Agricultural product supply chain quality management method based on big data

Abstract

The invention relates to the technical field of agricultural product supply chain quality management based on big data, and discloses an agricultural product supply chain quality management method based on big data. Transaction data is extracted to construct a node and directed edge set, and a graph structure is formed; counting shipment quantity and calculating propagation probability, and generating a probability matrix; calculating a propagation threshold based on the matrix, setting an initial seed node set, solving a maximum propagation probability by using a shortest path algorithm, iteratively optimizing seed nodes under resource constraints, and generating a detection priority sequence; meanwhile, path precursor nodes are recorded, and source path backtracking and probability verification are achieved. According to the method, a whole-process quality management system integrating risk identification, detection optimization and traceability analysis is constructed, the problems of low resource utilization, non-quantifiable propagation, non-traceability of paths and the like in traditional supervision are solved, and the supervision efficiency and accuracy are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural product supply chain quality management based on big data, and specifically to a method for agricultural product supply chain quality management based on big data. Background Art

[0002] As agricultural product supply chains continue to expand and scale, the challenges facing quality and safety management are becoming increasingly prominent. Traditional quality control methods often rely on static sampling, empirical parameter setting, and manual evaluation at each stage, making it difficult to dynamically characterize and accurately identify quality risks across the entire supply chain and across multiple dimensions.

[0003] First, there's the issue of independent node determination. In most current quality inspection schemes, each supply chain link (such as production, warehousing, wholesale, and retail) is often independently assessed, obtaining quality indicators through periodic sampling or fixed-point testing. These assessments are then made based on single-point data and fixed standards. These approaches ignore the interplay and interconnectedness between nodes, making it impossible to comprehensively assess quality risks at the network level. Instead, they produce isolated "pass / fail" conclusions, making it difficult to guide subsequent traceability and refined monitoring. Second, there's the issue of subjective threshold setting. Existing quality warning and anomaly detection methods generally use empirical thresholds or simple statistical thresholds (such as the mean ± k times the standard deviation). These thresholds are often derived from historical data or expert experience, lacking a formalized, automated threshold generation mechanism. Changes in market conditions, logistics patterns, or product variety make it difficult to adjust thresholds in a timely manner, leading to frequent false positives and false negatives. Third, the difficulty of quantifying parameters. Many methods incorporate subjective weights, factors, or implicit parameters to balance the influence of different quality indicators and upstream and downstream links. However, these weights often lack a traceable calculation process and require manual experience or repeated experiments to determine. They have black box properties and are not conducive to technology promotion and regulatory compliance. Then, there is the problem of low resource allocation efficiency. Most current testing plans are based on manual scheduling or fixed cycles. Limited testing resources are not dynamically matched with risk coverage, which easily leads to the phenomenon of "over-testing in low-risk areas and under-testing in high-risk areas." This not only wastes manpower costs, but may also miss anomalies in truly critical links. In addition, dynamic traceability is difficult. When a quality anomaly occurs at a node in the supply chain, traditional solutions require manual inquiries and on-site verification to trace the source. The process is complicated and time-consuming, and it is difficult to accurately track based on a complete logistics network structure.

[0004] To this end, this case aims to propose a big data-based quality management method for agricultural product supply chains, aiming to achieve digital and intelligent management of the entire process, including risk propagation modeling, impact coverage prediction, key detection node screening, and source tracing of quality issues in the agricultural product supply chain. By constructing a node-edge model, combining a propagation probability matrix, a path optimization algorithm, and a greedy point selection mechanism, a computable, traceable, and evaluable quality management system is formed. The aim is to achieve digital and intelligent management of the entire process, including risk propagation modeling, impact coverage prediction, key detection node screening, and source tracing of quality issues in the agricultural product supply chain. By constructing a node-edge model, combining a propagation probability matrix, a path optimization algorithm, and a greedy point selection mechanism, a computable, traceable, and evaluable quality management system is formed. Summary of the Invention

[0005] The present invention provides a method for agricultural product supply chain quality management based on big data, which promotes the solution of the problems mentioned in the above background technology.

[0006] The present invention provides the following technical solution: a method for agricultural product supply chain quality management based on big data, comprising:

[0007] Extract supply chain node data from the agricultural product transaction database to form a node set, and then count the flow of goods between nodes based on circulation records to construct a directed edge set;

[0008] The total shipment volume of each node is counted, and based on the quality impact probability of unit shipment volume, the propagation probability of each edge is calculated to form a propagation probability matrix;

[0009] Based on the propagation probability matrix, calculate the effective propagation probability threshold of the entire network;

[0010] Set the initial seed node set, and use the shortest path algorithm to solve the maximum path propagation probability based on the edge cost and the cumulative path cost.

[0011] Under the constraint of limiting the maximum number of seed nodes, the incremental contribution of candidate nodes to the coverage range is iteratively calculated, and the seed node set is gradually updated;

[0012] Based on the final selected seed node set, determine the nodes that affect coverage and generate a detection priority sequence;

[0013] During the path propagation process, the predecessor mapping of the optimal path is recorded, and the path probability is verified from the coverage node to the source node along the predecessor node, and the source path and its probability value are output.

[0014] Optionally, extracting supply chain node data from the agricultural product transaction database to form a node set, and statistically analyzing the flow of goods between nodes based on circulation records to construct a directed edge set, specifically includes:

[0015] Extract supply chain node data from the agricultural product transaction database and record it as a node set:

[0016] V={v1,v2,...,v N}; where V is the set of all nodes in the supply chain network; v i is the i-th node; N is the total number of nodes;

[0017] Extract agricultural product circulation records from the database and count the data from node v i To node v j The actual number of goods flows u ij , forming an edge set: E={(v i ,v j )|u ij >0}; where E is a set of directed edges, each (v i ,v j ) indicates the direction of cargo flow; u ij For slave node v i To node v j The actual number of shipments.

[0018] Optionally, the method of summing up the shipment volume of each node and calculating the propagation probability of each edge based on the quality impact probability of the unit shipment volume to form a propagation probability matrix specifically includes:

[0019] For node v i , calculate the total shipment volume of the node T i , specifically:

[0020] Among them, T i For node v i Total shipments;

[0021] For each edge (v i ,v j ), calculate the edge propagation probability p ij , specifically:

[0022] Among them, p ij The mass impact propagation probability corresponding to the unit freight;

[0023] All p ij Composed probability matrix {p ij}.

[0024] Optionally, the calculating of the effective propagation probability threshold of the entire network based on the propagation probability matrix specifically includes:

[0025] Calculate the arithmetic mean of the propagation probabilities of all edges in the entire supply chain network as the propagation threshold γ, specifically: |E| is the number of elements in the directed edge set E.

[0026] Optionally, loop over the relaxation updates:

[0027] Until all C v (S) until it no longer changes;

[0028] Back-calculate the maximum path probability from the cumulative cost value:

[0029] Among them, P k (S) is the node v k The maximum path probability propagating from S.

[0030] Optionally, under the constraint of limiting the maximum number of seed nodes, iteratively calculating the incremental contribution of candidate nodes to the coverage range and gradually updating the seed node set specifically includes:

[0031] Let the maximum number of seed nodes be K;

[0032] Initialize the seed node set

[0033] For each round t = {1, 2, ..., K}:

[0034] S1. Calculate the current cover set: R(S) = {v k ∈V|P k (S)≥γ}; where R(S) is the set of influence coverage nodes when the seed set S is given;

[0035] S2. For each candidate node v∈V\S, let S'=S∪{v} and calculate the covering set R(S');

[0036] S3. Calculate the incremental contribution value of each candidate node v to the impact coverage, specifically:

[0037] Δ(v|S)=|R(S')|-|R(S)|; where |·| is the cardinality function, representing the number of elements; Δ(v|S) is the incremental contribution of node v to the coverage size after it joins;

[0038] S4. Select And update S←S∪{v *}; where v * It is the node v in the set S that contributes the most to the increase in coverage size.

[0039] Optionally, determining the affected coverage nodes and generating a detection priority sequence based on the finally selected seed node set specifically includes:

[0040] For the final selected seed node set S, calculate the maximum path probability P of each node k (S);

[0041] Determine the final set of nodes that will have an impact on coverage, specifically: R(S) = {v k |P k (S)≥γ};

[0042] V k ∈R(S), according to P k (S) Sort from largest to smallest to generate a detection priority list sequence:

[0043]

[0044] Optionally, the process of recording the predecessor mapping of the optimal path during the path propagation, tracing back from the coverage node along the predecessor node to the source node, verifying the path probability, and outputting the source path and its probability value specifically includes:

[0045] During the relaxation update process, record the predecessor mapping of each update:

[0046] π(v j )=v i ,when Among them, π(v j ) is the precursor mapping function, which records the optimal relaxation source node;

[0047] For each covered node v k ∈R(S), since v k Backtrack along the π chain to a certain v s ∈S, forming a path:

[0048] v s →v s+1 →…→v k ;

[0049] Verify the path probability: Among them, Π is a multiplication operator used to calculate the path probability;

[0050] Calculate and check the probability of each path, sort out all paths and their probabilities, and output them.

[0051] The present invention has the following beneficial effects:

[0052] 1. This method constructs a set of nodes and edges based on agricultural product transaction data to form a supply chain graph structure, and introduces the actual number of freight shipments as edge weights. This approach breaks through the limitations of traditional modeling based on administrative hierarchies or linear links, is closer to the actual logistics flow, and embodies the concept of "data is relationship." By integrating the graph model with real transaction information, the system can quantitatively model the connectivity and strength between any two nodes in the supply chain. It has good scalability and data-driven capabilities, laying a solid foundation for subsequent transmission modeling and risk prediction. This "graphification" approach breaks down information silos and improves the ability to control quality issues in cross-regional and cross-entity supply chains.

[0053] 2. Unlike existing quality propagation models based on abstract weights, this method incorporates the concept of "quality propagation probability per unit of freight" into the calculation of propagation probability. This approach quantifies the "quality propagation weight" of each edge by normalizing the shipment volume at each node. This approach not only preserves the data foundation of actual circulation intensity but also incorporates the physical implications of propagation modeling, making the propagation of quality issues on the network more realistic. This method transforms quality-impacting events into random propagation processes within a graph structure and models them using a propagation probability matrix. This method outperforms traditional "qualitative description" or "fixed logic backtracking" approaches in terms of data integrity and scientificity, providing a theoretical foundation for subsequent path analysis and risk screening.

[0054] 3. Traditional methods often rely on fixed standards or expert judgments to determine whether a node is affected, lacking objective, unified quantitative metrics. This method automatically generates a propagation threshold by calculating the arithmetic mean of all propagation probabilities, which serves as the standard for determining whether a node is covered by risk impact. This approach is consistent, interpretable, and dynamically adjustable. The threshold setting, based on the full-graph propagation probability, automatically adapts to the structural differences of different supply chains, effectively avoiding bias caused by subjective settings. It also makes subsequent coverage judgments and detection planning more scientific, transparent, and data-driven.

[0055] 4. This method transforms the propagation probability space into a path cost space, enabling the use of classic shortest path algorithms to solve for the maximum propagation path probability. This algebraic mapping method retains the mathematical advantages of additive solutions in graph theory while accurately preserving the practical meaning of probability space. Compared to traditional algorithms that focus solely on the shortest distance or maximum flow, this method efficiently calculates the "maximum probability of risk propagation from the source to a node," thereby identifying the nodes most likely to be affected. This mechanism provides quantitative support for risk level assessment and detection priority sorting, addressing the lack of propagation capacity modeling in existing systems.

[0056] 5. Existing methods often use fixed key nodes or empirical settings, which cannot take into account both resource constraints and coverage effects. The present invention introduces a greedy iterative algorithm based on "incremental impact contribution". Under the premise of limiting the total amount of detection resources (i.e. the number of seed nodes), each round evaluates the marginal gain of candidate nodes on the coverage range, and dynamically updates the optimal node set. This method combines algorithm efficiency and practical feasibility, and can achieve a good balance between cost and effect; at the same time, it supports flexible setting of termination conditions (such as coverage thresholds, marginal benefit thresholds), and adapts to the detection goals of different enterprises or regulatory agencies. This algorithm is particularly suitable for large-scale supply chain scenarios with limited human resources and detection equipment, which is an application blind spot that is difficult to cover with existing methods.

[0057] 6. Based on the calculated results of the seed node propagation probability, this method prioritizes all affected nodes and generates a test list. Compared to traditional methods of random testing or testing in administrative order, this method has the significant advantage of "sorting by influence," prioritizing resources to the links most likely to be contaminated or have problems, thereby greatly improving detection efficiency and early warning levels. The detection sequence not only has the ability to be dynamically adjusted, but can also be directly converted into an on-site inspection plan, improving the standardization, transparency, and traceability of business execution, and facilitating rapid response and deployment in supervision, accountability, and emergency response.

[0058] 7. Traditional quality problem tracing is mostly based on subjective judgment or manual backtracking, which is inefficient and prone to omissions. This method records the path predecessor mapping in real time during the propagation process, constructs a propagation path map, and can immediately trace back to the source node from the bottom up after the affected node is discovered, and verify the validity based on the path probability. This path traceability technology not only improves the accuracy of traceability, but also makes each path auditable, providing a strong technical basis for subsequent responsibility division, problem reproduction, and emergency response. This method has significant application potential in areas such as food safety, pesticide residue tracking, and supply chain traceability, and is significantly superior to the traditional manual "pulling a single trace" model. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0060] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0061] Example, see Figure 1, a method for agricultural product supply chain quality management based on big data, including:

[0062] Extract supply chain node data from the agricultural product transaction database to form a node set, and then count the flow of goods between nodes based on circulation records to construct a directed edge set;

[0063] The total shipment volume of each node is counted, and based on the quality impact probability of unit shipment volume, the propagation probability of each edge is calculated to form a propagation probability matrix;

[0064] Based on the propagation probability matrix, calculate the effective propagation probability threshold of the entire network;

[0065] Set the initial seed node set, and use the shortest path algorithm to solve the maximum path propagation probability based on the edge cost and the cumulative path cost.

[0066] Under the constraint of limiting the maximum number of seed nodes, the incremental contribution of candidate nodes to the coverage range is iteratively calculated, and the seed node set is gradually updated;

[0067] Based on the final selected seed node set, determine the nodes that affect coverage and generate a detection priority sequence;

[0068] During the path propagation process, the predecessor mapping of the optimal path is recorded, and the path probability is verified from the coverage node to the source node along the predecessor node, and the source path and its probability value are output.

[0069] A method for agricultural product supply chain quality management based on big data is systematically proposed. Its core is to form a full-link quality management system that integrates risk perception, problem location, resource optimization and traceability through a series of steps such as constructing a graph structure, calculating the propagation probability, determining the detection path and tracing back to the source. Through the series connection of seven links, including "extracting nodes and edges from the database", "constructing a propagation probability model", "calculating the shortest path propagation", "incremental selection of seed nodes", "test priority sorting" and "tracing back to the source path", the present invention effectively solves the practical pain points of low detection resource utilization, unclear problem traceability path, inability to quantify propagation risk, and difficulty in building a global model in traditional agricultural product quality supervision. The advantages of this solution are that it is highly structured and can be automatically executed. It is particularly suitable for application in large-scale, high-mobility modern agricultural product supply systems. It can significantly improve the supervision response speed and decision-making accuracy, and realize intelligent support for the whole process "from risk propagation modeling to source location".

[0070] The process extracts supply chain node data from the agricultural product transaction database to form a node set, and then calculates the flow of goods between nodes based on circulation records to construct a directed edge set, specifically including:

[0071] Extract supply chain node data from the agricultural product transaction database and record it as a node set:

[0072] V={v1,v2,...,v N}; where V is the set of all link nodes in the supply chain network; v i is the i-th node, representing a specific link, including a farm or storage center; N is the total number of nodes; it forms the vertex basis of each link in the supply chain and provides nodes for the graph model;

[0073] Extract agricultural product circulation records from the database and count the data from node v i To node v j The actual number of goods flows u ij , forming an edge set: E={(v i ,v j )|u ij >0}; where E is a set of directed edges, each (v i ,v j ) indicates the direction of cargo flow; u ij For slave node v i To node v j The actual number of freight batches; obtain the real flow of goods and provide raw data for subsequent propagation probability calculation.

[0074] This paper proposes a specific technical approach for extracting supply chain nodes from an agricultural product transaction database and constructing directed edge sets based on circulation records. By abstracting agricultural product transaction data into "node sets" and "edge sets," this method maps the traditional static supply chain process into a graph model with directionality and propagation capabilities. Its beneficial effect lies in that by counting the actual number of freight transactions between different nodes as edge weights, it not only achieves a structured representation of the supply chain's operating status but also provides basic data support for the precise calculation of subsequent propagation probabilities. This graphing method truly reflects the path selection and transmission frequency of agricultural products during the circulation process, helping to improve the reliability and visibility of the model when simulating the propagation of quality issues. This addresses the problems of "isolated node information," "lack of upstream and downstream cargo flow," and "fragmented structural analysis" that existed in previous regulatory processes, thereby improving the feasibility of quality problem traceability and responsibility identification.

[0075] The total shipment volume of each node is counted, and based on the quality impact probability of the unit shipment volume, the propagation probability of each edge is calculated to form a propagation probability matrix, which specifically includes:

[0076] For node v i , calculate the total shipment volume of the node T i , specifically:

[0077] Among them, Ti For node v i Total shipments; summary node v i All shipments of are used as the denominator in subsequent normalization;

[0078] For each edge (v i ,v j ), calculate the edge propagation probability p ij , specifically:

[0079] Among them, p ij is the probability of quality impact propagation corresponding to unit freight; converting freight volume into probability measurement to achieve quantification of the "quality impact propagation" model;

[0080] All p ij Composed probability matrix {p ij}.

[0081] By counting the total shipment volume of a node and based on the quality impact probability per unit of shipment volume, the propagation probability of each edge is calculated and a propagation probability matrix is ​​constructed. The key to this method lies in the organic integration of "freight intensity" and "quality risk" to establish a quantitative risk propagation model. By normalizing the shipment volume, the propagation probability of each edge reflects not only the intensity of freight circulation but also the likelihood of risk diffusion. This constructed propagation probability matrix provides an accurate risk measurement foundation for the entire network, facilitating subsequent calculations for path propagation, impact coverage, and seed selection. Compared to traditional propagation models that rely on expert experience scoring or uniform distribution estimation, this method is more data-driven and trustworthy, accurately reflecting the actual impact relationships between nodes, thereby addressing the key challenges of "the difficulty in modeling the propagation of quality issues" and "the lack of quantification of impact paths."

[0082] The calculation of the effective propagation probability threshold of the entire network based on the propagation probability matrix specifically includes:

[0083] Calculate the arithmetic mean of the propagation probabilities of all edges in the entire supply chain network as the propagation threshold γ, specifically: Where γ is the effective propagation probability threshold; if the maximum path probability of a node is ≥γ, it is considered "affected"; |E| is the number of elements in the directed edge set E; the equivalent propagation probability threshold in the network is automatically obtained, and the "affected" determination standard is unified.

[0084] The average value of the propagation probability is introduced as the "propagation effective probability threshold" and used as the criterion for determining whether a node is "affected." This method provides a unified, quantitative impact determination mechanism, effectively replacing traditional manual setting or empirical judgment methods. By calculating the arithmetic mean of the propagation probability matrix to set the threshold, the system can automatically adapt to changes in network structure without human intervention, thus avoiding misjudgments and subjective bias. Its beneficial effect is to enhance the model's ability to adapt to risk diffusion boundaries, ensure the scientific and consistent determination of influencing nodes, and thus improve the rationality and comprehensiveness of subsequent testing object selection. This technology also enhances the system's automated operation capabilities, facilitating batch quality screening and testing plan generation in large-scale agricultural product networks.

[0085] The setting of the initial seed node set and solving the maximum path propagation probability by the shortest path algorithm based on the edge cost value and the cumulative path cost value specifically include:

[0086] Set any node collection As a set of seed nodes; where S is the initial set of seed nodes; pre-set according to the detection resource upper limit or determined by external services; identifies the initial impact source and provides a starting point for propagation simulation;

[0087] Set the path cost of each edge to c ij =-ln(p ij ); where c ij For the edge (v i ,v j ) cost value, converting high probability into low cost; mapping high probability to low cost, realizing additivity from probability space to cost space;

[0088] For each node v∈V, the cumulative path cost is defined as:

[0089]

[0090] Among them, C v (S) is the minimum cumulative cost from any seed node to v; prepares the initial state for the multi-source shortest path algorithm and distinguishes source nodes from non-source nodes;

[0091] For each edge (v i ,v j ), loop to perform the relaxation update:

[0092] Until all C v (S) does not change any more; iteratively solve the minimum cumulative cost from any source node to each node;

[0093] Back-calculate the maximum path probability from the cumulative cost value:

[0094] Among them, P k (S) is the node v k The maximum path probability of propagation from S; the minimum cost is reflected as the maximum path probability to achieve the calculation result of the propagation probability.

[0095] By introducing the shortest path concept and setting the cost value of the propagation path to the negative logarithm of the propagation probability, the maximum propagation probability from any initial node to any node can be solved. Its benefit lies in solving the computability problem when it is impossible to directly sum or weight the path in the probability space. By mapping to the cost space, classic shortest path algorithms (such as Dijkstra and Bellman-Ford) can be applied. This technology breaks through the limitation that traditional graph algorithms cannot be directly applied to probabilistic propagation scenarios, realizes the fusion of probabilistic models and path optimization algorithms, and greatly improves computational efficiency and reliability. By inferring the maximum probability path by accumulating the minimum cost, it not only improves the accuracy of propagation modeling, but also provides an accurate basis for judging which nodes are most likely to be contaminated or affected. It is a core step to ensure the effectiveness of system risk simulation.

[0096] Under the constraint of the maximum number of seed nodes, iteratively calculate the incremental contribution of candidate nodes to the coverage range and gradually update the seed node set, specifically including:

[0097] Assume that the maximum number of seed nodes is K; set a constant according to the detection resource capacity; constrain the number of seed nodes to ensure that the method is feasible; when K is large, there are more optional seed nodes, and after iteration, the final impact coverage set R(S) will be larger, and more potential risk nodes can be detected; but at the same time, each new seed node must be arranged for additional on-site or remote detection, and the detection manpower, time and cost increase linearly; when K is too small, the detection cost is low but the coverage is insufficient, and key nodes may be missed; if it is too large, the marginal benefits will decrease and the cost will be unbearable; each candidate node needs to be re-calculated and the coverage increment Δ(v|S) evaluated; the number of iterations is proportional to K. When K increases, the overall complexity and execution time of the algorithm will increase significantly. The value of K is based on: Detection resource constraints: the total amount of actual available detection resources such as manpower, equipment, budget, time, etc.; directly determines the maximum number of detection nodes that can be undertaken. Target coverage: set the minimum proportion α of the expected coverage (for example, 80% of the nodes need to be covered by "influence" detection); based on meeting The minimum required K is determined. Marginal coverage gain threshold: Calculate the maximum incremental coverage max in each iteration vΔ(v|S); When this increment falls below the preset threshold δ (for example, adding fewer than 5 nodes per day is valuable), iteration can be stopped, and K at this point is used as the upper limit. Risk tolerance and priority: Different quality management strategies have different tolerances for missed detection risks. If the business focuses on zero defects, K can be appropriately increased; if the focus is on cost control, it can be reduced to around the critical value.

[0098] Initialize the seed node set Clear the initial seed set, prepare for incremental selection, and provide a starting point for incremental iteration;

[0099] For each round t = {1, 2, ..., K}:

[0100] S1. Calculate the current cover set: R(S) = {v k ∈V|P k (S)≥γ}; where R(S) is the set of nodes affected by a given seed set S; the set of nodes that evaluates the impact of existing seeds on the network;

[0101] S2. For each candidate node v∈V\S, let S'=S∪{v} and calculate the covering set R(S'); evaluate the coverage effect of each candidate node after adding it separately;

[0102] S3. Calculate the incremental contribution value of each candidate node v to the impact coverage, specifically:

[0103] Δ(v|S)=|R(S')|-|R(S)|; where |·| is the cardinality function, representing the number of elements; Δ(v|S) is the incremental contribution of node v to the coverage size after it joins; quantifying the marginal contribution of each candidate node guides greedy selection;

[0104] S4. Select Apply the greedy strategy to gradually approach the maximum impact coverage; and update S←S∪{v *}, the optimal candidate node is included in the seed set and enters the next round of iteration; where v * It is the node v in the set S that contributes the most to the increase in coverage size.

[0105] Under resource constraints, the seed nodes that can bring the maximum incremental impact coverage are selected through greedy iteration to construct an optimal set of detection starting points. The greatest benefit of this scheme is that under the conditions of limited resources (detection manpower, time, budget, etc.), the coverage of overall quality inspection is maximized and the marginal benefits of inspection investment are improved. It dynamically evaluates the importance of candidate nodes by defining "incremental contribution value", avoiding the problems of coverage duplication and inefficiency that may be caused by random or empirical selection of points. This method not only improves the breadth of impact coverage, but also improves the scalability and execution efficiency of the algorithm. In reality, agricultural product supervision often faces the dilemma of resource constraints and insufficient coverage. This method provides regulatory agencies with a practical "high coverage, low investment" strategy model, significantly improving the controllability and scientific nature of quality risk control.

[0106] The process of determining the affected coverage nodes and generating a detection priority sequence based on the finally selected seed node set specifically includes:

[0107] For the final selected seed node set S, calculate the maximum path probability P of each node k (S); Update the maximum propagation probability of each node under the final seed;

[0108] Determine the final set of nodes that will have an impact on coverage, specifically: R(S) = {v k |P k (S)≥γ}; get the set of all affected nodes that should be detected in the end;

[0109] V k ∈R(S), according to P k (S) Sort from largest to smallest to determine the detection priority, ensure that the most critical nodes are detected first, and generate a detection priority list sequence:

[0110] Form an executable quality inspection plan list.

[0111] It is proposed to calculate the set of nodes affected by the propagation of seed nodes, and sort them by the maximum path probability to generate a detection priority sequence. This method significantly improves the strategy and execution of the detection link, making resource allocation directional and targeted. By placing nodes with high propagation probabilities in the forefront, the system prioritizes covering objects with greater risks, which helps to improve the problem detection rate and the timeliness of quality control. In the traditional supervision model, sampling often relies on historical experience, randomness or manual allocation, lacks quantitative basis, and has the problem of missing high-risk points. This method is based on the propagation model, making the detection plan more logical and data-supported, and has good deployment feasibility. The implementation of this method can greatly improve the accuracy of detection work and the efficiency of resource utilization, and is an important technical support for the concept of "precision supervision".

[0112] The process of recording the predecessor mapping of the optimal path during the path propagation, tracing back from the coverage node along the predecessor node to the source node, verifying the path probability, and outputting the source path and its probability value specifically includes:

[0113] During the relaxation update process, record the predecessor mapping of each update:

[0114] π(v j )=v i ,when Among them, π(v j ) is the precursor mapping function, which records the optimal relaxation source node; retains the optimal source path information for each cost update;

[0115] For each covered node v k ∈R(S), since v k Backtrack along the π chain to a certain v s ∈S, forming a path:

[0116] v s →v s+1 →…→v k ;Extract the specific propagation links from seeds to affected nodes;

[0117] Verify the path probability: Where π is a multiplication operator used to calculate the path probability; ensuring that the backtracking path strictly corresponds to the calculated result and ensuring the reliability of the method;

[0118] Calculate and verify the probability of each path, organize all paths and their probabilities, and output them; provide auditable quality risk traceability basis to support decision-making and responsibility division.

[0119] By recording the predecessor nodes of each node during the propagation path update process, and then tracing back along the path to the propagation source node after discovering the affected node, the path probability is calculated, and finally the complete risk source path is output. This mechanism combines the two processes of "forward propagation modeling" and "backward path tracing" to achieve closed-loop modeling of quality problem propagation. In actual supervision, problem tracing is a key link in resolving agricultural product quality incidents. Traditional methods rely on personnel investigation, paper records, or manual comparisons of distributed systems, which are inefficient, prone to misjudgment, and unauditable. This technology combines graph structure with probability calculation to automatically restore the propagation chain and its impact probability, not only improving the accuracy and efficiency of traceability, but also providing a credible chain of evidence for risk responsibility definition and event decision-making. It is a key component of the entire system's intelligent closed loop, greatly enhancing the system's visualization, traceability, and reviewability.

[0120] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0121] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for agricultural product supply chain quality management based on big data, characterized in that: include: Extract supply chain node data from the agricultural product transaction database to form a node set, and then count the flow of goods between nodes based on circulation records to construct a directed edge set; The total shipment volume of each node is counted, and based on the quality impact probability of unit shipment volume, the propagation probability of each edge is calculated to form a propagation probability matrix; Based on the propagation probability matrix, calculate the effective propagation probability threshold of the entire network; Set the initial seed node set, and use the shortest path algorithm to solve the maximum path propagation probability based on the edge cost and the cumulative path cost. Under the constraint of limiting the maximum number of seed nodes, the incremental contribution of candidate nodes to the coverage range is iteratively calculated, and the seed node set is gradually updated; Based on the final selected seed node set, determine the nodes that affect coverage and generate a detection priority sequence; During the path propagation process, the predecessor mapping of the optimal path is recorded, and the path probability is verified from the coverage node to the source node along the predecessor node, and the source path and its probability value are output.

2. The agricultural product supply chain quality management method based on big data according to claim 1, characterized in that: The process extracts supply chain node data from the agricultural product transaction database to form a node set, and then calculates the flow of goods between nodes based on circulation records to construct a directed edge set, specifically including: Extract supply chain node data from the agricultural product transaction database and record it as a node set: V={v1,v2,...,v N }; where V is the set of all nodes in the supply chain network; v i is the i-th node; N is the total number of nodes; Extract agricultural product circulation records from the database and count the data from node v i To node v j The actual number of goods flows u ij , forming an edge set: E={(v i ,v j )|u ij >0}; where E is a set of directed edges, each (v i ,v j ) indicates the direction of cargo flow; u ij For slave node v i To node v j The actual number of shipments.

3. The agricultural product supply chain quality management method based on big data according to claim 2 is characterized in that: The total shipment volume of each node is counted, and based on the quality impact probability of the unit shipment volume, the propagation probability of each edge is calculated to form a propagation probability matrix, which specifically includes: For node v i , calculate the total shipment volume of the node T i , specifically: Among them, T i For node v i Total shipments; For each edge (v i ,v j ), calculate the edge propagation probability p ij , specifically: Among them, p ij The mass impact propagation probability corresponding to the unit freight; All p ij Composed probability matrix {p ij }.

4. The agricultural product supply chain quality management method based on big data according to claim 3 is characterized in that: The calculation of the effective propagation probability threshold of the entire network based on the propagation probability matrix specifically includes: Calculate the arithmetic mean of the propagation probabilities of all edges in the entire supply chain network as the propagation threshold γ, specifically: |E| is the number of elements in the directed edge set E.

5. The agricultural product supply chain quality management method based on big data according to claim 4 is characterized in that: The setting of the initial seed node set and solving the maximum path propagation probability by the shortest path algorithm based on the edge cost value and the cumulative path cost value specifically include: Set any node collection As the seed node set; where S is the initial seed node set; Set the path cost of each edge to c ij =-ln(p ij ); where c ij For the edge (v i ,v j )’s cost value, turning high probability into low cost; For each node v∈V, the cumulative path cost is defined as: Among them, C v (S) is the minimum cumulative cost from any seed node to v; For each edge (v i ,v j ), loop to perform the relaxation update: Until all C v (S) until it no longer changes; Back-calculate the maximum path probability from the cumulative cost value: Among them, P k (S) is the node v k The maximum path probability propagating from S.

6. The agricultural product supply chain quality management method based on big data according to claim 5, characterized in that: Under the constraint of the maximum number of seed nodes, iteratively calculate the incremental contribution of candidate nodes to the coverage range and gradually update the seed node set, specifically including: Let the maximum number of seed nodes be K; Initialize the seed node set For each round t = {1, 2, ..., K}: S1. Calculate the current cover set: R(S) = {v k ∈V|P k (S)≥γ}; where R(S) is the set of influence coverage nodes when the seed set S is given; S2. For each candidate node v∈V\S, let S'=S∪{v} and calculate the covering set R(S'); S3. Calculate the incremental contribution value of each candidate node v to the impact coverage, specifically: Δ(v|S)=|R(S')|-|R(S)|; where |·| is the cardinality function, representing the number of elements; Δ(v|S) is the incremental contribution of node v to the coverage size after it joins; S4. Select And update S←S∪{v * }; where v * It is the node v in the set S that contributes the most to the increase in coverage size.

7. The agricultural product supply chain quality management method based on big data according to claim 6 is characterized in that: The process of determining the affected coverage nodes and generating a detection priority sequence based on the finally selected seed node set specifically includes: For the final selected seed node set S, calculate the maximum path probability P of each node k (S); Determine the final set of nodes that will have an impact on coverage, specifically: R(S) = {v k |P k (S)≥γ}; V k ∈R(S), according to P k (S) Sort from largest to smallest to generate a detection priority list sequence:

8. The agricultural product supply chain quality management method based on big data according to claim 7 is characterized in that: The process of recording the predecessor mapping of the optimal path during the path propagation, tracing back from the coverage node along the predecessor node to the source node, verifying the path probability, and outputting the source path and its probability value specifically includes: During the relaxation update process, record the predecessor mapping of each update: π(v j )=v i ,when Among them, π(v j ) is the precursor mapping function, which records the optimal relaxation source node; For each covered node v k ∈R(S), since v k Backtrack along the π chain to a certain v s ∈S, forming a path: in s →in s+1 →…→in k ; Verify the path probability: Among them, ∏ is a multiplication operator used to calculate the path probability; Calculate and check the probability of each path, sort out all paths and their probabilities, and output them.

Images