Agricultural product supply and marketing management system based on big data analysis

Abstract

The invention provides an agricultural product supply and marketing management system based on big data analysis, and the system comprises a digital label module which marks the basic information, extended information and unique identification of each single agricultural product; the dynamic adjustment module is used for obtaining an optimal distribution scheme by using multi-objective optimization and graph network path planning based on the result of the digital label module and the logistics information; the dynamic adjustment module updates the logistics information in real time in the transportation process of the agricultural products, analyzes the potential risk of the logistics information through an abnormity early warning model, and adjusts the optimal distribution scheme when the potential risk exceeds a threshold value; and the traceability management module is used for generating an A / B test scheme according to the optimal distribution scheme, implementing better optimization suggestions in the test scheme into the distribution scheme which is being executed, and completing continuous feedback and iteration of the distribution scheme. The method has the beneficial effects that distribution paths and inventory configuration are more reasonable, and the logistics cost and loss are remarkably reduced; the A / B test closed loop promotes continuous upgrading of the operation level.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of computer, and particularly relates to a management system for supply and marketing of agricultural products based on big data analysis. BACKGROUND

[0002] Before the management system for supply and marketing of agricultural products based on big data analysis and intelligent optimization is introduced, the traditional industry relies on the following several types of technology and process:

[0003] The first-line farmers or production bases register batches of agricultural products by means of paper labels, manual forms, etc., such as varieties, planting areas, simple quality inspection results, etc.; in the links of wholesale, transportation, warehousing, distribution, etc., information is often recorded by manual or scanning fixed two-dimensional codes, but only static information is often recorded, and real-time automatic collection cannot be realized; after information is collected, it is presented in Excel, database background or ERP system, but due to the fragmentation of data in each link, end-to-end transparency and visibility cannot be achieved.

[0004] The existing distribution center often uses experience rules or pre-set static routes to distribute goods. For example, according to the past historical delivery quantity, distance, vehicle fleet size and other factors, a fixed transportation scheme of "from A warehouse to B market" is manually formulated; the path planning usually only stops at the simple algorithm of the shortest distance or the least number of transfers, and lacks dynamic consideration of multi-dimensional constraints such as real-time road conditions (traffic congestion, road construction) and cold chain environment (temperature, humidity).

[0005] Part of the large agricultural bulk enterprises have installed GPS positioning and temperature and humidity sensors on the cold chain transport vehicles, but they are mostly used for post-query or regular alarm (such as alarm only when the temperature is higher than the preset upper limit); risk warning relies on manual monitoring platform, and once an abnormality is found, the relevant personnel are usually notified by telephone or short message, and the response speed and automation degree are limited. SUMMARY

[0006] Therefore, the present application aims to provide a management system for supply and marketing of agricultural products based on big data analysis, so as to solve at least one of the above technical problems.

[0007] To achieve the above purpose, the technical scheme of the present application is as follows:

[0008] The first aspect of the present application provides a management system for supply and marketing of agricultural products based on big data analysis, comprising:

[0009] A digital label module configured to label the basic information, extended information and unique identifier of each agricultural product single product;

[0010] a dynamic adjustment module configured to obtain an optimal distribution scheme using multi-objective optimization and graph network path planning based on the result of the digital tag module and the logistics information; wherein the dynamic adjustment module updates the logistics information in real time during the transportation of the agricultural products, analyzes potential risks of the logistics information through an abnormality early warning model, and adjusts the optimal distribution scheme when the potential risks exceed a threshold value;

[0011] a traceability management module configured to generate an A / B test scheme according to the optimal distribution scheme, implement better optimization suggestions in the test scheme to the distribution scheme being executed, and complete continuous feedback and iteration of the distribution scheme.

[0012] Further, the basic information includes the variety, origin, and harvesting time of the agricultural products.

[0013] The extended information includes the grade and primary processing conditions of the agricultural products.

[0014] The unique identifier is composed of a self-incrementing number, a batch number, and a serial number.

[0015] Further, the multi-objective optimization includes:

[0016] Based on the three objectives of minimizing transportation cost, shortening delivery delay, and reducing risk, a target function in the form of ε-constraint is constructed.

[0017] Based on four conditions of vehicle capacity constraint, flow balance and connectivity, time window restriction, and task unique assignment, the feasible solution space is described.

[0018] Further, the graph network path planning includes:

[0019] By constructing a graph network containing transportation cost, delay, and risk, and designing a weighted multi-objective heuristic function, optimization is performed in combination with the ε-constraint method.

[0020] During the path planning process using the graph network, the transportation cost, delay, and risk are calculated and updated in real time by expanding nodes through a priority queue, to ensure that the path meets the main objective of minimizing transportation cost while constraining the delay and risk within a preset upper limit.

[0021] Through constraint checking of capacity, flow balance, time window, and task unique assignment, it is ensured that each path meets the feasible solution space, thereby obtaining an optimal path that optimizes cost, timeliness, and risk at the same time.

[0022] Further, the working process of the abnormality early warning model is as follows:

[0023] The abnormal early warning model scores multiple source information in real time, if the score exceeds the set threshold, the affected subgraph is identified and local or global re-optimization is triggered, a new feasible scheme is generated by re-running the constraint process, and the scheduling execution is issued.

[0024] Further, the process of generating an A / B test scheme according to the optimal distribution scheme comprises:

[0025] Obtain relevant data of the optimal distribution scheme and perform feature processing;

[0026] Based on the data after feature processing, a binary classification model is trained, the data is divided into multiple batches by time slicing, and the complaint probability score of each batch is output;

[0027] An A / B test scheme is set according to the original quality inspection process and the newly added manual review process, and the high-risk batches with higher scores are randomly allocated to two groups.

[0028] Further, the process of implementing the better optimization suggestion in the test scheme to the executing distribution scheme comprises:

[0029] The A / B test scheme after the random grouping of the high-risk batches is executed, and the key features of the high-risk batches are extracted, the optimization scheme for the temperature control, quality inspection, packaging and route dimensions in the strategy library is matched according to the key features, and the intervention suggestion is generated;

[0030] According to the intervention suggestion, part of the parameters in the original optimal distribution scheme are replaced to obtain a new optimal distribution scheme, and the intervention suggestion is fed back to the strategy library, and the model is iterated by supplementing new labels and updating strategies.

[0031] The second aspect of the application provides an electronic device, comprising a processor and a memory connected with the processor and used for storing executable instructions of the processor, and the processor is used for executing the system of the first aspect.

[0032] The third aspect of the application provides a server, comprising at least one processor and a memory connected with the processor, the memory stores executable instructions of the at least one processor, and the instructions are executed by the processor to make the at least one processor execute the system of the first aspect.

[0033] The fourth aspect of the application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the system of the first aspect.

[0034] Compared with the prior art, the agricultural product supply and marketing management system based on big data analysis has the following beneficial effects:

[0035] The whole-process visualization and transparency of the supply chain improves the trust level.

[0036] The distribution path and inventory configuration are more reasonable, and the logistics cost and loss are significantly reduced.

[0037] Real-time early warning effectively guarantees product quality and safety.

[0038] A / B test closed loop promotes the continuous upgrading of operation level.

[0039] The system architecture has good compatibility and scalability, leaving enough space for subsequent technology iteration. BRIEF DESCRIPTION OF DRAWINGS

[0040] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application and are incorporated in and constitute a part of this application. The embodiments of the present application and its description together with the drawings make it clear that the present application is not limited to the described embodiments and can be practiced in a variety of ways. In the drawings:

[0041] Figure 1 The structure diagram of the agricultural product supply and marketing management system based on big data analysis according to the embodiments of the present application. DETAILED DESCRIPTION

[0042] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0043] In the description of the present application, it should be understood that the terms "center", "longitudinal", "transverse", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the purpose of facilitating the description of the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore it cannot be understood as a limitation of the present application. In addition, the terms "first", "second" and the like are only used for description purpose, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features limited by "first", "second" and the like can explicitly or implicitly include one or more of the features. In the description of the present application, unless otherwise specified, the meaning of "a plurality of" is two or more.

[0044] In the description of the present application, it should be noted that unless otherwise explicitly specified and limited, the terms "mounting", "connection", "linking" should be understood in a broad sense, for example, it can be fixed connection, or detachable connection, or integral connection; it can be mechanical connection, or electrical connection; it can be direct connection, or indirect connection through intermediate medium, or internal communication of two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0045] The present application will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0046] The agricultural product supply and marketing management system based on big data analysis comprises:

[0047] A digital tag module is configured to mark the basic information, extended information and unique identification of each agricultural product single product.

[0048] A dynamic adjustment module is configured to obtain the best distribution scheme by using multi-objective optimization and graph network path planning method based on the results of the digital tag module and logistics information. The dynamic adjustment module updates the logistics information in real time during the transportation of the agricultural products, analyzes the potential risks of the logistics information through an abnormal early warning model, and adjusts the best distribution scheme when the potential risks exceed a threshold.

[0049] A traceability management module generates a prediction report according to the best distribution scheme, and combines A / B testing to implement the optimized suggestions obtained by prediction into the distribution scheme being executed, thereby completing the continuous feedback and iteration of the distribution scheme.

[0050] The basic information includes the variety, origin and harvesting time of the agricultural product.

[0051] The extended information includes the grade and primary processing conditions of the agricultural product.

[0052] The unique identification is composed of a self-incrementing number, batch number and serial number.

[0053] In some embodiments, the workflow of the digital tag module is as follows:

[0054] I. Marking basic information:

[0055] 1. The variety and latitude and longitude of the origin are connected to the plot GIS map, and the plot boundary and soil type are marked in real time.

[0056] In traditional labels, only large categories such as "tomato" and "apple" are marked, but in the digital scenario, it is necessary to specify to the variety level, such as "Chilean red cherry tomato", "Yanfu No. 8 apple", "early spring cherry tomato", etc. At the same time, through the docking with the back-end system of farmers or cooperatives, the planting variety name, seed batch number and cultivation qualification (such as organic certification, green food certification, etc.) are recorded every time of sowing. In this way, when the terminal scans the label, consumers can directly see the variety source, cultivation qualification and related cultivation records behind this batch of products.

[0057] Further, the latitude and longitude of the production place are deeply integrated with the GIS map. When "latitude and longitude of the production place" is marked in the label, not only the geographical name such as "Xinjiang Uygur Autonomous Region Changji Hui Autonomous Prefecture" is written, but also the accurate coordinates such as "north latitude 44.0075°, east longitude 87.2968°" are written, and the coordinates are docked with the boundary data of the land in the GIS platform, so as to complete the "real-time marking":

[0058] Through the GIS map, the back-end can directly present the polygon boundary of the specific growth place of the product, so that the supervisors and consumers can see the product source at a glance. Based on the polygon boundary, the soil classification layer (such as sandy soil, clay, organic matter content, etc.) is superimposed, and the consumers can query the fertility level, acid-base degree and other information of the soil of this land, providing a scientific basis for the expectation of product quality.

[0059] 2. The harvesting date is accurate to the hour, and the harvesting temperature, humidity environment is recorded.

[0060] Traditional labels often only mark the date to the day, such as "May 20, 2025", but modern consumers, especially people with high requirements for fresh products, want to know "which day and time" the picking is, so as to evaluate the freshness. The "harvesting date" is accurate to the hour, for example, "May 20, 2025, 08:30 am" or "15:45 pm", which is particularly important for some temperature-sensitive crops (such as strawberries, blueberries); through the docking of intelligent handheld terminals (or agricultural machines) at the picking site, the operator triggers the time stamp on the spot after cutting the tie and picking, ensuring that the data no longer depends on human memory.

[0061] Further, the corresponding environmental data can be recorded during the picking process, including:

[0062] Harvesting temperature, during the picking operation, the handheld terminal automatically links with the nearby weather sensor (or temperature probe at the top of the vegetation), and real-time collects the air temperature data (unit: ℃) at the time of picking;

[0063] Harvest humidity, with a humidity sensor (air humidity meter), the relative humidity (RH%) can be recorded in real time. This data is crucial for subsequent transportation methods (ambient temperature, cold chain) and preservation scheme development;

[0064] Environmental illumination / ultraviolet index, under direct sunlight or rainy weather, the sugar content and flavor of crops will be different. You can choose to record the light intensity when picking important crops (such as tea, medicinal materials) as a reference for later fine quality assessment;

[0065] Application value, once a batch of products is found to have quality problems in the circulation link, you can trace back to the temperature and humidity environment at that time, analyze whether it is related to high temperature and humidity leading to bacterial growth, enzyme reaction, etc., and provide scientific basis for improving post-harvest links.

[0066] II. Label extension information:

[0067] 1. The grading results of product grades are refined to "weight interval" and "specification details".

[0068] Traditional grading labels generally have only "superior" "first class" "second class" such broad grades, while modern e-commerce and bulk wholesale scenarios require more detailed specification parameters. For example, for apples, not only "first-class fruit" needs to be indicated, but also the weight interval (such as "180-200g per fruit") and specification details (such as "diameter 80-85mm") of the batch of apples need to be reflected. For vegetables, it can be further divided into "fruit length 10-12cm" "single fruit net weight 15-18g", etc.

[0069] These parameters can be automatically collected and uploaded on the sorting line by intelligent sorting machines (equipped with weighing sensors and high-definition cameras), and finally displayed on the label as "grade: first class | weight: 180-200g | diameter: 80-85mm", allowing buyers or e-commerce sellers to more intuitively choose the required specifications, reducing the rate of returns and exchanges due to inaccurate descriptions.

[0070] When farmers or processing plants re-sort and re-pack, if the original record is found to be inconsistent with the actual situation on the sorting line, "dynamic correction" can be made through the background, for example, a batch of apples initially sorted as "weight 150-170g per fruit", but found to be actually "155-165g" in the circulation link. The operator can manually modify it in the system, and the label can be automatically updated to the latest grading result through secondary scanning, ensuring that the label information is consistent with the actual object and reducing disputes.

[0071] 2. Initial processing conditions include temperature control, cleaning, and sorting parameters.

[0072] Temperature control parameters, clearly define the cold chain temperature in primary processing, such as "pre-cooling temperature: 4℃; fresh-keeping warehouse temperature: 0-2℃; loading temperature for transportation: 2℃".

[0073] Cleaning parameters, including cleaning equipment type (tumbler / sprayer), water temperature (e.g. "12±2℃"), and water treatment method (ozone disinfection / sodium hypochlorite concentration 0.5ppm).

[0074] Sorting parameters, including sorting line speed (e.g. "2m / s"), sorting belt width, and sorting threshold (e.g. bad fruit color threshold).

[0075] Further, "sterilization method", "laser sorting", and "air cooling time" are added in primary processing conditions.

[0076] Sterilization method, traditional "high-temperature blanching" or "sodium hypochlorite disinfection" can damage fruit and vegetable tissue; modern technologies introduce ozone sterilization, ultrasonic sterilization, and electrolytic water sterilization. Record specific sterilization methods in the label, such as "ozone concentration 0.3mg / L×5min immersion", "ultrasonic power 40kHz×3min treatment".

[0077] Laser sorting, laser sorting machines use laser-induced breakdown spectroscopy (LIBS) or near-infrared technology to quickly determine whether there are diseased tissues, pesticide residues, or quality defects inside the fruit. Write parameters such as "laser wavelength range: 1064nm; trigger frequency: 50Hz; detection accuracy: ±0.5%" into the label to allow downstream detection agencies to trace and verify: for example, inspectors can query the specific sorting report at that time from the equipment manufacturer's background based on the laser sorting equipment number and parameters recorded on the label, and further compare the reliability of the results.

[0078] Air cooling time, air cooling refers to rapidly reducing the surface temperature of crops to a preset temperature after picking through forced wind speed to slow down the respiration rate. Air cooling parameters include wind speed (e.g. "≥3m / s"), wind temperature (e.g. "2℃±1℃"), and duration (e.g. "20min"). Label "air cooling after harvesting: 2℃×3m / s×20min" to facilitate consumers or terminal merchants to quickly determine whether the product has completed pre-cooling when receiving goods, and decide whether to immediately enter the cold storage or can be displayed at room temperature.

[0079] Three, label unique identification:

[0080] Self-incrementing number is UUIDv6, which is more optimized in sorting performance and can generate almost non-repeating 128-bit unique identifiers based on timestamps.

[0081] Batch number is composed of "origin code + production date + production line number", for example "XJ-CJ-20250520-01", which is convenient for directly locating the original production batch of the product in the platform system.

[0082] Serial number is within the same batch, and a number from "00001" to "99999" is assigned to each product. For example, "00037" represents the 37th product in the batch.

[0083] After combining the three, a complete identification is as follows:

[0084] 1e8c6aa0-d34a-6f4b-91c8-0e9e1d00c123 / XJ-CJ-20250520-01 / 00037;

[0085] If you quickly check the platform background after bundled use, you can get the whole link data of the product from sowing, fertilizing, pest and disease records to harvesting, sorting, loading, transportation, etc.

[0086] The multi-objective optimization includes:

[0087] Based on the three objectives of minimizing transportation cost, shortening delivery delay and reducing risk, the objective function in the form of ε-constraint is constructed;

[0088] Based on the four conditions of vehicle capacity constraint, flow balance and connectivity, time window restriction, and task unique assignment, the feasible solution space is described.

[0089] In some embodiments, the specific content of the above multi-objective optimization is as follows:

[0090] Define decision variables and parameters:

[0091] Decision variables: xijk, binary or continuous variable, indicating whether vehicle i performs transportation task from node j to node k; yi, binary variable indicating whether vehicle i starts (used for fixed cost calculation); ti, departure time or cumulative running time of vehicle i, used for time window constraint.

[0092] Parameters: djk, distance or time from node j to node k; cjk, transportation cost per unit distance (fuel, toll); Qi, maximum load of vehicle i; wp, weight of batch p.

[0093] Decision variables

[0094] The three types of path indicators include:

[0095] f1(x) = ∑ i Σ j Σ k cjk d jk x ijk ;

[0096]

[0097] f3(x)=∑ i ∑ j Σ k r jk x ijk ;

[0098] The ε-constraint method is used to minimize the transportation cost as the main objective, and the other two objectives are converted into constraints:

[0099]

[0100] where ε2, ε3 are the upper limits of time delay and risk given by the decision maker, by scanning different ε, the Pareto frontier solution set can be generated, or other objectives can be set as the main objective, and the remaining two indicators are ε-constrained.

[0101] The batch index z ip ∈{0, 1} represents whether the vehicle i is loaded with batch p, and the arrival time ti. The feasible region X is defined by the following four types of constraints:

[0102] The total load of each vehicle i does not exceed the limit:

[0103] Ensure the connectivity and closure of the route, that is, the inflow of any intermediate node is equal to the outflow:

[0104]

[0105] At the same time, sub-loop elimination (MTZ) or connectivity cutting plane can be added to eliminate sub-optimal loops;

[0106] If node j has a serviceable time window Then:

[0107]

[0108] And set the lower bound of the start time for the first vehicle i

[0109] Ensure that each distribution task (or each batch of goods from j to k) is only undertaken by one vehicle:

[0110]

[0111] The graph network path planning includes:

[0112] By constructing a graph network containing transportation cost, time delay and risk, and designing a weighted multi-objective heuristic function, combined with ε-constraint method for optimization;

[0113] In the path planning process using graph network, the node is expanded by priority queue, and the transportation cost, time delay and risk are calculated and updated in real time to ensure that the path meets the main objective of minimizing transportation cost while constraining the time delay and risk within the preset upper limit.

[0114] Through the constraint check of capacity, flow balance, time window and task unique allocation, it is ensured that each path meets the feasible solution space, so as to obtain an optimal path that optimizes cost, time efficiency and risk at the same time.

[0115] In some embodiments, the above graph network path planning is as follows:

[0116] Each state contains not only the current position node n, but also records: cumulative transportation cost g1(n), cumulative delivery time delay g2(n), cumulative risk cost g3(n), current timestamp t(n) (for time window check), and loaded goods batch situation (for capacity check);

[0117] For any edge (j→k), (cjk, djk, rjk) is pre-calculated and stored for updating g1, g2 and g3 respectively.

[0118] A weighted multi-objective heuristic is designed:

[0119] h(n) = αh1(n) + βh2(n) + γh3(n):

[0120] Where h1(n) is the lowest remaining transportation cost estimate from node n to the target (such as Euclidean distance x lowest unit cost); h2(n) is the lowest remaining time estimate (such as Euclidean distance ÷ highest vehicle speed); h3(n) is the lowest remaining risk estimate (based on the total risk of the safest path);

[0121] Before each state expansion, the following verification is performed:

[0122] g2(n) + h2(n) ≤ ε2, g3(n) + h3(n) ≤ ε3;

[0123] Only when the "current cumulative + estimated remaining" time delay and risk do not exceed the given threshold, the node is added to the "to be expanded" queue, otherwise it is pruned.

[0124] In this embodiment, A* algorithm is used for graph network path planning, and the workflow is as follows:

[0125] 1. Initialization:

[0126] Start state s: g1(s) = 0, g2(s) = 0, g3(s) = 0, load empty, t(s) = Tdepart;

[0127] Insert s into OpenList, ordered by f(s) = g1(s) + h(s).

[0128] 2. Loop expansion:

[0129] Pop state n from OpenList (with minimum f value), if n has reached the end and satisfies all constraints (capacity, flow balance, time window, task assignment), return the path; otherwise, for each outgoing edge (n→m):

[0130] 21. Update cumulative indicators:

[0131] g1(m) = g1(n) + cjk, g2(m) = g2(n) + djk, g3(m) = g3(n) + rjk, t(m) = t(n) + djk;

[0132] 22. Constraint verification:

[0133] Capacity, new load does not exceed Qi; flow balance & connectivity, ensure path continuity, no sub-loop; time window, Task uniqueness, no repeated assignment of the same batch;

[0134] 23. ε-constraint pruning:

[0135] If g2(m) + h2(m) > ε2 or g3(m) + h3(m) > ε3, discard state m;

[0136] 24. Heuristic ordering calculation:

[0137] f(m) = g1(m) + h(m), add m to OpenList.

[0138] 3. Termination condition:

[0139] Find a complete path that meets the main objective of minimizing (under the given ε condition) and all feasibility constraints; or OpenList is empty (no solution).

[0140] First, based on the product information provided by the digital label module and real-time logistics data, a graph network with three weights of cost, delay, and risk is constructed, and the A* heuristic search is strengthened by ε-constraint to minimize transportation cost as the main objective, while ensuring that the delivery delay and risk do not exceed the threshold;

[0141] During transportation, the abnormal early warning model scores multiple source information such as temperature, humidity, traffic, equipment, etc. in real time. Once the risk score exceeds the set threshold, the affected subgraph is identified and local or global re-optimization is triggered. A new feasible scheme is generated by re-running the A*+epsilon-constrained process, and the scheduling execution is issued. In the "perception-warning-decision-execution-re-perception" closed loop, the dynamic distribution of agricultural products is realized in a safe, efficient and low-cost manner.

[0142] In some embodiments, the working process of the above abnormal early warning model is as follows:

[0143] Input features:

[0144] Environmental monitoring, including vehicle / in-warehouse temperature, humidity, vibration, etc. sensor data; transportation status, including GPS position, speed, stay time; external information, including weather forecast, road congestion, traffic accident, holiday traffic flow; equipment health, including cold chain equipment voltage, current, compressor operating status;

[0145] The time series model trained based on historical transportation and risk event data outputs a risk score Rt, reflecting the cumulative or predicted potential risk intensity on the current transportation path.

[0146] The above input features are pushed to the stream processing platform (Kafka+Flink) at a fixed frequency (such as every 5 minutes), and simple cleaning and normalization are performed in the feature aggregation layer to form the model input vector Xt.

[0147] Online call risk early warning model, generate current risk score: Rt=m(Xt), if using multi-source model, the outputs of each sub-model can be fused by weighting or stacking to get the final Rt;

[0148] Set configuration parameters, risk upper limit Θm, when Rt>Θ, trigger "risk event" and record risk type (such as "temperature exceeds", "delay warning", "high congestion") for subsequent targeted adjustment.

[0149] Impact range identification:

[0150] Local path section: according to the risk GPS area, mark the affected node set and edge set in the graph network; batch status: identify the affected vehicle and goods batch, and the subsequent possible affected distribution center;

[0151] Re-optimization strategy selection:

[0152] Strategy one, local re-planning, in the affected subgraph, call local A* to search alternative path; keep the unaffected section in the original scheme unchanged, only replace the high-risk section;

[0153] Strategy two, global re-optimization, treats the unexecuted paths of all vehicles as new starting points, reloads the ε-constrained multi-objective model; re-runs the A* + ε-constrained process to generate a globally optimal or near-optimal solution.

[0154] The following checks are performed on the candidate new solution to ensure that it is still within the feasible domain:

[0155] Capacity and loading: does not exceed the remaining load capacity of the vehicle;

[0156] Flow balance: does not generate disconnected links or sub-loops;

[0157] Time window: the arrival time of the new path meets the time window of each node;

[0158] Task uniqueness: does not duplicate the assignment of the same batch.

[0159] The adjusted new path is issued through the scheduling system (Web console / driver APP); the monitoring panel is updated synchronously, and a new round of real-time monitoring and early warning begins; the trigger cause and adjustment effect of this event are recorded for subsequent model and strategy optimization.

[0160] The process of generating an A / B test scheme according to the optimal distribution scheme includes:

[0161] Obtain relevant data of the optimal distribution scheme and perform feature processing;

[0162] Train a binary classification model based on the processed data, divide the data into multiple batches using time slicing, and output the complaint probability score of each batch;

[0163] Set up an A / B test scheme according to the original quality inspection process and the newly added manual review process, and randomly assign high-risk batches with higher scores to two groups.

[0164] In some embodiments, the specific execution process of generating an A / B test scheme according to the optimal distribution scheme is as follows:

[0165] I. Data collection and preprocessing:

[0166] Aggregate multi-source data, including on-chain data, off-chain data, and external data;

[0167] On-chain data includes temperature, humidity, vibration, GPS location, transportation time, sorting level, summary hash and test results of quality inspection reports (pesticide residues, heavy metals, etc.), and detection results collected by sensors in real time; off-chain data includes weather data (temperature, precipitation, light, etc.), historical orders, promotion activity information, holiday markers, consumer ratings and text comments, customer service tickets, etc.; external data includes traffic conditions, fuel price index, supplier delivery records, and other auxiliary information that helps to predict logistics time and cost.

[0168] Interpolate or remove sensor data interruptions, GPS loss, or extreme outliers; filter out obvious fake orders and abnormal reviews in consumer ratings.

[0169] Align the "sorting completion time" on the chain with the "order receipt time" off the chain to calculate the transportation time; associate the quality inspection report results (qualified / unqualified, various indicator values) with the sorting temperature, transportation time, and meteorological conditions of the production area of ​​the corresponding batch.

[0170] Feature construction example:

[0171] Temperature control stability characteristics: such as “temperature fluctuation range during transportation” (max–min) / “standard deviation”;

[0172] Sorting quality characteristics: the average grade of goods shipped from the production area on the same day (such as the proportion of Grade A);

[0173] Timeliness characteristics: average time from sorting completion to receipt, number of delays, and delay percentage;

[0174] Historical quality feedback characteristics: customer complaint rate and return rate of the same production area in the past X weeks;

[0175] Meteorological derivative characteristics: the difference between the average temperature on the day of picking and the previous week, accumulated precipitation, etc.

[0176] Store the processed wide table (WideTable) with feature labels in a data warehouse (such as ClickHouse, Hive, ElasticSearch, etc.), and partition it according to "production area + batch number + date" to accelerate subsequent model training and report generation.

[0177] 2. Machine Learning Model Training and Prediction

[0178] Use the probability of customer complaints for prediction, that is, predict the probability that a certain batch will trigger a complaint (such as failed quality inspection, poor taste, etc.) after it finally reaches the hands of consumers.

[0179] The batches from the past three months are used as training samples: each sample contains the feature vector constructed by the above "feature engineering" and corresponding labels such as "true loss rate", "true complaint result (0 / 1)", and "actual time deviation".

[0180] Split the data by time to construct a dataset. For example, use the first two months of samples as the training set and the last month as the validation set to avoid information leakage. K-fold time series cross-validation can also be performed.

[0181] Use grid search and adjust parameters based on the MAPE (regression) and AUC / ROC (classification) indicators of the validation set.

[0182] Before the "Monthly / Quarterly AI Report" generation cycle, the background uniformly pulls the latest features of all batches of the last month, inputs the trained model, and calculates the following prediction results:

[0183] "Estimated loss rate" of each batch;

[0184] "Complaint probability" or "estimated complaint score" of each batch;

[0185] "Transportation delay risk score" of each batch.

[0186] Real-time / near real-time prediction (optional)

[0187] For new batches generated at the sorting or transportation node on the same day, real-time prediction can be triggered to immediately provide a "high-risk batch" list for the dispatch center or quality inspection personnel to quickly handle.

[0188] Three, from prediction to optimization suggestion A / B test:

[0189] Optimization strategies can be any of the following three strategies:

[0190] 1. Loss rate:

[0191] Identify high-risk production areas / batches: take all batches ranked in the top 5% of loss rate prediction values from the prediction results as the "experimental candidate set"; analyze feature drivers: if the loss risk of a batch is mainly affected by "transportation temperature fluctuations", then suggest the following in A / B testing for that batch:

[0192] Group A: maintain the original packaging + original temperature control scheme;

[0193] Group B: add thermal and shock-absorbing materials to the packaging, and increase temperature monitoring and compensation refrigeration equipment on the transportation vehicle.

[0194] The test indicator is to compare the actual "breakage rate / weight loss rate" of Group A vs. Group B.

[0195] 2. Complaint probability:

[0196] Identify high-risk batches: extract samples from the top 5% of complaint probability prediction values, further subdivide high-risk causes (such as "high quality inspection indicators" and "abnormal production area temperature"); if the high complaint risk is due to "quality inspection indicators close to the threshold but not rejected", then in A / B testing:

[0197] Group A: ship according to the original quality inspection pass standard;

[0198] Group B: add one manual re-inspection or additional herbicide residue re-measurement to samples close to the threshold, and discard or downgrade if they are high.

[0199] Test indicators: Comparison of the “actual complaint rate” and “return rate” of the two groups.

[0200] 3. Time deviation:

[0201] Identify batches with high delay risk: Take the batch with the highest predicted value for "transportation delay risk" and determine whether the main influencing factor is "logistics channel congestion" or "high number of cold chain transfers." Then, in the A / B test, we recommend:

[0202] Group A: Take the original direct logistics route;

[0203] Group B: Optimize the plan to "transfer once but directly connect with cold chain", or choose an alternative route and depart in advance to avoid peak hours.

[0204] Test indicators: "actual transportation time" and "timeliness achievement rate" of the two groups.

[0205] Based on the overall historical data, it is estimated that in the expected risk batches, if the "optimized loss rate" can be reduced from 8% to 5%, under the premise of significance level α = 0.05 and test power 0.8, about N batches will be needed as samples; if the minimum sample size is not reached, the test time will need to be appropriately extended or the production area will need to be expanded.

[0206] If the candidate batches are concentrated in the same production area or logistics channel, stratified randomization is required: first stratify by production area / channel, and then randomly sample from each stratum and assign them to groups A / B to ensure that the two groups are evenly distributed in key variables such as production area, batch size, and transportation season.

[0207] 4. Feedback loop and continuous iteration:

[0208] During the A / B test period, key indicators such as "actual loss rate", "actual complaint rate", and "actual transportation time" of each batch are entered into the data warehouse in real time;

[0209] Compare Group A and Group B using statistical methods:

[0210] For continuous indicators (such as loss rate and transportation time), use T test or U test;

[0211] For categorical / binary indicators (e.g., whether a complaint occurred or not), the chi-square test or Fisher's exact test was used.

[0212] Significance assessment: If Group B is significantly better than Group A in the predetermined key indicators (pvalue < 0.05 and the effect size is sufficient), it is recommended to fully promote the Group B solution and include the real labels such as "true loss rate", "complaint rate", and "transportation time" collected during the A / B trial into the next model training dataset to improve the model prediction accuracy.

[0213] If a certain feature is found to have a small contribution to the experimental results, it can be tried to remove the feature and test the new features (such as "package material type code" or "weather forecast fluctuation rate"), and for the B group strategy that performs well, it is standardized to form an "optimal practice SOP", and is written into the supply chain operation manual to guide the execution of all subsequent batches.

[0214] The process of implementing better optimization suggestions in the test scheme into the ongoing distribution scheme includes:

[0215] Perform A / B testing scheme after random grouping of high-risk batches, and extract key features of high-risk batches, match optimization schemes for temperature control, quality inspection, packaging, and route dimensions in the strategy library according to the key features, and generate intervention suggestions;

[0216] Replace part of the parameters in the original optimal distribution scheme according to the intervention suggestions to obtain a new optimal distribution scheme, and feed back the intervention suggestions to the strategy library to iterate the model by supplementing new labels and updating strategies.

[0217] In some embodiments, the specific implementation process of implementing the optimization suggestions into the ongoing distribution scheme is as follows:

[0218] I. Set input and trigger conditions: the multi-dimensional prediction of each batch has been completed in the process of generating the A / B test scheme, including:

[0219] Loss rate prediction value, the model outputs a continuous value representing the expected percentage loss from sorting to final delivery;

[0220] Complaint probability prediction value, the model outputs a probability between 0 and 1, indicating the likelihood of the batch triggering customer complaints eventually;

[0221] Transportation delay risk score, the model outputs a score or classification (such as "high, medium, low"), indicating the risk of delay that may occur for the batch under the current transportation scheme;

[0222] These prediction results are indexed according to "batch number + production area + date" and stored in the "prediction results table" in the data warehouse to provide original data for subsequent recommendation and testing.

[0223] Read the latest "prediction results table" regularly (such as every morning, once a week) and filter according to the current threshold:

[0224] High loss batch, when the predicted loss rate > set threshold (such as 5%), it is marked as a candidate batch;

[0225] High complaint batch, when the complaint probability > set threshold (such as 0.2), it is marked as a candidate batch;

[0226] High delay batch, marked as a candidate batch when the transportation delay risk score is determined as "high" by the model;

[0227] All three types of candidate batches are labeled with "risk tags" and written into the "candidate batch cache table". If the same batch meets multiple risk rules, the scores will be combined to determine the subsequent priority.

[0228] For important or urgent batches (such as large customer orders, holiday promotion period), the system can start the "real-time trigger" mode:

[0229] When the sorting is completed or the warehouse is out, the model interface is called immediately for prediction;

[0230] If the prediction result reaches the "high risk" threshold, an "automated recommendation" is generated immediately to notify the operation / logistics scheduling personnel to respond quickly; this mode is suitable for scenarios that require quick intervention, but attention should be paid to the fact that too frequent real-time recommendations may cause waste of manpower and false alarm risk.

[0231] II. Automatic recommendation generation:

[0232] For each data row marked as a "candidate batch", the system will call the corresponding "model explanation module" or "feature importance interface" to extract the main influencing factors behind the batch prediction result, such as:

[0233] For the loss rate prediction model, the top 3-5 features that contribute most to the prediction value of the batch (such as transportation temperature fluctuation, sorting level, manual operation speed, etc.) can be returned;

[0234] For the complaint probability prediction model, the main features (such as quality inspection index distance threshold, production area harvest temperature, packaging strength parameter, etc.) are returned.

[0235] The system has a built-in "optimization strategy library", which classifies a number of selectable strategies according to different risk types and features. Examples are as follows:

[0236] Large transportation temperature fluctuation → optimization strategy: suggest assigning the batch to a transportation vehicle that has installed real-time temperature control monitoring equipment; or add thermal insulation padding material in the packaging;

[0237] Low sorting level → optimization strategy: suggest adding secondary manual re-inspection in the sorting link; or reduce the sorting machine grabbing speed to reduce mechanical damage;

[0238] Quality inspection index threshold → optimization strategy: suggest performing "quick re-inspection" or "on-site rejection" on critical samples; use more stringent detection equipment for chemical residue samples that may exceed the standard.

[0239] High risk of congestion on the transportation route → optimization strategy: suggest enabling sub-optimal alternative routes or off-peak delivery at night; suggest switching to a more stable but slightly longer time-efficient partner carrier.

[0240] Each strategy is pre-configured with applicable scenarios (e.g., when "a certain feature" contributes ≥ X) and explicitly states "expected effect indicators" (e.g., "expected loss rate reduction ≥ 1%" or "expected complaint probability drop ≥ 0.05").

[0241] The system matches "candidate batches + key features" with the "strategy library" to generate specific recommendations. A complete "automated recommendation" includes:

[0242] Batch basic information: batch number, production area, sorting date, current estimated delivery time, etc.

[0243] Risk type and prediction value: loss prediction value / complaint probability / delay risk score

[0244] Main influencing features: for example, "transportation temperature fluctuation amplitude 4.5℃", "quality inspection chemical residue close to threshold 3%"

[0245] Optimization suggestions for this batch: such as "add insulation box during transportation and maintain temperature at 4-6℃ for this batch", "increase manual sampling by 10% in the sorting process and add 'fragile, do not press hard' label on the package"

[0246] Expected effect indicators: such as "expected loss rate from 6.8% to ≤5%", "expected complaint rate from 0.18 to ≤0.12".

[0247] Finally, all "automated recommendations" will be written into the "recommendation result table" in "batch units" and will be accompanied by "recommendation generation timestamps" for future tracking.

[0248] III. A / B test design and implementation:

[0249] Periodically (e.g., weekly), take the latest N batches tagged as "high risk" from the "recommendation result table" to form the "experimental candidate pool"; if the same batch has historical experimental records, determine whether the last experiment is over or assess whether it needs to be repeated (generally with an interval of at least one complete cycle).

[0250] According to the difference between historical baseline (historical loss / complaint average level) and expected effect, calculate the required minimum sample size M in advance using statistical methods (such as power analysis);

[0251] If the number of "experimental candidate pool" M' < M, you can: extend the experimental waiting period, wait for more "high-risk batches" to enter the candidate pool; or moderately relax the threshold (for example, increase the threshold from the top 5% to the top 8%) to expand the sample in the pool.

[0252] If the candidate batch source area, transportation channel is more concentrated, you need to use "stratified random" method: stratify according to "production area", "transportation route", "distribution carrier" and other key attributes first; Or randomly allocate to Group A (control group) or Group B (experimental group) in each layer to ensure that the two groups are evenly distributed in each layer.

[0253] If the candidate batch source distribution is relatively scattered, you can directly "randomly group" according to the 1:1 or 2:1 (A:B) ratio.

[0254] Control group (Group A): The control group continues to perform according to the original process, and is marked in the system as "only collect observation indicators, do not make additional interventions", and its data will be used as the baseline for comparison with the experimental group.

[0255] Experimental group (Group B): The batches in Group B apply the corresponding optimization strategies in "automated recommendations", such as: using the recommended new packaging materials; change to a more optimal route or a vehicle with lower temperature difference; increase the 10% inspection rate for this batch or use more sophisticated detection equipment.

[0256] The experimental group must record the "intervention strategy version number" automatically by the system before shipping / distribution, so that each batch can be clearly distinguished when it is traced back later.

[0257] During the A / B test, the following data is collected in real time and written into the "experimental results table":

[0258] Original indicators: actual "damage rate", "weight loss rate", "complaints", "time to sign (hours)", "number of delivery delays";

[0259] Supplementary indicators: packaging material cost, additional quality inspection labor cost, transportation cost changes, for subsequent cost-benefit analysis.

[0260] At the same time, the system will automatically generate a "mid-term experimental report" according to the pre-set "mid-term check point" (such as when the test reaches 50% sample size or the test has been running for 1 week):

[0261] Statistical analysis of whether the sample size of the two groups is balanced; whether there is a significant deviation in the key indicators (such as average loss rate, complaint rate, time to achieve rate); if significant covariate imbalance or abnormal samples in one group (such as sudden weather changes leading to widespread delays) are found, the "early warning mechanism" can be triggered, indicating that the test needs to be suspended or re-stratified.

[0262] When new data is sufficient, the model is fine-tuned or retrained with new samples to improve prediction accuracy. If new features are introduced (e.g., "courier driving behavior score"), feature selection and model validation are required. For batches that do not meet the expected results, the system records failure cases for subsequent model improvement.

[0263] Statistical analysis of different types of strategies in multiple batches / multiple rounds of experiments to identify the best "gold medal strategy" and the less effective "elimination strategy";

[0264] Forming "SOP templates" for outstanding strategies and adding them to the "advanced strategy set" in the strategy library for easy subsequent automated matching and priority invocation;

[0265] If new potential optimization directions are found in experiments (e.g., "the impact of using different thickness of bubble wrap on vibration loss"), they can be standardized as "candidate strategies" and verified in a small range before being included in the strategy library.

[0266] Through the closed loop of "automated recommendation → A / B testing → experimental feedback → model & strategy update", new data and new experience can be obtained at each stage to continuously improve the intelligence and refinement level of supply chain management.

[0267] The above system combines digital tags, dynamic optimization, and real-time tracking to bring the following main benefits in the entire process of agricultural product supply and marketing:

[0268] 1. Improve information transparency and traceability:

[0269] The digital tag module assigns a unique identifier to each agricultural product and records its basic information (such as variety, origin, batch, etc.) and extended information (such as production stages, quality inspection reports, etc.), ensuring that static data can be traced from the field to the terminal.

[0270] Based on this, the traceability management module can generate comparative test (A / B test) schemes according to the current distribution plan and feedback better suggestions to the execution process in a timely manner, forming a closed-loop traceability and continuous improvement.

[0271] Overall, information transparency has been greatly improved, which helps regulatory authorities, enterprises, and consumers to query the source, transportation track, and quality status of agricultural products in real time, reducing risks and improving trust.

[0272] 2. Optimize distribution plans and reduce logistics costs:

[0273] The dynamic adjustment module uses product information collected by digital tags and real-time logistics data as input, and combines multi-objective optimization algorithms (such as weighing indicators such as cost, timeliness, and loss rate) with graph network path planning technology to calculate the optimal distribution path and warehouse allocation strategy.

[0274] During transportation, the module continuously updates the logistics status (such as temperature, humidity, vehicle location, transportation time, etc.) and feeds the latest information back to the optimization model; when potential risks such as traffic congestion, sudden climate change or vehicle failure are detected, the system can quickly re-plan the route or adjust the distribution nodes to avoid unnecessary detours and delays.

[0275] Through the above mechanism, transportation costs and time losses (such as cold chain losses, detention costs, etc.) can be significantly reduced, and the overall supply chain efficiency can be improved:

[0276] The transportation routes are more reasonable, reducing empty trips and detours;

[0277] More accurate inventory allocation reduces the risk of slow sales and out-of-stock;

[0278] The cold chain environment is controllable, reducing quality losses caused by temperature fluctuations or transportation timeouts.

[0279] 3. Real-time risk warning to ensure the quality and safety of agricultural products:

[0280] During the transportation and distribution process, the dynamic adjustment module will access multi-dimensional logistics monitoring data (including sensor information such as location, temperature, humidity, vibration, etc.) and perform real-time analysis of this data based on the pre-trained anomaly warning model.

[0281] Once an abnormality in the transportation environment is detected (such as cold chain temperature exceeding the limit, excessive transportation detention time, abnormal route area, etc.), the system will immediately issue an early warning and automatically adjust the current distribution plan based on the risk threshold, such as:

[0282] When approaching a disaster point, the nearest transfer warehouse will be given priority for acceptance and redistribution;

[0283] Switch to backup vehicles or faster transportation routes in a timely manner;

[0284] In the event of an uncontrollable emergency, the manual intervention process will be automatically triggered, allowing operators to quickly develop an emergency plan.

[0285] This mechanism effectively reduces the probability of agricultural products being damaged, deteriorated or smuggled during circulation, and effectively ensures food safety and supply chain stability.

[0286] 4. Continuous feedback and iteration to improve operational level:

[0287] The traceability management module not only traces after the fact, but also continuously generates A / B test plans in actual operation: the system compares the current optimal distribution plan with alternative plans, and extracts better optimization suggestions.

[0288] When the test results show that a certain improvement path, a certain distribution strategy, or a certain inventory prediction model is better, the system will quickly issue it to the ongoing distribution process, enabling continuous iteration.

[0289] Through the "trial-test-improve" closed-loop mechanism, the platform can continuously accumulate high-quality scheduling experience and algorithm models, making the next round of decision-making more accurate and efficient, and gradually forming an intelligent supply and marketing system that learns and optimizes itself.

[0290] 5. Strong compatibility and scalability, leaving room for future upgrades:

[0291] The digital tag module uses a globally unique identification mechanism and can seamlessly integrate with technologies such as blockchain and Internet of Things (IoT), providing a technical foundation for further introduction of traceability and anti-counterfeiting, smart contracts, and other functions in the future.

[0292] The multi-objective optimization and graph network path planning framework of the dynamic adjustment module have good scalability, allowing for flexible addition or removal of optimization objectives (such as carbon emissions and vehicle scheduling) and the integration of more transportation modes (such as drones and cold-chain railway transportation).

[0293] The traceability management module is based on modular design, making it easy to integrate with third-party ERP, WMS, or TMS systems, facilitating collaboration among various agricultural, wholesale, and commercial entities, logistics companies, and participants from different industries on the same platform, reducing integration difficulty and maintenance costs.

[0294] An electronic device includes a processor and a memory communicatively connected to the processor and configured to store executable instructions for the processor, and the processor is configured to execute the above-mentioned big data analysis-based agricultural product supply and marketing management system.

[0295] A server includes at least one processor and a memory communicatively connected to the processor, the memory storing instructions executable by the at least one processor, and the instructions are executed by the processor to cause the at least one processor to execute a big data analysis-based agricultural product supply and marketing management system.

[0296] A computer-readable storage medium stores a computer program that, when executed by a processor, implements a big data analysis-based agricultural product supply and marketing management system.

[0297] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can still be modified, or some or all of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and they should be covered in the scope of the claims and the description of the present application.

[0298] The above only describes the preferred embodiments of the present application, and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. The agricultural product supply and marketing management system based on big data analysis is characterized by: include: A digital label module, which is configured to label basic information, extended information, and unique identifiers for each agricultural product; A dynamic adjustment module is configured to use multi-objective optimization and graph network path planning based on the results of the digital labeling module and logistics information to obtain the optimal distribution plan. The dynamic adjustment module updates the logistics information in real time during the transportation of agricultural products and analyzes the potential risks of the logistics information through an abnormal warning model. When the potential risk exceeds a threshold, the optimal distribution plan is adjusted; The traceability management module is configured to generate an A / B testing plan based on the optimal allocation plan, implement the better optimization suggestions in the test plan into the allocation plan being implemented, and complete continuous feedback and iteration of the allocation plan.

2. The agricultural product supply and marketing management system based on big data analysis according to claim 1 is characterized by: The basic information includes the variety, origin, and harvest time of the agricultural product; The extended information includes the grade and primary processing conditions of the agricultural product; The unique identifier consists of a self-incremental number, a batch number, and a serial number.

3. The agricultural product supply and marketing management system based on big data analysis according to claim 1 is characterized in that: The multi-objective optimization includes: Based on the three goals of minimizing transportation costs, shortening delivery delays, and reducing risks, an ε-constrained objective function is constructed. Based on four conditions: vehicle capacity constraint, flow balance and connectivity, time window restriction, and unique task allocation, the feasible solution space is characterized.

4. The agricultural product supply and marketing management system based on big data analysis according to claim 3 is characterized in that: The graph network path planning includes: By constructing a graph network that includes transportation costs, delays, and risks, and designing a weighted multi-objective heuristic function, we combine it with the ε-constraint method for optimization. When using a graph network for path planning, nodes are expanded through priority queues, and transportation costs, delays, and risks are calculated and updated in real time to ensure that the path meets the primary goal of minimizing transportation costs while constraining delays and risks within preset upper limits. By checking the constraints of capacity, flow balance, time window and unique task allocation, we ensure that each path conforms to the feasible solution space, thus obtaining an optimal path that optimizes cost, timeliness and risk at the same time.

5. The agricultural product supply and marketing management system based on big data analysis according to claim 1 is characterized in that: The working process of the abnormal warning model is as follows: The anomaly warning model scores multi-source information in real time. If the score exceeds the set threshold, the affected subgraph is identified and local or global re-optimization is triggered. A new feasible solution is generated by re-running the constraint process and dispatched for execution.

6. The agricultural product supply and marketing management system based on big data analysis according to claim 1 is characterized in that: The process of generating an A / B testing plan based on the optimal allocation plan includes: Obtain relevant data on the optimal allocation plan and perform feature processing; A binary classification model is used to train the feature-processed data, and the data is divided into multiple batches using time segmentation. The complaint probability score for each batch is output; An A / B testing plan was set up according to the original quality inspection process and the newly added manual re-inspection process, and high-risk batches with higher scores were randomly assigned to two groups.

7. The agricultural product supply and marketing management system based on big data analysis according to claim 6 is characterized in that: The process of implementing the better optimization suggestions from the test plan into the allocation plan being executed includes: Execute A / B testing plans after random grouping of high-risk batches, extract key features of high-risk batches, and generate intervention recommendations based on the optimization plans for temperature control, quality inspection, packaging, and routing dimensions in the key feature matching strategy library; According to the intervention suggestions, some parameters in the original optimal allocation plan are replaced to obtain a new optimal allocation plan. The intervention suggestions are fed back to the strategy library, and the model is iterated by adding new labels and updating strategies.

8. An electronic device comprising a processor and a memory in communication with the processor and configured to store instructions executable by the processor, wherein: The processor is used to execute the system described in any one of claims 1 to 7.

9. A server, characterized in that: The system comprises at least one processor and a memory communicatively connected to the processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the processor so that the at least one processor executes the system according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the system according to any one of claims 1 to 7 is implemented.

Images