Inventory information management method and device

Abstract

The invention discloses an inventory information management method and device, and relates to the technical field of intelligent supply chains. According to one specific embodiment of the invention, the method comprises the following steps: according to inventory data of articles and various supply chain data associated with the articles; generating inventory fusion features of the articles; predicting an article inventory state of the article based on the inventory fusion feature by using a dynamic inventory prediction model; and based on the inventory data of the articles and the predicted inventory state of the articles, determining a replenishment strategy of the articles. According to the embodiment of the invention, the dynamic inventory prediction model is utilized to predict the inventory state of the goods based on the inventory fusion features of various supply chain data, so that the replenishment strategy of the goods is determined and replenishment operation is carried out, and the accuracy, rationality and efficiency of goods inventory information management are improved.

Description

Technical Field

[0001] This invention relates to the field of smart supply chain technology, and in particular to an inventory information management method and apparatus. Background Technology

[0002] In the service process provided by modern e-commerce systems, inventory management in warehouses is one of the most important aspects. By managing warehouse inventory information, logistics efficiency and user shopping experience can be improved.

[0003] Currently, the replenishment quantity of items is mainly determined by the outbound data provided by the inventory management party. However, the existing methods have problems such as low intelligence in inventory information management, poor adaptability to the supply and demand of items, and low flexibility. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide an inventory information management method and apparatus, which can generate inventory fusion characteristics of items based on the inventory data of the items and various supply chain data associated with the items; predict the inventory status of the items using a dynamic inventory forecasting model based on the inventory fusion characteristics; and determine the replenishment strategy for the items based on the inventory data of the items and the predicted inventory status. Embodiments of the present invention improve the accuracy, rationality, and efficiency of inventory information management by using a dynamic inventory forecasting model based on the inventory fusion characteristics of various supply chain data to predict the inventory status of items and thereby determine the replenishment strategy for replenishment operations.

[0005] To achieve the above objectives, according to one aspect of the present invention, an inventory information management method is provided, comprising: acquiring inventory data of items in a warehouse and various supply chain data associated with the items; generating an inventory fusion feature of the items based on the various supply chain data; predicting the inventory status of the items using a dynamic inventory prediction model based on the inventory fusion feature; and determining a replenishment strategy for the items based on the inventory data and the predicted inventory status of the items.

[0006] Optionally, the supply chain data includes item supply data, item demand data, and the relationship between item suppliers and items; before generating the inventory fusion features of the items based on various supply chain data, the method further includes: using a preset semantic analysis model to parse the item demand text contained in the item demand data, and extracting item demand features from the item demand text; using a preset convolutional network model to analyze the item image information contained in the item demand data, and generating item visual features corresponding to the item image information; constructing a supply chain graph based on the relationship between item suppliers and items, and generating supply chain topology features based on the supply chain graph using a graph sampling aggregation model; and extracting item supplier features from the item supply data.

[0007] Optionally, generating the inventory fusion feature of the item based on multiple supply chain data includes: fusing the item supplier features, the item demand features, the item visual features, and the supply chain topology features to generate the inventory fusion feature.

[0008] Optionally, the supply chain data also includes item time-series data; using a preset hybrid neural network model, one or more time-dimensional item time-series features are extracted from the item time-series data; generating the item's inventory fusion feature based on multiple supply chain data includes: fusing the item supplier features, the item demand features, the item visual features, the supply chain topology features, and the item time-series features to generate the inventory fusion feature.

[0009] Optionally, the step of fusing the supplier characteristics, demand characteristics, visual characteristics, and supply chain topology characteristics of the goods to generate the inventory fusion feature includes: obtaining the initial weights of the supplier characteristics, demand characteristics, visual characteristics, and supply chain topology characteristics of the goods respectively; adjusting the initial weights of the supplier characteristics, demand characteristics, visual characteristics, and supply chain topology characteristics of the goods using a preset attention model; and fusing the adjusted weights and the corresponding supplier characteristics, demand characteristics, visual characteristics, and supply chain topology characteristics of the goods to generate the inventory fusion feature.

[0010] Optionally, the dynamic inventory forecasting model includes a supplier forecasting model and a demand forecasting model; the dynamic inventory forecasting model predicts the inventory status of the item based on the inventory fusion features, including: predicting the item supply status of the item supplier using the supplier forecasting model and the inventory fusion features; predicting the item demand status using the demand forecasting model and the inventory fusion features; and predicting the item inventory status by combining the item supply status and the item demand status.

[0011] Optionally, predicting the item's inventory status by combining the item's supply status and the item's demand status includes: obtaining current item demand information or current item supplier information; adjusting the prediction weights for the item's supply status and demand status using a gating fusion unit based on the current item demand information or current item supplier information; and predicting the item's inventory status based on the adjusted prediction weights and the corresponding item supply status and demand status.

[0012] Optionally, the following steps are taken: acquiring replenishment data executed according to the warehouse's replenishment strategy; determining the warehouse's current item conversion rate, inventory turnover rate, and item shortage rate based on the replenishment data; constructing a prediction reward function for a deep reinforcement learning model based on the item conversion rate, inventory turnover rate, item shortage rate, and weights; and optimizing the dynamic inventory prediction model using the deep reinforcement learning model and the real-time reward value calculated by the prediction reward function.

[0013] Optionally, the dynamic inventory forecasting model includes a supplier forecasting model and a demand forecasting model; the optimization of the dynamic inventory forecasting model using a deep reinforcement learning model and the calculation results of the forecasting reward function includes: comparing the real-time reward value with the historical best reward value, and generating a feedback signal based on the comparison result; the optimization of the forecasting weights of the supplier forecasting model and the demand forecasting model using a deep reinforcement learning model based on the feedback signal.

[0014] To achieve the above objectives, according to a second aspect of the present invention, an inventory information management device is provided, comprising: a data acquisition module, a feature extraction module, and a strategy determination module; wherein, The data acquisition module is used to acquire inventory data of items in the warehouse and various supply chain data associated with the items; The feature extraction module is used to generate inventory fusion features of the item based on various supply chain data. The strategy determination module is used to predict the inventory status of the item using a dynamic inventory forecasting model based on the inventory fusion features; and to determine the replenishment strategy for the item based on the inventory data and the predicted inventory status. Optionally, the supply chain data includes item supply data, item demand data, and the relationship between item suppliers and items; the inventory information management device, before generating the inventory fusion features of the items based on various supply chain data, further includes: using a preset semantic analysis model to parse the item demand text contained in the item demand data, and extracting item demand features from the item demand text; using a preset convolutional network model to analyze the item image information contained in the item demand data, and generating item visual features corresponding to the item image information; constructing a supply chain graph based on the relationship between item suppliers and items, and generating supply chain topology features based on the supply chain graph using a graph sampling aggregation model; and extracting item supplier features from the item supply data.

[0015] Optionally, the inventory information management device is used to generate inventory fusion features of the item based on various supply chain data, including: fusion of the item supplier features, the item demand features, the item visual features, and the supply chain topology features to generate the inventory fusion features.

[0016] Optionally, the supply chain data further includes item time-series data; an inventory information management device is used to extract one or more time-dimensional item time-series features from the item time-series data using a preset hybrid neural network model; the step of generating the inventory fusion feature of the item based on multiple supply chain data includes: fusing the item supplier features, the item demand features, the item visual features, the supply chain topology features, and the item time-series features to generate the inventory fusion feature.

[0017] Optionally, the inventory information management device is used to fuse the supplier characteristics, demand characteristics, visual characteristics, and supply chain topology characteristics of the goods to generate the inventory fusion feature, including: obtaining the initial weights of the supplier characteristics, demand characteristics, visual characteristics, and supply chain topology characteristics of the goods respectively; adjusting the initial weights of the supplier characteristics, demand characteristics, visual characteristics, and supply chain topology characteristics of the goods using a preset attention model; and fusing the adjusted weights and the corresponding supplier characteristics, demand characteristics, visual characteristics, and supply chain topology characteristics of the goods to generate the inventory fusion feature.

[0018] Optionally, the inventory information management device is used to utilize a dynamic inventory forecasting model, including a supplier forecasting model and a demand-side forecasting model; the use of the dynamic inventory forecasting model to predict the inventory status of the items based on the inventory fusion features includes: using the supplier forecasting model and the inventory fusion features to predict the supply status of the items from the supplier; using the demand-side forecasting model and the inventory fusion features to predict the demand status of the items; and combining the supply status and the demand status to predict the inventory status of the items.

[0019] Optionally, the inventory information management device is used to predict the inventory status of the item by combining the item supply status and the item demand status, including: obtaining current item demand information or current item supplier information; adjusting the prediction weights of the current item supply status and the item demand status using a gating fusion unit based on the current item demand information or the current item supplier information; and predicting the inventory status of the item based on the adjusted prediction weights and the corresponding item supply status and item demand status.

[0020] Optionally, the inventory information management device is used to acquire replenishment data executed according to the warehouse's replenishment strategy; determine the current item conversion rate, inventory turnover rate, and item shortage rate of the warehouse based on the replenishment data; construct a prediction reward function for a deep reinforcement learning model based on the item conversion rate, inventory turnover rate, item shortage rate, and weights; and optimize the dynamic inventory prediction model using the real-time reward value calculated by the deep reinforcement learning model and the prediction reward function.

[0021] Optionally, the inventory information management device is used to utilize a dynamic inventory forecasting model, including a supplier forecasting model and a demand forecasting model; the optimization of the dynamic inventory forecasting model using a deep reinforcement learning model and the calculation results of the forecasting reward function includes: comparing the real-time reward value with the historical best reward value, and generating a feedback signal based on the comparison result; the optimization of the forecasting weights of the supplier forecasting model and the demand forecasting model using a deep reinforcement learning model based on the feedback signal.

[0022] To achieve the above objectives, according to a third aspect of the present invention, an electronic device for inventory information management is provided, characterized in that it includes: one or more processors; and a storage device for storing one or more programs, wherein 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 above for inventory information management.

[0023] To achieve the above objectives, according to a fourth aspect of the present invention, a computer-readable medium is provided having a computer program stored thereon, characterized in that, when the program is executed by a processor, it implements any of the methods described above for inventory information management.

[0024] To achieve the above objectives, according to a fifth aspect of the present invention, a computer program product is provided, comprising a computer program, characterized in that, when executed by a processor, the computer program implements any of the methods described above for inventory information management.

[0025] One embodiment of the above invention has the following advantages or beneficial effects: it can generate inventory fusion characteristics of items based on inventory data and various supply chain data associated with the items; predict the inventory status of items using a dynamic inventory forecasting model based on the inventory fusion characteristics; and determine the replenishment strategy for items based on the inventory data and the predicted inventory status. This embodiment of the invention improves the accuracy, rationality, and efficiency of inventory information management by using a dynamic inventory forecasting model based on the inventory fusion characteristics of various supply chain data to predict the inventory status of items, thereby determining the replenishment strategy for replenishment operations; furthermore, it improves the accuracy and rationality of predicting the inventory status of items by dynamically adjusting the prediction weights of the supplier forecasting model and the demand forecasting model in the dynamic inventory forecasting model according to actual supply chain data.

[0026] The further effects of the aforementioned unconventional alternative methods will be explained below in conjunction with specific implementation methods. Attached Figure Description

[0027] The accompanying drawings are provided to better understand the invention and are not intended to unduly limit the scope of the invention. Wherein: Figure 1 This is a flowchart illustrating an inventory information management method provided in an embodiment of the present invention; Figure 2 is a schematic diagram of the structure of an inventory information management system provided in an embodiment of the present invention; Figure 3A This is a schematic diagram of a process for generating inventory fusion features according to an embodiment of the present invention; Figure 3B This is a schematic diagram of the structure of a dynamic inventory forecasting model provided in one embodiment of the present invention; Figure 3C This is a schematic diagram of an inventory information management system provided in one embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an inventory information management device according to an embodiment of the present invention; Figure 5 is an exemplary system architecture diagram in which embodiments of the present invention can be applied; Figure 6 This is a schematic diagram of the structure of a computer system suitable for implementing terminal devices or servers of the present invention. Detailed Implementation

[0028] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0029] It should be noted that the collection, use, storage, sharing and transfer of user personal information involved in the technical solution of the present invention all comply with the provisions of relevant laws and regulations, and require notification to users and obtaining their consent or authorization. When applicable, user personal information is subjected to de-identification and / or anonymization and / or encryption technical processing.

[0030] like Figure 1 As shown in the figure, an embodiment of the present invention provides an inventory information management method, which may include the following steps: Step S101: Obtain inventory data of items in the warehouse and various supply chain data associated with the items.

[0031] Specifically, inventory data includes the current inventory information of a particular item in the warehouse (e.g., inventory quantity, inventory batch, etc.); various supply chain data associated with the item include item supply data, item demand data, and the relationship between item suppliers and items; supply chain data can also be divided into structured data (e.g., purchase order data, inventory data, supplier user profile data, etc.) and unstructured data (e.g., item description information, item demand information, purchase record information, customer service record information, etc.); the relationship between item suppliers and items includes, for example, the relationship between one or more item suppliers for a particular item, and the upstream and downstream relationships in the supply chain.

[0032] Step S102: Generate the inventory fusion characteristics of the item based on various supply chain data.

[0033] Specifically, in this embodiment of the invention, multi-dimensional correlation features of items are extracted through different models to generate inventory fusion features of items.

[0034] The following is combined with Figure 3A The diagram illustrates the process of generating inventory fusion features, such as... Figure 3A As shown: Step S301A: Receive supply chain data; parse supply chain data of different types and from different sources, and determine the data type of the parsed data.

[0035] For text-based data, a pre-defined semantic analysis model is used to parse the item requirement text contained in the item requirement data and extract item requirement features from the text. This pre-defined semantic analysis model can be a BERT-Transformer hybrid model, where BERT stands for Bidirectional Encoder Representations from Transformers. The BERT-Transformer hybrid model includes a BERT pre-trained feature extraction layer and a Transformer encoding layer. The BERT pre-trained feature extraction layer performs contextual semantic encoding on the item requirement text, outputting an initial semantic vector. The Transformer encoding layer performs multiple rounds of self-attention interaction on the initial semantic vector to generate a feature vector containing the item requirement intent; the feature vector dimension is, for example, 512.

[0036] For image data (e.g., object image data), a pre-defined convolutional network model is used to analyze the object image information contained in the object demand data and generate the corresponding object visual features. The pre-defined convolutional network model can be a ResNet-50 convolutional network, where ResNet-50 stands for Residual Network with 50 layers. This model extracts features from the object image data, preserves shallow visual features through residual connections, and generates a high-dimensional encoding vector of the object visual features.

[0037] That is, the supplier features are extracted from the product supply data; these supplier features may include textual features and visual features of the product supplier. Specifically, the product supply data includes supplier user profile data; if the product supply data is textual, a preset semantic analysis model can be used to extract the product supplier textual features from the corresponding text; if the product supply data is image data, a preset convolutional network model can be used to analyze the image information contained in the product supply data and generate corresponding supplier visual features.

[0038] For data presented as supply chain graphs, a supply chain graph is constructed based on the relationships between the suppliers and the items. A graph sampling and aggregation model is then used to generate supply chain topology features based on this graph. The construction of the supply chain graph can be based on nodes contained in a supply chain industry knowledge graph. Different nodes correspond to various entity types, and the GraphSAGE algorithm (i.e., the graph sampling and aggregation model) is used to generate supply chain topology features. GraphSAGE (Graph Sample and Aggregate) samples the neighboring nodes of a given node in graph-structured data (such as knowledge graphs or topological networks), avoiding the computational cost of traversing the entire graph. The sampled neighboring node features are then fused with the target node features to generate a node feature vector containing topological relationships. In other words, the graph sampling and aggregation model extracts the topological features of the supply chain graph nodes and generates an encoded vector of the supply chain topology features.

[0039] Specifically, a preset semantic analysis model is used to parse the item demand text contained in the item demand data and extract item demand features from the item demand text; a preset convolutional network model is used to analyze the item image information contained in the item demand data and generate item visual features corresponding to the item image information; based on the relationship between the item supplier and the item, a supply chain graph is constructed, and a graph sampling aggregation model is used to generate supply chain topology features based on the supply chain graph; and item supplier features are extracted from the item supply data.

[0040] Step S302A: Generate the inventory fusion feature of the item based on the inventory data and various supply chain data.

[0041] Specifically, in this embodiment of the invention, the following methods are used to generate the inventory fusion features of the item: The first method: The inventory fusion feature is generated by fusing the characteristics of the product supplier, the characteristics of the product demand, the visual characteristics of the product, and the characteristics of the supply chain topology.

[0042] The second method involves fusing the characteristics of the product supplier, the product demand, the product visual characteristics, the supply chain topology characteristics, and the product procurement time series characteristics to generate the inventory fusion feature. Specifically, the supply chain data also includes product time series data, such as market price change data, logistics change data, and procurement time series data.

[0043] Furthermore, a pre-defined hybrid neural network model is used to extract one or more time-dimensional item time-series features from the item time-series data. This model can be an LSTM-TCN hybrid network. LSTM (Long Short-Term Memory) is used to capture long-range dependencies in time-series data, while TCN (Temporal Convolutional Network) efficiently captures multi-scale local time based on one-dimensional convolution through dilated convolution. The LSTM-TCN hybrid network, i.e., a hybrid network of long short-term memory and temporal convolution, allows one or more time dimensions to be 1 hour, 24 hours, 7 days, etc. Combining the long-range time-series modeling capability of LSTM with the multi-scale local feature extraction capability of TCN improves the dimensionality refinement of the extracted item time-series features. In other words, the supply chain data also includes item time-series data; a pre-defined hybrid neural network model is used to extract one or more time-dimensional item time-series features from this data.

[0044] Furthermore, generating the inventory fusion feature of the item based on various supply chain data includes: fusing the item supplier features, the item demand features, the item visual features, the supply chain topology features, and the item time-series features to generate the inventory fusion feature.

[0045] The third method: Based on the adjusted weights and the corresponding product supplier characteristics, product demand characteristics, product visual characteristics, and supply chain topology characteristics, the inventory fusion feature is generated by fusing them together. Specifically, the initial weights of the product supplier characteristics, product demand characteristics, product visual characteristics, and supply chain topology characteristics are obtained respectively; using a preset attention model, the initial weights of the product supplier characteristics, product demand characteristics, product visual characteristics, and supply chain topology characteristics are adjusted; based on the adjusted weights and the corresponding product supplier characteristics, product demand characteristics, product visual characteristics, and supply chain topology characteristics, the inventory fusion feature is generated by fusing them together.

[0046] Specifically, matrix multiplication calculations (e.g., the torch.matmul function) included in the preset attention model can be used to perform multiplication operations on feature tensors of different dimensions as the basic operation for calculating feature interactions, attention weights, and neural network layer outputs. Then, the weights of the item supplier features, the item demand features, the item visual features, and the supply chain topology features can be adjusted according to the calculation results. It can be understood that the sum of multiple weights is a fixed value, and any change in one weight means that the other weights also change accordingly.

[0047] Based on the adjusted weights and the corresponding characteristics of the item supplier, the item demand, the item visual features, and the supply chain topology features, the inventory fusion feature is generated.

[0048] The fourth method: a combination of the first to the third methods.

[0049] This invention generates inventory fusion features for items by extracting features of multiple types and dimensions, thereby improving the data richness for subsequent prediction of item inventory status and thus improving prediction accuracy and effect.

[0050] Step S103: Utilize the dynamic inventory forecasting model based on the inventory fusion features to predict the inventory status of the item; based on the inventory data and the predicted inventory status of the item, determine the replenishment strategy for the item.

[0051] Specifically, the dynamic inventory forecasting model includes a supplier forecasting model and a demand forecasting model; that is, the dynamic inventory forecasting model predicts the inventory status of the items based on the inventory fusion features, including: using the supplier forecasting model and the inventory fusion features to predict the supply status of the items; using the demand forecasting model and the inventory fusion features to predict the demand status of the items; and combining the supply status and the demand status to predict the inventory status of the items.

[0052] The following is combined with Figure 3B The diagram illustrates the model structure of the dynamic inventory forecasting model, such as... Figure 3B As shown, it includes: inventory fusion feature 301B, gating fusion unit 302B, goods supplier prediction model 303B, goods demander prediction model 304B, and joint loss function 305B.

[0053] Among them, inventory fusion feature 301B is used to generate inventory fusion features of the item based on various supply chain data.

[0054] The gating fusion unit 302B is used to adjust the prediction weights of the item supplier prediction model 303B for the item supply status and the item demander prediction model 304B for the item demand status based on the current item demand information or the current item supplier information.

[0055] The joint loss function 305B can be determined by the weights of the cross-entropy loss and mean squared error loss of the joint goods supplier forecasting model 303B and goods demander forecasting model 304B.

[0056] 1) For the prediction model of the supplier of goods, a deep cross network (DCN) trained based on the supplier's historical performance data can be used. The deep cross network includes the construction of a second-order feature interaction model, which constructs a non-linear relationship between the supplier's response speed (the supplier's response speed is a timeliness indicator that measures the supplier's response to the demand instructions initiated by the buyer, such as purchase orders, inquiry requests, change notices, etc.) and the type of goods.

[0057] Specifically, the second-order feature interaction model is shown in Equation (1): Xij = Xi × Xj × Wij (1) Wherein, Xij represents the second-order interaction feature value between supplier response speed and item type; Xi represents the basic feature vector (or feature value) of supplier response speed; Xj represents the basic feature vector (or feature value) of item type; Wij represents the interaction weight coefficient between supplier response speed (i.e., the speed at which items are provided) and item type; Xi and Xj can be normalized (with values ​​in the range of [0,1]), and Wij can be iteratively optimized through the backpropagation algorithm and converge to the interval [0,2] (which can be adjusted according to the actual business scenario) to ensure the stability and interpretability of the feature interaction results.

[0058] By using the "explicit interaction" of the second-order feature interaction model, we can clarify the strong correlation between Wij's learning process and the "supplier and item type" matching scenario.

[0059] Furthermore, by utilizing the aforementioned supplier prediction model and the aforementioned inventory fusion features, the supply status of the supplier (e.g., the supply response speed for the item) is predicted.

[0060] 2) For the demand-side prediction model for goods, a Transformer time-series prediction architecture can be adopted. During training, historical data with a certain time range (e.g., the past 30 days) is used, and an external attention mechanism is introduced to use market price fluctuations of goods as training data for feature enhancement. That is, the demand-side prediction model for goods and the inventory fusion features are used to predict the demand status of goods.

[0061] Furthermore, the step of predicting the item's inventory status by combining the item's supply status and the item's demand status includes: obtaining current item demand information or current item supplier information; adjusting the prediction weights for the item's supply status and demand status using a gating fusion unit based on the current item demand information or current item supplier information; and predicting the item's inventory status based on the adjusted prediction weights and the corresponding item supply status and demand status.

[0062] Specifically, the Gated Fusion Unit (GFU) is used for dynamically adaptive cross-domain feature fusion. For cross-domain features such as item supply status and item demand status (which have heterogeneous sources and large distribution differences), the GFU can calculate fusion weights in real time, overcoming the "demand-supply mismatch" problem caused by traditional fixed-weight fusion. For example, when the demand for a certain type of item surges (i.e., item demand status takes precedence), the GFU can automatically increase the prediction weight of the item demand status; when the supplier's ability to supply items fluctuates significantly, it can automatically adjust the prediction weight of the item supply status. That is, based on the current item demand information or the current item supplier information, the GFU adjusts the prediction weights for the item supply status and the item demand status.

[0063] Furthermore, the gating fusion unit can also dynamically offset the interference caused by cross-domain data distribution offsets. The predictive features of the demand status of goods originate from time-series data (such as market conditions, historical purchase volumes, etc.), while the predictive features of the supply status of goods may originate from structured data (such as historical supply data, supplier profiles, etc.). The distribution patterns of different types of data may change dynamically according to the market. The gating fusion unit can adjust the weight ratio in real time to make the prediction results of the inventory status of goods match the current actual inventory scenario.

[0064] By synergizing the logic of "item demand status forecasting" and "item supply status forecasting", we can overcome the problem of "over-replenishment (supplier cannot fulfill its obligations)" caused by relying solely on demand forecasting, as well as the problem of "insufficient replenishment (demand not met)" caused by relying solely on supply forecasting. This improves the overall accuracy and reliability of item inventory status forecasting, thereby enhancing the accuracy of replenishment decisions.

[0065] Furthermore, the specific method for adjusting the prediction weights of the current supply status and demand status of the item using the gating fusion unit is as follows: calculate the gate value gate of demand_feat (prediction feature of the demand status of the item) and supply_feat (prediction feature of the supply status of the item) using formula (2), and then adjust the prediction weights of the demand status and supply status of the item using the gate value.

[0066] gate = σ(Wg [demand_feat, supply_feat] + bg) (2)

[0067] Specifically, the preprocessed demand_feat and supply_feat can be concatenated in dimensions to generate a 2048-dimensional cross-domain feature vector [demand_feat, supply_feat]. The concatenated features are then linearly transformed into a 1024-dimensional vector using a learnable parameter matrix Wg (e.g., 1024×2048 in dimension), and then a bias term bg (e.g., 1024 in dimension) is added to generate a preliminary integration of the two features.

[0068] Furthermore, the linear transformation result is mapped to the [0,1] interval by normalizing the Sigmoid function (σ) to generate a gate value. The closer the gate value is to 1, the higher the importance of the predictive feature of the demand status of the inventory items; the closer it is to 0, the higher the importance of the predictive feature of the provision status of the items.

[0069] Furthermore, based on the adjusted prediction weights and the corresponding item supply status and item demand status, the item inventory status is predicted.

[0070] The adjusted prediction weights, the corresponding item supply status, and the item demand status are represented, for example, by formula (3). fused_output = gate demand_feat + (1-gate) supply_feat (3) Wherein, fused_output represents the result of fusing the regulated prediction weights and the corresponding item supply status and item demand status; gate represents the gate value calculated by formula (2). Demand_feat is positively weighted by gate, and supply_feat is negatively weighted by (1-gate), so that the sum of the weights of the two types of features is always 1, avoiding numerical drift in the fusion result.

[0071] Furthermore, during the feature fusion process, calculations can be performed separately for each dimension of the feature vector to achieve dynamic adaptation at the dimension level. For example, the gate value of the "short-term demand surge" dimension in the item demand status prediction feature is adjusted to 0.8 (higher weight), while the gate value of the "emergency fulfillment capability" dimension in the item supplier response prediction feature is adjusted to 0.2 (lower weight), thereby highlighting the short-term demand of the item while taking into account the basic supply capability of the item supplier.

[0072] Furthermore, the fused_output, as the final output of the dynamic inventory forecasting model (i.e., the dynamic inventory forecasting model that includes both the item supplier forecasting model and the item demander forecasting model), can be directly input into the real-time decision feedback system to determine the replenishment strategy for the item (such as replenishment quantity, replenishment priority, and the quantity that the item supplier can supply), thereby realizing a closed loop of inventory decision-making in the two dimensions of "demand-supply". That is, based on the predicted inventory status of the item, the replenishment strategy for the item is determined.

[0073] Figure 2 A schematic diagram of the structure of an inventory information management system is shown, such as... Figure 2 As shown, it includes: a data acquisition layer 201, a feature extraction engine 202, a dynamic inventory prediction model 203, a real-time decision feedback system 204, and a replenishment terminal 205; wherein, the dynamic inventory prediction model 203, the real-time decision feedback system 204, and the replenishment terminal 205 can achieve a closed loop of inventory decision-making through online learning of a preset prediction model during the prediction process. The data acquisition layer 201 is used to acquire inventory data of items in the warehouse and various supply chain data associated with the items; the feature extraction engine 202 is used to generate inventory fusion features of the items based on the various supply chain data; the dynamic inventory prediction model 203 is used to predict the inventory status of the items based on the inventory fusion features; the replenishment terminal 205 receives the replenishment strategy for the items determined based on the inventory data and the predicted inventory status of the items.

[0074] Figure 3C A schematic diagram of an inventory information management system is shown. (For example...) Figure 3CAs shown, it includes: a data stream processing layer 301C, a dynamic inventory prediction model adjustment layer 302C, a replenishment strategy determination layer 303C, and an online learning layer 304C. The data stream processing layer 301C is a real-time data processing pipeline built using an architecture combining Flink and Kafka, used to collect supply chain data. This architecture improves the efficiency of model prediction by achieving feature calculation latency <500ms, model inference response <200ms, and strategy execution triggering <100ms. The dynamic inventory prediction model adjustment layer 302C is used to obtain current item demand information or current item supplier information. Based on the current item demand information or current item supplier information, the prediction weights for the current item supply status and item demand status are adjusted using a gating fusion unit. Based on the adjusted prediction weights and the corresponding item supply status and item demand status, the item inventory status is predicted. The replenishment strategy determination layer 303C is used to determine the replenishment strategy for the item based on the item inventory data and the predicted item inventory status.

[0075] The following description Figure 2 and Figure 3C It includes a real-time decision feedback system and its online learning steps.

[0076] Specifically, Figure 2 Includes real-time decision feedback system 204 and Figure 3C The included online learning layer 304C provides feedback to the dynamic inventory prediction model, thereby dynamically optimizing the model. Specifically, it acquires replenishment data executed according to the warehouse's replenishment strategy; determines the warehouse's current item conversion rate, inventory turnover rate, and item shortage rate based on the replenishment data; constructs a prediction reward function for a deep reinforcement learning model based on the item conversion rate, inventory turnover rate, item shortage rate, and weights; and optimizes the dynamic inventory prediction model using the real-time reward value calculated by the deep reinforcement learning model and the prediction reward function.

[0077] Furthermore, the dynamic inventory forecasting model includes a supplier forecasting model and a demand forecasting model; the optimization of the dynamic inventory forecasting model using the deep reinforcement learning model and the calculation results of the forecasting reward function includes: optimizing the forecasting weights of the supplier forecasting model and the demand forecasting model using the deep reinforcement learning model and the real-time reward value.

[0078] The predicted reward function is shown in formula (4): R=α Item conversion rate + β Inventory turnover rate - γ Out-of-stock rate (4) Specifically, R represents the real-time reward value, the item conversion rate represents the proportion of replenished items that are purchased; the item inventory turnover rate represents the turnover efficiency of the inventory after item replenishment; the out-of-stock rate represents the frequency of out-of-stock still occurring after item replenishment; α, β, and γ represent weights, where the values of α, β, and γ can be determined through historical data. For example: α = 0.4, β = 0.3, γ = 0.3.

[0079] The state space of the deep reinforcement learning model can include: the current inventory data of items, the supply data of item suppliers, the current item demand data (such as market popularity), etc.

[0080] Furthermore, compare the real-time reward value R with the historical optimal reward value Rh to generate a feedback signal; if R > Rh, it means that the weight ratio of the current item supplier prediction model and the item demander prediction model is adapted to the current item replenishment business scenario, then a positive feedback signal is generated; if R < Rh, it means that there is a deviation in the weight ratio of the current item supplier prediction model and the item demander prediction model (for example: the branch weight of the item demander prediction model is too high, resulting in out-of-stock, or the branch weight of the item supplier prediction model is too high, resulting in inventory backlog), then a negative feedback signal is generated; if R ≈ Rh, a maintenance feedback signal is generated to maintain the current weight ratio.

[0081] Furthermore, input the feedback signal into the deep reinforcement learning model (such as DCN), that is, the dynamic inventory prediction model used includes an item supplier prediction model and an item demander prediction model; optimizing the dynamic inventory prediction model using the deep reinforcement learning model and the calculation result of the prediction reward function includes: comparing the real-time reward value with the historical optimal reward value, generating a feedback signal according to the comparison result; using the deep reinforcement learning model to optimize the prediction weights of the item supplier prediction model and the item demander prediction model based on the feedback signal.

[0082] For example, when the feedback signal is determined to be a negative feedback signal, it is determined that the prediction weights of the supply-side forecasting model and the demand-side forecasting model need to be adjusted. For instance, when determining the real-time reward value of the "out-of-stock rate" term in the prediction reward function, if it indicates that the feature contribution of the demand-side forecasting model branch is insufficient (i.e., it does not accurately capture demand) or the supply-side forecasting model branch excessively restricts replenishment, the deep reinforcement learning model will output a weight adjustment instruction to increase the prediction weight of the demand-side forecasting model branch (e.g., from 0.5 to 0.6) and simultaneously adjust the prediction weight of the supply-side forecasting model branch (e.g., from 0.5 to 0.4) to ensure that the replenishment volume can cover the actual demand. When it is determined that the "inventory turnover rate" term is too low, resulting in insufficient real-time reward value, if it indicates that the prediction weight of the demand-side forecasting model branch is too high (over-replenishment) or the prediction weight of the supply-side forecasting model branch is insufficient (failure to consider fulfillment capability), the deep reinforcement learning model will output a weight adjustment instruction, which can reduce the prediction weight of the demand-side forecasting model branch (e.g., from 0.6). Adjust to 0.45), and increase the prediction weight of the supplier prediction model branch (e.g., from 0.4 to 0.55) to avoid ineffective replenishment beyond the supplier's fulfillment capacity; when "conversion rate" is the core contribution item: it means that the current weight ratio of the two branches balances the demand and supply of the item, and the weight can be optimized through small iterations (e.g., ±0.01) to maintain the stability of item replenishment.

[0083] Preferably, the boundaries of the prediction weights are constrained. To overcome the extreme nature of the prediction weights of branches (e.g., the prediction weights of the supplier prediction model branches are ≥0.9 or ≤0.1), a weight adjustment range (e.g., [0.2, 0.8]) can be set. Furthermore, when the fluctuation range of the reward value R for three consecutive iterations is ≤5%, it is determined that the weight adjustment has entered a stable state. This allows the prediction weights to be maintained, and the step of re-triggering the adjustment of prediction weights can be triggered when there are significant changes in the replenishment data (e.g., sudden demand or supplier supply crisis).

[0084] By using online learning of deep reinforcement learning models (implemented through online learning layer 304C), incremental model updates can be performed at set time intervals (e.g., every hour); model stability can also be ensured through A / B testing and model switching in hot deployment mode (i.e., service without interruption), thus improving the real-time optimization of dynamic inventory forecasting models.

[0085] More preferably, the prediction weights of the item supplier prediction model and the item demand prediction model, adjusted by the deep reinforcement learning model, can be synchronized in real time to the gating fusion unit (GPU) through a parameter server architecture. This allows the gating fusion unit to regulate the fusion of the prediction weights for the current item supply status and the item demand status, thereby improving the prediction effect of the dynamic inventory prediction model through linkage.

[0086] like Figure 4 As shown, this embodiment of the invention provides an inventory information management device 400, including: a data acquisition module 401, a feature extraction module 402, and a strategy determination module 403; wherein, The data acquisition module 401 is used to acquire inventory data of items in the warehouse and various supply chain data associated with the items; The feature extraction module 402 is used to generate inventory fusion features of the item based on various supply chain data. The strategy determination module 403 is used to predict the inventory status of the item based on the inventory fusion features using a dynamic inventory prediction model; and to determine the replenishment strategy for the item based on the inventory data and the predicted inventory status.

[0087] This invention also provides an electronic device for inventory information management, comprising: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method provided in any of the above embodiments.

[0088] This invention also provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the method provided in any of the above embodiments.

[0089] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements any of the methods described above for inventory information management.

[0090] Figure 5 An exemplary system architecture 500 is shown for which the inventory information management method or inventory information management device of the present invention can be applied.

[0091] like Figure 5 As shown, system architecture 500 may include terminal devices 501, 502, and 503, a network 504, and a server 505. Network 504 serves as the medium for providing communication links between terminal devices 501, 502, and 503 and server 505. Network 504 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0092] Users can use terminal devices 501, 502, and 503 to interact with server 505 via network 504 to receive or send messages, etc. Various client applications can be installed on terminal devices 501, 502, and 503, such as e-commerce client applications, web browser applications, search applications, instant messaging tools, and email clients.

[0093] Terminal devices 501, 502, and 503 can be various electronic devices with displays and supporting various client applications, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0094] Server 505 can be a server that provides various services, such as a back-end management server that supports client applications used by users through terminal devices 501, 502, and 503. The back-end management server can perform feature extraction and replenishment prediction on received supply chain data and feed back replenishment strategies to the terminal devices.

[0095] It should be noted that the inventory information management method provided in this embodiment of the invention is generally executed by server 505, and correspondingly, the inventory information management device is generally set in server 505.

[0096] It should be understood that Figure 5 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0097] The following is for reference. Figure 6 It shows a schematic diagram of the structure of a computer system 600 suitable for implementing a terminal device of the present invention. Figure 6 The terminal device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0098] like Figure 6 As shown, the computer system 600 includes a central processing unit (CPU) 601, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 602 or programs loaded from storage section 608 into random access memory (RAM) 603. The RAM 603 also stores various programs and data required for the operation of the system 600. The CPU 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0099] The following components are connected to I / O interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to I / O interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 610 as needed so that computer programs read from it can be installed into storage section 608 as needed.

[0100] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 609, and / or installed from removable medium 611. When the computer program is executed by central processing unit (CPU) 601, it performs the functions defined above in the system of this invention.

[0101] It should be noted that the computer-readable medium shown in this invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0102] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0103] The modules and / or units described in the embodiments of the present invention can be implemented in software or hardware. The described modules and / or units can also be housed in a processor; for example, a processor can be described as including a data acquisition module, a feature extraction module, and a strategy determination module. The names of these modules do not necessarily limit the module itself; for example, a data acquisition module can also be described as "a module for acquiring inventory data of items in a warehouse and various supply chain data."

[0104] In another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiments; or it may exist independently and not assembled into the device. The computer-readable medium carries one or more programs that, when executed by the device, cause the device to: generate inventory fusion features of the items based on inventory data of the items and various supply chain data associated with the items; predict the inventory status of the items using a dynamic inventory forecasting model based on the inventory fusion features; and determine a replenishment strategy for the items based on the inventory data of the items and the predicted inventory status. The embodiments of the present invention improve the accuracy, rationality, and efficiency of inventory information management by using a dynamic inventory forecasting model based on inventory fusion features of various supply chain data to predict the inventory status of items, thereby determining a replenishment strategy for replenishment operations.

[0105] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. An inventory information management method, characterized in that, include: Acquire inventory data of items in the warehouse, as well as various supply chain data associated with those items; Based on various supply chain data, an inventory fusion feature for the item is generated; Based on the aforementioned inventory fusion features, the dynamic inventory forecasting model is used to predict the inventory status of the item. Based on the inventory data and the predicted inventory status of the items, a replenishment strategy for the items is determined.

2. The method according to claim 1, characterized in that, The supply chain data includes goods supply data, goods demand data, and the relationship between goods suppliers and goods; Before generating the inventory fusion characteristics of the item based on various supply chain data, the process further includes: Using a pre-defined semantic analysis model, the item demand data is parsed to extract the item demand text and the item demand features are extracted from the item demand text. Using a pre-defined convolutional network model, the item image information contained in the item demand data is analyzed, and the item visual features corresponding to the item image information are generated. Based on the relationships between the suppliers and the goods, a supply chain graph is constructed, and a graph sampling aggregation model is used to generate supply chain topology features based on the supply chain graph. Extract the characteristics of the item suppliers from the item supply data.

3. The method according to claim 2, characterized in that, The step of generating inventory fusion features for the item based on various supply chain data includes: The inventory fusion feature is generated by integrating the characteristics of the product supplier, the characteristics of the product demand, the visual characteristics of the product, and the characteristics of the supply chain topology.

4. The method according to claim 2, characterized in that, The supply chain data also includes item time-series data; Using a pre-defined hybrid neural network model, one or more time-dimensional features of the items are extracted from the item time-series data; The step of generating inventory fusion features for the item based on various supply chain data includes: The inventory fusion feature is generated by integrating the characteristics of the product supplier, the characteristics of the product demand, the visual characteristics of the product, the characteristics of the supply chain topology, and the characteristics of the product time sequence.

5. The method according to any one of claims 3 or 4, characterized in that, The process of generating the inventory fusion feature by fusing the product supplier characteristics, product demand characteristics, product visual characteristics, and supply chain topology characteristics includes: The initial weights of the product supplier characteristics, the product demand characteristics, the product visual characteristics, and the supply chain topology characteristics are obtained respectively. Using a preset attention model, adjust the initial weights of the item supplier characteristics, the item demand characteristics, the item visual characteristics, and the supply chain topology characteristics; Based on the adjusted weights and the corresponding characteristics of the item supplier, the item demand, the item visual features, and the supply chain topology features, the inventory fusion feature is generated.

6. The method according to claim 1, characterized in that, The dynamic inventory forecasting model includes a goods supplier forecasting model and a goods demand-side forecasting model. The method of using a dynamic inventory forecasting model to predict the inventory status of the item based on the inventory fusion features includes: Using the aforementioned supplier prediction model and the aforementioned inventory fusion features, the item supply status of the supplier is predicted; The demand status of goods is predicted using the demand-side forecasting model and the inventory fusion features. Based on the supply status and demand status of the item, predict the inventory status of the item.

7. The method according to claim 6, characterized in that, The step of predicting the inventory status of an item by combining its supply status and demand status includes: Obtain current item demand information or current item supplier information, and based on the current item demand information or current item supplier information, use the gating fusion unit to adjust the prediction weights of the current item supply status and the item demand status; Based on the adjusted prediction weights and the corresponding item supply status and item demand status, the item inventory status is predicted.

8. The method according to claim 1, characterized in that, Further includes: Obtain replenishment data executed according to the warehouse's item replenishment strategy; Based on the replenishment data, determine the current item conversion rate, inventory turnover rate, and item stockout rate of the warehouse; A prediction reward function for a deep reinforcement learning model is constructed based on the item conversion rate, the inventory turnover rate, the item stockout rate, and the weights. The dynamic inventory prediction model is optimized using a deep reinforcement learning model and the real-time reward value calculated by the prediction reward function.

9. The method according to claim 8, characterized in that, The dynamic inventory forecasting model includes a goods supplier forecasting model and a goods demand-side forecasting model. The optimization of the dynamic inventory prediction model using the calculation results of the deep reinforcement learning model and the prediction reward function includes: The real-time reward value is compared with the historical best reward value, and a feedback signal is generated based on the comparison result; The deep reinforcement learning model is used to optimize the prediction weights of the commodity supplier prediction model and the commodity demander prediction model based on the feedback signal.

10. An inventory information management device, characterized in that, include: The module comprises a data acquisition module, a feature extraction module, and a strategy determination module; among which, The data acquisition module is used to acquire inventory data of items in the warehouse and various supply chain data associated with the items; The feature extraction module is used to generate inventory fusion features of the item based on various supply chain data. The strategy determination module is used to predict the inventory status of the item based on the inventory fusion features using a dynamic inventory prediction model; and to determine the replenishment strategy for the item based on the inventory data and the predicted inventory status.

11. An electronic device, characterized in that, include: One or more processors; Storage device for storing 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 1-9.

12. A computer-readable medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-9.

13. A computer program product, comprising a computer program, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-9.

Images