A batch quality tracing and process abnormal intelligent early warning method for sea cucumber feed production

Abstract

The application belongs to the technical field of intelligent data processing in agricultural production management and aquatic feed industry, and discloses a batch quality tracing and process abnormal intelligent early warning method for sea cucumber feed production. The batch full life cycle data is bound to the unique digital identity encrypted by the national secret SM2, the multi-party alliance chain nodes such as raw material suppliers and regulatory departments are linked, and hierarchical permissions are set, so that the production full-process data is realized real-time on-chain storage, and a four-level linkage digital twin is built to copy the quality evolution process. Combined with the risk transmission map constructed by the cause-effect inference algorithm, the root cause can be quickly locked in reverse and the whole chain can be traced in one key when the quality is abnormal. By constructing the degradation kinetics equation of the heat-sensitive component, the quality failure mechanism is determined, the mechanism and data dual-dimensional abnormal feature library covering seven abnormal types are built, and the weak abnormal signal extraction method constrained by the degradation kinetics model is used to realize accurate extraction of weak abnormal signals.

Description

Technical Field

[0001] This invention belongs to the field of intelligent data processing technology in agricultural production management and aquatic feed industry, specifically a batch quality traceability and intelligent early warning method for process anomalies in sea cucumber feed production. Background Technology

[0002] As a core input for large-scale sea cucumber farming, the batch quality stability, nutritional accuracy, and safety of sea cucumber compound feed directly determine the growth performance, survival rate, and economic benefits of sea cucumber farming. Currently, the quality control and traceability systems of domestic sea cucumber feed production enterprises still have the following technical problems, making it difficult to adapt to the needs of full-chain production management and data governance under the digital upgrade of the aquaculture industry: The industry mostly uses manual paper ledgers and simple QR code entry for traceability, which only covers a few discrete nodes such as raw material entry and finished product exit. It cannot collect and link process parameters, equipment status, environmental and quality inspection data of the entire production process in real time, resulting in gaps in the traceability chain. Moreover, the traceability data is mostly manually entered after the fact, which is easy to tamper with and forge, and lacks credibility. When the finished product is unqualified, it can only be traced back linearly, and it is impossible to quickly locate the root cause of the quality through data correlation analysis, let alone establish a digital correlation between raw material properties, production process, finished product quality and breeding effect.

[0003] Sea cucumber feed is based on heat-sensitive components such as marine animal and plant-derived proteins, active polysaccharides, and unsaturated fatty acids. It requires high precision in processing and has strong coupling of multiple parameters. Existing technologies mostly use traditional single-point threshold alarms, which can only provide post-processing data alarms for parameters exceeding preset ranges. They do not combine the specific quality failure mechanism of sea cucumber feed to build a data-driven abnormal feature system, and cannot identify early weak abnormal signals when single parameters are normal but multiple parameters are coupled, resulting in prominent missed alarms and false alarms.

[0004] The existing management and control system only focuses on the compliance control of the production end, and has not opened up the digital data link between the biological feedback of the breeding end and the process optimization of the production end. It has not established a multi-party collaborative industrial-level production management and data sharing mechanism, which makes it difficult to support the digital needs of high-quality industrial upgrading and full-chain supervision of food safety. Summary of the Invention

[0005] The purpose of this invention is to provide a batch quality traceability and intelligent early warning method for process anomalies in sea cucumber feed production, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for batch quality traceability and intelligent early warning of process anomalies in sea cucumber feed production, comprising: Preferably, the parameter standardization stage targets the specific physiological characteristics and nutritional needs of sea cucumbers at different growth stages, and sorts out the entire production process from raw material acceptance, pretreatment, ingredient metering, mixing and conditioning, pelleting, drying and cooling, screening and packaging to finished product storage. Combined with the unique quality requirements of sea cucumber feed, it targets and anchors three major categories of core indicators: key quality attributes, key process parameters and key biological effect indicators. The key quality attributes include raw material freshness, nutrient content, retention rate of core heat-sensitive active components, finished product particle size, water stability, gelatinization degree, and limits for heavy metals and biotoxins. The key process parameters include crushing speed and grading accuracy, batching accuracy, mixing time and speed, conditioning temperature and steam volume, granulation pressure and ring die compression ratio, drying temperature gradient and wind speed, and cooling time and material temperature. The key biological effect indicators include feeding activity, protein digestibility, immune enzyme activity, body wall collagen content, growth rate, survival rate, disease incidence, and feed conversion ratio of sea cucumbers at the corresponding growth stages. For the anchored full-dimensional indicators, a unified standardized data dictionary is constructed to complete the format unification, dimension normalization and outlier cleaning of multi-source heterogeneous data. Based on fundamental research findings in sea cucumber nutrition and feed processing, as well as historical production data, a priori knowledge graph of raw material properties, process parameters, finished product quality, and aquaculture effects for sea cucumber feed was constructed, and the basic correlation rules between various indicators were clarified.

[0007] Preferably, the traceability system stage is based on standardized data and prior knowledge graphs constructed in the parameter standardization stage. Each production batch of sea cucumber feed is assigned a unique and tamper-proof distributed digital identity. The digital identity is generated using the national cryptographic SM2 asymmetric encryption algorithm and is bound to all data throughout the entire life cycle of the batch. A consortium blockchain node network is built, with the participation of raw material suppliers, feed production enterprises, aquaculture entities, germplasm resource providers, third-party testing institutions, food safety regulatory departments, and end consumers. Hierarchical permissions and smart contract rules are set for each node. The raw material genome traceability information, full-process process parameters, equipment status data, quality inspection results, warehousing and logistics information, operator information, and aquaculture biological effect data of the corresponding batch are uploaded to the blockchain in real time for evidence storage through the national cryptographic SM2 asymmetric encryption algorithm and smart contracts. The specific subcategories and collection standards for each type of data are as follows: 1. Raw material genomic traceability information: including raw material germplasm number, origin traceability code, and component detection genomic characteristic data, which is collected and uploaded by the raw material supplier during the raw material acceptance process and is a one-time data collection; 2. Full-process process parameters: including real-time operating parameters of each production process, collected by the production line sensor / PLC system at a frequency of seconds, and continuously uploaded throughout the entire production process; 3. Equipment status data: including equipment operating speed, pressure, temperature, fault codes, etc., collected by the equipment's IoT module and uploaded every minute, with immediate upload in case of fault; 4. Quality inspection results: including test data for raw material acceptance, process sampling, and finished product delivery, which shall be uploaded by the testing personnel within 10 minutes after the test is completed; 5. Warehousing and logistics information: including finished product storage temperature and humidity, storage location, outbound order number, transportation trajectory, and recipient information. Warehousing data is uploaded hourly, and logistics data is uploaded node by node (once for departure, transit, and arrival). 6. Operator Information: Includes operator ID, operation time, and operation content for each process, which is uploaded by the production management system immediately after the operation is completed; 7. Biological effect data at the aquaculture end: Upload according to the data collection interface specifications at the aquaculture end. All data must be labeled with the collection subject, collection time, and data source device number to ensure data traceability.

[0008] Based on the full-dimensional data on the blockchain, a four-level linked digital twin is constructed for each physical production batch, consisting of raw material attribute twins, production process twins, finished product quality twins, and aquaculture effect twins. This enables real-time dynamic correspondence between the physical production process and the digital twin, replicating the quality evolution process of the entire batch lifecycle. Based on prior knowledge graphs, the impact of process parameter fluctuations on finished product quality and aquaculture effect is simulated in real time.

[0009] Preferably, the root cause localization stage is based on the full-chain on-chain data and digital twin formed in the traceability system stage. For the full-chain data of the batches on the chain, the Do-Calculus causal inference algorithm with prior knowledge constraints is combined with Bayesian network and mediation effect analysis to eliminate irrelevant variables and confounding variables of the breeding environment. The causal relationship of the entire chain of raw material attributes, process parameters, finished product quality and breeding effect is explored. The influence weight and risk transmission threshold of each key parameter on finished product quality and breeding effect are quantified, the risk transmission path of each link is clarified, and a full-chain risk transmission causal map of sea cucumber feed batch quality is constructed. When finished products fail to meet quality standards or exhibit abnormal results in aquaculture, reverse tracing is achieved based on causal graphs and counterfactual reasoning. This allows for pinpointing the root cause, scope of impact, and key contributing factors of the quality issues. Simultaneously, the effects of different parameter correction schemes are simulated to provide optimal handling recommendations. Forward tracing is supported, enabling one-click access to information across the entire chain, from raw material germplasm, production and processing, warehousing and logistics to aquaculture applications and consumer terminals. Regulatory authorities can achieve full-chain supervision through consortium blockchain nodes, while consumers can query traceability information through batch digital identities, thus forming the ability to pinpoint the root cause.

[0010] Preferably, the abnormal characteristic stage, combined with the causal relationship of quality failure and the influence law of parameters clearly defined in the root cause localization stage, constructs the degradation kinetic equation of the core thermosensitive active component of sea cucumber feed under different process conditions through orthogonal experiments and accelerated degradation experiments, and clarifies the intrinsic quality failure mechanism of process parameter fluctuation, component degradation, quality deterioration and decline in aquaculture effect; Based on degradation kinetics models and quality failure mechanisms, and combined with historical batch anomaly cases and quality defect data, a dedicated anomaly feature library with two dimensions of mechanism-driven and data-driven approaches is constructed. It includes seven categories: single parameter over-limit anomaly, multi-parameter correlation anomaly, batch-to-batch fluctuation anomaly, equipment progressive degradation anomaly, component degradation risk anomaly, quality failure critical anomaly, and aquaculture effect correlation anomaly, covering the anomaly features corresponding to common quality defects in sea cucumber feed production. By employing an adaptive wavelet transform constrained by a degradation kinetic model and fusing it with a stacked autoencoder, the sensitive range of parameter fluctuations is first locked through the degradation kinetic model. Then, the original process data in the sensitive range is denoised and feature-enhanced to amplify the feature differences of weak anomalies, thereby enabling early anomaly signal extraction and forming anomaly feature extraction capability.

[0011] Preferably, the early warning model stage is based on a dedicated abnormal feature library and abnormal extraction capability built from the abnormal feature stage to construct an intelligent early warning model for process abnormalities and quality risks. It adopts a dual-drive architecture of mechanistic model pre-constraint and machine learning model identification. The pre-degradation kinetic model of thermosensitive components is used to perform initial screening of quality risks on the input real-time process parameters, and high-risk ranges and highly sensitive parameters are locked. A lightweight gradient booster is used as the base model, and a multi-scale attention mechanism is combined to focus on the correlation features with high impact weights and the fluctuation of sensitive range parameters. Finally, the abnormality level, quality deterioration risk level, probability of abnormality occurrence, degree of component degradation, potential impact on aquaculture effect, root cause location and recommended treatment measures are output. The early warning model establishes a dual-cycle dynamic update mechanism. Based on the newly added batch production data and disposal results, it achieves incremental dynamic updates. At the same time, based on the newly added experimental data and feedback data from the breeding end, it periodically calibrates the degradation kinetic model of heat-sensitive components to achieve coordinated updates of the mechanism model and the data model. After the early warning model is optimized for lightweight design, it is adapted for deployment at the edge of the production site, enabling real-time early warning and local handling, thus forming an early warning capability.

[0012] The hardware adaptation standard for edge deployment is as follows: it is compatible with industrial-grade edge computing gateways (main frequency ≥ 1.5GHz, memory ≥ 4G, storage ≥ 32G), supports industrial communication protocols such as RS485 / Modbus / Profinet, can directly connect to production line PLC systems, sensors, equipment IoT modules and other hardware without additional adapters, is compatible with working environments from -20℃ to 60℃ in the production site, and has industrial-grade characteristics such as anti-interference and vibration resistance. The data interaction rules at the edge are as follows: 1. With production line hardware: Second-level two-way data interaction, real-time acquisition of process parameters, and immediate issuance of handling instructions; 2. With the consortium blockchain cloud: Hourly incremental data synchronization, the early warning records, handling results and process data of the edge terminal are synchronized to the cloud for on-chain storage and evidence storage, and the data is synchronized to the regulatory node in real time when an anomaly is triggered; 3. After network outage recovery: The edge device automatically re-uploads the cached 72-hour data to the cloud in timestamp order. After the cloud completes data verification, the data is uploaded to the blockchain to ensure that there is no missing or duplicate data and that the data on the edge device and the cloud are consistent.

[0013] Preferably, the closed-loop control stage utilizes the early warning capabilities formed in the early warning model stage and the reliable data support of the traceability system to establish a deep collaborative link between the traceability system and the early warning model. This constructs a fully closed-loop control mechanism encompassing risk prediction, early warning triggering, digital twin backtracking, root cause localization, intelligent handling, effect verification, and on-chain archiving. This achieves full-process coverage of pre-event prevention, in-event control, and post-event traceability. When the early warning model identifies an abnormal signal or predicts a quality risk, it triggers the digital twin of the corresponding batch for real-time backtracking. Based on the full-link risk transmission causal graph, it identifies the process node where the abnormality occurred, the raw material batches involved, and the equipment and personnel information. Simultaneously, based on the dual dimensions of quality risk level and the scope of abnormal impact, it triggers the corresponding graded handling plan. The entire closed-loop system seamlessly connects each node according to the logic of triggering, execution, feedback, verification, and archiving. The execution time limits for each node are as follows: risk prediction is monitored in real time at the second level, the time from early warning triggering to twin backtracking completion is ≤10 seconds, root cause location is ≤30 seconds, and intelligent treatment instructions are issued within ≤1 minute. The effect verification adopts a dual-indicator judgment standard: first, process parameter verification: after treatment, the core process parameters return to the effective value range and operate stably for ≥5 minutes; second, quality risk verification: the component retention rate and quality indicators predicted by the degradation kinetic model have both recovered to the safe range. If both indicators meet the standards, the treatment is deemed effective. If they do not meet the standards, the root cause location and intelligent treatment process is retried until the verification is passed.

[0014] For minor component degradation risks and abnormal parameter fluctuations, the system automatically sends parameter fine-tuning instructions to the PLC system of the corresponding equipment to achieve adaptive correction of process parameters. For moderate anomalies, the system pushes early warning information, root cause analysis results, and optimal handling suggestions verified by digital twin simulation to the on-site operators, and simultaneously suspends downstream high-risk processes. Production is resumed only after the handling effect is verified. For severe anomalies, the system immediately triggers an emergency shutdown and simultaneously triggers full-chain traceability of the corresponding batch to lock the flow and scope of finished products that have been shipped. The system also synchronizes the anomaly information, traceability data, and handling plan to the regulatory nodes. All disposal processes, operation records, parameter adjustment results, and disposal effects are recorded on the blockchain in real time through smart contracts, and batch traceability files and digital twins are updated simultaneously to provide labeled data for model updates.

[0015] Preferably, the optimization and collaboration stage, based on the risk management and data storage achieved in the closed-loop control stage, establishes a standardized data collection interface for the aquaculture end based on the unique digital identity of each batch. The aquaculture entity uploads full-dimensional data on the aquaculture application of the corresponding batch of feed through the digital identity of the corresponding batch, including core performance indicators such as sea cucumber growth rate, survival rate, disease incidence rate, and feed conversion ratio, as well as biological phenotypic data such as sea cucumber immune enzyme activity and body wall quality. At the same time, basic data such as water temperature, salinity, and dissolved oxygen of the corresponding aquaculture environment are collected. By using a multi-scale causal inference algorithm to eliminate confounding variables of the aquaculture environment, the feedback data from the aquaculture end is correlated with the corresponding batch of production chain data for analysis. This allows us to uncover the intrinsic causal relationship between production process parameters, raw material properties and aquaculture effects, and to extract the optimal combination of process parameters and raw material formula optimization rules for different sea cucumber growth stages, different aquaculture environments and different aquaculture models. The optimization rules are then updated in reverse to the multi-dimensional parameter standardization system, prior knowledge graph, anomaly feature library and intelligent early warning model, and simultaneously updated to the digital twin of batch traceability.

[0016] The optimal rules are refined in a detailed way: In terms of process parameters, the growth stages are divided into four stages: juvenile sea cucumber, young sea cucumber, adult sea cucumber, and overwintering sea cucumber. Each stage is further subdivided into the aquaculture environment dimensions based on water temperature (below 10℃, 10-18℃, above 18℃) and salinity (28-32‰, above 32‰). The aquaculture mode dimensions are also divided into factory farming and bottom seeding farming, forming a three-dimensional combination of process parameters for growth stage, environment, and mode. In terms of raw material formulation, the above three-dimensional dimensions are matched simultaneously, and the ratio of marine animal and plant-derived proteins, active polysaccharides, and unsaturated fatty acids is adjusted to meet the nutritional needs of sea cucumbers in different scenarios.

[0017] The steps for implementing the rules are as follows: 1. Encode the optimization rules by dimension and input them into the production management system and prior knowledge graph; 2. Production enterprises select the corresponding dimensions based on the actual needs of the breeding end, and the system automatically retrieves the optimal process parameters and formula rules; 3. The production line completes parameter adjustments and ingredient preparation according to the rules, and tests the quality indicators of the finished products after small-batch trial production; 4. After successful trial production, formal mass production will commence, with continuous tracking of production data and aquaculture feedback data to provide a basis for rule iteration.

[0018] The beneficial effects of this invention are as follows: 1. This invention binds batch full lifecycle data to a unique digital identity encrypted with the national cryptographic standard SM2, links multiple alliance chain nodes such as raw material suppliers and regulatory authorities, and sets hierarchical permissions to achieve real-time on-chain storage of production process data. At the same time, it builds a four-level linkage digital twin to replicate the quality evolution process. Combined with the risk transmission map constructed by the causal inference algorithm, it can quickly lock the root cause in reverse when quality anomalies occur and trace the entire chain with one click in forward direction. This enables regulatory authorities to achieve full-chain supervision and consumers to query traceability information, improving the efficiency of quality problem handling and the credibility of traceability.

[0019] 2. This invention clarifies the quality failure mechanism by constructing a degradation kinetic equation for thermosensitive components, establishes a dual-dimensional anomaly feature library covering seven major anomaly types based on both mechanism and data, and achieves accurate extraction of weak anomaly signals by combining an early weak anomaly signal extraction method constrained by the degradation kinetic model. It employs a dual-drive early warning model combining mechanistic model and machine learning to first complete initial risk screening and then accurately identify anomalies. After lightweight optimization, it is adapted for edge deployment and can provide real-time early warnings, as well as output anomaly level, degradation degree, and disposal suggestions, realizing a shift from post-event alarm to pre-event prevention and effectively avoiding quality defects such as the loss of thermosensitive components.

[0020] 3. This invention establishes a standardized data collection interface for the aquaculture end to collect aquaculture effect and environmental data. Combined with a multi-scale causal inference algorithm to eliminate confounding variables, it explores the intrinsic causal relationship between production parameters and aquaculture effect, and extracts the optimal process and formula rules adapted to different growth stages and aquaculture scenarios. The optimized rules are then updated in reverse to the parameter standardization system, early warning model and other processes throughout the entire process. At the same time, a dual-loop dynamic update mechanism for the model is established to achieve two-way empowerment of production and aquaculture, build a multi-party collaborative industry-level management and control system, and support the high-quality upgrading of the sea cucumber aquaculture industry and the whole-chain supervision of food safety. Attached Figure Description

[0021] Fig. 1 This is a flowchart illustrating the whole-process quality control and early warning system for sea cucumber feed according to the present invention. Fig. 2 This is a flowchart illustrating the full-chain traceability process for batch quality of sea cucumber feed according to the present invention. Fig. 3 This is a flowchart illustrating the intelligent early warning process for abnormalities in sea cucumber feed production according to the present invention. Detailed Implementation

[0022] 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.

[0023] like Figs. 1 to 3 As shown, this embodiment of the invention provides a batch quality traceability and intelligent early warning method for process anomalies in sea cucumber feed production, including: The parameter standardization stage addresses the specific physiological characteristics and nutritional needs of sea cucumbers at different growth stages. It outlines the entire production process from raw material acceptance, pretreatment, ingredient metering, mixing and conditioning, pelleting, drying and cooling, screening and packaging to finished product storage. Combined with the unique quality requirements of sea cucumber feed, it targets and anchors three major categories of core indicators: key quality attributes, key process parameters, and key biological effect indicators. The sea cucumber growth stages include juvenile sea cucumbers, young sea cucumbers, adult sea cucumbers, and overwintering sea cucumbers. The key quality attributes include raw material freshness, nutrient content, retention rate of core heat-sensitive active components, finished product particle size, water stability, gelatinization degree, and limits for heavy metals and biotoxins. The key process parameters include crushing speed and grading accuracy, batching accuracy, mixing time and speed, conditioning temperature and steam volume, granulation pressure and ring die compression ratio, drying temperature gradient and wind speed, cooling time and material temperature. The key biological effect indicators include feeding activity, protein digestibility, immune enzyme activity, body wall collagen content, growth rate, survival rate, disease incidence, and feed conversion ratio of sea cucumbers at the corresponding growth stages. For the anchored full-dimensional indicators, a unified standardized data dictionary is constructed to complete the format unification, dimension normalization and outlier cleaning of multi-source heterogeneous data such as equipment PLC data, sensor data, laboratory test data, manually entered data, and feedback data from the breeding end. The standardized data dictionary is constructed according to seven dimensions: indicator category, indicator name, data type, unit of measurement, collection frequency, effective value range, and data source. It covers all sub-items of the three major categories of core indicators. Among them, numerical indicators are uniformly decimal with two decimal places, time indicators are uniformly in the format of "YYYY-MM-DDHH:MM:SS", and categorized indicators are uniformly in the form of numerical codes.

[0024] The data cleaning rules are as follows: 1. Missing value cleaning: When the number of missing items in a single data entry is ≤10%, it is filled with the average of the data in the same batch; when the number of missing items is >10%, it is directly removed. 2. Outlier cleaning: Outliers are identified using the 3σ principle. Outliers that are manually verified as having acquisition errors are directly removed, while those that are actually due to process fluctuations are retained and the cause of the anomaly is noted. 3. Dimensional normalization: The range standardization method is used to map all numerical indicators to the [0, 1] interval to eliminate the impact of dimensional differences on data analysis.

[0025] Based on fundamental research findings in sea cucumber nutrition and feed processing, as well as historical production data, a priori knowledge graph of raw material properties, process parameters, finished product quality, and aquaculture effects for sea cucumber feed was constructed, and the basic correlation rules between various indicators were clarified.

[0026] The construction process of prior knowledge graph is as follows: 1. Using the basic research findings in sea cucumber nutrition and feed processing as theoretical nodes and the correlation relationships of indicators in historical production data as data nodes, an initial map was constructed; 2. Standardize the coding of nodes and construct directed edges according to the logic of raw material attributes → process parameters → finished product quality → aquaculture effect. The weight of the edge is the Pearson correlation coefficient between the indicators. 3. The indirect associations between nodes are verified through mediation effect analysis, and the hidden association edges of the graph are supplemented.

[0027] The node association rules are as follows: nodes with a correlation coefficient ≥ 0.5 are assigned strong association edges and labeled as core influencing factors; nodes with a correlation coefficient ≤ 0.2 and < 0.5 are assigned weak association edges and labeled as secondary influencing factors; nodes with a correlation coefficient < 0.2 have no association edges and are removed from the graph.

[0028] The knowledge graph update mechanism is as follows: every quarter, based on the newly added production test data and aquaculture feedback data, the weights of the nodes' associated edges are recalculated, the newly added indicator relationships are added as new nodes and new associated edges, and invalid relationships are removed.

[0029] The traceability system stage is based on standardized data and prior knowledge graphs constructed in the parameter standardization stage. It assigns a unique and tamper-proof distributed digital identity to each production batch of sea cucumber feed. The digital identity is generated using the national cryptographic SM2 asymmetric encryption algorithm and is bound to all data throughout the entire life cycle of the batch. The process of generating distributed digital identities using the national cryptographic standard SM2 is as follows: using the feed production batch number, the raw material supplier's unique code, and the production line number as the original plaintext, a public key and a private key are generated using the SM2 algorithm. The public key is recorded on the blockchain as a publicly verifiable digital identity identifier for the batch, while the private key is exclusively managed by the production enterprise. All data throughout the entire lifecycle of the batch is asymmetrically encrypted using the SM2 algorithm, with "production stage + data collection timestamp + original data" as the encrypted plaintext. The encrypted data digest and the hash value of the original data are recorded on the blockchain together, and only nodes holding the corresponding private key can decrypt and view the original data.

[0030] A consortium blockchain node network is established, jointly participated in by raw material suppliers, feed production enterprises, breeding entities, germplasm resource providers, third-party testing institutions, food safety regulatory departments, and end consumers. Hierarchical permissions and smart contract rules are set for each node to ensure that data read and write permissions match business scenarios and prevent unauthorized tampering. Based on the above encryption and permission control mechanism, the genomic traceability information of the corresponding batch of raw materials, the whole process parameters, equipment status data, quality inspection results, warehousing and logistics information, operator information, and biological effect data of the breeding end are uploaded to the blockchain in real time through the national cryptographic SM2 asymmetric encryption algorithm and smart contracts to ensure that the data is traceable and tamper-proof throughout the entire process. The hierarchical permissions of each node in the consortium blockchain are as follows: 1. Raw material suppliers: can only read and write traceability information and batch data of the raw materials they supply, and cannot view data of other suppliers or core process parameters of production enterprises; 2. Feed production enterprises: have read and write permissions for the entire data chain, and can initiate data modification requests and record them on the blockchain; 3. Aquaculture entity: Can read and write data on the aquaculture effect and finished product quality testing data of the corresponding batch of feed, but cannot view core parameters of the production process; 4. Third-party testing institutions: can only read and write test data for a specified batch, and cannot modify data at other nodes; 5. Food safety regulatory authorities: Possess read-only access to data across the entire supply chain and can initiate traceability queries and anomaly verification commands; 6. Germplasm resource providers / end consumers: can only read and write their own associated germplasm information / product traceability information, and have no other data operation permissions.

[0031] The core triggering conditions for smart contracts are: hash value verification when data is uploaded to the blockchain, node permission change application, synchronization of regulatory node information when quality is abnormal, public key verification for batch traceability query, and on-chain evidence confirmation of disposal results. After being triggered, data encryption, permission verification, and information push operations are automatically executed.

[0032] Based on the full-dimensional data on the blockchain, a four-level linked digital twin is constructed for each physical production batch, consisting of raw material attribute twins, production process twins, finished product quality twins, and aquaculture effect twins. This enables real-time dynamic correspondence between the physical production process and the digital twins, replicating the quality evolution process of the entire batch lifecycle. Based on prior knowledge graphs, the impact of process parameter fluctuations on finished product quality and aquaculture effect is simulated in real time, achieving a digital upgrade of discrete ledger records, improving the traceability chain, and enhancing data credibility.

[0033] The four-level digital twin achieves a one-way drive and reverse traceability linkage based on the quality transmission logic of raw material attributes → production process → finished product quality → aquaculture effect. The construction dimensions of each twin correspond one-to-one with the physical entity: the raw material attribute twin covers 10 core indicators such as raw material freshness and nutrient content; the production process twin covers 15 key process parameters and equipment operating status throughout the entire production process; the finished product quality twin covers 8 key quality attribute test results; and the aquaculture effect twin covers 9 core biological effect indicators. The four-level twin adopts a second-level data synchronization mechanism. After the real-time data collected from the production end / aquaculture end is uploaded to the blockchain, the corresponding twin data is updated synchronously. Any data fluctuation in any link will simulate the quality impact results in the downstream twin.

[0034] The root cause localization stage is based on the full-chain on-chain data and digital twin formed in the traceability system stage. For the full-chain data of the batches on the chain, the Do-Calculus causal inference algorithm with prior knowledge constraints is combined with Bayesian network and mediation effect analysis to eliminate irrelevant variables and confounding variables of the breeding environment. It explores the causal relationship of the entire chain of raw material attributes, process parameters, finished product quality and breeding effect, quantifies the impact weight of each key parameter on finished product quality and breeding effect and the critical threshold of risk transmission, clarifies the risk transmission path of each link, and constructs a full-chain risk transmission causal map of sea cucumber feed batch quality. This algorithm combination uses standardized data from the entire sea cucumber feed production chain as input. The input data includes 10 dimensions of raw material attributes, 15 dimensions of process parameters, 8 dimensions of finished product quality, 9 dimensions of aquaculture effects, and 5 dimensions of aquaculture environment. First, the input data is filtered using the Do-Calculus causal inference algorithm, retaining key variables with causal effect coefficients ≥ 0.1 as input to the Bayesian network. The Bayesian network is structured with four layers: raw material layer, process layer, quality layer, and aquaculture layer. Nodes between layers are connected by directed edges. The conditional probability table is constructed based on the maximum likelihood estimation method using historical sea cucumber feed production data. The inference algorithm uses the clustered tree propagation algorithm, with an inference error threshold set at 0.001. Finally, the causal effect weights of each key parameter and the critical threshold for risk transmission are output, providing a quantitative basis for the construction of the risk transmission map.

[0035] The risk transmission causal map is constructed in five layers: raw material germplasm layer → raw material attribute layer → production process layer → finished product quality layer → aquaculture effect layer. The risk transmission relationship between layers is unidirectional, and the risk relationship within layers is parameter-related. The map nodes are divided into core risk nodes and secondary risk nodes. Core risk nodes are key parameters affecting finished product quality / aquaculture effect (such as conditioning temperature and retention rate of heat-sensitive components), marked in red. Secondary risk nodes are auxiliary influencing parameters, marked in blue. The critical threshold for risk transmission and the influence weight are marked next to the nodes. The steps for applying the graph are as follows: When locating the root cause in reverse, investigate layer by layer from the abnormal result node upstream, and identify the first node on the risk transmission path that exceeds the critical threshold as the root cause node. When tracing forward, display the risk transmission range and potential impact layer by layer from the target node downstream, and simultaneously mark the on-chain data traceability path of each node.

[0036] When finished products fail to meet quality standards or exhibit abnormal results in aquaculture, reverse tracing can be achieved based on causal graphs and counterfactual reasoning. This allows for rapid identification of the root cause and scope of the quality problem. Simultaneously, the effects of different parameter correction schemes are simulated to provide optimal handling recommendations. Forward tracing is supported, enabling one-click query of information across the entire chain, from raw material germplasm, production and processing, warehousing and logistics to aquaculture applications and consumer terminals. Regulatory authorities can achieve full-chain supervision through consortium blockchain nodes, and consumers can query traceability information through batch digital identities, thus forming the ability to pinpoint the root cause.

[0037] The abnormal characteristic stage, combined with the causal relationship and parameter influence law of quality failure clarified in the root cause localization stage, constructs the degradation kinetic equation of the core heat-sensitive active component of sea cucumber feed under different process conditions through orthogonal experiments and accelerated degradation experiments, clarifying the intrinsic quality failure mechanism of process parameter fluctuations, component degradation, quality deterioration, and decline in aquaculture effect; the core heat-sensitive active component includes marine animal and plant-derived proteins, active polysaccharides, unsaturated fatty acids, etc. The practical application process of the degradation kinetic equation for thermosensitive components is as follows: 1. Substitute the real-time collected process parameters such as drying temperature, conditioning temperature, and granulation pressure into the equation to calculate the predicted real-time degradation rate and retention rate of the heat-sensitive component under the current process conditions. 2. Compare the predicted values ​​with the pre-set component retention rate thresholds in the process to determine whether there is a risk of degradation; 3. If the predicted value is lower than the threshold, combine the correlation between the process parameters output by the equation and the degradation rate to identify the core process parameters that cause component degradation; 4. Feed back the fluctuation range of core parameters to the early warning model as the core basis for the initial screening of quality risks, and at the same time provide a targeted direction for adjusting process parameters.

[0038] Based on degradation kinetics models and quality failure mechanisms, and combined with historical batch anomaly cases and quality defect data, a dedicated anomaly feature library with two dimensions of mechanism-driven and data-driven approaches is constructed. It includes seven categories: single parameter over-limit anomalies, multi-parameter correlation anomalies, batch-to-batch fluctuation anomalies, equipment progressive degradation anomalies, component degradation risk anomalies, quality failure critical anomalies, and aquaculture effect correlation anomalies. It covers the abnormal features corresponding to common quality defects in sea cucumber feed production, such as stratification, pelleting, loss of active components, uneven nutrition, and deterioration of stability in water. The specific characteristics of each anomaly type are defined as follows: 1. Single parameter out-of-limit anomaly: A single process parameter / quality indicator exceeds the preset threshold by ±10% or more; 2. Multi-parameter association anomaly: Each parameter is within the threshold range individually, but the coupling coefficient between parameters exceeds the association threshold of the prior knowledge graph; 3. Abnormal fluctuations between batches: The core parameters of more than three consecutive batches under the same process fluctuate by more than 5%; 4. Gradual degradation anomaly in equipment: There are no sudden exceedances of equipment operating parameters, but the parameters show a linear deviation for 72 consecutive hours and the cumulative deviation exceeds 3%; 5. Abnormal risk of component degradation: The retention rate of heat-sensitive components decreases non-linearly with fluctuations in process parameters, failing to reach the quality threshold but decreasing at a rate exceeding 2% / hour; 6. Critical anomaly in quality failure: The quality indicators of the finished product are in the critical range of ±5% of the threshold, and are accompanied by continuous fluctuations in process parameters; 7. Abnormal correlation between aquaculture effects: The quality indicators of the finished feed product are qualified, but the biological effect indicators at the aquaculture end are more than 8% lower than the benchmark value.

[0039] By employing an adaptive wavelet transform constrained by a degradation kinetic model and fusing it with a stacked autoencoder, the sensitive range of parameter fluctuations is first locked through the degradation kinetic model. Then, the original process data in the sensitive range is denoised and feature-enhanced to amplify the feature differences of weak anomalies. This enables the accurate extraction of early abnormal signals when the core components begin to degrade but the quality has not yet become unqualified, thus forming an abnormal feature extraction capability.

[0040] This fusion model takes time-series data of process parameters in the sensitive range of sea cucumber feed production as input. The input data dimensions are real-time acquisition sequences of 12 core process parameters, such as drying temperature gradient, conditioning temperature, and pelleting pressure. The stacked autoencoder has 3 hidden layers with 64, 32, and 16 neurons respectively. The activation function is ReLU, the loss function is mean squared error, the training batch size is 32, and the number of iterations is 200. Pre-training uses a layer-by-layer greedy training method, and fine-tuning uses the backpropagation algorithm. The adaptive wavelet transform uses the db4 wavelet basis with 4 decomposition layers. After data denoising, the feature vector is input into the stacked autoencoder for feature enhancement. The final output is a 16-dimensional weak anomaly feature vector, and the vector dimension corresponds one-to-one with the feature dimensions of the seven anomaly types in sea cucumber feed.

[0041] The early warning model stage is based on a dedicated abnormal feature library and abnormal extraction capability built from the abnormal feature stage. It constructs an intelligent early warning model for process abnormalities and quality risks, and adopts a dual-drive architecture of mechanistic model pre-constraint and machine learning model identification. The pre-degradation kinetic model of thermosensitive components is used to perform initial screening of quality risks on the input real-time process parameters, and to lock in high-risk ranges and highly sensitive parameters. A lightweight gradient booster is used as the base model, and a multi-scale attention mechanism is combined to focus on the correlation features with high impact weights and the fluctuation of sensitive range parameters. Finally, the model outputs the abnormality level, quality deterioration risk level, probability of abnormality occurrence, degree of component degradation, potential impact on aquaculture effect, root cause location and recommended treatment measures. The multi-scale attention mechanism sets up three attention branches: time scale, parameter correlation scale, and risk level scale. Each branch is connected to the feature input layer of the lightweight gradient booster through a fully connected layer. The attention weights are normalized by the sigmoid function, and high-impact features are assigned weight coefficients of 0.8-1.0. The base learners of the lightweight gradient booster are decision trees with a maximum depth of 8, a learning rate of 0.05, and 100 base learners. Column sampling is used to reduce feature dimensionality with a sampling rate of 0.8, and the loss function is a logarithmic loss function. This early warning model uses high-risk parameter data initially screened by the degradation kinetics model and 16-dimensional weak anomaly feature vectors output by the stacked autoencoder as joint inputs. The input data is weighted by the multi-scale attention mechanism and then fed into the lightweight gradient booster for anomaly identification.

[0042] The early warning model establishes a dual-cycle dynamic update mechanism. Based on the newly added batch production data and disposal results, it achieves incremental dynamic updates and continuously optimizes model parameters. At the same time, based on the newly added experimental data and feedback data from the breeding end, it regularly calibrates the degradation kinetic model of thermosensitive components, realizes the coordinated update of the mechanism model and the data model, avoids model drift, and continuously improves the early warning accuracy and scenario adaptability. After the early warning model is optimized for lightweight design, it is adapted for deployment at the edge of the production site, enabling real-time early warning and local handling, thus forming an early warning capability.

[0043] The specific steps for incremental dynamic updates are as follows: 1. Collect new batch production data and disposal results, perform standardized preprocessing, and label abnormal / normal samples; 2. Employ the mini-batch incremental training method, adding no more than 20% of the total sample size each time, freezing the bottom feature extraction layer of the model, and only fine-tuning the parameters of the top fully connected layer; 3. When the cumulative number of new samples reaches 50% of the total number of samples, perform full model fine-tuning and verify the model accuracy. If the accuracy is lower than 95%, retrain the model.

[0044] The parameter calibration rule for the degradation kinetic model of thermosensitive components is as follows: every quarter, based on the biological effect data fed back from the breeding end and the new accelerated degradation test data, the parameters such as the rate constant and activation energy of the kinetic equation are corrected, and the correction error is controlled within 5%. After calibration, the new parameters are synchronized to the mechanism constraint layer of the early warning model.

[0045] The early warning model is optimized using a lightweight approach involving module pruning and weight quantization: redundant branches with weight coefficients <0.2 in the multi-scale attention mechanism are pruned, and decision tree nodes with feature importance <0.01 in the lightweight gradient booster are removed; model weights are quantized from 32-bit floating-point numbers to 16-bit fixed-point numbers, reducing the number of model parameters to 40% of the original model and increasing inference speed to three times that of the original model. After lightweighting, the model's anomaly detection accuracy remains above 93%. When deployed at the edge, the model is directly connected to the production site PLC system and sensor acquisition terminals. The input is second-level process parameter data collected in real time by sensors, and the output is millisecond-level anomaly judgment results and handling instructions. The edge device locally caches nearly 72 hours of model input and output data, enabling independent early warning and local handling even when the network is offline.

[0046] The closed-loop management phase utilizes the early warning capabilities formed in the early warning model phase and the reliable data support of the traceability system to establish a deep collaborative link between the traceability system and the early warning model. This constructs a fully closed-loop management mechanism encompassing risk prediction, early warning triggering, digital twin backtracking, root cause localization, intelligent handling, effect verification, and on-chain archiving. This achieves full-process coverage of pre-event prevention, in-event management, and post-event traceability. When the early warning model identifies an abnormal signal or predicts a quality risk, it triggers the digital twin of the corresponding batch for real-time backtracking. Based on the full-link risk transmission causal graph, it quickly identifies the process node where the abnormality occurred, the raw material batches involved, and the equipment and personnel information. Simultaneously, based on the dual dimensions of quality risk level and the scope of abnormal impact, it triggers the corresponding graded handling plan. The threshold for anomaly classification is: 1. Minor anomaly: Only a small fluctuation in a single parameter / slight degradation of a single component, which does not affect the quality indicators of the finished product, the risk level of quality deterioration is ≤30%, and the scope of the anomaly is limited to a single production process; 2. Moderate abnormality: Multiple parameters are correlated abnormally / component degradation rate exceeds the standard, finished product quality indicators are in the critical range, quality deterioration risk level is 30%-70%, the scope of the abnormality covers 2 or more production processes, and has not flowed out of the production workshop; 3. Serious anomaly: Parameters exceed limits / component degradation reaches quality failure standards, finished product quality indicators are unqualified, quality deterioration risk level is ≥70%, and the scope of the anomaly has affected finished product storage / shipment.

[0047] The PLC system parameter fine-tuning rules are as follows: For process parameters that can be adaptively adjusted, such as drying temperature, conditioning temperature, and mixing speed, fine-tuning instructions are issued according to the principle of "step-by-step correction of 20% of the deviation value". After each correction, process data is collected for 30 seconds to verify stability. If the parameter returns to the threshold range, the correction stops. If it does not return, the step-by-step adjustment continues. The maximum magnitude of a single fine-tuning shall not exceed 5% of the parameter threshold.

[0048] For minor component degradation risks and abnormal parameter fluctuations, the system automatically sends parameter fine-tuning instructions to the PLC system of the corresponding equipment to achieve adaptive correction of process parameters and avoid quality deterioration in advance. For moderate abnormalities, the system pushes early warning information, root cause analysis results, and optimal handling suggestions verified by digital twin simulation to the on-site operators, and simultaneously suspends downstream high-risk processes. Production is resumed only after the handling effect is verified. For severe abnormalities, the system immediately triggers an emergency shutdown and simultaneously triggers full-chain traceability of the corresponding batch to lock the flow and scope of finished products that have been shipped out of the factory. The abnormal information, traceability data, and handling plan are synchronized to the regulatory nodes to achieve rapid control of food safety risks. All disposal processes, operation records, parameter adjustment results, and disposal effects are recorded on the blockchain in real time through smart contracts, and batch traceability files and digital twins are updated simultaneously to achieve full traceability and verifiability of the disposal process, providing labeled data for model updates.

[0049] In the optimization and collaboration phase, based on the risk management and data storage achieved in the closed-loop control phase, a standardized data collection interface for the aquaculture end is built based on the unique digital identity of each batch. The aquaculture entity uploads full-dimensional data on the aquaculture application of the corresponding batch of feed through the digital identity of the corresponding batch, including core performance indicators such as sea cucumber growth rate, survival rate, disease incidence rate, and feed conversion ratio, as well as biological phenotypic data such as sea cucumber immune enzyme activity and body wall quality. At the same time, basic data such as water temperature, salinity, and dissolved oxygen of the corresponding aquaculture environment are collected. The aquaculture data acquisition interface adopts the MQTT IoT transmission protocol, supports wired / wireless dual-mode connection, and is compatible with mainstream IoT devices such as aquaculture water quality monitors, growth performance testing equipment, and environmental sensors. The interface data upload format is JSON, uploaded according to the fixed fields of "batch digital identity + data collection time + indicator name + indicator value". The interface is set to both scheduled upload and manual upload modes. Environmental data is uploaded at 1 hour / upload, and data such as sea cucumber growth performance and biological effects are uploaded at 1 day / upload. Abnormal aquaculture data can be manually uploaded immediately by the aquaculture owner. The interface has a local data caching function, which can cache 7 days of data when offline and automatically re-upload it to the on-chain node after reconnection.

[0050] By using a multi-scale causal inference algorithm to eliminate confounding variables of the aquaculture environment, the feedback data from the aquaculture end is correlated with the corresponding batch of production chain data for analysis. This allows us to uncover the intrinsic causal relationship between production process parameters, raw material properties and aquaculture effects, and to extract the optimal combination of process parameters and raw material formula optimization rules for different sea cucumber growth stages, different aquaculture environments and different aquaculture models. The optimization rules are updated in reverse to the multi-dimensional parameter standardization system, prior knowledge graph, abnormal feature library and intelligent early warning model, and at the same time updated to the digital twin of batch traceability, so as to realize the whole life cycle quality iteration closed loop from raw material entry, production and processing, finished product application to feedback optimization at the breeding end.

[0051] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0052] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for batch quality traceability and intelligent early warning of process anomalies in sea cucumber feed production, characterized in that, include: Parameter standardization phase: Anchoring three core indicators—quality attributes, process parameters, and biological effects—to construct a unified standardized data dictionary and a dedicated prior knowledge graph, and to clarify the basic association rules between each indicator; Traceability system stage: Based on standardized data and prior knowledge graph, a unique and tamper-proof distributed digital identity is assigned to each feed production batch, and relevant data throughout the entire life cycle of the batch is stored on the blockchain in real time to build a real-time correspondence between physical batches and digital twins; Root cause localization stage: Based on the on-chain data and digital twins of the entire chain, reverse root cause identification and forward full-chain tracing are realized when quality anomalies occur; Anomaly Feature Stage: Combining the causal relationship of quality failure with the influence law of parameters, a two-dimensional anomaly feature library is constructed to achieve early weak anomaly signal extraction; Early warning model stage: Based on the abnormal feature library and abnormal extraction results, a dual-drive architecture intelligent early warning model for process abnormalities and quality risks is constructed to realize abnormal identification and related information output; Closed-loop management phase: Construct a full-process closed-loop management mechanism. After an early warning is triggered, the abnormal information is traced back and locked through a digital twin, and corresponding disposal measures are taken. All disposal processes are recorded on the blockchain in real time for evidence storage. Optimization and Collaboration Phase: Explore the causal relationship between production parameters and aquaculture effects, extract the optimal combination of process parameters and raw material formulation optimization rules, and update them in reverse to relevant systems and models.

2. The batch quality traceability and intelligent early warning method for process anomalies in sea cucumber feed production according to claim 1, characterized in that, In the parameter standardization stage, the entire feed production process is analyzed based on the physiological characteristics and nutritional needs of sea cucumbers at different growth stages. Combined with the unique quality requirements of sea cucumber feed, the three categories of core indicators are determined. The entire production process includes all stages from raw material acceptance, pretreatment, ingredient metering, mixing and conditioning, pelleting, drying and cooling, screening and packaging to finished product storage. The three categories of core indicators specifically include key quality attributes, key process parameters, and key biological effect indicators. The key quality attributes include raw material freshness, nutrient content, retention rate of core heat-sensitive active components, finished product particle size, water stability, gelatinization degree, and limits for heavy metals and biotoxins; the key process parameters include crushing speed and grading accuracy, batching accuracy, mixing time and speed, conditioning temperature and steam volume, granulation pressure and ring die compression ratio, drying temperature gradient and wind speed, and cooling time and material temperature; the key biological effect indicators include feeding activity, protein digestibility, immune enzyme activity, body wall collagen content, growth rate, survival rate, disease incidence, and feed conversion ratio of sea cucumbers at the corresponding growth stages. A standardized data dictionary is used to unify the format, normalize the dimensions, and clean up outliers of multi-source heterogeneous data; a dedicated prior knowledge graph is constructed based on basic research results in sea cucumber nutrition and feed processing and historical production data, focusing on the correlation between raw material attributes, process parameters, finished product quality, and aquaculture effects.

3. The batch quality traceability and intelligent early warning method for process anomalies in sea cucumber feed production according to claim 2, characterized in that, In the traceability system phase, the distributed digital identity is generated using the national cryptographic SM2 asymmetric encryption algorithm and is bound to all data throughout the entire batch lifecycle. A consortium blockchain node network with multi-party participation is established and hierarchical permissions and smart contracts are set. The consortium blockchain node network includes raw material suppliers, feed production enterprises, breeding entities, germplasm resource providers, third-party testing institutions, food safety regulatory departments, and end consumers. Each node is configured with hierarchical permissions and smart contract rules. The on-chain evidence includes the genomic traceability information of the corresponding batch of raw materials, the process parameters of the whole process, the equipment status data, the quality inspection results, the warehousing and logistics information, the operator information, and the biological effect data of the breeding end. All data are uploaded to the chain in real time through the national cryptographic SM2 asymmetric encryption algorithm and smart contracts. The digital twin is a four-level linkage structure consisting of raw material attribute twins, production process twins, finished product quality twins, and aquaculture effect twins. It realizes real-time dynamic correspondence between the physical production process and the digital twin, replicates the quality evolution process, and simulates the impact of parameter fluctuations.

4. The batch quality traceability and intelligent early warning method for process anomalies in sea cucumber feed production according to claim 3, characterized in that, In the root cause localization stage, the Do-Calculus causal inference algorithm constrained by prior knowledge is adopted, combined with Bayesian network and mediation effect analysis, to eliminate irrelevant variables and confounding variables of the breeding environment, to explore the causal relationship of the whole link, to quantify the influence weight of key parameters on the quality of finished products and breeding effects and the critical threshold of risk transmission, to clarify the risk transmission path and to construct a causal map of the whole link risk transmission. When quality or breeding results are abnormal, reverse root cause identification is achieved based on the full-chain risk transmission causal map and counterfactual reasoning, which clarifies the root cause nodes, scope of impact and key causes, and simulates the effect of parameter correction schemes to provide optimal treatment suggestions; at the same time, it supports forward full-chain traceability, realizing one-click information query from raw material germplasm to consumer terminal, meeting the traceability needs of regulatory and consumer ends.

5. The batch quality traceability and intelligent early warning method for process anomalies in sea cucumber feed production according to claim 4, characterized in that, In the abnormal characteristic stage, the degradation kinetic equation of the core thermosensitive active component is constructed through experiments to clarify the intrinsic relationship between process parameter fluctuations, component degradation, quality deterioration and decline in breeding effect. Then, a two-dimensional abnormal characteristic library is constructed by combining the causal relationship of quality failure and the law of parameter influence. The two-dimensional abnormal characteristic library combines mechanism-driven and data-driven approaches and includes seven major abnormal categories: single parameter over-limit anomaly, multi-parameter correlation anomaly, batch-to-batch fluctuation anomaly, gradual equipment degradation anomaly, component degradation risk anomaly, quality failure critical anomaly, and breeding effect correlation anomaly. Early weak anomaly signal extraction adopts an adaptive wavelet transform constrained by degradation dynamics model and a stacked autoencoder fusion method. By locking the sensitive range of parameter fluctuations, denoising and feature enhancement, the differences in weak anomaly features are amplified, thereby improving the accuracy of early anomaly identification.

6. The batch quality traceability and intelligent early warning method for process anomalies in sea cucumber feed production according to claim 5, characterized in that, In the early warning model stage, the dual-drive architecture specifically combines the pre-constraint of the mechanism model with the identification of the machine learning model. The pre-degradation kinetic model of the thermosensitive component completes the initial screening of quality risks and locks in the high-risk range and highly sensitive parameters. The machine learning model uses a lightweight gradient booster as the base model and combines a multi-scale attention mechanism to focus on high-impact weight features and parameter fluctuations in sensitive intervals, thereby achieving anomaly identification and related information output. The early warning model outputs information including anomaly level, quality deterioration risk level, probability of anomaly occurrence, degree of component degradation, potential impact on aquaculture performance, root cause localization results, and recommended treatment measures. The dual-cycle dynamic update mechanism uses newly added production data, treatment results, experimental data, and aquaculture feedback data to achieve incremental model updates and mechanistic model calibration. After lightweight optimization, the early warning model is adapted for deployment at the edge of the production site, ensuring real-time early warning and local treatment capabilities.

7. The batch quality traceability and intelligent early warning method for process anomalies in sea cucumber feed production according to claim 6, characterized in that, The closed-loop management phase establishes a complete closed-loop mechanism encompassing risk prediction, early warning triggering, twin backtracking, root cause identification, intelligent handling, effect verification, and on-chain archiving, covering the entire process of pre-event prevention, in-event management, and post-event traceability. Once an early warning is triggered, the abnormal node, raw material batch, equipment, and personnel information are traced back through the digital twin. Based on the quality risk level and the scope of the abnormality's impact, a tiered response plan is triggered, and corresponding measures are taken for different degrees of abnormality: minor abnormalities automatically send parameter fine-tuning instructions to the equipment PLC system; moderate abnormalities push early warning information and response suggestions, and suspend downstream high-risk processes; severe abnormalities immediately shut down the machine, lock the flow of finished products, and synchronize information to the monitoring node; all relevant data are uploaded to the blockchain in real time for evidence storage, and batch traceability files and digital twins are updated.

8. The batch quality traceability and intelligent early warning method for process anomalies in sea cucumber feed production according to claim 7, characterized in that, In the optimization and collaboration phase, a data collection interface for the aquaculture end is built based on the unique digital identity of the batch to collect aquaculture application data and environmental data. The data includes core performance indicators such as sea cucumber growth rate and survival rate, biological phenotypic data, and aquaculture environmental data such as water temperature, salinity, and dissolved oxygen. A multi-scale causal inference algorithm is used to remove environmental confounding variables, correlate and analyze aquaculture feedback data with production chain data, and extract the optimal combination of process parameters and raw material formulation optimization rules for different growth stages, different aquaculture environments and aquaculture models. The optimization rules are then updated in reverse to the parameter standardization system, prior knowledge graph, abnormal feature library, early warning model and digital twin to achieve collaborative optimization of the entire system.

Images