Block chain-based inventory prediction and management system and method for cross-border e-commerce goods

Abstract

The invention discloses a blockchain-based inventory prediction and management system and method for cross-border e-commerce goods, and belongs to the technical field of inventory management. The method comprises the following steps: splitting alternative links according to the dimensions of a transportation mode and a transportation line, calculating the transportation time of each link, and constructing a transportation time set; based on the transportation time set, constructing an input feature matrix, training a convolutional neural network, generating an adjusted transportation time set, and screening an optimal link; generating a transportation list based on the optimal link; when the transportation list is confirmed to be executed, deducting the stock quantity of the SKU corresponding to the departure place warehouse; when the goods arrive at the destination area warehouse, updating the stock quantity of the SKU corresponding to the destination warehouse; obtaining the maximum daily purchase amount of the manufacturer, calculating the regional inventory demand priority, and formulating a purchase amount distribution rule; matching the optimal link with a purchase quantity distribution rule, adjusting the optimal link, and calculating an inventory turnover rate; and adjusting the daily purchase amount of the manufacturer based on the inventory turnover rate.

Description

Technical Field

[0001] This invention relates to the field of inventory management technology, specifically to a blockchain-based inventory forecasting and management system and method for cross-border e-commerce goods. Background Technology

[0002] With the deepening of globalization and the innovation of e-commerce models, the scale of cross-border e-commerce continues to expand, covering multiple countries, platforms, and warehouses. Its logistics links involve various modes of transport, including sea, air, and land. Inventory management has become a core element affecting operational efficiency and customer experience. On the one hand, cross-border e-commerce needs to address the differences in consumer demand in different regions; on the other hand, it needs to handle uncertainties such as long-cycle logistics and multi-stage customs clearance, leading to frequent inventory backlogs and stockouts.

[0003] Existing technologies largely rely on fixed logistics links, failing to differentiate between alternative links based on transportation methods and routes, and neglecting to dynamically adjust transportation times based on real-time weather forecasts and policy changes. This makes it impossible to ensure the shortest possible transportation time, leading to delayed inventory replenishment and exacerbating the risk of stockouts. Existing technologies, after obtaining the manufacturer's maximum daily purchase volume, often determine the purchase volume for each region based on averages or experience, without calculating regional inventory demand priorities. This results in frequent stockouts in high-priority regions and inventory buildup in low-priority regions. Furthermore, existing technologies do not integrate logistics link efficiency with purchase allocation results to calculate inventory turnover. Adjustments to manufacturer purchase volumes rely heavily on historical sales data or subjective judgment, failing to consider the impact of current inventory levels and logistics timeliness on turnover rates, leading to a disconnect between purchase volumes and actual turnover needs. Summary of the Invention

[0004] The purpose of this invention is to provide a blockchain-based inventory forecasting and management system and method for cross-border e-commerce goods, in order to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: Firstly, this application provides a blockchain-based method for inventory forecasting and management of cross-border e-commerce goods, including the following steps: Collect transportation mode data and transportation route data, and perform preprocessing; split the candidate links according to the dimensions of transportation mode and transportation route, calculate the transportation time of each link, and construct a transportation time set; Based on the set of transportation times, an input feature matrix is ​​constructed, a convolutional neural network is trained, an adjusted set of transportation times is generated, and the optimal link is selected; a transportation manifest is generated based on the optimal link. When the shipping manifest is confirmed for execution, a blockchain smart contract is triggered, automatically deducting the inventory quantity of the corresponding SKU from the origin warehouse; when the goods arrive at the destination warehouse, the arrival confirmation information is uploaded to the blockchain, triggering a smart contract to update the inventory quantity of the corresponding SKU at the destination warehouse. Obtain the manufacturer's maximum daily purchase volume, calculate the regional inventory demand priority, and formulate purchase volume allocation rules; match the optimal supply chain with the purchase volume allocation rules, adjust the optimal supply chain, and calculate the inventory turnover rate; adjust the manufacturer's daily purchase volume based on the inventory turnover rate.

[0006] In conjunction with the first aspect, in the first embodiment of the first aspect of this application, the step of splitting candidate links according to the dimensions of transportation mode and transportation route, calculating the transportation time of each link, and constructing a transportation time set includes: For each SKU, enumerate all feasible combinations of transportation methods and routes to form alternative links, assign a unique identifier to each alternative link, and form a set of alternative links; For each link in the candidate link set, the transportation time includes sea freight link time, air freight link time and land freight link time, which are all obtained by adding the basic transportation time and node operation time respectively; associate the unique identifier of each link with the corresponding transportation time to form a key-value pair; integrate all key-value pairs into a transportation time set.

[0007] In conjunction with the first aspect, in the second embodiment of the first aspect of this application, the step of constructing an input feature matrix based on the transportation time set, training a convolutional neural network, generating an adjusted transportation time set, and selecting the optimal link includes: The input features include basic transportation time features, weather impact features, and policy impact features. The transportation time of each link in the transportation time set is extracted as the transportation time feature. For each link, the probability of extreme weather occurring in the region it passes through within the next week and the expected duration of the impact are calculated, and multiplied to obtain the weather impact feature. For each link, the existence of any effective transportation-related policies in the countries or regions it passes through is marked (1 for effective, 0 for ineffective), and multiplied by the corresponding policy's expected duration of impact to obtain the policy impact feature. Corresponding feature values ​​are matched for each link, and the correspondence between all links and features is arranged in rows, with each row corresponding to one link and each column corresponding to one feature dimension, forming a structured matrix. All feature values ​​are mapped proportionally to the same interval. In the convolutional neural network model, historical feature matrices and historical actual transport times are collected for each link. The historical feature matrices are used as input samples to the model, and the historical actual transport times are used as sample labels, divided into training and validation sets proportionally. The input layer takes a structured matrix as input. The convolutional layer extracts the correlation between features through multiple sliding windows, generating multiple sets of feature combinations. The pooling layer retains key features from the output of the convolutional layer. The fully connected layer converts the pooled features into a single output value, which serves as the predicted transport time for the link. The model is trained and optimized with the deviation between the predicted transport time and the historical actual time as the optimization objective. The current feature matrix is ​​input, and the model calculates the estimated total transport time for each link under the current conditions, generating an adjusted transport time set. This set is then sorted by transport time to select the optimal link.

[0008] In conjunction with the first aspect, in the third embodiment of the first aspect of this application, the step of triggering a blockchain smart contract when the transport manifest is confirmed to be executed, and automatically deducting the inventory quantity of the corresponding SKU in the origin warehouse, includes: A consortium blockchain network is established, with participating nodes including warehouses, logistics providers, and platform providers. Pre-written smart contracts for inventory management are deployed. These contracts incorporate inventory deduction logic, triggering a deduction operation when the transport manifest is confirmed for execution, the initiator has the necessary permissions, and sufficient inventory is available. Initial inventory data for all SKUs at the origin warehouse is uploaded to the blockchain via warehouse nodes. The smart contract records this inventory data and associates it with the warehouse node's public key. A blockchain-recognizable structured transport manifest format is defined, including a unique manifest number, origin warehouse identifier, target SKU code, deducted inventory quantity, executor's identity, and confirmation timestamp. All fields must be encrypted and signed by both warehouse and logistics provider nodes. After the logistics provider completes loading the goods, it submits a loading confirmation voucher to the warehouse. The warehouse verifies the consistency between the voucher and the transport manifest. Once confirmed, authorized warehouse personnel initiate the transport manifest execution confirmation operation. The warehouse node associates the standardized transport manifest data with the hash value of the loading confirmation voucher, signs it using the warehouse node's private key, and uploads it to the blockchain network, simultaneously sending it to the logistics provider and the platform. After the blockchain network detects the confirmed transport manifest uploaded by the warehouse node, it automatically matches the trigger conditions of the inventory management smart contract and initiates the contract execution process. The contract calls the node permission list recorded in the blockchain ledger to verify whether the warehouse node initiating the manifest confirmation has the authority to change inventory. The contract compares the hash value of the transport manifest with the hash value of the manifest uploaded by the associated node to confirm that the manifest has not been tampered with and verifies whether the identity of the person performing the confirmation is on the warehouse's authorized personnel list. The contract retrieves the current inventory data of the corresponding SKU in the origin warehouse from the blockchain ledger and compares it with the deduction quantity in the manifest. If the current inventory is not less than the deduction quantity, the verification is successful; otherwise, an error message is triggered and the process is terminated. After successful verification, the smart contract executes the inventory deduction.

[0009] In conjunction with the first aspect, in the fourth embodiment of the first aspect of this application, the step of uploading arrival confirmation information to the blockchain when goods arrive at the destination warehouse, triggering a smart contract, and updating the inventory quantity of the corresponding SKU in the destination warehouse includes: Warehouse nodes generate arrival confirmation information according to a preset format. They then hash this information and sign the hash value using the destination warehouse's private key. The signed arrival confirmation information is uploaded to the blockchain consortium blockchain by the warehouse nodes and simultaneously pushed to the origin warehouse, logistics provider, and platform for network-wide verification. Once the blockchain network detects the upload of arrival confirmation information, it initiates the contract execution process when the information includes a transport manifest number, destination warehouse identifier, and private key signature, and the verification result shows consistency. Each node verifies whether the warehouse node uploading the information is an authorized node of the destination warehouse using the node permission list recorded in the blockchain ledger; it checks the completeness of the fields in the arrival confirmation information; it retrieves the historical records of the transport manifest in the blockchain to confirm that the actual quantity received is not greater than the quantity deducted by the origin warehouse; it decrypts the signature using the destination warehouse's public key and compares the decryption result with the hash value of the arrival information to confirm that the information has not been tampered with; after successful verification, the smart contract retrieves the current inventory data of the corresponding SKU from the destination warehouse in the blockchain ledger and executes the update.

[0010] In conjunction with the first aspect, in the fifth embodiment of the first aspect of this application, the step of obtaining the manufacturer's maximum daily purchase volume, calculating the regional inventory demand priority, and formulating purchase volume allocation rules includes: By connecting to the manufacturer's ERP system via API, we obtain the manufacturer's maximum daily purchase volume; we analyze customer contribution data for each region over the past month, calculate the regional share based on order volume, sort the regions by their share from highest to lowest, and initially categorize them into priority levels; for each SKU in each region, we subtract the current actual inventory quantity from the region's average daily demand for that SKU; a positive result indicates a stock shortage, while a negative result indicates excess inventory; and we adjust the priority level based on whether there is a current inventory shortage in each region. For a single SKU, first extract the total allocatable quantity of that SKU from the manufacturer's daily maximum purchase quantity table, and then allocate it in descending order of regional priority; set the upper limit for allocation per region per SKU and the upper limit for manufacturer purchase quantity to avoid over-allocation and inventory backlog.

[0011] In conjunction with the first aspect, in the sixth embodiment of the first aspect of this application, the step of matching the optimal link with the purchase quantity allocation rule, adjusting the optimal link, and calculating the inventory turnover rate includes: A directed weighted graph G=(V,E) is constructed. The node set V contains regional nodes, procurement allocation nodes, original optimal link nodes, and link nodes to be merged. The edge set E is used to establish logical relationships between nodes. Priority-related edges connect regional nodes and procurement allocation nodes, with weights equal to the proportion of the order volume of this SKU in the region to the total order volume in the region. Coverage-related edges are divided into two categories: the first category connects the original optimal link node and the regional node, with weights equal to the transportation capacity utilization rate of the original optimal link in the region; the second category connects the link nodes to be merged and the regional node, with weights having the same meaning as the first category, used to determine whether the link can cover the regional ordering demand. Merging evaluation edges connect the original optimal link node and the link nodes to be merged, with weights equal to the transportation efficiency improvement rate after the two links are merged. Procurement allocation is linked to regional priority based on priority-related edges, and the original optimal link is linked to high-priority regions based on coverage-related edges. The optimal link is adjusted based on the merged demand, and the inventory turnover rate is calculated based on the updated graph structure.

[0012] In conjunction with the first aspect, in the seventh embodiment of the first aspect of this application, the step of adjusting the optimal link based on the merged demand and calculating the inventory turnover rate based on the updated graph structure includes: Traverse all initially matched original optimal link nodes and check the sum of edge weights of their coverage-related edges with high-priority region nodes. If the sum of edge weights is not greater than 100%, retain the original optimal link as the final link; if the sum of edge weights is greater than 100%, trigger the link merging process and proceed to the next step. Traverse the link nodes to be merged in the graph and filter out links whose coverage includes unmatched high-priority regions. Combine the original optimal link nodes with the matching link nodes to be merged, calculate the total capacity after merging, specifically the sum of the maximum capacity of the original optimal link and the maximum capacity of the link to be merged, ensuring that the total capacity is not less than the total order volume of high-priority regions. Based on the edge weights of the coverage-related edges, allocate the order volume of high-priority regions proportionally to the original optimal link and the link to be merged. Mark the merged link combination as the new optimal link node and update the graph structure. Extract the current inventory quantity I, the average daily sales volume S of the target region over the past week, and the estimated arrival volume Q of the region via the new optimal link node from the regional node; extract the transportation time D of the merged link from the new optimal link node; calculate the remaining inventory before arrival. Calculate the average inventory before delivery. Calculate the remaining inventory after delivery. ,in, To calculate the average inventory level after delivery, based on the shipping time recorded after the goods arrive. Calculate total average inventory Calculate inventory turnover rate .

[0013] In conjunction with the first aspect, in the eighth embodiment of the first aspect of this application, adjusting the manufacturer's daily purchase volume based on inventory turnover includes: Thresholds are set according to SKU type and regional priority, and calibrated with reference to the industry average level. They are dynamically adjusted quarterly based on changes in demand. Inventory turnover rate is compared with the corresponding threshold to screen low-turnover combinations with turnover rates below the lower limit of the threshold and high-turnover combinations with turnover rates above the upper limit of the threshold. Purchase volume adjustment strategies are formulated for low-turnover and high-turnover scenarios respectively to adjust the daily purchase volume of manufacturers.

[0014] Secondly, this application provides a blockchain-based inventory forecasting and management system for cross-border e-commerce goods, including: The transportation time calculation module includes a data acquisition unit and a transportation time calculation unit. The data acquisition unit collects transportation mode data and transportation route data and performs preprocessing. The transportation time calculation unit splits the candidate links according to the dimensions of transportation mode and transportation route, calculates the transportation time of each link, and constructs a transportation time set. The transport manifest generation module includes an optimal link selection unit and a transport manifest generation unit. The optimal link selection unit constructs an input feature matrix based on the transport time set, trains a convolutional neural network, generates an adjusted transport time set, and selects the optimal link. The transport manifest generation unit generates a transport manifest based on the optimal link. Inventory update module: includes: origin SKU update unit and destination SKU update unit; wherein, when the shipping manifest is confirmed and executed, the origin SKU update unit triggers the blockchain smart contract to automatically deduct the inventory quantity of the corresponding SKU in the origin warehouse; when the goods arrive at the destination warehouse, the destination SKU update unit uploads the arrival confirmation information to the blockchain, triggers the smart contract, and updates the inventory quantity of the corresponding SKU in the destination warehouse. The daily purchase volume adjustment module includes: a purchase volume allocation rule formulation unit, an inventory turnover rate calculation unit, and a daily purchase volume adjustment unit. The purchase volume allocation rule formulation unit obtains the manufacturer's maximum daily purchase volume, calculates regional inventory demand priority, and formulates purchase volume allocation rules. The inventory turnover rate calculation unit matches the optimal supply chain with the purchase volume allocation rules, adjusts the optimal supply chain, and calculates the inventory turnover rate. The daily purchase volume adjustment unit adjusts the manufacturer's daily purchase volume based on the inventory turnover rate.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention first splits candidate links according to transportation methods and routes, constructs a set of transportation times, and then uses a convolutional neural network combined with weather forecasts and policy influences to dynamically adjust transportation times and select the optimal link with the shortest transportation time. In cases where the order volume of high-priority areas exceeds the capacity of a single link, it can also merge other links to improve efficiency, ensure timely inventory replenishment, and reduce the risk of stockouts from the source.

[0016] 2. This invention first calculates the regional inventory demand priority, then allocates the manufacturer's maximum daily purchase volume according to the priority, and matches the allocation result with the optimal supply chain; ensuring that high-priority regions receive sufficient inventory first, and low-priority regions are allocated according to demand, avoiding resource waste and improving customer satisfaction.

[0017] 3. This invention matches the optimal supply chain with the purchase allocation rules, accurately calculates the inventory turnover rate, and then adjusts the daily purchase volume of the manufacturer based on the turnover rate. It reduces the purchase volume for low-turnover combinations and increases the purchase volume for high-turnover combinations, effectively improving inventory turnover efficiency and realizing intelligent and refined inventory management in cross-border e-commerce. Attached Figure Description

[0018] Figure 1 This is a schematic diagram illustrating the steps of the blockchain-based cross-border e-commerce goods inventory prediction and management method of the present invention. Figure 2 This is a system architecture diagram of the blockchain-based inventory prediction and management system for cross-border e-commerce goods. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example: Figures 1-2 As shown, the present invention provides a technical solution: like Figure 1 As shown, this application provides a blockchain-based method for inventory forecasting and management of cross-border e-commerce goods, including the following steps: Step S100: Collect transportation mode data and transportation route data, and perform preprocessing; split the candidate links according to the dimensions of transportation mode and transportation route, calculate the transportation time of each link, and construct a transportation time set; Specifically, for each SKU, all feasible combinations of transportation methods and routes are enumerated to form alternative links, and a unique identifier is assigned to each alternative link to form a set of alternative links; For each link in the candidate link set, the transportation time includes sea freight link time, air freight link time and land freight link time, which are all obtained by adding the basic transportation time and node operation time respectively; associate the unique identifier of each link with the corresponding transportation time to form a key-value pair; integrate all key-value pairs into a transportation time set.

[0021] In one specific embodiment, taking the popular cross-border e-commerce SKU "SKU001: 50ml Moisturizing Essence" as an example, the following operations are performed: Enumerate alternative routes and assign identifiers: Based on the attributes of this SKU "non-hazardous chemical, shelf life of 3 years", select 3 feasible "transportation mode + route" combinations and assign unique identifiers: L1 (air freight - Guangzhou Baiyun Airport → North American Los Angeles Airport - SKU001), L2 (sea freight - Guangzhou Port → European Rotterdam Port - SKU001), L3 (land freight - Guangzhou bonded warehouse → Southeast Asian Bangkok land port - SKU001), forming the alternative route set {L1, L2, L3}.

[0022] Calculate the transportation time for each link: L1 basic flight time 16 hours + 8 hours for cargo inspection and loading at the port of origin + 4 hours for transit customs clearance + 6 hours for unloading and picking up at the port of destination = 34 hours; L2 basic sailing time 336 hours (14 days) + 12 hours for loading and unloading at the port of origin + 24 hours for stay at the transit port + 12 hours for unloading and picking up at the port of destination = 408 hours; L3 basic sailing time 48 hours + 4 hours for loading at the port of origin + 6 hours for border clearance + 2 hours for unloading at the destination = 60 hours.

[0023] Construct a set of transportation times: Associate the link identifier with the total transportation time as key-value pairs and integrate them to obtain the set {(L1, 34 hours), (L2, 408 hours), (L3, 60 hours)}.

[0024] Step S200: Based on the transportation time set, construct the input feature matrix, train the convolutional neural network, generate the adjusted transportation time set, and select the optimal link; generate a transportation manifest based on the optimal link; Specifically, the input features include basic transportation time features, weather impact features, and policy impact features; the transportation time of each link in the transportation time set is extracted as the transportation time feature; for each link, the probability of extreme weather occurrence and the expected duration of impact in the areas it passes through in the next week are calculated, and multiplied to obtain the weather impact feature; for each link, it is marked whether there are any effective transportation-related policies in the countries or regions it passes through, with effective policies marked as 1 and ineffective policies marked as 0, and multiplied with the expected duration of impact of the corresponding policies to obtain the policy impact feature; corresponding feature values ​​are matched for each link, and the correspondence between all links and features is arranged in rows, with each row corresponding to one link and each column corresponding to one feature dimension, forming a structured matrix, and all feature values ​​are mapped to the same interval proportionally; In the convolutional neural network model, historical feature matrices and historical actual transport times are collected for each link. The historical feature matrices are used as input samples to the model, and the historical actual transport times are used as sample labels, divided into training and validation sets proportionally. The input layer takes a structured matrix as input. The convolutional layer extracts the correlation between features through multiple sliding windows, generating multiple sets of feature combinations. The pooling layer retains key features from the output of the convolutional layer. The fully connected layer converts the pooled features into a single output value, which serves as the predicted transport time for the link. The model is trained and optimized with the deviation between the predicted transport time and the historical actual time as the optimization objective. The current feature matrix is ​​input, and the model calculates the estimated total transport time for each link under the current conditions, generating an adjusted transport time set. This set is then sorted by transport time to select the optimal link.

[0025] In one specific embodiment, taking the three links (L1, L2, L3) of SKU001 in step S100 as the objects, the following operations are performed: Input feature construction and standardization: Extract feature values ​​for each link and calculate: L1 (Air transport Guangzhou → Los Angeles) basic transport time 34 hours, weather impact feature 0.4 (20% probability of typhoon in the South China Sea × 2 hours of impact), policy impact feature 3 (new US customs clearance regulations take effect × 3 hours of impact); L2 (Sea transport Guangzhou → Rotterdam) basic transport time 408 hours, weather impact feature 0.4 (10% probability of heavy rain in the Indian Ocean × 4 hours of impact), policy impact feature 0 (no new EU regulations); L3 (Land transport Guangzhou → Bangkok) basic transport time 60 hours, weather impact feature 0.15 (15% probability of heavy rain in the Indochina Peninsula × 1 hour of impact), policy impact feature 0 (no new Thai regulations). Standardize the feature values ​​to the [0,1] interval to form a matrix: [L1:0.00,0.86,1.00; L2:1.00,0.86,0.00; L3:0.07,0.00,0.00].

[0026] Model Training and Adjusted Time Generation: Collect historical data from the past 12 months for three links (e.g., L1 historical feature matrix and actual transportation time of 32-38 hours), divide them into training and validation sets in a 7:3 ratio, and train the CNN model until the deviation on the validation set is stable. Input the current matrix, and the model outputs adjusted times: L1≈37.4 hours, L2≈408.4 hours, L3≈60.15 hours, forming the adjusted set {(L1,37.4 hours),(L2,408.4 hours),(L3,60.15 hours)}.

[0027] Optimal link selection and manifest generation: Sorting L1 by time to find the best, generating a transport manifest: SKU001, link L1 (Guangzhou Baiyun Airport → Los Angeles Airport), quantity 500 pieces, estimated transport time 37.4 hours, departure time November 1, 2025.

[0028] Step S300: When the shipping manifest is confirmed and executed, the blockchain smart contract is triggered to automatically deduct the inventory quantity of the corresponding SKU in the origin warehouse; when the goods arrive at the destination warehouse, the arrival confirmation information is uploaded to the blockchain, triggering the smart contract to update the inventory quantity of the corresponding SKU in the destination warehouse. Specifically, a consortium blockchain network is built, with participating nodes including warehouses, logistics providers, and platform providers. Pre-written smart contracts for inventory management are deployed. These contracts contain built-in inventory deduction logic, triggering a deduction operation when the transport manifest has been confirmed for execution, the initiator has the necessary permissions, and sufficient inventory is available. Initial inventory data for all SKUs at the origin warehouse is uploaded to the blockchain via warehouse nodes. The smart contract records this inventory data and associates it with the warehouse node's public key. A blockchain-recognizable structured transport manifest format is defined, including a unique manifest number, origin warehouse identifier, target SKU code, deducted inventory quantity, executor's identity, and confirmation timestamp. All fields must be encrypted and signed by both warehouse and logistics provider nodes. After the logistics provider completes loading the goods, it submits a loading confirmation voucher to the warehouse. The warehouse verifies the consistency between the voucher and the transport manifest. Once confirmed, authorized warehouse personnel initiate the transport manifest execution confirmation operation. The warehouse node associates the standardized transport manifest data with the hash value of the loading confirmation voucher, signs it using the warehouse node's private key, and uploads it to the blockchain network, simultaneously sending it to the logistics provider and the platform. After the blockchain network detects the confirmed transport manifest uploaded by the warehouse node, it automatically matches the trigger conditions of the inventory management smart contract and initiates the contract execution process. The contract calls the node permission list recorded in the blockchain ledger to verify whether the warehouse node initiating the manifest confirmation has the authority to change inventory. The contract compares the hash value of the transport manifest with the hash value of the manifest uploaded by the associated node to confirm that the manifest has not been tampered with and verifies whether the identity of the person performing the confirmation is on the warehouse's authorized personnel list. The contract retrieves the current inventory data of the corresponding SKU in the origin warehouse from the blockchain ledger and compares it with the deduction quantity in the manifest. If the current inventory is not less than the deduction quantity, the verification is successful; otherwise, an error message is triggered and the process is terminated. After successful verification, the smart contract executes the inventory deduction.

[0029] Furthermore, warehouse nodes generate arrival confirmation information according to a preset format. The warehouse nodes hash the arrival confirmation information and sign the hash value using the destination warehouse's private key. The signed arrival confirmation information is uploaded to the blockchain consortium blockchain by the warehouse nodes and simultaneously pushed to the origin warehouse, logistics provider, and platform for verification across the entire network. After the blockchain network detects the upload of arrival confirmation information, if the information includes a transport manifest number, destination warehouse identifier, and private key signature, and the verification result shows consistency, the contract execution process is initiated. Each node verifies whether the warehouse node uploading the information is an authorized node of the destination warehouse using the node permission list recorded in the blockchain ledger; verifies whether the fields of the arrival confirmation information are complete; retrieves the historical records of the transport manifest in the blockchain to confirm that the actual arrival quantity is not greater than the quantity deducted by the origin warehouse; decrypts the signature using the destination warehouse's public key and compares the decryption result with the hash value of the arrival information to confirm that the information has not been tampered with; after successful verification, the smart contract retrieves the current inventory data of the corresponding SKU from the destination warehouse in the blockchain ledger and executes the update.

[0030] In one specific embodiment, a consortium blockchain is established, with nodes including a Guangzhou Baiyun Airport warehouse (origin), a Los Angeles airport warehouse (destination), logistics provider A, and the platform provider, deploying an inventory management smart contract. The Guangzhou warehouse uploads an initial inventory of 800 units of SKU001 through the node, and the contract associates its public key record data; the standardized format of the transport manifest includes a unique number Q001, an origin identifier GC001, SKU001, a deducted quantity of 500 units, a confirmer ID (warehouse administrator Zhang), and a timestamp field, requiring signatures from both nodes.

[0031] After logistics provider A loads 500 units of SKU001, it submits a loading photo (hash value H001) to the Guangzhou warehouse. Zhang verifies that the shipment matches the manifest Q001, initiates confirmation, associates the manifest data with H001, signs it with the Guangzhou warehouse's private key, and uploads it to the consortium blockchain, synchronizing it to all nodes. The blockchain triggers a contract, verifying that Zhang is the authorized person, the manifest hash matches the logistics provider's upload, and the Guangzhou warehouse's SKU001 inventory is ≥500 (800 ≥ 500). A deduction is then executed, updating the Guangzhou warehouse inventory to 300 units. The deduction transaction is recorded on the blockchain (Q001, deduction of 500 units, time 2025-11-10 8:00).

[0032] After the goods arrived at the Los Angeles warehouse, warehouse manager Li counted 500 items and confirmed they were correct. He then generated arrival information (including Q001, destination identifier LA001, actual arrival of 500 items, and timestamp 2025-11-21 8:00), hashed it to obtain H002, and uploaded it using the Los Angeles warehouse's private key. The blockchain triggered a contract to verify that Li was an authorized node, that the fields were complete, that 500 items were less than or equal to the deduction quantity of 500, and that H002 matched the decrypted signature. The initial inventory of SKU001 in the Los Angeles warehouse was retrieved (50 items), and after the update, the inventory was 550 items. The transaction record was synchronized on the blockchain.

[0033] Step S400: Obtain the manufacturer's maximum daily purchase volume, calculate the regional inventory demand priority, and formulate purchase volume allocation rules; match the optimal supply chain with the purchase volume allocation rules, adjust the optimal supply chain, and calculate the inventory turnover rate; adjust the manufacturer's daily purchase volume based on the inventory turnover rate.

[0034] Specifically, the system connects to the manufacturer's ERP system via API to obtain the manufacturer's maximum daily purchase volume; it then analyzes customer contribution data for each region over the past month, calculates the regional share based on order volume, sorts regions by their share from highest to lowest, and initially assigns priority levels; for each SKU in each region, it subtracts the current actual inventory quantity from the region's average daily demand for that SKU; a positive result indicates a stock shortage, while a negative result indicates excess inventory; and the priority level is adjusted based on whether a region currently has an inventory shortage. For a single SKU, first extract the total allocatable quantity of that SKU from the manufacturer's daily maximum purchase quantity table, and then allocate it in descending order of regional priority; set the upper limit for allocation per region per SKU and the upper limit for manufacturer purchase quantity to avoid over-allocation and inventory backlog.

[0035] Furthermore, a directed weighted graph G=(V,E) is constructed. The node set V includes regional nodes, procurement allocation nodes, original optimal link nodes, and link nodes to be merged. The edge set E is used to establish logical connections between nodes. Priority-related edges connect regional nodes and procurement allocation nodes, with weights equal to the proportion of the order volume of this SKU in the region to the total order volume in the region. Coverage-related edges are divided into two categories: the first category connects the original optimal link node and regional nodes, with weights equal to the transportation capacity utilization rate of the original optimal link in the region; the second category connects the link nodes to be merged and regional nodes, with weights having the same meaning as the first category, used to determine whether the link can cover the regional ordering demand. Merging evaluation edges connect the original optimal link node and the link nodes to be merged, with weights equal to the transportation efficiency improvement rate after the two links are merged. Procurement allocation is associated with regional priority based on priority-related edges, and the original optimal link is associated with high-priority regions based on coverage-related edges. The optimal link is adjusted based on the merged demand, and the inventory turnover rate is calculated based on the updated graph structure.

[0036] Furthermore, iterate through all initially matched original optimal link nodes and check the sum of edge weights of their coverage-related edges with high-priority region nodes. When the sum of edge weights is not greater than 100%, retain the original optimal link as the final link; when the sum of edge weights is greater than 100%, trigger the link merging process and proceed to the next step. Traverse the link nodes to be merged in the graph and filter out links whose coverage includes unmatched high-priority regions. Combine the original optimal link nodes with the matching link nodes to be merged, calculate the total capacity after merging, specifically the sum of the maximum capacity of the original optimal link and the maximum capacity of the link to be merged, ensuring that the total capacity is not less than the total order volume of high-priority regions. Based on the edge weights of the coverage-related edges, allocate the order volume of high-priority regions proportionally to the original optimal link and the link to be merged. Mark the merged link combination as the new optimal link node and update the graph structure. Extract the current inventory quantity I, the average daily sales volume S of the target region over the past week, and the estimated arrival volume Q of the region via the new optimal link node from the regional node; extract the transportation time D of the merged link from the new optimal link node; calculate the remaining inventory before arrival. Calculate the average inventory before delivery. Calculate the remaining inventory after delivery. ,in, To calculate the average inventory level after delivery, based on the shipping time recorded after the goods arrive. Calculate total average inventory Calculate inventory turnover rate .

[0037] Furthermore, thresholds are set according to SKU type and regional priority, and the thresholds are calibrated with reference to the industry average level. They are dynamically adjusted every quarter according to changes in demand. The inventory turnover rate is compared with the corresponding threshold to screen low turnover combinations with a turnover rate below the lower limit of the threshold and high turnover combinations with a turnover rate above the upper limit of the threshold. Purchase volume adjustment strategies are formulated for low turnover scenarios and high turnover scenarios respectively to adjust the daily purchase volume of manufacturers.

[0038] In one specific embodiment, taking cross-border e-commerce SKU "SKU002: 200ml Shampoo" (fast-moving consumer goods, monthly turnover threshold: 15 times for high-priority regions, 12 times for medium-priority regions) and cooperating manufacturer B as examples, the specific process is as follows: By connecting to Manufacturer B's ERP system via API, we obtained the maximum daily supply of SKU002, which is 1500 units (including packaging and transportation loss allowance, the actual allocable quantity is 1500 units). The order volume distribution over the past month is as follows: North America 45% (highest order volume), Europe 30%, Southeast Asia 25%, preliminarily divided into: North America (high), Europe (medium), Southeast Asia (medium).

[0039] North America: SKU002 has an average daily demand of 200 units, and the current warehouse inventory is 80 units. The shortage is 200 - 80 = 120 units (positive number, out of stock). The priority is raised to "highest". Europe: Daily demand is 150 units, current inventory is 100 units, shortage = 150 - 100 = 50 units (positive number, out of stock), priority remains "medium"; Southeast Asia: Daily demand is 100 units, current inventory is 150 units, gap = 100 - 150 = -50 units (negative, surplus), priority downgraded to "low".

[0040] Extract the total allocable quantity of BSKU002 from the manufacturer, 1500 units. Allocate according to priority: North America (highest): prioritize filling the gap with 120 units (allocated); Europe (medium): fill the gap with 50 units (allocated); Southeast Asia (lowest): set a "3-day maximum demand" limit (100 units / day × 3 days = 300 units), but due to current excess inventory, only 300 units will be allocated (to avoid stockpiling); Remaining quantity: 1500 - 120 - 50 - 300 = 1030 units, temporarily stored as "contingency inventory" (to cope with sudden demand).

[0041] Define the nodes and edges in the graph structure: Region nodes (V): V1 (North America), V2 (Europe), V3 (Southeast Asia), with attributes including priority and shortage / surplus; Inventory allocation nodes (A): A1 (North America 120 pieces), A2 (Europe 50 pieces), A3 (Southeast Asia 300 pieces); Original optimal link nodes (L): L5 (Air freight Guangzhou → North America Los Angeles, maximum capacity 300 pieces, transportation time 2 days), L6 (Air freight Guangzhou → Europe Paris, maximum capacity 500 pieces, transportation time 3 days) Edge set (E): Priority associated edge: E1 (V1-A1) weight = 120 / (120+50+300) = 26% (the proportion of North American allocation to the total allocation), E2 (V2-A2) ≈ 11%, E3 (V3-A3) ≈ 63%; Coverage associated edge: E4 (L5-V1) weight = 120 / 300 = 40% (L5's utilization rate of North American capacity), E5 (L6-V2) weight = 50 / 500 = 10% (L6's utilization rate of European capacity).

[0042] Check the sum of the coverage associated edge weights of the original optimal link and the high priority region: E4 (40%) + E5 (10%) = 50% ≤ 100% (total capacity can cover the demand). There is no need to merge the links. Keep L5 and L6 as the final optimal links and update the "new optimal link nodes" in the graph structure to L5 (North America) and L6 (Europe).

[0043] Taking North America (V1) as an example: Data extracted: Current inventory I=80 units, average daily sales S=200 units, estimated arrival quantity Q=120 units (A1 allocation quantity), link transportation time D=2 days (L5 transportation time), post-arrival statistical period Dpost=7 days.

[0044] Calculation by stage: Remaining inventory before delivery: max(80-200×2,0)=max(80-400,0)=0 units; Average inventory before arrival: (80+0) / 2=40 pieces; Remaining inventory after delivery: max(0+120-200×7,0)=max(120-1400,0)=0 pieces; Average inventory after arrival: (0+120+0) / 2=60 pieces; Total average inventory: (40×2+60×7) / (2+7)=(80+420) / 9≈55.6 pieces; Inventory turnover rate: (200×(2+7)) / 55.6≈1800 / 55.6≈32.4 times.

[0045] North America is classified as "highest priority + FMCG," with a threshold of 15 times. The actual turnover rate is 32.4 times, exceeding 15 times, classifying it as a "high-turnover combination" (requiring increased order volume). Based on a 20% increase in current demand, the new order volume = 1500 × (1 + 20%) = 1800 units (not exceeding Manufacturer B's maximum capacity of 2000 units), prioritizing North America. Europe's calculated turnover rate is 18 times (above the mid-priority threshold of 12 times), maintaining the original allocation. Southeast Asia's turnover rate is 8 times (below the low-priority threshold of 10 times), maintaining the 300-unit limit. The 1800 new units will be uploaded to the blockchain "Manufacturer Order Volume Table" and synchronized to the inventory allocation system and warehouse WMS to ensure subsequent allocations are based on the new quantity.

[0046] like Figure 2 As shown, this application provides a blockchain-based inventory forecasting and management system for cross-border e-commerce goods, including: The transportation time calculation module includes a data acquisition unit and a transportation time calculation unit. The data acquisition unit collects transportation mode data and transportation route data and performs preprocessing. The transportation time calculation unit splits the candidate links according to the dimensions of transportation mode and transportation route, calculates the transportation time of each link, and constructs a transportation time set. The transport manifest generation module includes an optimal link selection unit and a transport manifest generation unit. The optimal link selection unit constructs an input feature matrix based on the transport time set, trains a convolutional neural network, generates an adjusted transport time set, and selects the optimal link. The transport manifest generation unit generates a transport manifest based on the optimal link. Inventory update module: includes: origin SKU update unit and destination SKU update unit; wherein, when the shipping manifest is confirmed and executed, the origin SKU update unit triggers the blockchain smart contract to automatically deduct the inventory quantity of the corresponding SKU in the origin warehouse; when the goods arrive at the destination warehouse, the destination SKU update unit uploads the arrival confirmation information to the blockchain, triggers the smart contract, and updates the inventory quantity of the corresponding SKU in the destination warehouse. The daily purchase volume adjustment module includes: a purchase volume allocation rule formulation unit, an inventory turnover rate calculation unit, and a daily purchase volume adjustment unit. The purchase volume allocation rule formulation unit obtains the manufacturer's maximum daily purchase volume, calculates regional inventory demand priority, and formulates purchase volume allocation rules. The inventory turnover rate calculation unit matches the optimal supply chain with the purchase volume allocation rules, adjusts the optimal supply chain, and calculates the inventory turnover rate. The daily purchase volume adjustment unit adjusts the manufacturer's daily purchase volume based on the inventory turnover rate.

[0047] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A blockchain-based method for inventory forecasting and management of cross-border e-commerce goods, characterized in that, Includes the following steps: Collect transportation mode data and transportation route data, and perform preprocessing; The candidate links are split according to the dimensions of transportation mode and transportation route, the transportation time of each link is calculated, and a set of transportation times is constructed. Based on the set of transportation times, an input feature matrix is ​​constructed, a convolutional neural network is trained, an adjusted set of transportation times is generated, and the optimal link is selected; a transportation manifest is generated based on the optimal link. When the shipping manifest is confirmed for execution, the blockchain smart contract is triggered, automatically deducting the inventory quantity of the corresponding SKU from the warehouse at the point of origin; When the goods arrive at the destination warehouse, the arrival confirmation information is uploaded to the blockchain, triggering a smart contract to update the inventory quantity of the corresponding SKU in the destination warehouse; Obtain the manufacturer's maximum daily purchase volume, calculate the priority of regional inventory demand, and formulate purchase volume allocation rules; match the optimal supply chain with the purchase volume allocation rules, adjust the optimal supply chain, and calculate the inventory turnover rate; Adjust the manufacturer's daily purchase volume based on inventory turnover rate.

2. The method for inventory forecasting and management of cross-border e-commerce goods based on blockchain according to claim 1, characterized in that, The process of splitting candidate links according to transportation mode and route, calculating the transportation time for each link, and constructing a transportation time set includes: For each SKU, enumerate all feasible combinations of transportation methods and routes to form alternative links, assign a unique identifier to each alternative link, and form a set of alternative links; For each link in the candidate link set, the transportation time includes sea freight link time, air freight link time and land freight link time, which are all obtained by adding the basic transportation time and node operation time respectively; associate the unique identifier of each link with the corresponding transportation time to form a key-value pair; integrate all key-value pairs into a transportation time set.

3. The method for inventory forecasting and management of cross-border e-commerce goods based on blockchain according to claim 1, characterized in that, The process of constructing an input feature matrix based on a set of transportation times, training a convolutional neural network, generating an adjusted set of transportation times, and selecting the optimal link includes: The input features include basic transportation time features, weather impact features, and policy impact features. The transportation time of each link in the transportation time set is extracted as the transportation time feature. For each link, the probability of extreme weather occurring in the region it passes through within the next week and the expected duration of the impact are calculated, and multiplied to obtain the weather impact feature. For each link, the existence of any effective transportation-related policies in the countries or regions it passes through is marked (1 for effective, 0 for ineffective), and multiplied by the corresponding policy's expected duration of impact to obtain the policy impact feature. Corresponding feature values ​​are matched for each link, and the correspondence between all links and features is arranged in rows, with each row corresponding to one link and each column corresponding to one feature dimension, forming a structured matrix. All feature values ​​are mapped proportionally to the same interval. In the convolutional neural network model, historical feature matrices and historical actual transport times are collected for each link. The historical feature matrices are used as input samples to the model, and the historical actual transport times are used as sample labels, divided into training and validation sets proportionally. The input layer takes a structured matrix as input. The convolutional layer extracts the correlation between features through multiple sliding windows, generating multiple sets of feature combinations. The pooling layer retains key features from the output of the convolutional layer. The fully connected layer converts the pooled features into a single output value, which serves as the predicted transport time for the link. The model is trained and optimized with the deviation between the predicted transport time and the historical actual time as the optimization objective. The current feature matrix is ​​input, and the model calculates the estimated total transport time for each link under the current conditions, generating an adjusted transport time set. This set is then sorted by transport time to select the optimal link.

4. The method for inventory forecasting and management of cross-border e-commerce goods based on blockchain according to claim 1, characterized in that, When the shipping manifest is confirmed for execution, the blockchain smart contract is triggered to automatically deduct the inventory quantity of the corresponding SKU from the origin warehouse, including: A consortium blockchain network is established, with participating nodes including warehouses, logistics providers, and platform providers. Pre-written smart contracts for inventory management are deployed. These contracts incorporate inventory deduction logic, triggering a deduction operation when the transport manifest is confirmed for execution, the initiator has the necessary permissions, and sufficient inventory is available. Initial inventory data for all SKUs at the origin warehouse is uploaded to the blockchain via warehouse nodes. The smart contract records this inventory data and associates it with the warehouse node's public key. A blockchain-recognizable structured transport manifest format is defined, including a unique manifest number, origin warehouse identifier, target SKU code, deducted inventory quantity, executor's identity, and confirmation timestamp. All fields must be encrypted and signed by both warehouse and logistics provider nodes. After the logistics provider completes loading the goods, it submits a loading confirmation voucher to the warehouse. The warehouse verifies the consistency between the voucher and the transport manifest. Once confirmed, authorized warehouse personnel initiate the transport manifest execution confirmation operation. The warehouse node associates the standardized transport manifest data with the hash value of the loading confirmation voucher, signs it using the warehouse node's private key, and uploads it to the blockchain network, simultaneously sending it to the logistics provider and the platform. After the blockchain network detects the confirmed transport manifest uploaded by the warehouse node, it automatically matches the trigger conditions of the inventory management smart contract and initiates the contract execution process. The contract calls the node permission list recorded in the blockchain ledger to verify whether the warehouse node initiating the manifest confirmation has the authority to change inventory. The contract compares the hash value of the transport manifest with the hash value of the manifest uploaded by the associated node to confirm that the manifest has not been tampered with and verifies whether the identity of the person performing the confirmation is on the warehouse's authorized personnel list. The contract retrieves the current inventory data of the corresponding SKU in the origin warehouse from the blockchain ledger and compares it with the deduction quantity in the manifest. If the current inventory is not less than the deduction quantity, the verification is successful; otherwise, an error message is triggered and the process is terminated. After successful verification, the smart contract executes the inventory deduction.

5. The method for inventory forecasting and management of cross-border e-commerce goods based on blockchain according to claim 1, characterized in that, When goods arrive at the destination warehouse, the arrival confirmation information is uploaded to the blockchain, triggering a smart contract to update the inventory quantity of the corresponding SKU in the destination warehouse, including: Warehouse nodes generate arrival confirmation information according to a preset format. They then hash this information and sign the hash value using the destination warehouse's private key. The signed arrival confirmation information is uploaded to the blockchain consortium blockchain by the warehouse nodes and simultaneously pushed to the origin warehouse, logistics provider, and platform for network-wide verification. Once the blockchain network detects the upload of arrival confirmation information, it initiates the contract execution process when the information includes a transport manifest number, destination warehouse identifier, and private key signature, and the verification result shows consistency. Each node verifies whether the warehouse node uploading the information is an authorized node of the destination warehouse using the node permission list recorded in the blockchain ledger; it checks the completeness of the fields in the arrival confirmation information; it retrieves the historical records of the transport manifest in the blockchain to confirm that the actual quantity received is not greater than the quantity deducted by the origin warehouse; it decrypts the signature using the destination warehouse's public key and compares the decryption result with the hash value of the arrival information to confirm that the information has not been tampered with; after successful verification, the smart contract retrieves the current inventory data of the corresponding SKU from the destination warehouse in the blockchain ledger and executes the update.

6. The method for inventory forecasting and management of cross-border e-commerce goods based on blockchain according to claim 1, characterized in that, The process of obtaining the manufacturer's maximum daily purchase volume, calculating regional inventory demand priority, and formulating purchase volume allocation rules includes: By connecting to the manufacturer's ERP system via API, we obtain the manufacturer's maximum daily purchase volume; we analyze customer contribution data for each region over the past month, calculate the regional share based on order volume, sort the regions by their share from highest to lowest, and initially categorize them into priority levels; for each SKU in each region, we subtract the current actual inventory quantity from the region's average daily demand for that SKU; a positive result indicates a stock shortage, while a negative result indicates excess inventory; and we adjust the priority level based on whether there is a current inventory shortage in each region. For a single SKU, first extract the total allocatable quantity of that SKU from the manufacturer's daily maximum purchase quantity table, and then allocate it in descending order of regional priority; set the upper limit for allocation per region per SKU and the upper limit for manufacturer purchase quantity to avoid over-allocation and inventory backlog.

7. The method for inventory forecasting and management of cross-border e-commerce goods based on blockchain according to claim 1, characterized in that, The process of matching the optimal supply chain with the purchase volume allocation rules, adjusting the optimal supply chain, and calculating the inventory turnover rate includes: A directed weighted graph G=(V,E) is constructed. The node set V contains regional nodes, procurement allocation nodes, original optimal link nodes, and link nodes to be merged. The edge set E is used to establish logical relationships between nodes. Priority-related edges connect regional nodes and procurement allocation nodes, with weights equal to the proportion of the order volume of this SKU in the region to the total order volume in the region. Coverage-related edges are divided into two categories: the first category connects the original optimal link node and the regional node, with weights equal to the transportation capacity utilization rate of the original optimal link in the region; the second category connects the link nodes to be merged and the regional node, with weights having the same meaning as the first category, used to determine whether the link can cover the regional ordering demand. Merging evaluation edges connect the original optimal link node and the link nodes to be merged, with weights equal to the transportation efficiency improvement rate after the two links are merged. Procurement allocation is linked to regional priority based on priority-related edges, and the original optimal link is linked to high-priority regions based on coverage-related edges. The optimal link is adjusted based on the merged demand, and the inventory turnover rate is calculated based on the updated graph structure.

8. The method for inventory forecasting and management of cross-border e-commerce goods based on blockchain according to claim 7, characterized in that, The process of adjusting the optimal link based on merged demand and calculating inventory turnover based on the updated graph structure includes: Traverse all initially matched original optimal link nodes and check the sum of edge weights of their coverage-related edges with high-priority region nodes. If the sum of edge weights is not greater than 100%, retain the original optimal link as the final link; if the sum of edge weights is greater than 100%, trigger the link merging process and proceed to the next step. Traverse the link nodes to be merged in the graph and filter out links whose coverage includes unmatched high-priority regions. Combine the original optimal link nodes with the matching link nodes to be merged, calculate the total capacity after merging, specifically the sum of the maximum capacity of the original optimal link and the maximum capacity of the link to be merged, ensuring that the total capacity is not less than the total order volume of high-priority regions. Based on the edge weights of the coverage-related edges, allocate the order volume of high-priority regions proportionally to the original optimal link and the link to be merged. Mark the merged link combination as the new optimal link node and update the graph structure. Extract the current inventory quantity I, the average daily sales volume S of the target region over the past week, and the estimated arrival volume Q of the region via the new optimal link node from the regional node; extract the transportation time D of the merged link from the new optimal link node; calculate the remaining inventory before arrival. Calculate the average inventory before delivery. Calculate the remaining inventory after delivery. ,in, To calculate the average inventory level after delivery, based on the shipping time recorded after the goods arrive. Calculate total average inventory Calculate inventory turnover rate .

9. The method for inventory forecasting and management of cross-border e-commerce goods based on blockchain according to claim 1, characterized in that, The adjustment of the manufacturer's daily purchase volume based on inventory turnover rate includes: Thresholds are set according to SKU type and regional priority, and calibrated with reference to the industry average level. They are dynamically adjusted quarterly based on changes in demand. Inventory turnover rate is compared with the corresponding threshold to screen low-turnover combinations with turnover rates below the lower limit of the threshold and high-turnover combinations with turnover rates above the upper limit of the threshold. Purchase volume adjustment strategies are formulated for low-turnover and high-turnover scenarios respectively to adjust the daily purchase volume of manufacturers.

10. A blockchain-based inventory forecasting and management system for cross-border e-commerce goods, using the blockchain-based inventory forecasting and management method for cross-border e-commerce goods as described in any one of claims 1-9, characterized in that... include: The transportation time calculation module includes a data acquisition unit and a transportation time calculation unit. The data acquisition unit collects transportation mode data and transportation route data and performs preprocessing. The transportation time calculation unit splits the candidate links according to the dimensions of transportation mode and transportation route, calculates the transportation time of each link, and constructs a transportation time set. The transport manifest generation module includes an optimal link selection unit and a transport manifest generation unit. The optimal link selection unit constructs an input feature matrix based on the transport time set, trains a convolutional neural network, generates an adjusted transport time set, and selects the optimal link. The transport manifest generation unit generates a transport manifest based on the optimal link. Inventory update module: includes: origin SKU update unit and destination SKU update unit; wherein, when the shipping manifest is confirmed and executed, the origin SKU update unit triggers the blockchain smart contract to automatically deduct the inventory quantity of the corresponding SKU in the origin warehouse; when the goods arrive at the destination warehouse, the destination SKU update unit uploads the arrival confirmation information to the blockchain, triggers the smart contract, and updates the inventory quantity of the corresponding SKU in the destination warehouse. The daily purchase volume adjustment module includes: a purchase volume allocation rule formulation unit, an inventory turnover rate calculation unit, and a daily purchase volume adjustment unit. The purchase volume allocation rule formulation unit obtains the manufacturer's maximum daily purchase volume, calculates regional inventory demand priority, and formulates purchase volume allocation rules. The inventory turnover rate calculation unit matches the optimal supply chain with the purchase volume allocation rules, adjusts the optimal supply chain, and calculates the inventory turnover rate. The daily purchase volume adjustment unit adjusts the manufacturer's daily purchase volume based on the inventory turnover rate.

Images