Commodity transportation management method based on online shopping platform

Abstract

The invention belongs to the technical field of logistics management, and discloses a commodity transportation management method based on an online shopping platform, which integrates real-time traffic data through space-time two-dimensional path optimization, dynamically adjusts a distribution route, greatly improves the distribution time efficiency, and effectively reduces the vehicle no-load rate. The intelligent scheduling system matches the transport capacity in real time according to order changes and vehicle states based on a dynamic resource allocation mechanism, thereby not only helping merchants to reduce the management cost, but also shortening the in-transit time of commodities. Even if the order quantity is greatly increased, the system can avoid invalid transportation and resource waste through accurate resource regulation and control, and dual optimization of transportation efficiency improvement and cost control is realized; the whole process of transportation data can be traced through a blockchain and Internet of Things fused evidence storage system, and a user can check the positions, temperature, humidity and other states of goods in real time; a multi-level abnormity emergency response mechanism pointedly processes transportation abnormity, and the commodity loss rate is reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of logistics management, and specifically relates to a commodity transportation management method based on an online shopping platform. Background Art

[0002] Against the backdrop of the continued expansion of online shopping, commodity transportation management faces multiple challenges, including order fragmentation, complex delivery scenarios, and diversified user needs. Traditional logistics management systems mostly use static planning models, which are difficult to cope with dynamically changing transportation networks and sudden order fluctuations. Existing transportation management methods have problems such as delivery delays, commodity losses, and insufficient information transparency.

[0003] After searching, the Chinese invention patent with announcement number CN120258664A proposed a cross-border e-commerce logistics order management system based on big data. Although the route planning was optimized through clustering algorithms, there were three significant limitations: first, dynamic factors such as real-time road conditions and extreme weather were not integrated, resulting in a significant lag in route adjustment; second, there was a lack of a monitoring mechanism for the entire life cycle of goods, resulting in a high loss rate for fresh goods; third, data storage relied on centralized servers, which posed a risk of information tampering and made it difficult to enhance user trust. Summary of the Invention

[0004] The purpose of the present invention is to provide a commodity transportation management method based on an online shopping platform to solve the problems raised in the above-mentioned background technology. Compared with the existing technology with announcement number CN120258664A, the difference of the present invention is: ① Integrating real-time road conditions and extreme weather dynamic factors, and realizing dynamic path adjustment through an improved DDPG algorithm to solve the problem of path lag; ② Introducing blockchain-Internet of Things integrated evidence storage, covering the entire life cycle of commodities monitoring, and reducing the fresh food loss rate; ③ Using digital twins to achieve global optimization, get rid of the dependence on centralized servers, and improve data credibility.

[0005] In order to achieve the above-mentioned object, the present invention provides the following technical solution: a commodity transportation management method based on an online shopping platform, the specific steps of the method are as follows: S1. Multi-source heterogeneous data fusion preprocessing: Order, environment, and logistics network data are collected, cleaned and integrated through federated learning, and the urgency of goods is quantified using a product urgency quantification algorithm model to provide a data foundation for subsequent planning and scheduling; S2, spatiotemporal dual-dimensional path optimization: Based on preprocessed data, a multidimensional state space is constructed and a path is generated using an improved DDPG algorithm. The improved DDPG algorithm includes an attention mechanism layer, a dynamic reward pool mechanism, and edge computing to improve response speed, providing an optimal path reference for step S3. S3, Adaptive Resource Scheduling Engine: Based on the route optimization results, the submodules prioritize orders and match vehicles, set thresholds to trigger resource reallocation, and call on backup resources to match orders with capacity, continuing the route planning results of step S2. S4. Blockchain-IoT Fusion Evidence Storage System: Equipment is deployed at dispatched transport nodes, and data is stored in a consortium chain via a dedicated network. Users can access data across the entire chain, and certificates are pushed when standards are exceeded, providing trusted data support and traceability for the transport process. S5. Multi-level emergency response to abnormalities: Based on stored evidence data, neural networks are used for prevention and sensors for abnormalities. Decision trees are used for handling and optimization to ensure smooth transportation and respond to abnormalities that may arise in the evidence system. In the decision tree model, in the event of a vehicle failure, backup vehicles within 5 km of the fault point and with a range of 100 km or more are prioritized. If no resources are available, shared logistics pool scheduling is triggered. S6. Full-link service quality iteration: Combined with exception response results, user verification data and evaluations are collected, service strategies are adjusted using algorithms, and reports are generated to assist merchants in decision-making. Service quality is continuously improved based on exception handling feedback. S7. Digital Twin Logistics Network Construction: This integrates the aforementioned data and optimization results to build a virtual twin. This digital twin synchronizes data with the physical system every 30 minutes, simulates and predicts order fluctuations, optimizes warehouse networks, transportation capacity, and scheduling, and achieves intelligent evolution of the logistics network, integrating optimization effects across the entire chain. The digital twin data synchronization threshold is set at a deviation of more than 10% between virtual and physical data. This threshold is determined based on the historical fluctuation range of key parameters of the logistics network (such as vehicle positioning error and inventory count differences) to ensure the necessity and timeliness of synchronization calibration; the dynamic calibration algorithm uses the least squares method to correct the virtual model parameters so that the deviation returns to within the threshold.

[0006] Preferably, the specific steps of multi-source heterogeneous data fusion preprocessing in step S1 are as follows: S11. Synchronous Multi-Dimensional Data Collection: Build a distributed data collection architecture to simultaneously acquire shopping platform order data (including product ID, weight, delivery address, etc.), environmental data (road construction information, traffic accident warnings, regional precipitation probability, etc.), and logistics network data (warehouse inventory turnover rate, vehicle status, delivery driver historical performance, etc.), achieving comprehensive multi-source information aggregation and providing data support for subsequent links; S12. Data cleaning and feature quantification: A federated learning framework is used for cross-domain data cleaning. Feature fusion is completed while ensuring data privacy. The product urgency quantification algorithm model is used to quantify the product urgency, providing data basis for steps S2 and S3. The federated learning framework includes three types of participants (platform, warehouse node, and distribution node). The local model uses random forest. Model parameters are exchanged instead of original data in each round of iteration. A fixed number of iterations is set to ensure the accuracy of feature fusion.

[0007] Preferably, the specific steps of the spatiotemporal dual-dimensional path optimization in step S2 are as follows: S21. Multi-dimensional state space construction and path generation: Relying on pre-processed multi-source data, we break through the time static limitations of traditional path planning and construct a multi-dimensional state space containing spatial coordinates, time parameters, and vehicle status. We use the improved Deep Deterministic Policy Gradient (DDPG) algorithm to generate dynamic paths. The Actor network of the improved deep deterministic policy gradient algorithm includes an input layer, an attention layer, and an output layer. The number of neurons in each layer and the type of activation function are set according to the path planning requirements. The improved DDPG algorithm specifically adds an attention mechanism layer to the Actor network, assigns a 0.3x weight to the product urgency feature, and introduces a double-delay update strategy to the Critic network to avoid overestimation bias and achieve real-time adaptation of path planning to spatiotemporal changes. The improved DDPG algorithm's attention mechanism assigns a 0.3x weight to the urgency feature. This weighting was determined by analyzing the delivery delay risk coefficients of 10 typical product categories (including fragile items and fresh produce). Compared to the traditional DDPG algorithm's equal weighting, this algorithm can optimize the routing logic for high-priority orders. S22. Dynamic Reward Mechanism and Response Optimization: An innovative dynamic reward pool mechanism is designed: the base reward is inversely proportional to the route duration, and additional rewards are included for avoiding congested roads, energy conservation (calculated based on vehicle energy consumption models), and service quality (linked to on-time delivery rate). Local deployment of the model through edge computing nodes allows for regular route recalculation, significantly improving response speed compared to traditional methods and ensuring efficient route execution. In the dynamic reward pool mechanism, energy conservation rewards are calculated based on the GB / T36980-2018 electric vehicle energy consumption standard, and congestion avoidance rewards correspond to a congestion index ≥ 0.7 published by the traffic management department; Preferably, the specific steps of the adaptive resource scheduling engine in step S3 are as follows: S31. Order Prioritization and Vehicle Matching: Based on the optimization path generated in step S2, develop an intelligent "order-resource" matching system. This system uses an order priority algorithm model to weight the remaining shelf life of products (to ensure the freshness of perishable goods such as fresh produce) and the user waiting time ratio (to improve service timeliness). Fresh produce is given an additional weight. The weighting coefficient δ for fresh produce ranges from 1.2 to 1.8 and is dynamically adjusted based on the perishability of the product, enabling dynamic calculation of order priorities. The vehicle matching submodule improves the genetic algorithm crossover operator, incorporating constraints such as battery life and maintenance status to enhance matching accuracy. S32. Dynamic response and resource reallocation: Set multiple trigger thresholds such as order volume fluctuations, sudden changes in road conditions, and vehicle anomalies. Once triggered, quickly initiate resource reallocation, call on spare vehicles or share the logistics resource pool (sign real-time scheduling agreements with multiple logistics companies in the region), realize dynamic adjustment of resource scheduling, and ensure a smooth transportation process.

[0008] Preferably, the specific steps of the blockchain-IoT integrated evidence storage system in step S4 are as follows: S41. IoT Device Deployment and Data Collection: Multiple IoT devices are deployed at dispatched logistics nodes (warehouses, transport vehicles, and delivery terminals): UHF RFID readers at warehouse entrances, temperature and humidity sensors on transport vehicles, and positioning modules on delivery personnel's handheld terminals. These devices collect real-time product status and transportation data, transmit it via a dedicated network, and provide a data source for full-process evidence storage. S42. Blockchain evidence storage and user query mechanism: Device data is transmitted to alliance chain nodes, the number of which is 3-7. The consensus mechanism is PBFT or DPoS. A distributed file system is used to store unstructured data (such as photos of product packaging status). Key operations (completion of warehouse sorting, warehousing at transfer stations, etc.) must be confirmed by multi-node consensus. A user-side query tool is developed to support full-chain order data query. When monitoring data exceeds the standard and persists for a certain period of time, abnormal certificates with blockchain timestamps are automatically pushed to ensure that the data is credible and traceable.

[0009] Preferably, the specific steps of the multi-level abnormal emergency response in step S5 are as follows: S51. Abnormality Prevention and Real-Time Monitoring: Based on the real-time data support from step S4, a "prevention-monitoring" closed loop is established: the prevention layer uses neural networks to predict potential risks such as extreme weather and road abnormalities; the monitoring layer uses vibration sensors to identify abnormal handling of goods, distinguishing acceleration thresholds for fragile goods and ordinary goods and recording the status. The vibration sensor's acceleration threshold for fragile goods is 8-12g, and for ordinary goods is 18-22g, so as to promptly capture abnormal signals during transportation; S52. Intelligent handling and replay optimization: The handling layer has a built-in decision tree model to generate response plans (such as dispatching replacement vehicles and triggering dedicated connection resources) for anomalies such as vehicle failure and out-of-control fresh food environment parameters. The replay layer stores the anomaly processing data in the knowledge base and optimizes the decision model through knowledge graph technology to continuously improve the processing efficiency of similar anomalies and strengthen the risk resistance of the transportation network.

[0010] Preferably, the specific steps of link service quality iteration in step S6 are as follows: S61. User feedback data collection: Based on the processing results of step S5, an "experience-improvement" feedback mechanism is designed. When users sign for goods, they complete product verification through intelligent scanning. The system automatically collects visual data such as packaging integrity and product integrity. A sentiment analysis algorithm is used to parse user review texts, extract service-related keywords, and quantify them into service defect values ​​to comprehensively capture service quality feedback. S62. Dynamic optimization of service strategies: Dynamically adjust service strategies based on defect values: Strengthen delivery staff training in areas with higher defect values; provide exclusive services (such as timeliness guarantees) to high-value users; regularly generate regional service quality reports and provide merchants with inventory pre-positioning suggestions (such as allocating goods to order-intensive areas), converting user feedback into service optimization motivation to improve overall service levels.

[0011] Preferably, the specific steps of constructing the digital twin bioflow network in step S7 are: S71. Virtual Twin Construction: Integrate full-chain optimization results and build a digital twin of the logistics network based on a professional engine. This accurately maps physical elements such as warehouse layout, vehicle trajectories, and personnel flows into the virtual space, creating a digital mirror and providing a visual and simulatable virtual carrier for global optimization. The data synchronization mechanism between the digital twin and the physical system is as follows: full data synchronization is performed every 30 minutes, using a dual primary key of 'timestamp + device ID' to avoid conflicts; when the deviation between virtual and physical data exceeds a set threshold, an abnormality warning is automatically triggered and the model is recalibrated; The digital twin data synchronization threshold is set at a deviation of more than 10% between virtual and physical data. This threshold is determined based on the historical fluctuation range of key parameters of the logistics network (such as vehicle positioning error and inventory count differences) to ensure the necessity and timeliness of synchronization calibration; the dynamic calibration algorithm uses the least squares method to correct the virtual model parameters so that the deviation returns to within the threshold.

[0012] The synchronous calibration mechanism of the digital twin adopts a combined strategy of "least squares method + dynamic threshold". Different from the fixed-cycle synchronization of existing technologies, it can adjust the calibration frequency according to the real-time load of the logistics network to improve the accuracy of virtual-physical mapping.

[0013] S72. Intelligent Forecasting and Parameter Optimization: Future order volume fluctuations are predicted through simulation algorithms, and core parameters are optimized accordingly: warehouse network layout (exceeding expected areas triggers new warehouse site selection evaluation), transportation capacity reserves (adjusting the ratio of owned and outsourced vehicles according to peak and off-season), and personnel scheduling (generating the optimal plan based on the work status model); the digital twin regularly synchronizes data with the physical system to ensure that decision simulations are consistent with actual operations, thereby achieving continuous evolution of the logistics network.

[0014] The beneficial effects of the present invention are as follows: 1. This invention integrates real-time traffic data through dual-dimensional space-time path optimization to dynamically adjust delivery routes, significantly improving delivery efficiency while effectively reducing vehicle empty load rates. The intelligent scheduling system relies on a dynamic resource allocation mechanism to match transportation capacity in real time based on order changes and vehicle status, which not only helps merchants reduce management costs but also shortens the time goods are in transit. Even in the case of a significant increase in order volume, the system can avoid ineffective transportation and resource waste through precise resource regulation, achieving the dual optimization of improved transportation efficiency and cost control.

[0015] 2. The present invention makes the entire transportation data traceable through the evidence storage system that integrates blockchain and the Internet of Things. Users can view the location, temperature, humidity and other status of the goods in real time, breaking the "information island" of traditional logistics and solving the problem of data credibility; the multi-level abnormal emergency response mechanism handles transportation anomalies in a targeted manner, reduces the loss rate of goods, and especially ensures the safety of fragile goods and fresh products; transparent logistics information and reliable product status improve user satisfaction with logistics services, enhance consumer trust, and thus promote user repurchase behavior.

[0016] 3. The present invention combines digital twins with a full-link feedback mechanism. The prediction of digital twins can achieve effective estimation of order volume fluctuations. After the warehouse network layout is optimized, the cross-domain delivery time is shortened, inventory backlogs are reduced, the retention rate of high-value users is improved, and the number of customer service inquiries is reduced, thereby realizing the continuous evolution of the logistics network and enabling the logistics network to have self-iteration capabilities; through virtual simulation, the regional warehouse network layout is optimized to shorten cross-domain delivery time; relying on predictive replenishment to reduce inventory backlogs; the service quality dynamic adjustment mechanism optimizes services based on user feedback, improves the retention rate of high-value users, and at the same time reduces the number of customer service inquiries, optimizes enterprise operating costs from multiple dimensions, and realizes the continuous evolution of the logistics network. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of the commodity transportation management method based on the online shopping platform of the present invention. DETAILED DESCRIPTION

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0019] like Figure 1 As shown, an embodiment of the present invention provides a commodity transportation management method based on an online shopping platform. The specific steps of the method are as follows: S1. Multi-source heterogeneous data fusion preprocessing: Order, environment, and logistics network data are collected, cleaned and integrated through federated learning, and the urgency of goods is quantified using a product urgency quantification algorithm model to provide a data foundation for subsequent planning and scheduling; S2, spatiotemporal dual-dimensional path optimization: Based on preprocessed data, a multidimensional state space is constructed, paths are generated using an improved DDPG algorithm, a dynamic reward pool mechanism is designed, and edge computing is used to improve response speed, providing an optimal path reference for step S3; S3, Adaptive Resource Scheduling Engine: Based on the route optimization results, the submodules prioritize orders and match vehicles, set thresholds to trigger resource reallocation, and call on backup resources to match orders with capacity, continuing the route planning results of step S2. S4. Blockchain-IoT Fusion Evidence Storage System: Equipment is deployed at dispatched transport nodes, and data is stored in a consortium chain via a dedicated network. Users can access data across the entire chain, and certificates are pushed when standards are exceeded, providing trusted data support and traceability for the transport process. S5. Multi-level emergency response to abnormalities: Based on stored evidence data, neural networks are used for prevention, sensors are used for monitoring abnormalities, and decision trees are used for handling and review and optimization to ensure smooth transportation and respond to abnormal situations that may arise in the evidence storage system. S6. Full-link service quality iteration: Combined with exception response results, user verification data and evaluations are collected, service strategies are adjusted using algorithms, and reports are generated to assist merchants in decision-making. Service quality is continuously improved based on exception handling feedback. S7. Construction of a digital twin logistics network: Integrate the aforementioned data and optimization results to build a virtual twin, simulate and predict order fluctuations, optimize warehouse networks, transportation capacity, and scheduling, realize the intelligent evolution of the logistics network, and integrate the optimization effects of the entire chain. Order fluctuation prediction uses the LSTM algorithm. Input features include order volume in the past 30 days, regional population density, holiday factors, etc., and the time window is set to 72 hours.

[0020] The multi-source heterogeneous data fusion preprocessing in step S1 includes two key links: first, by building a distributed data collection architecture to achieve synchronous multi-dimensional data collection, synchronously obtain the shopping platform's order data (covering product ID, weight, delivery address, etc.), environmental data (including road construction information, traffic accident warnings, regional precipitation probability, etc.) and logistics network data (involving warehouse inventory turnover rate, vehicle status, delivery personnel's historical performance, etc.), so as to achieve comprehensive aggregation of multi-source information and provide basic data support for the subsequent implementation of each link.

[0021] On this basis, a federated learning framework is used for cross-domain data cleaning, and feature fusion is completed under the premise of effectively protecting data privacy. At the same time, the urgency of goods is quantified using a product urgency quantification algorithm model, thereby providing accurate and reliable data basis for the spatiotemporal dual-dimensional path optimization in step S2 and the adaptive resource scheduling engine in step S3, ensuring that subsequent links can carry out work based on high-quality data.

[0022] The fixed number of iterations in the federated learning framework is set to 20, which is determined by testing the stability of feature fusion under different iterations. The local random forest model is set to 100 trees with a maximum depth of 10 layers. The hyperparameter selection is determined by balancing the model convergence speed and the risk of overfitting. The local random forest training sample size is 5,000+ order data items per node. The feature selection method uses the Gini coefficient method to screen out 12 key features such as product weight and delivery distance. The algorithm model expression for quantifying the urgency of a product is: ; Where: E is the urgency of the product (range: 1-10, the larger the value, the higher the urgency); L is the user membership level coefficient (1.0 for ordinary users, 1.2 for silver members, 1.5 for gold members, 2.0 for diamond members); K is the urgency coefficient of customer service notes (1.0 for no notes, 1.5 for general urgency, 2.0 for high urgency); α and β are weight coefficients ( , default α=0.6, β=0.4, can be adjusted dynamically according to platform rules); The underlying logic of this formula involves the idea of ​​weighted summation in mathematics, which achieves quantitative evaluation of complex phenomena by assigning weights to different influencing factors. This formula is used to quantify the urgency of the product in step S1. By integrating the user's membership level (reflecting user value) and customer service notes (reflecting special needs), it can achieve a refined division of order priority. For example, the urgent order of a Diamond member (L=2.0, K=2.0) is calculated as: E=0.6×2.0+0.4×2.0=2.0; significantly higher than the regular orders of ordinary users, providing a quantitative basis for subsequent route planning and resource scheduling.

[0023] The spatiotemporal dual-dimensional path optimization in step S2, based on preprocessed multi-source data, breaks the static limitations of traditional path planning, which only considers spatial factors, and constructs a multidimensional state space encompassing spatial coordinates, temporal parameters, and the vehicle's real-time status. By introducing an improved Deep Deterministic Policy Gradient (DDPG) algorithm, a path is dynamically generated to adapt to spatiotemporal changes, ensuring that path planning can respond in real time to changes in time and the spatial environment, achieving precise alignment with the actual scenario.

[0024] Then, an innovative dynamic reward pool mechanism was designed, directly linking basic rewards to route duration. The shorter the route duration, the higher the reward. At the same time, additional rewards were incorporated, such as avoiding congested roads, energy savings calculated based on vehicle energy consumption models, and service quality associated with on-time delivery rates, forming a multi-dimensional incentive. By deploying localized models at edge computing nodes, the system can regularly recalculate routes, significantly improving response speed compared to traditional methods, effectively ensuring the efficient execution of optimized routes and effectively improving transportation timeliness and resource utilization efficiency. In the dynamic reward pool mechanism, energy conservation rewards are calculated based on electric vehicle energy consumption standards. Congestion avoidance rewards are based on the real-time congestion index released by the traffic management department (triggered when the set congestion level is reached). The on-time delivery rate-related reward value in the service quality reward is set based on the results of regression analysis of historical service data. Energy conservation rewards are calculated using the basic formula in the GB / T36980-2018 Electric Vehicle Energy Consumption Standard. Congestion avoidance rewards correspond to the congestion index grading standards published by the traffic management department (triggered when the congestion index is ≥0.7). The association rules between service quality rewards and on-time delivery rates are determined based on statistical analysis of historical service data. Dynamic reward pool mechanism expression: ;

[0025] Where: is the reward value for a single-step decision; is the path length penalty term ( is the current path length, the smaller the value, the higher the reward); Energy consumption reward item ( Estimate the energy consumption for the current path, is the historical maximum energy consumption of this section); is the urgency of the product; S is the on-time rate bonus (1.0 for on-time delivery, 0 for delayed delivery, and 0-1 for early delivery based on the proportion of the early delivery time); is the weight coefficient ( , the default values ​​are 0.3, 0.2, 0.3, 0.2 respectively); Weight coefficient Based on the historical delivery data statistics of three scenarios (urban core area, suburbs, and remote areas), the areas with high road complexity are (Congestion avoidance weight) takes a higher value; The basis of this formula comes from deep reinforcement learning theory and function construction knowledge in mathematics; By comprehensively considering four dimensions: path length (shortening delivery distance), energy consumption (reducing transportation costs), product urgency (giving priority to high-urgency orders), and punctuality (improving service quality), the deep reinforcement learning model is guided to generate the optimal path.

[0026] For example, a short-distance, low-energy route that can deliver high-urgency goods on time will receive a higher reward value (larger R). The model maximizes cumulative rewards through continuous learning, achieving the optimal balance between efficiency, cost and service quality in route planning, significantly optimizing delivery time and vehicle empty load rate.

[0027] The adaptive resource scheduling engine in step S3 refers to the development of an intelligent "order-resource" matching system based on the optimized path generated in step S2. Order priority is dynamically calculated using a specialized algorithmic model: a weighted fusion is performed between the remaining shelf life of the product (to ensure the freshness of perishable goods such as fresh produce) and the user's waiting time (to improve service timeliness). Fresh produce is also given an additional weight to ensure that high-priority orders receive priority resources. Regarding vehicle matching, the crossover operator of the genetic algorithm is improved to take into account actual operational constraints such as vehicle range and maintenance status, further improving the accuracy of order-vehicle matching.

[0028] Then, a flexible scheduling mechanism was built: multiple trigger thresholds were set for factors such as order volume fluctuations, sudden road condition changes, and vehicle anomalies. Once these thresholds were reached, the system would quickly initiate resource reallocation, promptly mobilizing spare vehicles or the regional shared logistics resource pool (real-time scheduling agreements had been signed with multiple logistics companies). This enabled dynamic adjustments to transportation resources, effectively responding to various emergencies, ensuring the smooth and efficient operation of the entire transportation process, and ensuring that optimized routes were effectively implemented. Order priority algorithm model expression: ; Where: is the order priority coefficient (the value range is 0-1, the larger the value, the higher the priority); The remaining percentage of shelf life ( The remaining shelf life of the product. is the total shelf life in days); is the proportion of waiting time ( The user has been waiting for a long time, The longest delivery time promised by the platform); is the weight coefficient (the default is 0.6, balancing product freshness and user waiting experience); is the weighting coefficient for fresh products (1.5 for fresh products and 1.0 for non-fresh products); This formula combines the ideas of proportional calculation and weight distribution in mathematics; The reason for setting δ=1.5 in the order priority algorithm is that the loss characteristics of fresh goods are significantly different from those of ordinary goods. This weighting coefficient has been verified through experiments to balance the demand for fresh food preservation and service timeliness.

[0029] For example, if a fresh food order has a remaining shelf life of 1 day (total shelf life of 3 days), the user has waited for 4 hours (the maximum promised time is 8 hours). P=0.6×(1 / 3)+0.4×(4 / 8)×1.5=0.2+0.3=0.5; Priority is higher than that of similar orders of ordinary goods, ensuring that resources are tilted towards high-priority orders.

[0030] The blockchain-IoT integrated evidence storage system in step S4 involves first deploying IoT devices and collecting data. Multiple IoT devices are deployed at all logistics nodes after scheduling: UHF RFID readers are installed at warehouse entrances to accurately identify incoming and outgoing goods; temperature and humidity sensors are installed on transport vehicles to monitor the storage environment in real time; and handheld terminals equipped with positioning modules for delivery personnel record the delivery trajectory of goods. These devices collect data on product status, transportation routes, environmental parameters, and other aspects, which are transmitted via dedicated network lines. This creates a raw data chain covering the entire product process from shipment to delivery, providing a comprehensive and real-time data source for subsequent evidence storage.

[0031] The temperature and humidity sensors have set accuracy ranges (temperature and humidity each have corresponding error ranges), and the deployment density is two per vehicle (one at the front and one at the rear). The recognition range and reading rate of the UHF RFID reader / writer meet the requirements of efficient identification at logistics nodes. The temperature and humidity sensors comply with GB / T2887-2011 General Specifications for Data Center Sites, with a temperature measurement range of -20°C to 60°C and a humidity measurement range of 20% to 80% RH. The UHF RFID reader / writer complies with ISO18000-6C standards and has an identification distance of 5-8 meters (test conditions: unobstructed, ambient humidity ≤85% RH). The temperature and humidity sensor accuracy for refrigerated trucks is ±0.3°C / ±2%RH (meets the GB / T24406-2012 cold chain logistics standard), and for ordinary trucks it is ±0.5°C / ±3%RH (meets the GB / T19779-2005 general transportation standard).

[0032] Data transmitted by devices is synchronized to the alliance chain nodes in real time. Unstructured data, such as photos of product packaging status, is stored in a distributed file system to ensure security. Key operations, such as warehouse sorting completion and transfer station warehousing, must be jointly verified and confirmed by multiple nodes on the chain to ensure that the data cannot be tampered with. At the same time, a user-side query tool has been developed to support consumers in querying full-link data by order number. When parameters such as temperature and humidity are detected to exceed the standard and persist for a certain period of time, the system automatically generates an exception certificate with a blockchain timestamp and pushes it to the user. This not only makes data traceable throughout the entire process, but also allows users to understand the status of goods in real time, effectively ensuring the credibility and transparency of transportation data.

[0033] The alliance chain consists of 5 nodes (1 platform, 3 logistics companies, and 1 regulatory agency), adopts the PBFT consensus mechanism, the block time is set to a fixed value, the IPFS protocol is used for unstructured data storage, and the hash value is written into the blockchain to ensure that it cannot be tampered with; The blockchain block time is set to 3 seconds, based on the average network latency test results between alliance chain nodes. IPFS data shards use a fixed size of 256KB and the hash algorithm is SHA-256 to ensure efficient data storage and verification. Among them, the multi-level abnormal emergency response in step S5 refers to first building a "prevention-monitoring" linkage mechanism based on the real-time data provided by step S4: the prevention layer uses neural networks to analyze historical data, predict potential risks such as extreme weather and road abnormalities in advance, and reserve preparation time for response; the monitoring layer uses vibration sensors to capture the status of commodity transportation in real time, set different acceleration thresholds according to the differences in the characteristics of fragile goods and ordinary goods, and immediately record the abnormal status once the threshold is exceeded to ensure that abnormal signals during transportation can be discovered in time.

[0034] Intelligent handling and review optimization further strengthen response capabilities: The handling layer incorporates a built-in decision tree model to automatically generate targeted solutions for various abnormal situations, such as vehicle failure and out-of-control parameters of fresh produce storage environments. These solutions can include dispatching spare vehicles to replace faulty vehicles and triggering dedicated cold chain reconnection resources to ensure fresh produce quality. The review layer incorporates detailed data from each abnormal handling into a knowledge base and, combined with knowledge graph technology, continuously optimizes the decision model, continuously improving the efficiency of handling similar abnormalities and thus enhancing the risk resistance of the entire transportation network. The duration of data exceeding the standard is set at 5 minutes, which is determined based on the industry's general quality maintenance requirements for fresh product transportation and the quality change patterns of perishable products in non-standard environments.

[0035] The link service quality iteration in step S6 refers to the design of an "experience-improvement" feedback mechanism, closely integrated with the exception handling results of step S5. When users sign for goods, the product is verified through the intelligent scanning function, and the system automatically records visual information such as packaging integrity and product integrity. At the same time, a sentiment analysis algorithm is used to sort out user review texts, extract service-related keywords such as "slow delivery" and "damaged packaging", and convert them into quantifiable service defect values ​​to comprehensively capture problem points in the service process. The service defect value trigger threshold is 0.8, which is determined based on the correlation analysis between regional service quality and user feedback, and corresponds to the critical point of service level that requires intervention and optimization.

[0036] Service plans are then flexibly adjusted based on defect counts: For areas with high defect counts, delivery personnel training is strengthened to improve service standards; exclusive services such as guaranteed delivery times are provided to high-value users; and regular regional service quality reports are compiled to provide merchants with inventory pre-positioning recommendations, such as allocating goods to areas with high order density. In this way, user feedback is directly translated into concrete service optimization measures, continuously improving overall service levels and forming a closed loop of continuous iterative upgrades in service quality.

[0037] The quantitative standard for service defect values ​​is: corresponding defect values ​​are set according to the type of service problem (such as delivery time, packaging status, etc.). Regions where the defect value reaches the set threshold trigger a training mechanism, and the training cycle and duration are determined based on the severity of the regional service problem.

[0038] Among them, the construction of the digital twin logistics network in step S7 refers to first integrating the results of the entire chain optimization, relying on a professional engine to build a digital twin of the logistics network, accurately mapping the core elements such as warehouse layout, vehicle real-time trajectory, and personnel flow status in the physical world to the virtual space, and constructing a digital mirror consistent with the actual logistics network, providing a visual and simulatable virtual operation carrier for global optimization, allowing managers to intuitively grasp the overall operation status in a virtual environment.

[0039] Simulation algorithms then analyze historical data to predict future order volume fluctuations, enabling targeted optimization of core operational parameters. For warehouse network layout, areas with consistently exceeding order volumes trigger new warehouse site selection evaluations. For transportation capacity reserves, the ratio of owned and outsourced vehicles is adjusted based on seasonal patterns. For staff scheduling, optimal attendance plans are generated based on work status models. The digital twin regularly synchronizes real-time data with the physical system to ensure that decisions made in the virtual simulation align with actual operational status, enabling dynamic adjustments and continuous evolution of the logistics network.

[0040] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0041] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A commodity transportation management method based on an online shopping platform, characterized by: The specific steps of this method are as follows: S1. Multi-source heterogeneous data fusion preprocessing: Order, environment, and logistics network data are collected, cleaned and integrated through federated learning, and the product urgency quantification algorithm model is used to quantify the product urgency; S2. Spatiotemporal Path Optimization: Based on preprocessed data, a multidimensional state space is constructed and a path is generated using an improved DDPG algorithm. The improved DDPG algorithm includes an attention mechanism layer, a dynamic reward pool mechanism, and edge computing to improve response speed and output an optimal path reference. S3, Adaptive Resource Scheduling Engine: Based on route optimization results, submodules prioritize orders, match vehicles, set thresholds to trigger resource reallocation, and call on backup resources to match orders with capacity. S4. Blockchain-IoT integrated evidence storage system: Equipment is deployed at the transportation nodes after scheduling, and data is stored in the alliance chain via a dedicated network. Users can check the data of the entire chain, and certificates are pushed when the standards are exceeded; S5. Multi-level emergency response to abnormalities: Based on stored evidence data, neural network prevention and sensor monitoring are used to detect abnormalities, and decision trees are used to handle and optimize the abnormalities. This is done to deal with abnormalities that may occur in the evidence storage system. S6. Full-link service quality iteration: Combined with exception response results, user verification data and evaluations are collected, and algorithms are used to adjust service strategies and generate reports to assist merchants in decision-making. S7. Construction of a digital twin logistics network: Integrate the aforementioned data and optimization results to build a virtual twin. The digital twin synchronizes data with the physical system every 30 minutes, simulates and predicts order fluctuations, optimizes warehouse networks, transportation capacity, and scheduling, and integrates the optimization effects of the entire chain.

2. A commodity transportation management method based on an online shopping platform according to claim 1, characterized in that: The specific steps of multi-source heterogeneous data fusion preprocessing in step S1 are as follows: S11. Multi-dimensional data synchronous acquisition: Build a distributed data acquisition architecture to synchronously obtain shopping platform order data, environmental data, and logistics network data; S12. Data cleaning and feature quantification: Use the federated learning framework for cross-domain data cleaning, complete feature fusion while ensuring data privacy, and use the product urgency quantification algorithm model to quantify the product urgency.

3. The method for managing commodity transportation based on an online shopping platform according to claim 1, characterized in that: The specific steps of the spatiotemporal dual-dimensional path optimization in step S2 are as follows: S21. Multidimensional State Space Construction and Path Generation: Based on preprocessed multi-source data, a multidimensional state space containing spatial coordinates, time parameters, and vehicle status is constructed. A dynamic path is generated using an improved deep deterministic policy gradient algorithm. The actor network of the improved deep deterministic policy gradient algorithm includes an input layer, an attention layer, and an output layer. The number of neurons and activation function type in each layer are set according to the path planning requirements. S22. Dynamic Reward Mechanism and Response Optimization: Innovatively design a dynamic reward pool mechanism: the basic reward is inversely proportional to the path duration, and additional rewards are included for avoiding congested roads, energy conservation, and service quality; the model is locally deployed through edge computing nodes, and path recalculation is completed regularly.

4. The method for managing commodity transportation based on an online shopping platform according to claim 1, characterized in that: The specific steps of the adaptive resource scheduling engine in step S3 are as follows: S31. Order Prioritization and Vehicle Matching: Based on the optimized path generated in step S2, an intelligent order-resource matching system is developed. This system uses an order priority algorithm model to weight the remaining shelf life of products and the proportion of user waiting time, and additionally weights fresh products. The weighting coefficient δ for fresh products ranges from 1.2 to 1.8 and is dynamically adjusted based on the perishability of the product, enabling dynamic calculation of order priorities. The vehicle matching submodule improves the genetic algorithm crossover operator, incorporating constraints such as battery life and maintenance status to enhance matching accuracy. S32. Dynamic response and resource reallocation: Set multiple trigger thresholds for order volume fluctuations, sudden changes in road conditions, and vehicle anomalies. Once triggered, quickly initiate resource reallocation and call on spare vehicles or shared logistics resource pools.

5. The method for managing commodity transportation based on an online shopping platform according to claim 1, characterized in that: The specific steps of the blockchain-IoT integrated evidence storage system in step S4 are as follows: S41. IoT device deployment and data collection: Multiple IoT devices are deployed at dispatched logistics nodes: UHF RFID readers at warehouse entrances, temperature and humidity sensors on transport vehicles, and positioning modules on delivery personnel's handheld terminals to collect real-time product status and transportation data. S42. Blockchain evidence storage and user query mechanism: Device data is transmitted to alliance chain nodes. The number of alliance chain nodes is 3-7, and the consensus mechanism is PBFT or DPoS. A distributed file system is used to store unstructured data. Key operations must be confirmed by multi-node consensus. A user-side query tool is developed to support full-chain order data query. When the monitoring data exceeds the standard and persists for a certain period of time, the abnormal certificate with the blockchain timestamp is automatically pushed.

6. The method for managing commodity transportation based on an online shopping platform according to claim 1, characterized in that: The specific steps of the multi-level abnormal emergency response in step S5 are as follows: S51, Abnormality Prevention and Real-time Monitoring: Based on the real-time data support of step S4, a prevention-monitoring closed loop is established: the prevention layer predicts extreme weather and potential risks of road abnormalities through neural networks; The monitoring layer uses vibration sensors to identify abnormal handling of goods, distinguishing acceleration thresholds for fragile goods and ordinary goods and recording the status. The acceleration thresholds of the vibration sensors for fragile goods are 8-12g, and for ordinary goods are 18-22g, thus promptly capturing abnormal signals during transportation. S52. Intelligent handling and review optimization: The handling layer has a built-in decision tree model to generate response plans for vehicle failures and abnormal out-of-control parameters of fresh food environments; the review layer stores the exception processing data in the knowledge base and optimizes the decision model through knowledge graph technology.

7. The method for managing commodity transportation based on an online shopping platform according to claim 1, characterized in that: The specific steps of link service quality iteration in step S6 are as follows: S61. User feedback data collection: Based on the processing results of step S5, design an experience-improved feedback mechanism: When the user signs for the product, the product is verified through intelligent scanning, and the system automatically collects visual data on packaging integrity and product integrity; A sentiment analysis algorithm is used to parse user review texts, extract service-related keywords, and quantify them into service defect values ​​to comprehensively capture service quality feedback. The service defect value trigger threshold is 0.8, which is determined based on the correlation analysis between regional service quality and user feedback. S62. Dynamic optimization of service strategies: Dynamically adjust service strategies based on defect values: Strengthen delivery staff training in areas with high defect values; Provide exclusive services for high-value users; Generate regional service quality reports regularly to provide merchants with inventory pre-positioning suggestions.

8. The method for managing commodity transportation based on an online shopping platform according to claim 1, characterized in that: The specific steps of constructing the digital twin bioflow network in step S7 are: S71. Virtual Twin Construction: Integrate full-chain optimization results and build a digital twin of the logistics network based on a professional engine. Accurately map physical warehouse layouts, vehicle trajectories, and personnel flow elements to the virtual space to create a digital mirror. The digital twin data synchronization threshold is a deviation of more than 10% between virtual and physical data. Synchronization calibration uses the least squares method to correct model parameters. S72. Intelligent prediction and parameter optimization: Future order volume fluctuations are predicted through simulation algorithms. The order fluctuation prediction adopts the LSTM algorithm. The input features include 30-day order volume, regional population density, and holiday factors. The core parameters are optimized based on this: warehouse network layout, transportation capacity reserves, and personnel scheduling; the digital twin regularly synchronizes data with the physical system.

Images