Intelligent inventory distribution method and system based on multi-platform sales volume prediction
By constructing a multi-dimensional time series dataset and a deep learning model, combined with supply chain cost parameters, the problems of prediction bias and static scheduling in multi-platform inventory allocation were solved, realizing the dynamic adjustment and refined configuration of intelligent inventory, and improving prediction accuracy and supply chain agility.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- AGZZX OPTOELECTRONICS TECH CO LTD
- Filing Date
- 2026-01-08
- Publication Date
- 2026-04-17
AI Technical Summary
Existing inventory allocation methods cannot effectively capture non-linear sales pulses in multi-platform sales environments, resulting in large forecasting biases and a lack of dynamic scheduling mechanisms with a global perspective, leading to stockouts or inventory backlogs, increasing warehousing costs and lost sales opportunities.
By constructing a multidimensional time series dataset, performing preprocessing and feature engineering, and using a deep learning prediction model to output sales forecasts, a multi-objective inventory optimization model is constructed by combining supply chain cost parameters, and deviations are monitored in real time to trigger dynamic adjustments, thereby achieving intelligent inventory allocation.
It significantly improves forecasting accuracy, enables global intelligent coordination and refined allocation of inventory resources, reduces warehousing costs and ineffective cross-warehouse transfer fees, and empowers the supply chain with agile response capabilities.
Smart Images
Figure CN121882373A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the fields of supply chain management and artificial intelligence technology, and in particular relates to an intelligent inventory allocation method and system based on multi-platform sales forecasting. Background Technology
[0002] With the increasing popularity of omnichannel retail, merchants typically need to sell on multiple e-commerce platforms such as Tmall, JD.com, and Douyin. Currently, existing inventory allocation methods mainly rely on manual experience or static rules based on historical average sales. Companies usually use ERP systems to record historical sales, estimate future demand using a simple weighted average method, and pre-allocate inventory to regional distribution centers or forward warehouses for each channel.
[0003] However, existing technologies have significant shortcomings in dealing with the complex and ever-changing e-commerce environment. First, linear statistical forecasting methods cannot effectively capture non-linear sales pulses caused by live-streaming e-commerce, platform promotions, or sudden public opinion crises, resulting in large forecasting errors. Second, inventory allocation across platforms is often static and fragmented, lacking a dynamic scheduling mechanism based on a global perspective. When a surge in traffic on one platform leads to stockouts, other platforms may be facing inventory backlogs. This supply-demand mismatch not only increases warehousing costs but also results in significant lost sales opportunities. Therefore, this invention proposes a technical solution for an intelligent inventory allocation method and system based on multi-platform sales forecasting to address the aforementioned problems. Summary of the Invention
[0004] The purpose of this invention is to provide an intelligent inventory allocation method and system based on multi-platform sales forecasting, aiming to solve the problems mentioned in the background art.
[0005] This invention is implemented as follows: On one hand, a smart inventory allocation method based on multi-platform sales forecasting, the method comprising:
[0006] Acquire historical business data and external influencing factor data of the target product on multiple sales platforms, and construct a multidimensional time series dataset;
[0007] The multidimensional time series dataset is preprocessed and feature engineering is performed to generate a standardized feature vector matrix containing multiple influencing factors.
[0008] The standardized feature vector matrix is input into a pre-trained deep learning prediction model, which outputs the sales forecast values for each sales platform within a preset future period.
[0009] Based on the sales forecast, a multi-objective inventory optimization model is constructed by combining supply chain cost parameters, and the inventory allocation scheme for each platform is obtained by solving the model.
[0010] The inventory allocation plan is executed, and the deviation between actual sales and forecast values is monitored in real time. Dynamic adjustments are triggered when preset conditions are met.
[0011] As a further aspect of the present invention, the preprocessing and feature engineering operations performed on the multidimensional time series dataset to generate a standardized feature vector matrix containing multiple influencing factors specifically include:
[0012] The isolated forest algorithm is used to identify and remove abnormal noise points in the historical business data, and the Lagrange interpolation method is used to fill in the missing data.
[0013] The cleaned data is normalized to map data of different dimensions to a unified range;
[0014] Extract time features, including periodic factors of day, week, and month, and holiday markers; extract marketing features, including product discount rates, number of live broadcasts, and advertising investment; extract competitive features, including price fluctuation indices of similar competing products.
[0015] The extracted time features and competition features are concatenated into vectors to generate the standardized feature vector matrix.
[0016] As a further aspect of the present invention, the step of inputting the standardized feature vector matrix into a pre-trained deep learning prediction model and outputting the sales forecast values for each sales platform within a preset future period specifically includes:
[0017] A long short-term memory network based on an attention mechanism is constructed as a prediction model, in which the LSTM layer is used to capture the temporal dependence of long sequences, and the attention mechanism layer is used to calculate the weights of the impact of features at different time steps on future sales.
[0018] The standardized feature vector matrix is then segmented by a time sliding window and input into the prediction model;
[0019] The model output is decoded by a fully connected layer to obtain the quantile regression prediction results for each sales platform, and the output includes the predicted mean and confidence interval of the sales forecast.
[0020] As a further aspect of the present invention, the step of constructing a multi-objective inventory optimization model based on the sales forecast value and combining it with supply chain cost parameters, and solving for the inventory allocation scheme for each platform specifically includes:
[0021] The objective function is set as minimizing the total supply chain cost, which includes holding costs, stockout loss costs, cross-warehouse transfer costs, and last-mile delivery costs.
[0022] Set constraints, including hard constraints on total inventory, minimum safety stock levels for each platform, and maximum warehouse throughput.
[0023] Substitute the sales forecast and the current inventory status of each warehouse into the multi-objective inventory optimization model;
[0024] The model is iteratively solved using a mixed-integer linear programming algorithm or a particle swarm optimization algorithm to output the optimal replenishment quantity and allocation path for each platform, thus forming the inventory allocation scheme.
[0025] As a further aspect of the present invention, the execution of the inventory allocation scheme and the real-time monitoring of the deviation between actual sales and forecast values, triggering dynamic adjustments when preset conditions are met, specifically includes:
[0026] The inventory allocation plan is converted into an inbound order or transfer order that can be recognized by the warehouse management system and then issued for execution.
[0027] The actual real-time sales of each platform are collected at a preset frequency, and the deviation rate between the actual sales and the predicted sales value is calculated.
[0028] If the deviation rate exceeds a preset threshold, the remaining unallocated inventory is locked. Using the latest actual sales data as input, the sales forecast values for each sales platform in the future preset period are re-output and the inventory allocation scheme for each platform is re-solved to generate a corrected inventory replenishment strategy.
[0029] As a further aspect of the present invention, another option is an intelligent inventory allocation system based on multi-platform sales forecasting, the system comprising:
[0030] The data acquisition module is used to connect to various e-commerce platforms and external databases through API interfaces to obtain historical business data and external influencing factor data of target products, and aggregate the data into a multi-dimensional time series dataset.
[0031] The data processing module is used to receive the multidimensional time series dataset, perform denoising, completion, and normalization on it, and extract multidimensional features for vector concatenation to generate a standardized feature vector matrix.
[0032] The sales forecasting module is used to load a pre-trained deep learning forecasting model, take the standardized feature vector matrix as input, calculate and output the sales forecast value of each sales platform in the future preset period.
[0033] The allocation decision module is used to store supply chain cost parameters and constraints, construct a multi-objective inventory optimization model based on the sales forecast value and solve it to generate inventory allocation schemes for each platform.
[0034] The execution monitoring module is used to issue the inventory allocation plan and calculate the deviation rate between the actual sales and the predicted value in real time. When the deviation rate meets the trigger condition, a recalculation signal is sent to the sales prediction module to generate a corrected replenishment strategy.
[0035] As a further aspect of the present invention, the data processing module specifically includes:
[0036] The anomaly cleaning unit is used to run the Isolation Forest algorithm to identify and remove anomalous noise points in the data, and to fill in missing data using Lagrange interpolation.
[0037] The feature construction unit is used to normalize the cleaned data, extract time period features, marketing activity features and competitive features from the data, and concatenate the features to generate the standardized feature vector matrix that is adapted to the input of the model.
[0038] As a further aspect of the present invention, the sales forecasting module specifically includes:
[0039] The model inference unit is internally deployed with a long short-term memory network with an attention mechanism, which is used to perform forward computation on the input feature matrix, use LSTM layers to capture time dependencies, and use attention mechanism layers to compute feature weights.
[0040] The result decoding unit is used to convert the tensor data output by the network into numerical values with business meaning, and output the sales forecast value including the mean and confidence interval.
[0041] As a further aspect of the present invention, the allocation decision module specifically includes:
[0042] The model building unit is used to construct a mathematical programming model with the objective function of minimizing total supply chain cost, and with constraints such as total inventory, safety level and warehouse throughput capacity, combined with real-time inventory status.
[0043] The operations research unit integrates a mixed-integer linear programming solver or a particle swarm optimization algorithm engine to iteratively optimize the mathematical programming model and output the inventory allocation scheme containing the optimal replenishment quantity and transfer path.
[0044] As a further aspect of the present invention, the execution monitoring module specifically includes:
[0045] The instruction conversion unit is used to parse and convert the inventory allocation scheme into a standard format of warehouse management system instructions for receiving or transferring goods.
[0046] The deviation analysis unit is used to compare the actual sales volume of each platform with the predicted sales volume at a preset frequency, calculate the deviation rate, and lock the remaining inventory and trigger the dynamic adjustment mechanism when the deviation rate exceeds a preset threshold.
[0047] This invention provides an intelligent inventory allocation method and system based on multi-platform sales forecasting. By integrating multi-source heterogeneous data and deep learning technology, it overcomes the shortcomings of traditional linear models in accurately capturing complex sales fluctuations, significantly improving forecast accuracy. Combining a multi-objective operations optimization model with a dynamic closed-loop monitoring mechanism, it achieves global intelligent coordination and refined allocation of inventory resources across multiple platforms. This not only effectively balances stockout risks and warehousing holding costs, significantly reducing ineffective cross-warehouse transfer fees, but also endows the supply chain with agile response capabilities to market changes and non-linear demand, realizing a fundamental shift from static experience-based decision-making to data-driven intelligent decision-making. Attached Figure Description
[0048] Figure 1 This is the main flowchart of an intelligent inventory allocation method based on multi-platform sales forecasting.
[0049] Figure 2 This is a flowchart illustrating the preprocessing and feature engineering operations performed on the multi-dimensional time series dataset to generate a standardized feature vector matrix containing multiple influencing factors, as described in the intelligent inventory allocation method based on multi-platform sales forecasting.
[0050] Figure 3 This is a flowchart illustrating how a smart inventory allocation method based on multi-platform sales forecasting inputs the standardized feature vector matrix into a pre-trained deep learning prediction model to output the sales forecast values for each sales platform within a preset future period.
[0051] Figure 4 This is a flowchart illustrating how a multi-platform sales forecast-based intelligent inventory allocation method constructs a multi-objective inventory optimization model based on the predicted sales value and supply chain cost parameters, and then solves the model to derive the inventory allocation scheme for each platform.
[0052] Figure 5 This is a flowchart illustrating the execution of the inventory allocation scheme and real-time monitoring of the deviation between actual sales and predicted values in an intelligent inventory allocation method based on multi-platform sales forecasting.
[0053] Figure 6 This is the main structure diagram of an intelligent inventory allocation system based on multi-platform sales forecasting. Detailed Implementation
[0054] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0055] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.
[0056] The intelligent inventory allocation method and system based on multi-platform sales forecasting provided by this invention solves the technical problems in the background art.
[0057] like Figure 1 The diagram shown is the main flowchart of an intelligent inventory allocation method based on multi-platform sales forecasting provided in an embodiment of the present invention. The intelligent inventory allocation method based on multi-platform sales forecasting includes:
[0058] Step S100: Obtain historical business data and external influencing factor data of the target product on multiple sales platforms, and construct a multidimensional time series dataset;
[0059] Step S200: Perform preprocessing and feature engineering operations on the multidimensional time series dataset to generate a standardized feature vector matrix containing multiple influencing factors;
[0060] Step S300: Input the standardized feature vector matrix into the pre-trained deep learning prediction model and output the sales forecast value of each sales platform in the future preset period;
[0061] Step S400: Based on the sales forecast, construct a multi-objective inventory optimization model in conjunction with supply chain cost parameters, and solve for the inventory allocation scheme of each platform;
[0062] Step S500: Execute the inventory allocation plan and monitor the deviation between actual sales and forecast values in real time, triggering dynamic adjustment when preset conditions are met;
[0063] In application, this embodiment breaks down data silos between e-commerce platforms through a multi-source heterogeneous data acquisition mechanism, integrating previously scattered sales records, traffic logs, and external environmental factors into a panoramic decision-making basis. Building on this, the method abandons the extensive model of traditional ERP systems that rely on manual experience or simple linear statistics for prediction, instead introducing deep learning technology to handle nonlinear characteristics. This allows the system to keenly capture sales pulses triggered by sudden factors such as live-streaming e-commerce and holiday promotions. Furthermore, this solution transforms the inventory allocation problem into a global optimization problem in the field of mathematical operations research. By constructing a multi-objective cost function, it finds a precise mathematical balance point between holding costs, stockout risk, and logistics allocation costs, thereby replacing the local and static quota allocation strategy. Finally, through an embedded dynamic feedback mechanism, the system can monitor the execution effect in real time and make self-corrections. This end-to-end data-driven design not only significantly improves inventory turnover and reduces capital occupation costs, but also endows the supply chain with extremely strong resilience in the face of market uncertainty, achieving a qualitative leap from passive response to proactive prediction and dynamic scheduling.
[0064] like Figure 2As shown, in a preferred embodiment of the present invention, the preprocessing and feature engineering operations performed on the multidimensional time series dataset to generate a standardized feature vector matrix containing multiple influencing factors specifically include:
[0065] Step S201: Isolation forest algorithm is used to identify and remove abnormal noise points in the historical business data, and Lagrange interpolation is used to fill in the missing data;
[0066] Step S202: Normalize the cleaned data to map data of different dimensions to a unified range;
[0067] Step S203: Extract time features, including periodic factors of day, week, and month, and holiday markers; extract marketing features, including product discount rates, number of live broadcasts, and advertising investment; extract competitive features, including price fluctuation index of similar competing products;
[0068] Step S204: Concatenate the extracted time features and competition features into vectors to generate the standardized feature vector matrix.
[0069] In this embodiment, the Isolation Forest algorithm is first introduced. Leveraging its density-based unsupervised learning characteristics, it accurately identifies and removes anomalous noise deviating from the normal distribution in high-dimensional data space, effectively preventing dirty data from interfering with model training. Simultaneously, Lagrange interpolation is employed to scientifically fill in missing data based on the continuity principle of time series, ensuring the integrity of time-series information. Furthermore, to address the issue of significant differences in the dimensions of the original data, normalization processing accelerates the convergence speed of subsequent models. This not only extracts basic time-cycle features but also transforms the domain knowledge of business experts into machine-understandable vector language, deeply exploring the potential impact of marketing factors such as discount intensity and livestream duration, as well as competitive factors such as competitor price fluctuations, on sales volume, greatly enriching the model's ability to perceive micro-level market changes.
[0070] like Figure 3 As shown, in a preferred embodiment of the present invention, the step of inputting the standardized feature vector matrix into a pre-trained deep learning prediction model and outputting the sales forecast values for each sales platform within a preset future period specifically includes:
[0071] Step S301: Construct a long short-term memory network based on attention mechanism as a prediction model, where the LSTM layer is used to capture the temporal dependence of long sequences, and the attention mechanism layer is used to calculate the weights of the influence of features at different time steps on future sales.
[0072] Step S302: The standardized feature vector matrix is divided into segments according to a time sliding window and then input into the prediction model;
[0073] Step S303: Decode the model output through a fully connected layer to obtain the quantile regression prediction results for each sales platform, and output the sales forecast value including the predicted mean and confidence interval.
[0074] In this embodiment, a Long Short-Term Memory (LSTM) network with an integrated attention mechanism is used as the prediction engine. The LSTM network, with its unique gating mechanism, effectively overcomes the vanishing gradient problem in traditional recurrent neural networks, thus accurately capturing seasonal sales patterns and long-term trends spanning weeks or even months. The introduced attention mechanism simulates the human visual focusing process, dynamically calculating the weights of various features in historical time steps based on the current prediction target, automatically focusing on high-weight features at key time points such as Singles' Day (November 11th) and top livestreamers' broadcasts, while suppressing irrelevant noise. Furthermore, the model does not stop at outputting a single numerical value but uses quantile regression technology to output a probability distribution result including the predicted mean and confidence interval.
[0075] like Figure 4 As shown, in a preferred embodiment of the present invention, the step of constructing a multi-objective inventory optimization model based on the sales forecast value and combining it with supply chain cost parameters, and solving for the inventory allocation scheme for each platform, specifically includes:
[0076] Step S401: Set the objective function to minimize the total supply chain cost, which includes holding costs, stockout loss costs, cross-warehouse transfer costs, and last-mile delivery costs;
[0077] Step S402: Set constraints, including hard constraints on total inventory, minimum safety stock level constraints on each platform, and maximum throughput constraints on the warehouse.
[0078] Step S403: Substitute the sales forecast and the current inventory status of each warehouse into the multi-objective inventory optimization model;
[0079] Step S404: Iteratively solve the model using a mixed-integer linear programming algorithm or a particle swarm optimization algorithm, output the optimal replenishment quantity and allocation path for each platform, and form the inventory allocation scheme.
[0080] It should be understood that a multi-objective mathematical programming model with multiple constraints was constructed, with minimizing the total supply chain cost as the objective function. This model comprehensively considers the holding costs of inventory backlog, potential sales losses due to stockouts, logistics costs from cross-warehouse transfers, and last-mile delivery costs. By setting hard constraints on total inventory, soft constraints on safety levels, and warehouse throughput constraints, the generated allocation scheme is ensured to be physically feasible and business-safe. At the solution level, mixed-integer linear programming or particle swarm optimization algorithms are employed, enabling rapid searching for the globally optimal solution in a large and non-convex solution space. This technique changes the previous localized scheduling model that relied on manual experience and piecemeal solutions. It can automatically calculate the optimal replenishment quantity and transfer path between platforms and warehouses, minimizing ineffective transportation and inventory redundancy, and achieving refined allocation and maximization of supply chain resources.
[0081] like Figure 5 As shown, in a preferred embodiment of the present invention, the execution of the inventory allocation scheme and the real-time monitoring of the deviation between actual sales and forecast values, triggering dynamic adjustments when preset conditions are met, specifically includes:
[0082] Step S501: Convert the inventory allocation plan into an inbound order or transfer order that can be recognized by the warehouse management system and issue it for execution;
[0083] Step S502: Collect the actual real-time sales of each platform at a preset frequency, and calculate the deviation rate between the actual sales and the predicted sales value;
[0084] Step S503: Determine whether the deviation rate exceeds the preset threshold. If so, lock the remaining unallocated inventory, and use the latest actual sales data as input to re-output the sales forecast value of each sales platform in the future preset period and re-solve to obtain the inventory allocation scheme of each platform, generating the corrected inventory replenishment strategy.
[0085] In application, this embodiment seamlessly transforms the optimized inventory plan into standard instructions executable by the WMS system, ensuring the implementation of decisions. Furthermore, the system establishes a two-tiered monitoring system, frequently collecting real-time sales data from various platforms and calculating the deviation rate from the predicted values. Once this deviation rate exceeds a preset safety threshold, indicating that the original forecast assumptions have failed, a dynamic adjustment mechanism is immediately triggered to lock remaining inventory resources and re-run the forecasting and optimization process based on the latest real-time data. This endows the supply chain with exceptional agility, enabling it to quickly mitigate losses or seize fleeting sales opportunities when market fluctuations occur, controlling inventory risks at their inception and ensuring the system always operates smoothly on the optimal track.
[0086] like Figure 6As shown, in another preferred embodiment of the present invention, an intelligent inventory allocation system based on multi-platform sales forecasting is provided, the system comprising:
[0087] The data acquisition module 100 is used to connect to various e-commerce platforms and external databases through API interfaces to obtain historical business data and external influencing factor data of target products, and aggregate the data into a multi-dimensional time series dataset.
[0088] The data processing module 200 is used to receive the multidimensional time series dataset, perform denoising, completion, and normalization on it, and extract multidimensional features for vector concatenation to generate a standardized feature vector matrix.
[0089] The sales forecasting module 300 is used to load a pre-trained deep learning forecasting model, take the standardized feature vector matrix as input, calculate and output the sales forecast value of each sales platform in the future preset period.
[0090] The allocation decision module 400 is used to store supply chain cost parameters and constraints, construct a multi-objective inventory optimization model based on the sales forecast value and solve it to generate inventory allocation schemes for each platform.
[0091] The execution monitoring module 500 is used to issue the inventory allocation plan and calculate the deviation rate between the actual sales and the predicted value in real time. When the deviation rate meets the trigger condition, a recalculation signal is sent to the sales prediction module to generate a corrected replenishment strategy.
[0092] In this embodiment, the system is divided into five core modules: data acquisition, data processing, sales forecasting, allocation decision-making, and execution monitoring. When a new e-commerce platform is connected or the forecasting algorithm is upgraded, only the corresponding module needs to be updated without reconstructing the entire system. Specifically, it connects the entire chain from external data sources to the internal ERP / WMS system, forming an intelligent organic whole. Through the deep integration of hardware computing power and software algorithms, this system can support the high-concurrency computing needs of massive SKUs, providing a solid technical foundation for enterprises to build a standardized, replicable, and self-evolving intelligent supply chain platform, ensuring that the technical solutions in the method claims can be stably and efficiently implemented in industrial scenarios.
[0093] In another preferred embodiment of the present invention, the data processing module 200 specifically includes:
[0094] The anomaly cleaning unit 201 is used to run the isolated forest algorithm to identify and remove abnormal noise points in the data, and to fill in missing data using the Lagrange interpolation method.
[0095] The feature construction unit 202 is used to normalize the cleaned data, extract time period features, marketing activity features and competitive features from the data, and concatenate the features to generate the standardized feature vector matrix that is adapted to the input of the model.
[0096] In this embodiment, the data processing module 200 integrates an anomaly cleaning unit 201 and a feature construction unit 202, which work together to achieve data standardization and value creation. The anomaly cleaning unit 201 is configured to automatically run statistical algorithms such as Isolation Forest, capable of processing high-concurrency data streams within milliseconds. It can remove dirty data in real time and intelligently repair data breakpoints using interpolation algorithms, ensuring the continuity and purity of the data stream. The feature construction unit 202 acts as a computational engine, internalizing complex business logic into vector operation rules. It can automatically extract multi-dimensional features such as time periods, marketing activities, and market competition from the cleaned data and concatenate them into a standardized tensor adapted to the input of a deep learning model.
[0097] In another preferred embodiment of the present invention, the sales forecasting module 300 specifically includes:
[0098] The model inference unit 301 has a long short-term memory network with an attention mechanism deployed inside, which is used to perform forward computation on the input feature matrix, use LSTM layers to capture time dependencies, and use attention mechanism layers to calculate feature weights.
[0099] The result decoding unit 302 is used to convert the tensor data output by the network into numerical values with business meaning, and output the sales forecast value including the mean and confidence interval.
[0100] In this embodiment, the sales forecasting module 300 is internally divided into a model inference unit 301 and a result decoding unit 302, achieving high-performance online forecasting through hardware and software co-design. The model inference unit 301 is not simply a computational carrier, but a neural network inference engine optimized for the Attention-LSTM network structure. It is equipped with parallel computing acceleration logic, enabling it to quickly process the forward propagation operations of high-dimensional feature matrices and accurately capture nonlinear patterns and long-term / short-term dependencies in time series data. The result decoding unit 302 performs semantic translation, mapping the abstract tensor or probability distribution data output by the neural network back to sales figures with actual business meaning, and simultaneously calculating confidence intervals.
[0101] In another preferred embodiment of the present invention, the allocation decision module 400 specifically includes:
[0102] Model building unit 401 is used to construct a mathematical programming model with the objective function of minimizing total supply chain cost, and with constraints such as total inventory, safety level and warehouse throughput capacity, combined with real-time inventory status.
[0103] The operations research unit 402 integrates a mixed integer linear programming solver or a particle swarm optimization algorithm engine to iteratively optimize the mathematical programming model and output the inventory allocation scheme containing the optimal replenishment quantity and transfer path.
[0104] In this embodiment, the allocation decision module 400 includes a model building unit 401 and an operations research and solution unit 402, which together form a complete automated decision-making chain. The model building unit 401 is a dynamic logic generator with a pre-set library of various cost parameters and business constraint rules. It can assemble a system of mathematical programming equations reflecting the actual business scenario in real time based on the current inventory status and market environment. The operations research and solution unit 402 integrates a high-performance mathematical solver and is equipped with an acceleration algorithm for mixed integer programming problems. It can quickly search for the globally optimal solution that satisfies all constraints in a very large solution space. Its technical advantage lies in replacing manual experience-based judgment with the powerful computing power of a computer, enabling the completion of complex overall planning involving multiple warehouses, platforms, and cost items within seconds, ensuring that the output inventory allocation scheme is optimal in terms of economic benefits.
[0105] In another preferred embodiment of the present invention, the execution monitoring module 500 specifically includes:
[0106] The instruction conversion unit 501 is used to parse and convert the inventory allocation scheme into a standard format of warehouse management system instruction for receiving or transferring goods.
[0107] The deviation analysis unit 502 is used to compare the actual sales of each platform with the predicted sales value at a preset frequency, calculate the deviation rate, and lock the remaining inventory and trigger the dynamic adjustment mechanism when the deviation rate exceeds a preset threshold.
[0108] In this embodiment, the execution monitoring module 500 consists of an instruction conversion unit 501 and a deviation analysis unit 502, which respectively undertake the dual functions of execution and supervision. The instruction conversion unit 501 is equipped with a standardized protocol interface, capable of translating the abstract inventory allocation matrix into standard operation messages such as inbound orders and transfer orders that the WMS system can directly recognize, achieving seamless automated flow from algorithmic decision-making to warehousing operations. The deviation analysis unit 502 acts as a real-time watchdog process, configured to continuously poll actual sales data and perform rigorous statistical difference tests. Once the deviation rate between actual sales and predicted values exceeds the warning threshold, this unit, possessing the highest priority system privileges, can immediately block the current process and trigger a recalculation signal.
[0109] The above embodiments of the present invention provide an intelligent inventory allocation method based on multi-platform sales forecasting, and provide an intelligent inventory allocation system based on multi-platform sales forecasting, through..., + technical effects.
[0110] In order for the above methods and systems to operate smoothly, the system may include more or fewer components than those described above, or combine certain components, or different components, in addition to the various modules mentioned above. For example, it may include input / output devices, network access devices, buses, processors, and memory.
[0111] The processor can be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (OPGs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. This processor is the control center of the system, connecting various parts via various interfaces and lines.
[0112] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0113] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.
[0114] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for intelligent inventory allocation based on multi-platform sales forecast, characterized in that, The method includes: Acquire historical business data and external influencing factor data of the target product on multiple sales platforms, and construct a multidimensional time series dataset; The multidimensional time series dataset is preprocessed and feature engineering is performed to generate a standardized feature vector matrix containing multiple influencing factors. The standardized feature vector matrix is input into a pre-trained deep learning prediction model, which outputs the sales forecast values for each sales platform within a preset future period. Based on the sales forecast, a multi-objective inventory optimization model is constructed by combining supply chain cost parameters, and the inventory allocation scheme for each platform is obtained by solving the model. The inventory allocation plan is executed, and the deviation between actual sales and forecast values is monitored in real time. Dynamic adjustments are triggered when preset conditions are met.
2. The method for intelligent inventory allocation based on multi-platform sales forecasting as claimed in claim 1, wherein, The preprocessing and feature engineering operations performed on the multidimensional time series dataset to generate a standardized feature vector matrix containing multiple influencing factors specifically include: The isolated forest algorithm is used to identify and remove abnormal noise points in the historical business data, and the Lagrange interpolation method is used to fill in the missing data. The cleaned data is normalized to map data of different dimensions to a unified range; Extract time features, including periodic factors of day, week, and month, and holiday markers; extract marketing features, including product discount rates, number of live broadcasts, and advertising investment; extract competitive features, including price fluctuation indices of similar competing products. The extracted time features and competition features are concatenated into vectors to generate the standardized feature vector matrix.
3. The intelligent inventory allocation method based on multi-platform sales forecasting according to claim 1, characterized in that, The step of inputting the standardized feature vector matrix into a pre-trained deep learning prediction model and outputting the sales forecast values for each sales platform within a preset future period specifically includes: A long short-term memory network based on an attention mechanism is constructed as a prediction model, in which the LSTM layer is used to capture the temporal dependence of long sequences, and the attention mechanism layer is used to calculate the weights of the impact of features at different time steps on future sales. The standardized feature vector matrix is then segmented by a time sliding window and input into the prediction model; The model output is decoded by a fully connected layer to obtain the quantile regression prediction results for each sales platform, and the output includes the predicted mean and confidence interval of the sales forecast.
4. The intelligent inventory allocation method based on multi-platform sales forecasting according to claim 1, characterized in that, The process of constructing a multi-objective inventory optimization model based on the sales forecast and combining it with supply chain cost parameters, and then solving for the inventory allocation scheme for each platform, specifically includes: The objective function is set as minimizing the total supply chain cost, which includes holding costs, stockout loss costs, cross-warehouse transfer costs, and last-mile delivery costs. Set constraints, including hard constraints on total inventory, minimum safety stock levels for each platform, and maximum warehouse throughput. Substitute the sales forecast and the current inventory status of each warehouse into the multi-objective inventory optimization model; The model is iteratively solved using a mixed-integer linear programming algorithm or a particle swarm optimization algorithm to output the optimal replenishment quantity and allocation path for each platform, thus forming the inventory allocation scheme.
5. The intelligent inventory allocation method based on multi-platform sales forecasting according to claim 1, characterized in that, The execution of the inventory allocation plan and the real-time monitoring of the deviation between actual sales and forecasts, triggering dynamic adjustments when preset conditions are met, specifically include: The inventory allocation plan is converted into an inbound order or transfer order that can be recognized by the warehouse management system and then issued for execution. The actual real-time sales of each platform are collected at a preset frequency, and the deviation rate between the actual sales and the predicted sales value is calculated. If the deviation rate exceeds a preset threshold, the remaining unallocated inventory is locked. Using the latest actual sales data as input, the sales forecast values for each sales platform in the future preset period are re-output and the inventory allocation scheme for each platform is re-solved to generate a corrected inventory replenishment strategy.
6. An intelligent inventory allocation system based on multi-platform sales forecasting, characterized in that, The system employs the intelligent inventory allocation method based on multi-platform sales forecasting as described in any one of claims 1-5, wherein the system comprises: The data acquisition module is used to connect to various e-commerce platforms and external databases through API interfaces to obtain historical business data and external influencing factor data of target products, and aggregate the data into a multi-dimensional time series dataset. The data processing module is used to receive the multidimensional time series dataset, perform denoising, completion, and normalization on it, and extract multidimensional features for vector concatenation to generate a standardized feature vector matrix. The sales forecasting module is used to load a pre-trained deep learning forecasting model, take the standardized feature vector matrix as input, calculate and output the sales forecast value of each sales platform in the future preset period. The allocation decision module is used to store supply chain cost parameters and constraints, construct a multi-objective inventory optimization model based on the sales forecast value and solve it to generate inventory allocation schemes for each platform. The execution monitoring module is used to issue the inventory allocation plan and calculate the deviation rate between the actual sales and the predicted value in real time. When the deviation rate meets the trigger condition, a recalculation signal is sent to the sales prediction module to generate a corrected replenishment strategy.
7. The intelligent inventory allocation system based on multi-platform sales forecasting according to claim 6, characterized in that, The data processing module specifically includes: The anomaly cleaning unit is used to run the Isolation Forest algorithm to identify and remove anomalous noise points in the data, and to fill in missing data using Lagrange interpolation. The feature construction unit is used to normalize the cleaned data, extract time period features, marketing activity features and competitive features from the data, and concatenate the features to generate the standardized feature vector matrix that is adapted to the input of the model.
8. The intelligent inventory allocation system based on multi-platform sales forecasting according to claim 6, characterized in that, The sales forecasting module specifically includes: The model inference unit is internally deployed with a long short-term memory network with an attention mechanism, which is used to perform forward computation on the input feature matrix, use LSTM layers to capture time dependencies, and use attention mechanism layers to compute feature weights. The result decoding unit is used to convert the tensor data output by the network into numerical values with business meaning, and output the sales forecast value including the mean and confidence interval.
9. The intelligent inventory allocation system based on multi-platform sales forecasting according to claim 6, characterized in that, The allocation decision module specifically includes: The model building unit is used to construct a mathematical programming model with the objective function of minimizing total supply chain cost, and with constraints such as total inventory, safety level and warehouse throughput capacity, combined with real-time inventory status. The operations research unit integrates a mixed-integer linear programming solver or a particle swarm optimization algorithm engine to iteratively optimize the mathematical programming model and output the inventory allocation scheme containing the optimal replenishment quantity and transfer path.
10. The intelligent inventory allocation system based on multi-platform sales forecasting according to claim 6, characterized in that, The execution monitoring module specifically includes: The instruction conversion unit is used to parse and convert the inventory allocation scheme into a standard format of warehouse management system instructions for receiving or transferring goods. The deviation analysis unit is used to compare the actual sales volume of each platform with the predicted sales volume at a preset frequency, calculate the deviation rate, and lock the remaining inventory and trigger the dynamic adjustment mechanism when the deviation rate exceeds a preset threshold.