A cloud computing-based supply chain management method and system

Abstract

The application relates to a cloud-computing-based supply chain management method and system, and relates to the technical field of supply chain management. The cloud-computing-based supply chain management method comprises the following steps: acquiring and normalizing multi-channel data of products, combining data collection sources to construct a plurality of product data sets, and transmitting the product data sets to the cloud; constructing and correcting a product supply-demand model according to the product data sets and product market distribution, and outputting product simulation supply-demand data; generating a product inventory allocation strategy according to the product simulation supply-demand data and product inventory management indexes; collaboratively managing all enterprises of a supply chain according to the product inventory allocation strategy and enterprise shared information, and formulating a supply chain collaborative scheduling plan; and reducing inventory costs, reducing out-of-stock losses and overstock risks through accurate prediction data of the product supply-demand model and optimized inventory management, and reducing procurement costs, transportation costs and production costs through optimized supply chain processes, so that the economic benefits of enterprises are improved.

Description

Technical Field

[0001] This invention relates to the field of supply chain management technology, and in particular to a cloud computing-based supply chain management method and system. Background Technology

[0002] With the continuous development of information technology, more and more domestic enterprises are applying technologies such as cloud computing, big data, artificial intelligence, and the Internet of Things to their supply chain FCST systems to improve the accuracy and timeliness of forecasts. Cloud platforms, as efficient data storage, computing, and sharing tools, provide new technical support for enterprise procurement management. For example, Lianbao Technology, by building automated order management systems, intelligent scheduling systems, and other digital management systems, and leveraging the capabilities of the FineReport & FineBI platform, has achieved automation of supply chain order processing, highly accurate and intelligent planning, and a new model of supply chain ecosystem collaboration.

[0003] Enterprises have placed higher demands on the efficient operation and risk management of their supply chains, prompting continuous upgrades and improvements to supply chain FCST (Fulfilled Supply Chain Task) systems. Supplier collaboration platforms such as Xieke Cloud SRM (Supplier Relationship Management) possess FCST planning collaboration functions, enabling features such as supplier delivery plan dashboards, FCST production plan dashboards, and supplier incoming material tracking dashboards, helping enterprises better manage suppliers and production plans.

[0004] Application number CN202311379555;7 discloses a supply chain management system based on a cloud service platform, including a supply and demand planning module, an inventory management module, a supplier management module, an order processing module, a transportation and logistics management module, a quality management module, a cost management module, a reporting and analysis module, a risk management module, a visibility and collaboration module, a mobile application module, a security and permissions module, and a cloud service platform. The invention is flexible and scalable, allowing enterprises to easily expand or shrink their resources as needed, enabling them to cope with seasonal or sudden supply and demand fluctuations without large-scale investment in infrastructure. Furthermore, the invention uses a cloud service platform to reduce hardware and maintenance costs, allowing enterprises to pay based on actual usage, avoiding expensive initial investments and periodic upgrade costs.

[0005] The existing technical solutions mentioned above have the following drawbacks: 1. In terms of procurement supply and demand forecasting, traditional methods often rely on a single data source (such as historical procurement data), ignoring the impact of multi-dimensional data such as inventory levels, seasonal factors, market price fluctuations and sudden events on supply and demand forecasting, resulting in poor accuracy and real-time performance of the forecast results.

[0006] In terms of supplier collaboration, traditional supply chain management systems have significant limitations in supplier performance evaluation, supply chain visualization monitoring, and intelligent supplier recommendation, and cannot achieve efficient supplier matching and real-time tracking.

[0007] In terms of procurement execution and risk response, traditional systems typically rely on manual intervention for order generation and approval, which cannot respond quickly to emergencies and poses significant supply chain risks. Summary of the Invention

[0008] To address the shortcomings of existing technologies, the purpose of this invention is to provide a cloud-based supply chain management method and system. Through accurate forecasting data from product supply and demand models and optimized inventory management, it can reduce inventory costs, minimize stockout losses and backlog risks. At the same time, by optimizing supply chain processes, it can reduce procurement costs, transportation costs, and production costs, thereby improving the economic efficiency of enterprises.

[0009] The objective of this invention is achieved through the following technical solution:

[0010] A cloud-based supply chain management method includes:

[0011] Acquire and normalize multi-channel product data, construct several product datasets by combining data collection sources, and transmit them to the cloud;

[0012] Based on the product dataset and product market distribution, construct and revise the product supply and demand model, and output simulated product supply and demand data.

[0013] Based on the simulated supply and demand data of the product and the product inventory management indicators, a product inventory allocation strategy is generated.

[0014] Based on the product inventory allocation strategy and shared information among enterprises, a collaborative supply chain scheduling plan is formulated to manage all enterprises in the supply chain.

[0015] By adopting the above technical solutions, a standardized product dataset is constructed and transmitted to cloud storage by integrating internal ERP, sales system, and external market research data through multi-source data collection and normalization algorithms (such as Min-Max standardization and Z-Score standardization). Based on time series forecasting models (such as Prophet and ARIMA) and machine learning algorithms (such as random forests and gradient boosting trees), the supply and demand model is dynamically corrected in conjunction with product market distribution to output high-precision simulated supply and demand data. Furthermore, inventory allocation strategies are generated through safety stock and economic order quantity (EOQ) models. Finally, the Shapley value algorithm from game theory, combined with shared enterprise information (such as real-time inventory and transportation costs), is used to fairly allocate allocation costs and formulate collaborative scheduling plans. Multi-dimensional data fusion improves forecast accuracy, dynamic model correction enhances market adaptability, scientific inventory strategies reduce the risk of stockouts and slow sales, and a contribution-based cost-sharing mechanism optimizes supply chain collaboration efficiency.

[0016] The present invention is further configured such that: the specific steps of acquiring and normalizing multi-channel product data, constructing several product datasets by combining data collection sources, and transmitting them to the cloud include:

[0017] Extract product-related data from the company's internal product-related systems and external market research institutions to obtain historical sales data, inventory data, product production schedules, market sales change tables, and customer order volumes, and mark the corresponding product source identifiers;

[0018] Based on the information collection sources, all product-related data are classified and aggregated to construct several product datasets;

[0019] The entire product dataset is divided according to a preset data segmentation mechanism to obtain several product data blocks, and the segmentation timestamps are marked.

[0020] Based on the block timestamps and the parallel transmission mechanism, the product data blocks are transmitted to the cloud sequentially.

[0021] The product data blocks are aggregated in the cloud based on the product source identifier to obtain the product dataset.

[0022] By employing the above technical solution, multi-channel product data (including historical sales, inventory, orders, etc.) is obtained from internal enterprise systems (ERP, production schedules) and external market research institutions through data extraction algorithms (such as ETL tools), and product source identifiers are marked. Subsequently, data cleaning algorithms (such as interpolation and KNN imputation) are used to remove noise, fill missing data, and check for duplicates, resulting in a deduplicated dataset. Then, normalization algorithms (such as Min-Max standardization and Z-Score standardization) are used to transform the heterogeneous data into standardized product data. Finally, data is classified and aggregated according to data source characteristics (such as K-means clustering or rule-based methods). The system constructs a multi-dimensional product dataset; it uses hash sharding or consistent hashing algorithms to divide the data into blocks and marks the block timestamps; it uploads the data blocks to the cloud through parallel transmission mechanisms (such as HDFS or Spark's distributed transmission framework); finally, it uses distributed computing frameworks (such as Flink or MapReduce) to aggregate the dataset in the cloud based on the product source identifier to form a standardized product data pool; it improves data quality through fully automated cleaning and normalization, optimizes the efficiency of multi-source data integration through parallel transmission and distributed aggregation, and supports flexible expansion to adapt to the dynamic needs of different data sources and business scenarios.

[0023] The present invention is further configured such that: the specific steps of constructing and revising the product supply and demand model based on the product dataset and product market distribution, and outputting simulated product supply and demand data, include:

[0024] Based on the time series and product market distribution, feature extraction is performed on the product dataset to obtain regional time series feature parameters. Then, based on distributed computing rules, horizontal or vertical sharding is performed to obtain several feature parameter blocks and generate several data computing nodes.

[0025] Decision trees are deployed on each data computing node to build a distributed tree, and a feature subset sampling mechanism is generated by combining the preset number of iterations and learning rate.

[0026] Based on the feature parameter block, the feature subset sampling mechanism trains the local decision tree, calculates the predicted values ​​of the leaf nodes, and calculates the gradient error value of each leaf node in combination with the learning rate.

[0027] The gradient error values ​​are aggregated according to the full reduction operation mechanism to generate a node gradient error matrix. The matrix is ​​then compared and judged one by one with the preset node error threshold range to calculate the error compliance value.

[0028] If the error compliance value is less than the preset compliance threshold, the feature subset sampling mechanism optimizes the global decision weight matrix of the decision tree by calling the learning rate based on the node gradient error matrix to obtain the global corrected weight matrix.

[0029] Based on the implicit modeling mechanism of the splitting rules of the distributed tree, the product dataset is associated with features to generate a hybrid feature association formula, and the local feature is filtered for each decision tree by combining the global correction weight matrix and the local feature weight is calculated.

[0030] Based on the tree splitting gain mechanism and regularization function, the local feature weights are corrected and cross-node aligned to obtain a local comprehensive weight matrix. Combined with relevant constraint parameters, supply and demand calculations are performed on each leaf node to obtain local supply and demand simulation values.

[0031] The absolute supply-demand difference is calculated based on the historical supply and demand data of the product, and compared and judged in conjunction with the preset supply and demand error range.

[0032] If the absolute supply-demand difference is within the supply-demand error range, then all the local comprehensive weight matrices are weighted or integrated by voting to obtain the global comprehensive weight matrix and generate the product supply-demand model.

[0033] Based on the product supply and demand model and the entire product lifecycle, supply and demand calculations are performed at each node of the supply chain, and corresponding simulated product supply and demand data are output.

[0034] By adopting the above technical solution, firstly, a time-series feature extraction algorithm (such as tsfresh's statistical features and deep learning features) is used to extract sales time-series feature parameters from the product dataset, and spatial analysis is performed in conjunction with regional distribution. Then, distributed sharding technology (horizontal / vertical sharding) is used to divide the regional time-series feature parameters into feature parameter blocks and distribute them to multiple computing nodes. Decision tree models are deployed on each node, and local decision trees are trained using a feature subset sampling mechanism (similar to XGBoost feature sampling), and the gradient error values ​​of leaf nodes are calculated. The global node gradient error matrix is ​​aggregated through AllReduce, and the global weights are optimized by calling the learning rate η (ΔW=W-ηH) after judging based on the error threshold. The local feature weights are corrected using tree splitting gain and regularization function, and then a local comprehensive weight matrix is ​​generated through cross-node alignment. Supply and demand simulation is performed in conjunction with business constraints and compared with historical data for verification. If the absolute supply and demand difference meets the preset error range, the global comprehensive weight matrix is ​​weighted and integrated to construct a product supply and demand model, and finally, simulated supply and demand data of the entire life cycle supply chain nodes are output. High-precision spatiotemporal feature modeling is achieved by integrating distributed computing and ensemble learning. Prediction robustness is improved through dynamic weight optimization and error feedback mechanisms, and the practicality of supply and demand simulation is ensured by relying on business rule constraints.

[0035] The present invention is further configured such that: the specific steps of training the local decision tree based on the feature parameter block and the feature subset sampling mechanism, calculating the predicted value of the leaf node, and calculating the gradient error value of all the leaf nodes in combination with the learning rate include:

[0036] Based on the preset sampling ratio in the feature subset sampling mechanism, a corresponding number of sample features are extracted from the feature parameter block as a parameter feature subset, and a split node is generated.

[0037] For each sample feature in the parameter feature subset, the continuous values ​​are discretized into histogram intervals bin, and the data samples within each histogram interval are counted. The sum of the first-order gradients Gb and the sum of the second-order gradients Hb;

[0038] ;

[0039] in, For sample feature number, For actual data samples, The cross-entropy loss function;

[0040] The histogram intervals of all the sample features are traversed, and the information gain of each split point is calculated. ;

[0041] ;

[0042] in, This is the left split point in the local decision tree. This is the right split point in the local decision tree. These are leaf nodes in a local decision tree. The regularization coefficient is used.

[0043] The information gain between all the split points is compared, and the sample feature and split point with the largest gain are selected for node splitting. This process is repeated recursively until the stopping condition is met.

[0044] Based on the sum of the first-order gradients Gb and the sum of the second-order gradients Hb, combined with the information gain... Calculate the predicted value for each leaf node. And calculate the cumulative value of the local decision tree. ;

[0045] ;

[0046] in, The leaf node number;

[0047] Based on the cumulative value of the current local decision tree, calculate the gradient error of all sample features, and recalculate the gradient error value of each leaf node in conjunction with the learning rate.

[0048] By adopting the above technical solution, the computational complexity is reduced through feature subset sampling. The sum of the first-order gradient (Gb) and the sum of the second-order gradient (Hb) of samples within each feature interval are discretized using histograms. Based on the principle of maximizing information gain, the optimal split point is selected for recursive node splitting. Finally, the predicted value of the leaf node is calculated and the gradient error is updated by combining the regularization term and the learning rate. This significantly improves the training efficiency of large-scale data (histograms accelerate the evaluation of split points). At the same time, overfitting is suppressed through gradient statistics and regularization, balancing model accuracy and generalization ability.

[0049] The present invention is further configured such that the specific steps for generating a product inventory allocation strategy based on the simulated supply and demand data of the product and product inventory management indicators include:

[0050] Based on historical benchmark supply and demand values ​​combined with seasonal factors, the simulated supply and demand data of products are divided to determine the product supply and demand type and calculate the safety redundancy inventory.

[0051] Based on product value and supply and demand stability, and combined with product inventory management indicators, each product is allocated and graded to determine the priority of product allocation.

[0052] Based on the aforementioned safety redundancy inventory and the supply and demand during the lead time, the product reorder quantity is determined, and the product transportation costs, stockout losses, and unsold risks are assessed. In addition, the product allocation cost is calculated based on the difference between transportation distance and supply and demand.

[0053] Track all products in the warehouse, calculate the product storage age and product stockout rate, and combine this with the product transfer costs to generate a product inventory allocation strategy.

[0054] By adopting the above technical solutions, product supply and demand types are classified and safety redundancy inventory is calculated by combining seasonal decomposition algorithms (such as STL decomposition). ABC classification or Pareto analysis is used to allocate and classify products to determine priorities. Reorder quantity is calculated based on the reorder point model (ROP = average demand × lead time + safety stock). Transfer costs are evaluated using transportation optimization algorithms (such as linear programming or minimum spanning tree algorithm) and quantitative analysis is performed by combining transportation distance with the difference between supply and demand. Inventory turnover rate formula (turnover rate = total sales / average inventory value) is used to track inventory age and stockout rate. Finally, multi-objective optimization algorithms (such as NSGA-II or particle swarm optimization) are used to comprehensively generate inventory allocation strategies based on transfer costs, inventory age, and stockout rate. By accurately classifying supply and demand types, dynamically adjusting priorities, scientifically calculating reorder quantity, and conducting multi-dimensional cost assessments, inventory turnover efficiency is improved while reducing stockout risk and unsold costs, achieving optimal allocation of supply chain resources.

[0055] The present invention is further configured such that: the specific steps for formulating a supply chain collaborative scheduling plan by combining the product inventory allocation strategy with shared enterprise information and collaboratively managing all enterprises in the supply chain include:

[0056] For each product, all enterprises in the supply chain are identified as nodes, a data sharing network is built, and the scope of data sharing, decision-making authority, and cost-sharing mechanism are defined.

[0057] The product simulation supply and demand data of each node in the data sharing network are weighted and calculated to generate a consensus supply and demand plan.

[0058] Based on the consensus supply and demand plan and the supply and demand fluctuation coefficient, the safety redundancy inventory is adjusted in a coordinated manner to obtain a coordinated safety inventory.

[0059] Based on the product inventory allocation strategy, combined with the collaborative safety stock and the distance between warehouses, cascade scheduling is performed on each product, and the integrated economic batch size and product allocation cost are calculated.

[0060] The cost allocation of the product is determined according to the cost sharing mechanism, and a supply chain collaborative scheduling plan is constructed.

[0061] By adopting the above technical solutions, supply chain enterprises are nodeized and a data-sharing network is constructed using graph theory algorithms (such as node clustering). Linear regression or weighted average algorithms are then used to weight macroeconomic indicators and node supply and demand data to generate a consensus supply and demand plan. Dynamic adjustment algorithms (such as those based on safety stock formulas) are used to collaboratively adjust safety stock in conjunction with supply and demand fluctuation coefficients. Furthermore, cascaded scheduling and allocation cost calculations are performed using economic batch size (EOQ) models and transportation optimization algorithms (such as shortest path algorithms or linear programming) combined with inventory allocation strategies, warehouse distances, and collaborative safety stock. Finally, product allocation costs are fairly allocated based on the Shapley Value algorithm from game theory, constructing a supply chain collaborative scheduling plan. The data-sharing network achieves full-link information transparency, dynamically adjusts safety stock to reduce supply and demand fluctuation risks, integrates multi-objective optimization algorithms to improve scheduling efficiency, and leverages the fair allocation mechanism of Shapley Value to enhance enterprise collaboration willingness, thereby minimizing overall supply chain costs and maximizing response speed.

[0062] Secondly, the present invention also provides a cloud computing-based supply chain management system, which adopts the following technical solution:

[0063] A cloud-based supply chain management system includes:

[0064] The data acquisition and management module is used to acquire and normalize multi-channel data of products, combine data acquisition sources to construct several product datasets, and transmit them to cloud storage.

[0065] The supply and demand construction module is used to construct and revise the product supply and demand model based on the product dataset and product market distribution, and output simulated product supply and demand data.

[0066] The inventory management module is used to generate product inventory allocation strategies based on the simulated supply and demand data of the product and product inventory management indicators.

[0067] The collaborative scheduling module is used to collaboratively manage all enterprises in the supply chain and formulate a supply chain collaborative scheduling plan based on the product inventory allocation strategy and shared enterprise information.

[0068] The interface integration module is used for deep integration with internal product-related systems and to develop interfaces for connecting to external market research institutions.

[0069] The visual reporting module is used to display the simulated supply and demand data and real-time supply and demand data of the products of each enterprise in the supply chain, and to generate various supply chain reports.

[0070] By adopting the above technical solutions, ETL data extraction and data cleaning algorithms (such as interpolation and KNN filling) are used to integrate multi-channel product data (historical sales, inventory, production schedules, etc.) in the data acquisition and management module and normalize it to the cloud. The supply and demand construction module uses time series decomposition algorithms (such as STL) and machine learning models (such as XGBoost or random forest) combined with product market distribution to dynamically correct the supply and demand model and output simulated supply and demand data. The inventory management module uses the safety stock formula and economic batch size (EOQ) model to generate inventory allocation strategies. The collaborative scheduling module uses multi-objective optimization algorithms (such as linear programming and NSGA-II) to integrate inventory strategies, enterprise shared information, and warehouse distances, and combines the Shapley value algorithm to fairly allocate allocation costs and formulate collaborative scheduling plans. The interface integration module adopts standardized API design and system docking technology to achieve deep integration of internal and external systems. The visual reporting module uses data visualization tools (such as ECharts and Tableau) to display real-time and simulated supply and demand data and generate multi-dimensional supply chain reports. By automating the entire process to improve data quality and real-time performance, by using dynamic model correction and multi-objective optimization to reduce the risk of supply and demand fluctuations, by combining the fair allocation mechanism of Shapley value to enhance supply chain collaboration, and finally by using visual analysis to assist enterprises in making accurate decisions.

[0071] Thirdly, the present invention also provides an electronic device, comprising:

[0072] One or more processors;

[0073] Memory, used to store one or more programs;

[0074] When the one or more programs are executed by the one or more processors, the one or more processors implement any of the methods described in the above scheme.

[0075] Fourthly, the present invention also provides a storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the cloud computing-based supply chain management method as described above.

[0076] In summary, the beneficial technical effects of the present invention are as follows:

[0077] Accurate forecasting data and optimized inventory management can reduce inventory costs, minimize stockout losses and overstock risks, while optimizing supply chain processes can reduce procurement, transportation and production costs, thereby improving the company's economic efficiency.

[0078] By integrating and analyzing massive amounts of data in the supply chain, we can uncover the value and patterns behind the data, providing data support and decision-making basis for enterprises. This enables enterprise managers to make scientific and reasonable decisions based on objective data, improving the accuracy and timeliness of decisions and avoiding the risks associated with subjective assumptions and blind decision-making.

[0079] By optimizing supply chain networks and operational processes, reducing resource waste and environmental pollution; by rationally planning transportation routes, reducing energy consumption and carbon emissions during transportation; and by optimizing inventory management, reducing the environmental impact of inventory backlog, companies can establish a positive social image and enhance their sustainable development capabilities. Attached Figure Description

[0080] Figure 1 This is a schematic flowchart of a supply chain management method according to one embodiment of the present invention.

[0081] Figure 2 This is a flowchart illustrating step S2 in a supply chain management method according to one embodiment of the present invention.

[0082] Figure 3 This is a flowchart illustrating step S3 in a supply chain management method according to one embodiment of the present invention.

[0083] Figure 4 This is a schematic diagram of the supply chain management system according to one embodiment of the present invention. Detailed Implementation

[0084] The present invention will be further described in detail below with reference to the accompanying drawings.

[0085] Reference Figure 1The present invention discloses a cloud computing-based supply chain management method, comprising:

[0086] S1: Acquire and normalize multi-channel product data, combine data collection sources to construct several product datasets, and transmit them to the cloud;

[0087] S2: Based on the product dataset and product market distribution, construct and revise the product supply and demand model, and output simulated product supply and demand data;

[0088] S3: Generate a product inventory allocation strategy based on the simulated supply and demand data of the product and the product inventory management indicators;

[0089] S4: Based on the product inventory allocation strategy and combined with shared enterprise information, collaboratively manage all enterprises in the supply chain and formulate a supply chain collaborative scheduling plan.

[0090] The implementation principle of this embodiment is as follows: Product data is collected from multiple channels (such as online stores, offline stores, wholesalers, etc.) and normalized based on a multimodal database to construct a standardized product dataset, which is then transmitted to the cloud. Combining product market distribution characteristics with checks on data integrity, consistency, and timeliness, the product supply and demand model is dynamically revised to output accurate simulated supply and demand data. Furthermore, core indicators such as inventory turnover rate, stockout rate, and safety stock are integrated to generate product inventory allocation strategies. Finally, relying on shared enterprise information (such as supplier capacity, logistics network, and distributor demand), cost-sharing algorithms (such as Shapley value) and dynamic optimization techniques (such as reinforcement learning) are used to formulate a collaborative scheduling plan for the entire supply chain. Based on multi-source data fusion, through model iteration and strategy generation, combined with real-time shared information and collaborative algorithms, a closed-loop management system is formed from data collection, demand forecasting, inventory optimization to global scheduling, thereby improving supply chain response efficiency and cost control capabilities.

[0091] Step S1 includes:

[0092] Extract product-related data from the company's internal product-related systems and external market research institutions to obtain historical sales data, inventory data, product production schedules, market sales change tables, and customer order volumes, and mark the corresponding product source identifiers;

[0093] Based on the business logic, noise removal, missing data filling, and duplicate checking are performed on all product-related data to obtain duplicate product data.

[0094] Based on the information type, all the aforementioned weightless product data are normalized to obtain the corresponding standard product data;

[0095] Based on the information collection source, all the standard product data are classified and aggregated to construct several product datasets;

[0096] The entire product dataset is divided according to a preset data segmentation mechanism to obtain several product data blocks, and the segmentation timestamps are marked.

[0097] Based on the block timestamps and the parallel transmission mechanism, the product data blocks are transmitted to the cloud sequentially.

[0098] The product data blocks are aggregated in the cloud based on the product source identifier to obtain the product dataset.

[0099] The implementation principle of this embodiment is as follows: Historical sales data, inventory data, production schedules, market sales change tables, and customer order volumes are extracted from internal enterprise product systems (such as ERP and CRM) and external market research institutions (such as industry reports and social media sentiment analysis), and the data is labeled based on a unique product source identifier; then, noise filtering algorithms (such as IsolationForest), missing value imputation algorithms (such as KNN interpolation and time series ARIMA prediction), and duplicate checking algorithms (such as hash deduplication) are used to clean the data, generating a dataset of products without duplicates; then, normalization algorithms (such as Min-Max standardization and Z-Score standardization) are applied to different information types (such as sales volume and inventory volume) to eliminate differences in units. Standardized product data is then generated. Based on the data collection sources (such as internal systems and third-party platforms), classification and aggregation algorithms (such as rule-based grouping or K-means clustering) are used to construct a multi-dimensional product dataset. The dataset is then segmented using a pre-defined data segmentation mechanism (such as time window segmentation or hash segmentation) and the segment timestamps are marked. Combined with parallel transmission algorithms (such as MapReduce distributed transmission), the data blocks are uploaded to the cloud sequentially. Finally, all data blocks are aggregated in the cloud based on the product source identifier using data fusion algorithms (such as hash mapping or database primary key association) to reconstruct the complete product dataset. This forms a closed-loop process from data extraction, cleaning, standardization to cloud integration, providing a high-quality, multi-source collaborative unified data foundation for subsequent supply and demand models and inventory optimization.

[0100] Reference Figure 2 Step S2 includes:

[0101] A: Extract features from the product dataset based on the time series data to obtain the time series feature parameters of sales.

[0102] B: Analyze the sales time-series characteristic parameters based on the product market distribution to obtain the regional time-series characteristic parameters;

[0103] C: Based on the distributed computing rules, the regional time-series feature parameters are horizontally or vertically partitioned to obtain several feature parameter blocks and generate several data computing nodes;

[0104] D: Deploy a decision tree for each data computing node, construct a distributed tree, and generate a feature subset sampling mechanism by combining the preset number of iterations and learning rate;

[0105] E: Based on the feature parameter block, the feature subset sampling mechanism trains the local decision tree, calculates the predicted values ​​of the leaf nodes, and calculates the gradient error values ​​of each leaf node in combination with the learning rate;

[0106] F: Aggregate all the gradient error values ​​according to the full reduction operation mechanism to generate a node gradient error matrix;

[0107] G: The node gradient error matrix is ​​compared and judged one by one according to the preset node error threshold range, and the error compliance value is calculated.

[0108] If the error compliance value is less than the preset compliance threshold, the feature subset sampling mechanism optimizes the global decision weight matrix of the decision tree by calling the learning rate based on the node gradient error matrix to obtain the global corrected weight matrix.

[0109] H: Based on the implicit modeling mechanism of the splitting rules of the distributed tree, feature association is performed on the product dataset to generate a hybrid feature association formula;

[0110] I: Based on the global corrected weight matrix and the hybrid feature correlation formula, perform local feature filtering on each decision tree and calculate the local feature weights;

[0111] J: Based on the tree splitting gain mechanism and the regularization function, the local feature weights are corrected, and the feature correction weights are calculated.

[0112] K: Based on the product supply and demand pattern, the weights of all the aforementioned features are adjusted and aligned across nodes to obtain a local comprehensive weight matrix;

[0113] L: Based on the enterprise's business supply and demand and relevant constraint parameters, the supply and demand calculations are performed on each leaf node through the local comprehensive weight matrix to obtain the local supply and demand simulation values;

[0114] M: Calculate the absolute supply-demand difference based on the historical supply and demand data of the product, and compare and judge it with the preset supply and demand error range;

[0115] If the absolute supply-demand difference is within the supply-demand error range, then all the local comprehensive weight matrices are weighted or integrated by voting to obtain the global comprehensive weight matrix and generate the product supply-demand model.

[0116] N: Based on the product supply and demand model and the entire product lifecycle, calculate the supply and demand for each node in the supply chain and output the corresponding simulated product supply and demand data.

[0117] The implementation principle of this embodiment is as follows: Based on time series feature extraction algorithms (such as statistical feature calculation, Fourier transform, or wavelet analysis), sales time series feature parameters (such as mean, variance, and trend term) are parsed from the product dataset. Then, combined with product market distribution information (such as regional sales share and channel penetration rate), regional time series feature parameters are generated through spatial correlation models (such as geographic weighted regression or spatial autocorrelation analysis). Subsequently, according to distributed computing rules (such as horizontal partitioning by time / region and vertical partitioning by feature type), the regional time series feature parameters are divided into feature parameter blocks and distributed to multiple data computing nodes. Decision trees (such as XGBoost or Lig) are deployed on each node. The distributed version of htGBM constructs a feature subset sampling mechanism (such as random feature selection in random forests or column sampling in gradient boosting trees) by presetting the number of iterations (e.g., 100 rounds) and the learning rate (η, controlling the contribution of each tree to the global model, usually set to 0; 01~0; 3). It trains local decision trees using feature parameter blocks and the sampling mechanism, calculates the predicted values ​​of leaf nodes (e.g., through negative gradient approximation optimization in gradient boosting), and calculates the gradient error value in conjunction with the learning rate (e.g., the gradient error formula in gradient boosting). It then aggregates all gradient error values ​​through full reduction operations (e.g., the AllReduce algorithm) to generate a node gradient error matrix (recording the error distribution of each leaf node). The system is structured as follows: A node error threshold range is defined (e.g., 0.01 to 0.1, to measure whether the error is within an acceptable range). If the error compliance value (i.e., the error value of each node in the node gradient error matrix) is less than the compliance threshold (e.g., 0.05), a dynamic learning rate adjustment mechanism (e.g., AdaBoost or adaptive learning rate strategy) is invoked to optimize the global decision weight matrix of the decision tree (e.g., the update direction and magnitude of model parameters), generating a global corrected weight matrix. Furthermore, implicit modeling based on the splitting rules of the distributed tree (e.g., splitting strategies based on information gain or Gini index) is used to extract feature associations (e.g., interaction terms between time series and regional features) from the product dataset, generating hybrid feature relationships. The process involves: combining a global modified weight matrix with a hybrid feature correlation formula to perform local feature filtering on each decision tree (e.g., ranking based on feature importance), calculating local feature weights (e.g., weighted values ​​of split gain); modifying local feature weights using tree split gain mechanisms (e.g., maximizing information gain) and regularization functions (e.g., L1 / L2 regularization or complexity penalty terms) to obtain modified feature weights; aligning modified feature weights across nodes based on product supply and demand patterns (e.g., correlation models of demand-inventory-supply cycles) (e.g., aligning time series features from different warehouses or channels), and generating a local comprehensive weight matrix (recording the modified feature weights of each node).Finally, combining enterprise business supply and demand constraints (such as maximum transportation capacity and minimum safety stock) and local comprehensive weight matrices, a global comprehensive weight matrix is ​​generated through weighted operations (such as linear weighting or voting integration). This constructs a product supply and demand model (such as a gradient boosting regression tree or a deep learning model), and performs cross-node supply and demand calculations based on the dynamic demand changes throughout the product's entire lifecycle (such as introduction, growth, maturity, and decline phases). Simulated supply and demand data is output, enabling collaborative decision-making throughout the entire process from feature extraction to model optimization.

[0118] Step E includes:

[0119] Based on the preset sampling ratio in the feature subset sampling mechanism, a corresponding number of sample features are extracted from the feature parameter block as a parameter feature subset, and a split node is generated.

[0120] For each sample feature in the parameter feature subset, the continuous values ​​are discretized into histogram intervals bin, and the data samples within each histogram interval are counted. The sum of the first-order gradients Gb and the sum of the second-order gradients Hb;

[0121] ;

[0122] in, For sample feature number, For actual data samples, The cross-entropy loss function;

[0123] The histogram intervals of all the sample features are traversed, and the information gain of each split point is calculated. ;

[0124] ;

[0125] in, This is the left split point in the local decision tree. This is the right split point in the local decision tree. These are leaf nodes in a local decision tree. The regularization coefficient is used.

[0126] The information gain between all the split points is compared, and the sample feature and split point with the largest gain are selected for node splitting. This process is repeated recursively until the stopping condition is met.

[0127] Based on the sum of the first-order gradients Gb and the sum of the second-order gradients Hb, combined with the information gain... Calculate the predicted value for each leaf node. And calculate the cumulative value of the local decision tree. ;

[0128] ;

[0129] in, The leaf node number;

[0130] Based on the cumulative value of the current local decision tree, calculate the gradient error of all sample features, and recalculate the gradient error value of each leaf node in conjunction with the learning rate.

[0131] The implementation principle of this embodiment is as follows: A feature subset is extracted from the feature parameter block using a feature subset sampling mechanism (e.g., the column sampling ratio is preset to 0; 8, meaning 80% of the features are randomly selected each time). Continuous sample features (e.g., sales figures) in the subset are discretized and divided into histogram interval bins (default 256 bins). The sum of the first and second gradients of the data samples within each bin is calculated. All feature split points are traversed, and the information gain is calculated. The split point with the largest gain is selected for node splitting, and this process is recursively executed until the stopping condition is met (e.g., the tree depth reaches a preset value of 6, or the number of leaf node samples is less than the minimum value of 100). The predicted value of the leaf node and the cumulative value of the local decision tree are calculated based on the gradient statistics. Finally, the sample gradient error is recalculated based on the cumulative value, and the gradient error value of the leaf node is updated to drive subsequent iterative optimization.

[0132] Reference Figure 3 Step S3 includes:

[0133] S31: Based on historical benchmark supply and demand values ​​and seasonal factors, classify the simulated supply and demand data of products, determine the product supply and demand type, and calculate the safety redundancy inventory.

[0134] S32: Based on product value and supply and demand stability, and combined with product inventory management indicators, allocate and classify each product to determine the priority of product allocation;

[0135] S33: Determine the product reorder quantity based on the aforementioned safety redundancy inventory and the supply and demand during the lead time period;

[0136] S34: Assess the transportation costs, stockout losses, and unsold inventory risks of the products based on the reorder quantity, and calculate the product transfer costs by combining the transportation distance and the difference between supply and demand.

[0137] S35: Track all products in the warehouse and calculate the product storage age and product stockout rate;

[0138] S36: Generate a product inventory allocation strategy based on the product transfer cost, the product storage age, and the product stockout rate.

[0139] The implementation principle of this embodiment is as follows: Based on historical benchmark supply and demand values ​​and the Holt-Winters seasonal forecasting model, products are classified to determine whether they belong to stable (stable demand), seasonal (demand fluctuates with a specific cycle), impulsive (demand bursts intermittently), or stochastic (demand fluctuations are irregular). Then, the safety redundancy inventory is calculated as Z-score (service level) * standard deviation of demand fluctuation * (supply lead time). 1 / 2 The formula calculates the safety margin inventory; then, a multi-objective optimization algorithm (such as the linear weighted method) is used to comprehensively consider product value (such as gross profit margin), supply and demand stability (such as demand fluctuation coefficient), and inventory management indicators (such as turnover days) to assign a priority value to each product (such as priority = α * gross profit margin + β * (1 / turnover days) - γ * stockout rate), where α, β, and γ are weighting coefficients; next, the reorder quantity is calculated through a regression model, and cost-benefit analysis is used in conjunction with the transportation cost function (allocation cost = unit transportation cost * transportation distance * demand + stockout loss cost * stockout probability + slow-moving risk cost * (allocation quantity - predicted sales volume)) for evaluation; finally, the real-time statistical product storage age (i.e., inventory days) and stockout rate are used as dynamic constraints, and the final inventory allocation strategy is generated through a constraint optimization algorithm (such as the Lagrange multiplier method), realizing closed-loop optimization from risk identification, cost assessment to dynamic decision-making.

[0140] Step S4 includes:

[0141] For each product, all enterprises in the supply chain are identified as nodes, a data sharing network is built, and the scope of data sharing, decision-making authority, and cost-sharing mechanism are defined.

[0142] Based on macroeconomic indicators, the simulated supply and demand data of products at each node in the data sharing network are weighted and calculated to generate a consensus supply and demand plan.

[0143] Based on the consensus supply and demand plan and the supply and demand fluctuation coefficient, the safety redundancy inventory is adjusted in a coordinated manner to obtain a coordinated safety inventory.

[0144] Based on the product inventory allocation strategy, combined with the collaborative safety stock and the distance between warehouses, cascade scheduling is performed on each product, and the integrated economic batch size and product allocation cost are calculated.

[0145] Based on the cost-sharing mechanism and the Shapley value, the product allocation cost is allocated to construct a supply chain collaborative scheduling plan.

[0146] The implementation principle of this embodiment is as follows: By treating supply chain enterprises as nodes (e.g., modeling suppliers, manufacturers, and distributors as nodes in a graph network) and constructing a data sharing network (e.g., based on graph neural networks or supply chain network topology), the scope of data sharing (e.g., field-level access control), decision-making authority (e.g., role-based access control, RBAC), and cost-sharing mechanisms (e.g., contribution-based allocation rules) are defined. Combined with macroeconomic indicators (e.g., GDP growth rate, industry prosperity index), fuzzy comprehensive evaluation or weighted average algorithms (weights can be determined by industry experience or historical data importance analysis) are applied to the simulated supply and demand data of each node to generate a consensus supply and demand plan. Subsequently, a dynamic adjustment algorithm (e.g., an error correction model based on time series forecasting) is used in conjunction with supply and demand fluctuations. The volatility coefficient (defined as the variance or standard deviation of historical demand and supply, measuring market uncertainty) is used to collaboratively adjust safety stock levels (e.g., introducing a volatility coefficient to adjust the safety factor Z value in the safety stock formula); based on inventory allocation strategies (e.g., greedy algorithms or linear programming) and integrating economic order quantity (EOQ model) and allocation cost calculation (e.g., transportation cost function = unit freight rate * distance * allocation quantity + stockout loss * stockout probability), products are cascaded and scheduled; finally, the Sharpe ratio (a game theory measure of each node's contribution to cost) is used to fairly allocate allocation costs, generating a supply chain collaborative scheduling plan, forming a full-process collaborative decision-making process from network construction, data weighting, inventory adjustment to cost allocation, improving the overall supply chain response efficiency and the fairness of cost allocation.

[0147] Reference Figure 4 A cloud-based supply chain management system, applied to the aforementioned fault detection method, includes:

[0148] The data acquisition and management module is used to acquire and normalize multi-channel data of products, combine data acquisition sources to construct several product datasets, and transmit them to cloud storage.

[0149] The supply and demand construction module is used to construct and revise the product supply and demand model based on the product dataset and product market distribution, and output simulated product supply and demand data.

[0150] The inventory management module is used to generate product inventory allocation strategies based on the simulated supply and demand data of the product and product inventory management indicators.

[0151] The collaborative scheduling module is used to collaboratively manage all enterprises in the supply chain and formulate a supply chain collaborative scheduling plan based on the product inventory allocation strategy and shared enterprise information.

[0152] The interface integration module is used for deep integration with internal product-related systems and to develop interfaces for connecting to external market research institutions.

[0153] The visual reporting module is used to display the simulated supply and demand data and real-time supply and demand data of the products of each enterprise in the supply chain, and to generate various supply chain reports.

[0154] The implementation principle of this embodiment is as follows: Based on multi-source data acquisition and normalization algorithms (such as Min-Max standardization and Z-Score standardization), product data (such as historical sales, inventory, and production plans) is extracted and integrated from internal enterprise systems (ERP, MES) and external market research institutions to construct a multi-dimensional product dataset and upload it to the cloud through a parallel transmission mechanism; subsequently, time series forecasting (such as ARIMA and Prophet) and machine learning models (such as random forest and gradient boosting tree) are used to generate a dynamic product supply and demand model and output simulated supply and demand data; the inventory management module combines supply and demand data with inventory management indicators (such as safety stock threshold Ss=T*σ*L) 1 / 2 The supply chain coordination module formulates allocation strategies based on shared information (such as real-time inventory and production plans) and uses the Shapley Value algorithm from game theory to fairly allocate allocation costs (measuring the contribution of each node to the cost) and generate a supply chain coordination plan. The interface integration module realizes data linkage between internal systems (ERP, WMS) and external partner systems through APIs or middleware technologies (such as ETL tools). The visual reporting module uses data visualization algorithms (such as chart rendering in Tableau or Power BI) and statistical analysis methods (such as mean squared error (MSE) and mean absolute error (MAE)) to display simulated and real-time supply and demand data and generate supply chain performance reports, forming a closed loop from data collection, model building, inventory optimization to coordination scheduling, thereby improving the intelligent decision-making and resource allocation efficiency of the supply chain. Example

[0155] Data is collected from multiple channels, including the company's internal ERP system, sales system, production system, and external market research institutions and partners, including historical sales data, inventory data, production plans, market trends, and customer orders.

[0156] The collected data is cleaned to remove noise and outliers, handle missing and duplicate values, and undergo preprocessing operations such as standardization and normalization to improve data quality and provide a reliable foundation for subsequent analysis and prediction.

[0157] Establish an efficient data storage architecture, such as a data warehouse or database, to classify, store, and manage processed data for quick retrieval and access;

[0158] By employing statistical methods, machine learning algorithms, and deep learning technologies, demand forecasting models suitable for the characteristics of enterprise products and markets can be constructed, such as time series analysis models, regression models, neural network models, and random forest models, in order to achieve demand forecasting for different products, different regions, and different time periods.

[0159] Based on the company's business needs and data characteristics, set relevant parameters for the forecasting model, such as forecast period, confidence interval, seasonal factors, and trend factors, to improve the accuracy and reliability of the forecast.

[0160] Establish an evaluation index system for forecast results, such as mean squared error, mean absolute error, and accuracy, to evaluate and verify the forecast results, and adjust and optimize the forecast model in a timely manner based on actual sales data and market changes in order to continuously improve forecast accuracy.

[0161] Based on demand forecasts and the company's inventory management objectives, formulate reasonable inventory strategies, such as safety stock strategies, economic order quantity strategies, and periodic inventory count strategies, to balance inventory costs and stockout risks.

[0162] Real-time monitoring of inventory levels; timely warning signals are issued when inventory levels fall below safety stock or exceed the set upper limit, reminding relevant personnel to replenish stock or adjust inventory strategies.

[0163] By analyzing and mining inventory data, we can optimize inventory layout and allocation plans, such as rationally allocating inventory to different warehouses, determining the best replenishment points and replenishment quantities, in order to improve inventory turnover and capital utilization.

[0164] Establish an information sharing platform among upstream and downstream enterprises in the supply chain to enable real-time information exchange and sharing among suppliers, manufacturers, distributors, retailers, and other links, including demand information, inventory information, production plans, and logistics information, in order to improve the collaborative efficiency and response speed of the supply chain.

[0165] Based on shared information, we develop collaborative plans and scheduling schemes for the supply chain, coordinate the production, procurement, and distribution activities of various enterprises, and ensure that products can be delivered to customers on time, in the correct quantity, and with the correct quality, while reducing inventory backlog and stockouts.

[0166] Evaluate, select and manage partners in the supply chain, establish long-term and stable cooperative relationships, and incentivize partners to jointly optimize supply chain processes and improve the overall competitiveness of the supply chain by signing cooperation agreements, sharing benefits and risks.

[0167] Deeply integrate the supply chain FCST system with other internal information systems of the enterprise, such as ERP, MES, WMS, etc., to achieve seamless data flow and automated connection of business processes, avoid information silos and duplication of work, and improve the operational efficiency of the enterprise.

[0168] Develop interfaces with external partner systems, such as supplier procurement systems and logistics provider transportation management systems, to enable information exchange and business collaboration with external enterprises and expand the coverage and influence of the supply chain.

[0169] By using data visualization technologies, such as bar charts, line charts, pie charts, and maps, key data and information in the supply chain can be displayed in an intuitive and easy-to-understand way, enabling enterprise managers to understand the operation status of the supply chain at a glance and identify problems and potential risks in a timely manner.

[0170] Based on the management needs of enterprises, it automatically generates various supply chain reports, such as demand forecast reports, inventory reports, procurement reports, and sales reports, and provides data analysis and mining functions to help managers gain a deeper understanding of supply chain performance and problems, providing strong support for decision-making.

[0171] An electronic device, comprising:

[0172] One or more processors;

[0173] Memory, used to store one or more programs;

[0174] When the one or more programs are executed by the one or more processors, the one or more processors implement any of the methods described in the above scheme.

[0175] A storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement a cloud-based supply chain management method as described above.

[0176] The embodiments described herein are preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape, and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A cloud computing-based supply chain management method, characterized in that, include: Acquire and normalize multi-channel product data, construct several product datasets by combining data collection sources, and transmit them to the cloud; The specific steps for constructing and refining the product supply and demand model based on the product dataset and product market distribution, and outputting simulated product supply and demand data, include: Based on the time series and product market distribution, feature extraction is performed on the product dataset to obtain regional time series feature parameters. Then, based on distributed computing rules, horizontal or vertical sharding is performed to obtain several feature parameter blocks and generate several data computing nodes. Decision trees are deployed on each data computing node to build a distributed tree, and a feature subset sampling mechanism is generated by combining the preset number of iterations and learning rate. The specific steps of training the local decision tree based on the feature parameter block, the feature subset sampling mechanism, calculating the predicted values ​​of the leaf nodes, and calculating the gradient error values ​​of all the leaf nodes in conjunction with the learning rate include: Based on the preset sampling ratio in the feature subset sampling mechanism, a corresponding number of sample features are extracted from the feature parameter block as a parameter feature subset, and a split point is generated. For each sample feature in the parameter feature subset, the continuous values ​​are discretized into histogram intervals, and the sum of the first-order gradients Gb and the sum of the second-order gradients Hb of the data samples in each histogram interval are calculated. The histogram intervals of all the sample features are traversed, and the information gain Gain of each split point is calculated. Where left is the left split point in the local decision tree, right is the right split point in the local decision tree, leaf is the leaf node in the local decision tree, and λ is the regularization coefficient; The information gain between all the split points is compared, and the sample feature and split point with the largest gain are selected for node splitting. This process is repeated recursively until the stopping condition is met. Based on the sum of the first-order gradients Gb and the sum of the second-order gradients Hb, combined with the information gain Gain, the predicted value w for each leaf node is calculated. j And calculate the cumulative value of the local decision tree. Where j is the leaf node number; Based on the cumulative value of the current local decision tree, calculate the gradient error of all sample features, and recalculate the gradient error value of each leaf node in combination with the learning rate. The gradient error values ​​are aggregated according to the full reduction operation mechanism to generate a node gradient error matrix. The matrix is ​​then compared and judged one by one with the preset node error threshold range to calculate the error compliance value. If the error compliance value is less than the preset compliance threshold, the feature subset sampling mechanism optimizes the global decision weight matrix of the decision tree by calling the learning rate based on the node gradient error matrix to obtain the global corrected weight matrix. Based on the simulated supply and demand data of the product and the product inventory management indicators, a product inventory allocation strategy is generated. Based on the product inventory allocation strategy and shared information among enterprises, a collaborative supply chain scheduling plan is formulated to manage all enterprises in the supply chain.

2. The cloud-based supply chain management method according to claim 1, characterized in that, The specific steps for acquiring and normalizing multi-channel product data, constructing several product datasets based on data collection sources, and transmitting them to the cloud include: Extract product-related data from the company's internal product-related systems and external market research institutions to obtain historical sales data, inventory data, product production schedules, market sales change tables, and customer order volumes, and mark the corresponding product source identifiers; Based on the information collection sources, all product-related data are classified and aggregated to construct several product datasets; The entire product dataset is divided according to a preset data segmentation mechanism to obtain several product data blocks, and the segmentation timestamps are marked. Based on the block timestamps and the parallel transmission mechanism, the product data blocks are transmitted to the cloud sequentially. The product data blocks are aggregated in the cloud based on the product source identifier to obtain the product dataset.

3. The supply chain management method based on cloud computing according to claim 1, characterized in that, The specific steps of constructing and refining the product supply and demand model based on the product dataset and product market distribution, and outputting simulated product supply and demand data, further include: Based on the implicit modeling mechanism of the splitting rules of the distributed tree, the product dataset is associated with features to generate a hybrid feature association formula, and the local feature is filtered for each decision tree by combining the global correction weight matrix and the local feature weight is calculated. Based on the tree splitting gain mechanism and regularization function, the local feature weights are corrected and cross-node aligned to obtain a local comprehensive weight matrix. Combined with relevant constraint parameters, supply and demand calculations are performed on each leaf node to obtain local supply and demand simulation values. The absolute supply-demand difference is calculated based on the historical supply and demand data of the product, and compared and judged in conjunction with the preset supply and demand error range. If the absolute supply-demand difference is within the supply-demand error range, then all the local comprehensive weight matrices are weighted or integrated by voting to obtain the global comprehensive weight matrix and generate the product supply-demand model. Based on the product supply and demand model and the entire product lifecycle, supply and demand calculations are performed at each node of the supply chain, and corresponding simulated product supply and demand data are output.

4. The cloud computing-based supply chain management method according to claim 1, characterized in that, The specific steps for generating a product inventory allocation strategy based on the simulated supply and demand data and product inventory management indicators include: Based on historical benchmark supply and demand values ​​combined with seasonal factors, the simulated supply and demand data of products are divided to determine the product supply and demand type and calculate the safety redundancy inventory. Based on product value and supply and demand stability, and combined with product inventory management indicators, each product is allocated and graded to determine the priority of product allocation. Based on the aforementioned safety redundancy inventory and the supply and demand during the lead time, the product reorder quantity is determined, and the product transportation costs, stockout losses, and unsold risks are assessed. In addition, the product allocation cost is calculated based on the difference between transportation distance and supply and demand. Track all products in the warehouse, calculate the product storage age and product stockout rate, and combine this with the product transfer costs to generate a product inventory allocation strategy.

5. The cloud-based supply chain management method according to claim 1, characterized in that, The specific steps for formulating a supply chain collaborative scheduling plan based on the product inventory allocation strategy and shared enterprise information, and collaboratively managing all enterprises in the supply chain, include: For each product, all enterprises in the supply chain are identified as nodes, a data sharing network is built, and the scope of data sharing, decision-making authority, and cost-sharing mechanism are defined. The product simulation supply and demand data of each node in the data sharing network are weighted and calculated to generate a consensus supply and demand plan. Based on the consensus supply and demand plan and the supply and demand fluctuation coefficient, the safety redundancy inventory is adjusted in a coordinated manner to obtain a coordinated safety inventory. Based on the product inventory allocation strategy, combined with the collaborative safety stock and the distance between warehouses, cascade scheduling is performed on each product, and the integrated economic batch size and product allocation cost are calculated. The cost allocation of the product is determined according to the cost sharing mechanism, and a supply chain collaborative scheduling plan is constructed.

6. A cloud-based supply chain management system applied to the method of any one of claims 1 to 5, characterized in that, include: The data acquisition and management module is used to acquire and normalize multi-channel data of products, combine data acquisition sources to construct several product datasets, and transmit them to cloud storage. The supply and demand construction module is used to construct and revise the product supply and demand model based on the product dataset and product market distribution, and output simulated product supply and demand data. The inventory management module is used to generate product inventory allocation strategies based on the simulated supply and demand data of the product and product inventory management indicators. The collaborative scheduling module is used to collaboratively manage all enterprises in the supply chain and formulate a supply chain collaborative scheduling plan based on the product inventory allocation strategy and shared enterprise information. The interface integration module is used for deep integration with internal product-related systems and to develop interfaces for connecting to external market research institutions. The visual reporting module is used to display the simulated supply and demand data and real-time supply and demand data of the products of each enterprise in the supply chain, and to generate various supply chain reports.

7. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-5.

8. A storage medium storing at least one instruction, at least one program, a code set, or an instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the cloud-based supply chain management method as described in any one of claims 1 to 5.

Images