follows:

[0081] For each differential batch, all direct upstream and direct downstream nodes corresponding to its key process nodes are determined based on the connection relationship of the production resource flow network; for the directed edge formed by the key process node and each of the direct upstream or direct downstream nodes, the actual flow data represented by the differential batch on the directed edge is extracted.

[0082] The actual flow data is compared with the flow ratio parameter on the same directed edge in the preset resource topology to obtain the relationship deviation value. When the relationship deviation value is greater than zero, it is recorded as a positive deviation, and when the relationship deviation value is less than zero, it is recorded as a negative deviation. If the absolute value of the relationship deviation value is greater than the preset judgment threshold, it is determined that the inter-process resource flow relationship corresponding to the directed edge has changed, and the node pair with the changed directed edge and its corresponding relationship deviation value and deviation direction are recorded.

[0083] For each changed node, the directed edge and its relationship deviation value and direction are associated with the unit performance parameter set corresponding to the source key process node of the directed edge; according to the topology of the production resource flow network, the quality output parameters and production energy consumption parameters contained in the unit performance parameter set are combined with all the change relationship deviation values ​​issued by the source key process node; the combined parameter and deviation value data are integrated to generate a joint optimization parameter set.

[0084] First, for each identified batch of discrepancies, such as a batch marked due to excessive moisture in the raw materials, the direct upstream and downstream nodes corresponding to the key process nodes of that batch are determined based on the pre-constructed production resource flow network. Taking the "drying" node as an example of a key node in this batch, according to the network diagram, its direct upstream node might be the "aging" node, and its direct downstream node might be the "packaging" node. These nodes can be determined based on network connections because the network is a true mapping of the entire production process topology, and the connecting edges between nodes clearly encode the actual flow of materials and energy. Next, for the directed edges formed by the key process nodes and each of the direct upstream or downstream nodes, the actual flow data representing the discrepancy batch on those directed edges is extracted. For example, the edge pointing from "aging" to "drying" might be defined as a "material flow edge," so the specific weight data of the semi-finished product transferred from the aging process to the drying process is extracted from the batch's production data; for the "energy flow edge" pointing to the "drying" node, the specific electrical or hot air energy consumed in the drying process of that batch is extracted. Extracting this data is feasible because the cross-process flow information was recorded and associated with batches during the construction of the trajectory library. Then, a core comparative analysis is performed: the actual flow data is compared with the flow ratio parameter on the same directed edge in the preset resource topology to obtain the relationship deviation value. For example, the preset topology specifies that "drying" energy consumption should account for 30% of the total system heat energy consumption, but the actual data for this batch calculates a ratio of 45%, then the relationship deviation value is +15%. This deviation value greater than zero is recorded as a positive deviation, meaning that the actual resource consumption of that edge exceeds the optimal baseline; if it is less than zero, it is a negative deviation, meaning insufficient consumption. The reason for obtaining the deviation value through difference calculation is that the preset topology represents an ideal proportional balance, and any deviation of the actual data from it quantitatively characterizes the degree of disruption to the system's equilibrium state. If the absolute value of the deviation value of a directed edge is greater than a preset judgment threshold (e.g., 5%), it is determined that the resource flow relationship between the two process nodes connected by that edge has undergone a substantial "change," and this "aging -> drying" edge, its deviation value +15%, and the direction of the positive deviation are recorded. This signifies the identification of a specific abnormal transmission link. Subsequently, deep information fusion is performed: each recorded "directed edge of the changed node and its relationship deviation value and direction" is associated with the unit performance parameter set corresponding to the source of that directed edge—that is, the critical process node. For example, the edge ("aging -> drying", +15%) is associated with the unit performance parameter set of the critical node "drying" itself (including the output moisture, absolute energy consumption, etc. of the drying process in that batch).Based on the overall topology of the production resource flow network, the quality output parameters (such as finished product moisture content) and production energy consumption parameters (such as drying power consumption) recorded in the unit performance parameter set are combined with all change relationship deviation values ​​originating from the key node "drying" (possibly including its change edges with other nodes). For example, the combined information might be: "Drying node: its own parameters (moisture content X, power consumption Y), while its upstream aging incoming material relationship deviation is +15%, and its downstream packaging receiving relationship deviation is -5%." Finally, by summarizing and integrating such combined information generated from all differential batches, the cascade joint optimization parameter set is generated. It is not a simple list, but a structured knowledge base, where each record indicates that when a node exhibits a specific self-performance state (quality and energy consumption), it will be accompanied by a specific pattern of resource flow relationship changes with surrounding nodes (described by deviation values ​​and directions). Every "abnormality" or "deviation" event in production is transformed from a problem signal into a learnable "knowledge unit" containing complete causal logic. Its direct benefit is that it no longer views the high or low of a certain process parameter in isolation, but rather captures the dynamic picture of how local anomalies trigger chain reactions in the process network. For example, it not only knows that "drying energy consumption is high", but also knows that "at the same time as the drying energy consumption is high, the state relationship of the aging incoming material upstream has undergone specific changes". This connects the isolated problem of the drying process with the deep-seated reasons of the upstream process. In essence, it is a "case library" and "experience library" extracted from historical failures or suboptimal operations.

[0085] In another preferred embodiment of the present invention, the specific process for determining the quantization mapping relationship in step S5 is as follows:

[0086] A training dataset is constructed by collecting the joint optimization parameter sets corresponding to all differential batches. The quality output parameters and relationship deviation values ​​recorded in the training dataset are used as input features, and the corresponding production energy consumption data are used as target variables. The model is trained using regression analysis. The trained model is determined as the quantitative mapping relationship between the quality output parameters and the production energy consumption data.

[0087] In another preferred embodiment of the present invention, the specific construction process of the cross-unit parameter collaborative decision-making model in step S5 is as follows:

[0088] Using the aforementioned joint optimization parameter set as model training samples, the quality output parameters and production energy consumption parameters contained in the unit performance parameter set corresponding to multiple different batches recorded in the samples are used as basic features, and the directed edges, relationship deviation values ​​and deviation directions of the node pairs corresponding to the changes in inter-process resource flow relationships identified in each batch are used as association features.

[0089] Using the gradient boosting tree algorithm, a regression model is obtained by supervising learning with the basic features and related features as input feature vectors and the production energy consumption parameters as target variables. The model hyperparameters are adjusted through cross-validation, and the error in the model's prediction of production energy consumption is verified to be within a preset tolerance range using independent historical batch data. The validated regression model is then serialized and stored to obtain a cross-unit parameter collaborative decision-making model.

[0090] The construction of a cross-unit parameter collaborative decision-making model begins with the in-depth utilization of the established hierarchical joint optimization parameter set. This parameter set is essentially a systematically organized knowledge base, where each record fully describes the unique state pattern exhibited by a historically differentiated batch under specific raw material fluctuations. The construction process first transforms these records into training samples that a machine learning model can recognize. Specifically, each differentiated batch record is decomposed into two parts: the first part is the basic characteristics of the key process nodes within that batch. For example, in a batch marked for high amylose content, the "outlet gelatinization degree" and "steam consumption" of the "gelatinization" node are quality and energy consumption parameters directly extracted from its unit performance parameter set; the second part is the correlation features between this node and other parts of the network, such as information identified in the same batch like "a positive deviation of +5% occurred in the state flow edge from the gelatinization node to the aging node" and "a negative deviation of -8% occurred in the energy flow edge from the gelatinization node to the drying node." Combining these two types of features constitutes a complete vector describing the core characteristics of the batch. The reason for this splitting and combination is that the cascaded joint optimization parameter set is explicitly linked to the node's own performance and the network relationship changes it causes during construction, thus ensuring that each sample can simultaneously reflect both the "local state" and the "global impact." Subsequently, algorithms such as gradient boosting trees are used for model training. The principle of this algorithm is to learn complex patterns in the data by constructing a series of sequentially connected decision trees, with each new tree dedicated to correcting the residuals predicted by the previous tree. During training, the algorithm takes the aforementioned combined feature vectors as input and the corresponding production energy consumption parameters (such as the total energy consumption of the batch or the energy consumption of key nodes) as the target values ​​to be predicted. The algorithm automatically searches for and learns those patterns that have a decisive impact on energy consumption from massive features. For example, it may discover that when the "bleaching degree" feature is within a certain range and the "state deviation to aging" is positive, the total energy consumption tends to decrease. Hyperparameters are adjusted through cross-validation. The principle is to divide the training data into several parts and use one part in rotation to validate the model's performance on unseen data, thereby finding a configuration that makes the model neither overfit (rote memorization) nor oversimplify (unlearnable). The model is then validated using a batch of historical data that was not used in the training process to ensure that the error in its energy consumption prediction remains within an acceptable engineering range. Finally, this validated model, which has a series of complex decision-making rules internally fixed, is saved in a fixed format, resulting in a cross-unit parameter collaborative decision-making model that can be called up at any time.

[0091] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A quality management and control method for the production of low-GI instant rice, characterized in that, Includes the following steps: S1. Based on historical production data, extract the initial characteristic parameters, full-process process parameters, quality output parameters and production energy consumption data of each batch of materials, and establish a production quality-energy consumption correlation trajectory library. S2, perform process node analysis and resource flow chain reconstruction on the production quality-energy consumption correlation trajectory library, construct a production resource flow network containing multi-level processing units based on the resource flow chain, and determine the preset resource topology of the production resource flow network; S3, based on the preset resource topology, filter out the difference batches that deviate from the preset resource topology from historical production data, extract the quality output parameters and production energy consumption parameters of key process nodes in the difference batches and integrate them to obtain a set of unit performance parameters; S4, deconstruct the production resource flow network model, identify changes in the inter-process resource flow relationship corresponding to different batches, and establish a joint optimization parameter set in combination with the unit performance parameter set; The specific process of establishing the joint optimization parameter set is as follows: For each differential batch, all direct upstream and direct downstream nodes corresponding to its key process nodes are determined based on the connection relationship of the production resource flow network; for the directed edge formed by the key process node and each of the direct upstream or direct downstream nodes, the actual flow data represented by the differential batch on the directed edge is extracted. The actual flow data is compared with the flow ratio parameter on the same directed edge in the preset resource topology to obtain the relationship deviation value. When the relationship deviation value is greater than zero, it is recorded as a positive deviation, and when the relationship deviation value is less than zero, it is recorded as a negative deviation. If the absolute value of the relationship deviation value is greater than the preset judgment threshold, it is determined that the inter-process resource flow relationship corresponding to the directed edge has changed, and the node pair with the changed directed edge and its corresponding relationship deviation value and deviation direction are recorded. For each changed node, the directed edge and its relationship deviation value and direction are associated with the unit performance parameter set corresponding to the source key process node of that directed edge; according to the topology of the production resource flow network, the quality output parameters and production energy consumption parameters contained in the unit performance parameter set are combined with all change relationship deviation values ​​issued by the source key process node; the combined quality output parameters, production energy consumption parameters and change relationship deviation value data are integrated to generate a joint optimization parameter set; S5. Based on the joint optimization parameter set, determine the quantitative mapping relationship between quality output parameters and production energy consumption data, and construct a cross-unit parameter collaborative decision-making model based on the quantitative mapping relationship. S6. Obtain the set of joint optimization parameters constructed based on real-time material initial characteristic data and input it into the cross-unit parameter collaborative decision-making model to generate the process parameter instruction set of each processing unit and drive each processing unit to perform collaborative production.

2. The quality management and control method for low-GI instant rice production according to claim 1, characterized in that, In S1, the specific process of establishing the production quality-energy consumption correlation trajectory library is as follows: Extract the initial material characteristic parameters of the corresponding batch from the raw material inspection records, extract the full process parameters of the corresponding batch through each process from the distributed control system log, extract the quality output parameters of the corresponding batch from the laboratory information management system and finished product inspection report, and extract the total production energy consumption data of the corresponding batch from the energy management system. The initial characteristic parameters, full-process process parameters, quality output parameters, and total production energy consumption data of the materials under the same batch identifier are associated and merged to form a batch production trajectory with a unified timestamp sequence; the batch production trajectories of all batches are aggregated and stored to obtain the production quality-energy consumption associated trajectory library.

3. The quality management and control method for low-GI instant rice production according to claim 1, characterized in that, In S2, the specific construction process of the production resource circulation network is as follows: Based on the production quality-energy consumption correlation trajectory library, each independent process recorded in chronological order is extracted as a network node; the upstream and downstream connection relationship between nodes is determined according to the fixed process execution sequence in the production order and batch record. Analyze the data fields related to each pair of upstream and downstream processes in the production quality-energy consumption correlation trajectory library: if the data field represents the quantity or composition change of physical raw materials or semi-finished products, then define the connection between these nodes as a material flow edge; if the data field represents the consumption and measurement of electricity, steam or fuel, then define the connection between these nodes as an energy flow edge; if the data field represents process state indicators such as temperature, viscosity or pH value, then define the connection between these nodes as a state flow edge. Traverse all process sequences and associated data fields in the database, create a directed edge with a clear type identifier for each pair of upstream and downstream nodes with actual flow relationship, and then generate a production resource flow network composed of process nodes and typed directed edges.

4. The quality management and control method for low-GI instant rice production according to claim 1, characterized in that, In step S2, the specific process of determining the preset resource topology of the production resource circulation network is as follows: The batch with the smallest total production energy consumption data value is selected from the production quality-energy consumption correlation trajectory library as the benchmark batch. The complete production data of the benchmark batch is parsed. For each directed edge in the production resource circulation network, the circulation volume data of the resource type corresponding to the directed edge in the batch is extracted, and the proportion of the circulation volume data to the total consumption data of the same type of resource in the batch is calculated. The calculated proportion value is assigned as the preset weight parameter of the directed edge. After the preset weight parameter is assigned to all directed edges in the production resource circulation network, the weighted directed network structure generated is determined as the preset resource topology.

5. A quality management and control method for the production of low-GI instant rice according to claim 1, characterized in that, In S3, the specific screening process for the difference batches is as follows: Traverse each historical production batch in the production quality-energy consumption correlation trajectory library, and for each batch, calculate the actual circulation ratio of that batch on each resource circulation edge in the production resource circulation network based on its production data; The actual flow ratio of the batch on each side is compared with the flow ratio parameter of the corresponding side in the preset resource topology, and the overall deviation is calculated; the batch with the overall deviation greater than the preset deviation threshold is marked as the difference batch.

6. A quality management and control method for the production of low-GI instant rice according to claim 1, characterized in that, In S3, the specific process for generating the unit performance parameter set is as follows: For each selected batch with discrepancies, the actual circulation ratio of the batch on each resource circulation edge in the production resource circulation network is compared with the circulation ratio parameter of the corresponding edge in the preset resource topology, and the ratio deviation value is calculated. The process nodes connected to edges whose proportional deviation values ​​exceed the preset node screening threshold are marked as key process nodes of the difference batch. The quality output parameters and production energy consumption parameters corresponding to all marked key process nodes in the difference batch are extracted from the production quality-energy consumption correlation trajectory library. Using the unique identifier of the key process node as an index, the extracted node parameters are collected and stored in a structured manner to generate the unit performance parameter set corresponding to the batch with the difference.

7. A quality management and control method for the production of low-GI instant rice according to claim 1, characterized in that, In S5, the specific process for determining the quantization mapping relationship is as follows: A training dataset is constructed by collecting the joint optimization parameter sets corresponding to all differential batches. The quality output parameters and relationship deviation values ​​recorded in the training dataset are used as input features, and the corresponding production energy consumption data are used as target variables. The model is trained using regression analysis. The trained model is determined as the quantitative mapping relationship between the quality output parameters and the production energy consumption data.

8. A quality management and control method for the production of low-GI instant rice according to claim 1, characterized in that, In S5, the specific construction process of the cross-unit parameter collaborative decision-making model is as follows: Using the aforementioned joint optimization parameter set as model training samples, the quality output parameters and production energy consumption parameters contained in the unit performance parameter set corresponding to multiple different batches recorded in the samples are used as basic features, and the directed edges, relationship deviation values ​​and deviation directions of the node pairs corresponding to the changes in inter-process resource flow relationships identified in each batch are used as association features. Using the gradient boosting tree algorithm, a regression model is obtained by supervising learning with the basic features and related features as input feature vectors and the production energy consumption parameters as target variables. The model hyperparameters are adjusted through cross-validation, and the error in the model's prediction of production energy consumption is verified to be within a preset tolerance range using independent historical batch data. The validated regression model is then serialized and stored to obtain a cross-unit parameter collaborative decision-making model.

Images