Inventory management optimization method and device, storage medium and electronic equipment

Abstract

The invention discloses an inventory management optimization method and device, a storage medium and electronic equipment. The method comprises the following steps: acquiring historical inventory data of a plurality of cycles; according to the historical inventory data, determining the historical optimal order quantity of each period in the plurality of periods; and inputting the historical optimal order quantity and the target feature data into a multi-product inventory optimization neural network model to obtain a target optimal order quantity, the multi-product inventory optimization neural network model is integrated with at least one of a multi-layer perceptron module, a long short-term memory network algorithm module, a graph learning module, a graph convolutional network module and an attention mechanism module. Through application of the method and the device, the problem of relatively low inventory management optimization efficiency caused by lack of analysis of relevance between modeled commodities in a single-cycle mode in related technologies is solved.

Description

Background Art

[0002] In related technologies, for the multi-cycle and multi-product inventory management problem in e-commerce scenarios, previous data-driven inventory management research mostly considered demand estimation and inventory optimization separately, which led to suboptimal solutions due to the inconsistency of the two-stage optimization objectives. Although some studies have introduced integrated optimization methods into inventory management in recent years, most of them consider single-cycle problems, lack analysis of the correlation between modeled products, and do not fully utilize the rich and available non-demand information in e-commerce scenarios.

[0003] Currently, no effective solution has been proposed to address the problem that in the single-cycle mode, related technologies lack analysis of the correlation between modeled commodities, which leads to low efficiency in inventory management optimization. Summary of the invention

[0004] The main purpose of the present application is to provide an inventory management optimization method, device, storage medium and electronic device to solve the problem of low inventory management optimization efficiency in related technologies.

[0005] In order to achieve the above-mentioned purpose, according to the first aspect of the present application, an inventory management optimization method is provided. The method comprises: obtaining historical inventory data of multiple periods; determining the historical optimal order quantity of each period in the multiple periods according to the historical inventory data; inputting the historical optimal order quantity and the target feature data into a multi-product inventory optimization neural network model to obtain the target optimal order quantity, wherein the target feature data comprises at least one of the following: inventory feature data, commodity feature data, historical commodity demand data, promotion feature data, and derivative feature data, wherein the multi-product inventory optimization neural network model integrates at least one of the following: a multi-layer perceptron module, a long short-term memory network algorithm module, a graph learning module, a graph convolutional network module, and an attention mechanism module.

[0006] Furthermore, the historical optimal order quantity of each cycle in multiple cycles is determined based on the historical inventory data, including: determining the initial inventory data and the actual order quantity of each cycle based on the historical inventory data; and determining the order demand information and the current cycle inventory data of two adjacent cycles in the multiple cycles; determining the historical optimal order quantity based on the initial inventory data, the actual order quantity, the order demand information, and the current cycle inventory data.

[0007] Furthermore, the historical optimal order quantity is determined based on the initial inventory data, the actual order quantity, the order demand information, and the current cycle inventory data, including: calculating the current commodity cost between orders based on the initial inventory data, the actual order quantity, the order demand information, and the current cycle inventory data, wherein the current commodity cost includes: cycle holding cost, cycle out-of-stock cost, wherein the current commodity cost is the minimum cost generated from the delivery time of the commodity order within the target cycle to all subsequent cycles; determining the minimum total cost of the commodity in each cycle based on the current commodity cost; and calculating the minimum total cost of the commodity through a reverse algorithm to obtain the historical optimal order quantity.

[0008] Furthermore, the multi-product inventory optimization neural network model integrates a graph learning module and a graph convolutional network module. The target feature data is the commodity feature data. The historical optimal order quantity and the target feature data are input into the multi-product inventory optimization neural network model to obtain the target optimal order quantity, including: determining the association information between each commodity according to the graph learning module; inputting the association information and the historical optimal order quantity into the graph convolutional network module to obtain the target optimal order quantity.

[0009] Furthermore, the association information between each commodity is determined according to the graph learning module, including: determining the commodity node corresponding to each commodity; calculating the embedding vector of each commodity node, wherein the embedding vector is used to represent the commodity node feature; calculating the adjacency matrix between each commodity based on the embedding vector; and determining the association information based on the adjacency matrix.

[0010] Furthermore, the multi-product inventory optimization neural network model integrates a long short-term memory network algorithm module, a multi-layer perceptron module, and an attention mechanism module. The target feature data are historical commodity demand data and promotion feature data. The historical optimal order quantity and the target feature data are input into the multi-product inventory optimization neural network model to obtain the target optimal order quantity, including: inputting the historical commodity demand data into the long short-term memory network algorithm module to obtain a commodity time series vector, wherein the commodity time series vector is used to capture the correlation between commodities in the time dimension; and, inputting the promotion feature data into the multi-layer perceptron module to obtain a promotion feature vector; inputting the commodity time series vector and the promotion feature vector into the attention mechanism module to obtain attention weight information; and determining the target optimal order quantity according to the attention weight information.

[0011] Furthermore, the target optimal order quantity is determined based on the attention weight information, including: normalizing the attention weight information to obtain processed attention weight information, weighting the hidden state of the product time series vector based on the processed attention weight information to obtain the target product time series vector; inputting the target product time series vector into a graph convolutional network module to obtain the target optimal order quantity.

[0012] In order to achieve the above-mentioned purpose, according to the second aspect of the present application, an inventory management optimization device is provided. The device includes: an acquisition unit, which is used to acquire historical inventory data of multiple periods; a first determination unit, which is used to determine the historical optimal order quantity of each period in the multiple periods according to the historical inventory data; a second determination unit, which is used to input the historical optimal order quantity and the target feature data into a multi-product inventory optimization neural network model to obtain the target optimal order quantity, wherein the target feature data includes at least one of the following: inventory feature data, commodity feature data, historical commodity demand data, promotion feature data, and derivative feature data, wherein the multi-product inventory optimization neural network model integrates at least one of the following: a multi-layer perceptron module, a long short-term memory network algorithm module, a graph learning module, a graph convolution network module, and an attention mechanism module.

[0013] Furthermore, the first determination unit includes: a first determination module, used to determine the initial inventory data and actual order quantity of each cycle based on historical inventory data; and a second determination module, used to determine the order demand information and current cycle inventory data of two adjacent cycles among multiple cycles; a third determination module, used to determine the historical optimal order quantity based on the initial inventory data, actual order quantity, order demand information, and current cycle inventory data.

[0014] Furthermore, the third determination module includes: a first calculation submodule, which is used to calculate the current commodity cost between orders based on the initial inventory data, the actual order quantity, the order demand information, and the current cycle inventory data, wherein the current commodity cost includes: cycle holding cost, cycle out-of-stock cost, wherein the current commodity cost is the minimum cost generated from the delivery time of the commodity order within the target cycle to all subsequent cycles; the first determination submodule, which is used to determine the minimum total cost of the commodity in each cycle based on the current commodity cost; the second calculation submodule, which is used to calculate the minimum total cost of the commodity through the reverse algorithm to obtain the historical optimal order quantity.

[0015] Furthermore, the multi-product inventory optimization neural network model integrates a graph learning module and a graph convolutional network module. The target feature data is the commodity feature data. The second determination unit includes: a fourth determination module, which is used to determine the association information between each commodity according to the graph learning module; and a fifth determination module, which is used to input the association information and the historical optimal order quantity into the graph convolutional network module to obtain the target optimal order quantity.

[0016] Furthermore, the fourth determination module includes: a second determination submodule, used to determine the product node corresponding to each product; a third calculation submodule, used to calculate the embedding vector of each product node, wherein the embedding vector is used to represent the product node characteristics; a fourth calculation submodule, used to calculate the adjacency matrix between each product based on the embedding vector; and a third determination submodule, used to determine the association information based on the adjacency matrix.

[0017] Furthermore, the multi-product inventory optimization neural network model integrates a long short-term memory network algorithm module, a multi-layer perceptron module, and an attention mechanism module. The target feature data are historical commodity demand data and promotion feature data. The second determination unit includes: a first data input module, used to input the historical commodity demand data into the long short-term memory network algorithm module to obtain a commodity time series vector, wherein the commodity time series vector is used to capture the correlation between commodities in the time dimension; and a second data input module, used to input the promotion feature data into the multi-layer perceptron module to obtain a promotion feature vector; a third data input module, used to input the commodity time series vector and the promotion feature vector into the attention mechanism module to obtain attention weight information; and a sixth determination module, used to determine the target optimal order quantity according to the attention weight information.

[0018] Furthermore, the sixth determination module includes: a normalization submodule, which is used to normalize the attention weight information to obtain the processed attention weight information, and weight the hidden state of the product time series vector based on the processed attention weight information to obtain the target product time series vector; a vector input submodule, which is used to input the target product time series vector into the graph convolutional network module to obtain the target optimal order quantity.

[0019] According to a third aspect of an embodiment of the present application, there is provided an electronic device, comprising: a processor; and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, any one of the above-mentioned inventory management optimization methods is implemented.

[0020] According to a fourth aspect of an embodiment of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, an inventory management optimization method according to any one of the above items is implemented.

[0021] Through this application, the following steps are adopted: This application discloses an inventory management optimization method, device, storage medium and electronic device. The method includes: obtaining historical inventory data of multiple periods; determining the historical optimal order quantity of each period in multiple periods according to the historical inventory data; inputting the historical optimal order quantity and the target feature data into the multi-product inventory optimization neural network model to obtain the target optimal order quantity, wherein the target feature data includes at least one of the following: inventory feature data, commodity feature data, historical commodity demand data, promotion feature data, and derivative feature data, wherein the multi-product inventory optimization neural network model integrates at least one of the following: multi-layer perceptron module, long short-term memory network algorithm module, graph learning module, graph convolution network module, and attention mechanism module. Through this application, the problem of the lack of analysis of the correlation between modeled commodities in the single-period mode in the related technology is solved, which leads to low inventory optimization efficiency. By obtaining historical inventory data of multiple periods, the historical inventory data is analyzed based on the multi-product inventory optimization neural network model to obtain the target optimal order quantity, thereby achieving the effect of improving inventory optimization efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0023] Figure 1 is a flow chart of an inventory management optimization method provided according to an embodiment of the present application;

[0024] Figure 2 is a schematic diagram of a multi-product inventory optimization neural network model framework provided according to an embodiment of the present application;

[0025] Figure 3 It is a schematic diagram of commodity dependency provided according to an embodiment of the present application;

[0026] Figure 4 is a flow chart of an inventory management optimization device provided according to an embodiment of the present application;

[0027] Figure 5 It is a schematic diagram of the network architecture of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0028] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0029] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0030] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] According to an embodiment of the present application, a method for optimizing inventory management is provided.

[0032] Figure 1 Flowchart of the inventory management optimization method according to an embodiment of the present application. Figure 1 As shown, the method comprises the following steps:

[0033] Step S101, obtaining historical inventory data of multiple periods.

[0034] Among them, the cycles in the multiple cycles can be different time periods of a month, a quarter or a year. For example, this application selects historical inventory data for 12 months a year to conduct inventory analysis, predict future demand, evaluate inventory management efficiency, etc.

[0035] Step S102, determining the historical optimal order quantity for each of the multiple cycles based on the historical inventory data.

[0036] Among them, historical inventory data includes commodity prices, commodity sales (commodity demand), commodity inventory, inventory costs, etc. at each decision cycle stage. This application dynamically solves the historical optimal order quantity of each cycle through historical inventory data, and can flexibly and efficiently obtain the optimal solution for each cycle, providing strong data support for subsequent inventory optimization forecasts.

[0037] In some embodiments, the historical optimal order quantity for each of multiple cycles is determined based on historical inventory data, which can be obtained through the following steps: determining the initial inventory data and the actual order quantity for each cycle based on the historical inventory data; and determining the order demand information and current cycle inventory data of two adjacent cycles in the multiple cycles; determining the historical optimal order quantity based on the initial inventory data, the actual order quantity, the order demand information, and the current cycle inventory data.

[0038] For example, for any historical inventory item i, the following dynamic programming model can be established: Each cycle of the dynamic programming model is regarded as a decision day, each decision day t is regarded as a stage, t∈{1, 2, ...T}, there are T stages in total, and the initial inventory data I of each stage is obtained. i，t , and use the initial inventory data as the state of the current stage, where I i，t is a non-negative integer, that is, I i，t ∈{0, 1, 2, ...}. The actual order quantity is the order quantity on each decision day. Assume that the decision variable on that decision day is q i，t ,q i，t is a non-negative integer, that is, q i，t ∈{0, 1, 2, 3...}, two adjacent cycles will generate a certain amount of demand d i，t , that is, the order demand information, which reduces the inventory. At the same time, the orders from the previous stage will arrive, which increases the inventory. The current inventory data is determined based on the order demand information. The inventory data of the current cycle can be solved by the state transfer equation, which can be obtained by the following formula:

[0039]

[0040] Among them, I i，t+1 is the current inventory data, I i，t is the initial inventory data, d i，t is the order demand information, q i，t is the actual order quantity, t is the decision date, t k is the tth decision cycle, and L is the lead time.

[0041] This application determines the historical optimal order quantity based on the initial inventory data, actual order quantity, order demand information, and current cycle inventory data, which can reduce inventory costs while achieving efficient optimization of inventory management.

[0042] In some embodiments, the historical optimal order quantity is determined based on the initial inventory data, the actual order quantity, the order demand information, and the current cycle inventory data, which can be obtained through the following steps: the current commodity cost between orders is calculated based on the initial inventory data, the actual order quantity, the order demand information, and the current cycle inventory data, wherein the current commodity cost includes: cycle holding cost, cycle out-of-stock cost, wherein the current commodity cost is the minimum cost generated from the delivery time of the commodity order within the target cycle to all subsequent cycles; the minimum total cost of the commodity in each cycle is determined based on the current commodity cost; the minimum total cost of the commodity is calculated by a reverse algorithm to obtain the historical optimal order quantity.

[0043] For example, in order to better represent the recursive relationship between costs, the cost generated when two orders arrive is used as a bridge to establish an indicator function, which is used to represent the current commodity cost. Suppose the kth order of commodity i is in a i，k When it arrives, the indicator function represents the i，k The holding costs and out-of-stock costs generated after It means that from t = a i，k Initially, the inventory is The lowest cost incurred in all subsequent stages (i.e., the current commodity cost in this case) is calculated by calculating the holding cost and out-of-stock cost during this period. The current commodity cost can be expressed by an indicator function formula as follows:

[0044]

[0045] in, is the current commodity cost, h i is the holding cost of commodity i, For the current product i in a i，k Inventory at the moment, q i，k The actual order quantity of commodity i in the kth period, For a i，k The order demand at the moment, b i is the out-of-stock cost of commodity i, From time i to a i，k The order demand generated at any time, V i，k+1 It means that from t = a i，k+1 Initially, the inventory is The lowest cost in all subsequent stages. From time i to a i，k+1 Order requirements generated at any time.

[0046] Among them, the minimum total cost of the product is also the optimal value of the above indicator function. Based on this, the minimum total cost of the product can be obtained by the following formula:

[0047]

[0048] in, is the minimum cost of the product, is the optimal value function, h i is the holding cost of commodity i, For the current product i in a i，k Inventory at the moment, q i，k The actual order quantity of commodity i in the kth period, For a i，k The order demand at the moment, b i is the out-of-stock cost of commodity i, V i，k+1 It means that from t = a i，k+1 Initially, the inventory is The lowest cost in all subsequent stages. From time i to a i，k Order requirements generated at any time, From time i to a i,k+1 Order requirements generated at any time.

[0049] That is, after obtaining the optimal value function, the inverse method can be used to solve the optimal order quantity for each stage, and then used as a label for neural network training to achieve the effect of inputting features to output the historical optimal order quantity.

[0050] Step S103, input the historical optimal order quantity and the target feature data into the multi-product inventory optimization neural network model to obtain the target optimal order quantity, wherein the target feature data includes at least one of the following: inventory feature data, commodity feature data, historical commodity demand data, promotion feature data, and derivative feature data, wherein the multi-product inventory optimization neural network model integrates at least one of the following: a multi-layer perceptron module, a long short-term memory network algorithm module, a graph learning module, a graph convolutional network module, and an attention mechanism module.

[0051] For example, Figure 2 As one of the implementation methods of the above steps that can be implemented in the embodiment of the present application, the following multi-product inventory optimization neural network (Multi-Product Inventory Optimization Neural Network, MPIONN) is proposed. The structure of the model is as follows: Figure 2 As shown in the figure: The training target / output of the MPIONN model is the optimal order quantity previously obtained through the dynamic programming model, and the network input is divided into five parts: inventory feature data, product feature data, historical product demand data, promotion feature data, and other derived feature data.

[0052] (1) Inventory characteristics: including the current inventory quantity, inventory holding costs, inventory out-of-stock costs, etc.

[0053] (2) Product characteristics: including product category, product brand, product attributes, etc.

[0054] (3) Historical demand: includes the daily sales volume of the product in the past few days, reflecting the sales trend of the product over a period of time.

[0055] (4) Promotional characteristics: types of promotional activities, promotion intensity, number of days until the start / end of the promotional activities, price, and other characteristics designed for various promotional activities in e-commerce scenarios.

[0056] (5) Other features: Other features include some of the features established in the previous article based on feature engineering, including date-derived features, demand sliding window features, category feature encoding, etc.

[0057] Among them, for inventory features, in order to enable the model to make the optimal order quantity decision, it is necessary to understand the current inventory level and the inventory holding cost and out-of-stock cost information. Therefore, this application uses MLP (multi-layer perceptron module) to map it to a high-dimensional hidden layer vector, and send it to the feature fusion module of the last layer to splice it with the high-dimensional hidden layer vectors generated by other modules for fusion learning.

[0058] Regarding product features, the present invention takes into account the possible correlation between multiple products. First, the static features related to the products are input into a graph learning module to identify whether there is a correlation between each product. Then, an adjacency matrix (product dependency graph) is output for use by GCN (graph convolutional network module). In the GCN module, the constructed adjacency matrix and the hidden layer vectors of each product are used as input to model the correlation between multiple products, so as to obtain better inventory optimization results.

[0059] For historical demand, since historical demand is a time series data, it contains the real demand for goods in the past period of time and reflects the sales trend of goods in the past period of time. Therefore, the present invention uses LSTM (long short-term memory network algorithm module) suitable for sequence feature modeling to capture the temporal dependence of commodity demand.

[0060] For promotional features, they are first sent to the MLP module and mapped to a high-dimensional hidden layer vector. Then, the vector output by the LSTM module is weighted in the Attention module as a query to further extract the demand pattern of current products by utilizing the rich promotional activity information in the e-commerce scenario.

[0061] For other features established by feature engineering, they are also sent to the MLP module and mapped to a high-dimensional hidden layer vector. These features established based on feature engineering are helpful to help the learning of the neural network. Finally, the output of the module is directly sent to the top layer of the network and interacts with the outputs of other modules.

[0062] By inputting the above-mentioned feature data into the multi-product inventory optimization neural network, the present application can better utilize the various information of various commodities and capture the correlation between multiple products to improve the final inventory optimization effect.

[0063] In some embodiments, the multi-product inventory optimization neural network model integrates a graph learning module and a graph convolutional network module. The target feature data is the product feature data. The historical optimal order quantity and the target feature data are input into the multi-product inventory optimization neural network model to obtain the target optimal order quantity, which can be obtained through the following steps: determining the association information between each product according to the graph learning module; inputting the association information and the historical optimal order quantity into the graph convolutional network module to obtain the target optimal order quantity.

[0064] Exemplarily, considering the possible correlation between multiple products, the static features related to the products are first input into a graph learning module to identify whether there is a correlation between each product. If so, the correlation information is used as an adjacency matrix and input into the graph convolutional network module. The historical optimal order quantity is used as a training label and input into the graph convolutional network module. The correlation information is mapped to the historical optimal order quantity to obtain the expected target optimal order quantity. Of course, the target optimal order quantity in this case can also be obtained by adding multiple feature data and training labels, which is not limited here.

[0065] More specifically, the association information between each product can be determined according to the graph learning module through the following steps: determining the product node corresponding to each product; calculating the embedding vector of each product node, wherein the embedding vector is used to represent the product node feature; calculating the adjacency matrix between each product based on the embedding vector; and determining the association information based on the adjacency matrix.

[0066] For example, Figure 3 As shown in the figure, the products are taken as nodes V of the graph, and the associations between products are taken as edges E between nodes. A product dependency network graph G(V,E) is constructed, and the embedding vector of each product node is calculated. Static attributes (such as product ID, category, attributes, etc.) can be used to calculate the embedding of the node. Each product node will eventually get two embedding vectors, namely embedding vector E1 and embedding vector E2.

[0067] After that, the two embedded vector matrices of the products are used to perform matrix multiplication, and then the results are subtracted, so that the elements on the diagonal are all set to 0. At the same time, the adjacency matrix obtained is also asymmetric, which is used to reflect the unidirectional dependency relationship between products. The formula of the adjacency matrix A is:

[0068]

[0069] Among them, α is a hyperparameter used to control the saturation rate of the tanh activation function to prevent the gradient vanishing problem of the tanh activation function. ReLU maps the final adjacency matrix value to a number greater than 0, representing the degree of association between products. E1 and E2 are embedding vectors. as well as is the transpose of E1 and E2.

[0070] Furthermore, in order to ensure the sparsity of the adjacency matrix, it is necessary to make some adjustments to the previously obtained adjacency matrix. For each product node, only the k nodes with the strongest association are selected to be connected to it, and the adjacency matrix weights of other nodes are set to 0, which are regarded as having no association. k is also a hyperparameter, representing the maximum number of edges for each product, which can be used to control the maximum number of nodes that a certain product can be connected to.

[0071] There is a correlation between other products.

[0072] Finally, after learning with the graph learning module, an adjacency matrix reflecting the association relationship between products is obtained. A value of 0 on the adjacency matrix means that there is no association relationship between the two products, and a value other than 0 on the matrix means that there is an association relationship between the products. The value of the matrix represents the degree of association between the products. A value of 0 means that there is no association relationship between the products. Finally, this adjacency matrix will be used as the subsequent GCN

[0073] The input of the module is used to further learn the correlation between the demands of commodities.

[0074] That is, the graph learning module of the present application can adaptively generate an adjacency matrix as the network learns to characterize the correlation between different commodities. This end-to-end learning method does not need to rely on prior expert knowledge or any statistical methods. Instead, during the training process of the model, as the neural network parameters are optimized, a graph structure that can reflect the correlation between commodities and is most suitable for the input of the graph neural network is established, which can effectively model the target. At the same time, in the process of model learning, the graph can be sampled and constructed as needed to construct subgraphs for learning, thus avoiding problems with computing performance. On the other hand, in order to make the final commodity association relationship directed, the commodity association graph finally constructed is asymmetric and can reflect unidirectional commodity association relationships as needed.

[0075] In some embodiments, a multi-product inventory optimization neural network model integrates a long short-term memory network algorithm module, a multi-layer perceptron module, and an attention mechanism module. The target feature data are historical commodity demand data and promotion feature data. The historical optimal order quantity and the target feature data are input into the multi-product inventory optimization neural network model to obtain the target optimal order quantity, which can be obtained through the following steps: inputting the historical commodity demand data into the long short-term memory network algorithm module to obtain a commodity time series vector, wherein the commodity time series vector is used to capture the correlation between commodities in the time dimension; and, inputting the promotion feature data into the multi-layer perceptron module to obtain a promotion feature vector; inputting the commodity time series vector and the promotion feature vector into the attention mechanism module to obtain attention weight information; and determining the target optimal order quantity according to the attention weight information.

[0076] For example, historical commodity demand data includes the daily sales volume of commodities in the past few days, reflecting the sales trend of commodities over a period of time. The processing of historical commodity demand data in the LSTM module is as follows: the commodity demand in t periods in history is taken as input, and then modeled in LSTM, and finally the feature representation at each time step is output. Due to the recursiveness of the LSTM model, each time step contains the information of the previous time step, thereby capturing the dependency of commodities in the time dimension. The specific formula is as follows:

[0077] O LSTM，t =LSTM(x demand，1 , x demand，2 ..x demand，t );

[0078] Among them, O LSTM，t is the vector output by the LSTM module, x demand，1 , x demand，2 ..x demand，t represents the demand for goods in each of t periods.

[0079] After passing through the LSTM module, the important information of historical demand at each time step will be saved in a vector, that is, the commodity time series vector in this case, which can reflect important patterns such as the trend and periodicity of commodity demand at that time step, and well model the dependency of each commodity demand in the time dimension.

[0080] After that, the promotion information is extracted, and the promotion feature vector is obtained by inputting the promotion related features into a multi-layer perceptron module to capture the global relationship of various promotion features, which can be obtained by the following formula:

[0081] O promotion =MLP(xpromotion );

[0082] Among them, O promotion is the vector output by the MLP module, x promotion Represents the relevant features of promotion activities. MLP is a multi-layer perceptron with two hidden layers, and the activation function uses ReLU. Each type of feature is directly concatenated from the discrete features and continuous features after the embedding layer.

[0083] After the promotion activity features are mapped by the MLP module to obtain a high-dimensional hidden layer promotion feature vector, this vector is used as the query vector. After normalization, the dot product similarity between it and the normalized hidden layer vector (key vector) output by the LSTM module is calculated. This dot product similarity is then used as the weight distribution of each time step of the LSTM hidden layer. Finally, this weight distribution is used to weight the hidden layer vector (value vector) output by the LSTM module, thereby capturing the impact of the promotion on demand in the future. The specific calculation steps are as follows:

[0084] Calculate the promotion feature vector O promotion The LSTM output requires a hidden state H at each time step t The degree of association, the attention function here uses dot product attention:

[0085]

[0086] Among them, d is the dimension of the hidden layer vector, O promotion , is the promotion feature vector output by the MLP module, O LSTM，t Output the demand feature vector of period t for the LSTM module, a(O promotion , O LSTM,t ) represents the promotion feature vector O promotion Attention weights for the required features with the LSTM output.

[0087] Based on the above-mentioned attention function, the attention weight information is determined, and the target optimal order quantity is determined according to the attention weight information. By determining the attention function, the present application achieves a better characterization of the current commodity demand pattern when subsequently determining the optimal order quantity, thereby improving inventory management efficiency.

[0088] Furthermore, the target optimal order quantity can be determined according to the attention weight information through the following steps: normalizing the attention weight information to obtain the processed attention weight information, weighting the hidden state of the product time series vector based on the processed attention weight information to obtain the target product time series vector; inputting the target product time series vector into the graph convolutional network module to obtain the target optimal order quantity.

[0089] Exemplarily, the calculated attention weights of the promotion feature vector and the LSTM output vector are normalized using a softmax function to obtain processed attention weight information, wherein the processed attention weight information is obtained by the following formula:

[0090]

[0091] Among them, a(O promotion , O LSTM,t ) represents the promotion feature vector O promotion The attention weight of the required feature output by LSTM, α(O promotion , O LSTM，t ) represents the normalized attention weight.

[0092] The normalized attention weights are used to weight the hidden state of each time step of the LSTM output, and the target product time series vector is obtained by the following formula:

[0093]

[0094] Among them, α(O promotion , O LSTM,t ) represents the normalized attention weight, O LSTM，t is the vector output by the LSTM module, O attention The feature vector representing the output of the Attention module is the LSTM module vector enhanced by the promotion activity features.

[0095] Afterwards, the vectors of each product are input into the GCN module, and the GCN module's network learning ability is used to learn the correlation between multiple products, and then train the target optimal order quantity.

[0096] Among them, the working mechanism of the GCN module in this case is obtained through the following steps:

[0097] The GCN module input is a product adjacency matrix A output by a graph learning module and a hidden layer vector O of the product output by the LSTM and Attention modules. attention , GCN will automatically pass messages between those commodity nodes that have associated relationships, and integrate the information of other commodity nodes that have associated relationships into each commodity node to model the possible associations between commodities. The output of GCN is O gcn as follows:

[0098]

[0099] in, A is the adjacency matrix output by the previous graph learning module, I is the identity matrix, for The degree matrix, O attention is the vector output by the attention module in the previous step, W gcn is the parameter matrix of the graph convolutional network.

[0100] Ultimately, a vector is learned for each product, which not only contains information about its own demand changes and promotional activities, but also incorporates information about other products through the graph convolutional network, effectively modeling the correlation between multiple products, which will be conducive to solving the optimal order quantity.

[0101] In an optional embodiment, if Figure 2 As shown in the fusion module, the inventory features modeled by the MLP module, other features, and features output by the GCN module are finally fused and output. While fusing information from multiple sources, the outputs of the previous modules are also integrated to serve the final modeling goal. In order to model inventory features, this application constructs the intrinsic relationship between features through feature engineering, and maps it from low-dimensional space to high-dimensional space to facilitate the model to learn more complex patterns. The present invention also uses the MLP module to model these two partial features separately, map them to high-dimensional vectors, and then fuse them with other modules, as shown in the following formula:

[0102] O invebtory =MLP(x invebtory );

[0103] O other =MLP(x other );

[0104] Among them, x inventory represents the input inventory features, x other Represents other features of the input, O inventory represents the inventory feature vector output by the inventory MLP module, O other Represents other feature vectors output by other feature MLP modules.

[0105] Similarly, the MLP module here is a multi-layer perceptron with two hidden layers. The activation function uses ReLU, and each type of feature is directly concatenated from the discrete features and continuous features through the embedding layer.

[0106] Finally, after different information of commodities is processed by different deep learning modules, three highly compressed feature vectors are obtained for each commodity, corresponding to the output of inventory MLP module O inventory , the GCN module output O gcnAnd other feature MLP modules O other The output of the inventory MLP module O inventory Reflecting the current inventory and inventory cost information, the output of the GCN module O gcn It not only integrates the historical demand and promotion information of the current product, but also integrates the information of related products. The output of other feature MLP modules O other It is high-level information obtained based on human feature engineering.

[0107] Therefore, this application designs a fusion module to fuse and output information from different sources. First, the three feature vectors are spliced ​​and then output to the structure of the multi-layer perceptron. With the help of the fitting ability of the multi-layer perceptron, the complex relationship between information from multiple sources can be fully learned, and the output of each module can produce a synergistic effect to jointly serve the final modeling goal. The specific form of the fusion module is as follows: After splicing the three feature vectors, they are passed through a two-layer multi-layer perceptron module, and then the optimal order quantity is output. The specific formula is as follows:

[0108]

[0109] Among them, O inventory represents the inventory MLP module output vector, O gcn Represents the output vector of the GCN module, O other Represents the output vector of other feature MLP modules, represents the optimal order quantity.

[0110] Finally, in the fusion module, the multilayer perceptron fuses the information from three different sources, integrates the outputs of the previous modules, and integrates the heterogeneous characteristics of the information, so as to more effectively learn the final optimal order quantity.

[0111] This application models different features in different ways, and with the help of the coordination between various modules, after training with a large amount of historical data, the MPIONN model can learn the complex rules of the historical optimal order quantity. When making future decisions, various product characteristics at the current time point can be input into the model, thereby outputting the optimal order quantity for each decision variable.

[0112] The inventory management optimization method provided by the embodiment of the present application obtains historical inventory data of multiple periods; determines the historical optimal order quantity of each period in multiple periods according to the historical inventory data; inputs the historical optimal order quantity and the target feature data into the multi-product inventory optimization neural network model to obtain the target optimal order quantity, wherein the target feature data includes at least one of the following: inventory feature data, commodity feature data, historical commodity demand data, promotion feature data, and derivative feature data, wherein the multi-product inventory optimization neural network model integrates at least one of the following: multi-layer perceptron module, long short-term memory network algorithm module, graph learning module, graph convolution network module, and attention mechanism module. Through this application, the problem of lack of analysis of the correlation between modeled commodities in the single-period mode in the related technology is solved, which leads to low inventory optimization efficiency. By obtaining historical inventory data of multiple periods, the historical inventory data is analyzed based on the multi-product inventory optimization neural network model to obtain the target optimal order quantity, thereby achieving the effect of improving inventory optimization efficiency.

[0113] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0114] The embodiment of the present application also provides an inventory management optimization device. It should be noted that the inventory management optimization device of the embodiment of the present application can be used to execute the inventory management optimization method provided in the embodiment of the present application. The inventory management optimization device provided in the embodiment of the present application is introduced below.

[0115] Figure 4 Schematic diagram of an inventory management optimization device according to an embodiment of the present application. Figure 4 As shown, the device includes: an acquisition unit 401, a first determination unit 402, and a second determination unit 403.

[0116] Specifically, the acquisition unit 401 is used to acquire historical inventory data of multiple periods;

[0117] A first determining unit 402 is used to determine the historical optimal order quantity of each period in a plurality of periods according to the historical inventory data;

[0118] The second determination unit 403 is used to input the historical optimal order quantity and the target feature data into the multi-product inventory optimization neural network model to obtain the target optimal order quantity, wherein the target feature data includes at least one of the following: inventory feature data, commodity feature data, historical commodity demand data, promotion feature data, and derivative feature data, wherein the multi-product inventory optimization neural network model integrates at least one of the following: a multi-layer perceptron module, a long short-term memory network algorithm module, a graph learning module, a graph convolutional network module, and an attention mechanism module.

[0119] The inventory management optimization device provided in the embodiment of the present application is used to obtain historical inventory data of multiple periods through an acquisition unit 401; a first determination unit 402 is used to determine the historical optimal order quantity of each period in multiple periods according to the historical inventory data; a second determination unit 403 is used to input the historical optimal order quantity and the target feature data into the multi-product inventory optimization neural network model to obtain the target optimal order quantity, wherein the target feature data includes at least one of the following: inventory feature data, commodity feature data, historical commodity demand data, promotion feature data, and derivative feature data, wherein the multi-product inventory optimization neural network model integrates at least one of the following: a multi-layer perceptron module, a long short-term memory network algorithm module, a graph learning module, a graph convolution network module, and an attention mechanism module, which solves the problem that the correlation between modeled commodities is lacking in the single-period mode in the related technology, thereby resulting in low inventory optimization efficiency. By acquiring historical inventory data of multiple periods, the historical inventory data is analyzed based on the multi-product inventory optimization neural network model to obtain the target optimal order quantity, thereby achieving the effect of improving inventory optimization efficiency.

[0120] Optionally, in the inventory management optimization device provided in the embodiment of the present application, the first determination unit includes: a first determination module, used to determine the initial inventory data and actual order quantity of each cycle based on historical inventory data; and, a second determination module, used to determine the order demand information and current cycle inventory data of two adjacent cycles among multiple cycles; a third determination module, used to determine the historical optimal order quantity based on the initial inventory data, the actual order quantity, the order demand information, and the current cycle inventory data.

[0121] Optionally, in the inventory management optimization device provided in the embodiment of the present application, the third determination module includes: a first calculation sub-module, used to calculate the current commodity cost between orders based on the initial inventory data, the actual order quantity, the order demand information, and the current cycle inventory data, wherein the current commodity cost includes: cycle holding cost, cycle out-of-stock cost, wherein the current commodity cost is the minimum cost generated from the delivery time of the commodity order within the target cycle to all subsequent cycles; the first determination sub-module, used to determine the minimum total cost of the commodity in each cycle based on the current commodity cost; the second calculation sub-module, used to calculate the minimum total cost of the commodity through a reverse algorithm to obtain the historical optimal order quantity.

[0122] Optionally, in the inventory management optimization device provided in the embodiment of the present application, the multi-product inventory optimization neural network model integrates a graph learning module and a graph convolutional network module, the target feature data is the commodity feature data, and the second determination unit includes: a fourth determination module, which is used to determine the association information between each commodity according to the graph learning module; and a fifth determination module, which is used to input the association information and the historical optimal order quantity into the graph convolutional network module to obtain the target optimal order quantity.

[0123] Optionally, in the inventory management optimization device provided in the embodiment of the present application, the fourth determination module includes: a second determination submodule, used to determine the product node corresponding to each product; a third calculation submodule, used to calculate the embedding vector of each product node, wherein the embedding vector is used to represent the product node characteristics; the fourth calculation submodule, used to calculate the adjacency matrix between each product based on the embedding vector; the third determination submodule, used to determine the association information based on the adjacency matrix.

[0124] Optionally, in the inventory management optimization device provided in the embodiment of the present application, the multi-product inventory optimization neural network model integrates a long short-term memory network algorithm module, a multi-layer perceptron module, and an attention mechanism module, and the target feature data is historical commodity demand data and promotion feature data. The second determination unit includes: a first data input module, which is used to input the historical commodity demand data into the long short-term memory network algorithm module to obtain a commodity time series vector, wherein the commodity time series vector is used to capture the correlation between commodities in the time dimension; and a second data input module, which is used to input the promotion feature data into the multi-layer perceptron module to obtain a promotion feature vector; a third data input module, which is used to input the commodity time series vector and the promotion feature vector into the attention mechanism module to obtain attention weight information; and a sixth determination module, which is used to determine the target optimal order quantity according to the attention weight information.

[0125] Optionally, in the inventory management optimization device provided in the embodiment of the present application, the sixth determination module includes: a normalization sub-module, used to normalize the attention weight information to obtain the processed attention weight information, and weight the hidden state of the product time series vector based on the processed attention weight information to obtain the target product time series vector; a vector input sub-module, used to input the target product time series vector into the graph convolutional network module to obtain the target optimal order quantity.

[0126] The inventory management optimization device includes a processor and a memory. The acquisition unit 401, the first determination unit 402, the second determination unit 403, etc. are all stored in the memory as program units, and the processor executes the program units stored in the memory to implement corresponding functions.

[0127] The processor contains a kernel, which calls the corresponding program unit from the memory. One or more kernels can be set, and inventory management can be optimized by adjusting kernel parameters.

[0128] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0129] In the exemplary embodiment of the present application, a computer storage medium capable of implementing the above method is also provided. A program product capable of implementing the above method of the present specification is stored thereon. In some possible embodiments, various aspects of the present application can also be implemented in the form of a program product, which includes a program code. When the program product is run on a terminal device, the program code is used to enable the terminal device to execute the steps according to various exemplary embodiments of the present application described in the above "Exemplary Method" section of this specification, such as, the following steps can be performed: obtaining historical inventory data of multiple periods; determining the historical optimal order quantity of each period in multiple periods based on the historical inventory data; inputting the historical optimal order quantity and the target feature data into the multi-product inventory optimization neural network model to obtain the target optimal order quantity, wherein the target feature data includes at least one of the following: inventory feature data, commodity feature data, historical commodity demand data, promotion feature data, and derivative feature data, wherein the multi-product inventory optimization neural network model integrates at least one of the following: a multi-layer perceptron module, a long short-term memory network algorithm module, a graph learning module, a graph convolutional network module, and an attention mechanism module.

[0130] In an optional implementation: the initial inventory data and actual order quantity of each period are determined based on historical inventory data; and the order demand information and current period inventory data of two adjacent periods among multiple periods are determined; the historical optimal order quantity is determined based on the initial inventory data, the actual order quantity, the order demand information, and the current period inventory data.

[0131] In an optional implementation: cycle holding cost, cycle out-of-stock cost, where the current commodity cost is the minimum cost generated from the delivery time of the commodity order within the target period to all subsequent periods; the minimum total cost of the commodity in each period is determined based on the current commodity cost; the minimum total cost of the commodity is calculated by a reverse algorithm to obtain the historical optimal order quantity.

[0132] In an optional implementation: determining the association information between each commodity according to the graph learning module; inputting the association information and the historical optimal order quantity into the graph convolutional network module to obtain the target optimal order quantity.

[0133] In an optional implementation: determine the product node corresponding to each product; calculate the embedding vector of each product node, wherein the embedding vector is used to represent the product node feature; calculate the adjacency matrix between each product based on the embedding vector; and determine the association information based on the adjacency matrix.

[0134] In an optional implementation: historical commodity demand data is input into a long short-term memory network algorithm module to obtain a commodity time series vector, wherein the commodity time series vector is used to capture the correlation between commodities in the time dimension; and promotion feature data is input into a multi-layer perceptron module to obtain a promotion feature vector; the commodity time series vector and the promotion feature vector are input into an attention mechanism module to obtain attention weight information; and the target optimal order quantity is determined based on the attention weight information.

[0135] In an optional implementation: the attention weight information is normalized to obtain processed attention weight information, the hidden state of the product time series vector is weighted based on the processed attention weight information to obtain a target product time series vector; the target product time series vector is input into a graph convolutional network module to obtain a target optimal order quantity.

[0136] In an optional embodiment, the embodiment of the present application may also include a program product for implementing the above method, which may adopt a portable compact disk read-only memory (CD-ROM) and include program code, and can be run on a terminal device, such as a personal computer. However, the program product of the present application is not limited thereto, and in this document, a readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, apparatus, or device.

[0137] The program product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0138] Computer readable signal media may include data signals propagated in baseband or as part of a carrier wave, in which readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Readable signal media may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0139] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

[0140] Program code for performing the operations of the present application may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0141] In addition, in an exemplary embodiment of the present application, an electronic device capable of implementing the above method is also provided.

[0142] Those skilled in the art will appreciate that various aspects of the present application may be implemented as a system, method or program product. Therefore, various aspects of the present application may be specifically implemented in the following forms, namely: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which may be collectively referred to as "circuit", "module" or "system" herein.

[0143] Refer to the following Figure 5 The electronic device 500 according to this embodiment of the present application is described. Figure 5 The electronic device 500 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0144] like Figure 5 As shown, the electronic device 500 is in the form of a general computing device. The components of the electronic device 500 may include, but are not limited to: the at least one processing unit 510, the at least one storage unit 520, a bus 530 connecting different system components (including the storage unit 520 and the processing unit 510), and a display unit 540.

[0145] The storage unit stores program codes, which can be executed by the processing unit 510, so that the processing unit 510 executes the steps described in the "Exemplary Method" section of the specification according to various exemplary embodiments of the present application. For example, the processing unit 510 can execute the following steps: obtaining historical inventory data of multiple periods; determining the historical optimal order quantity of each period in the multiple periods based on the historical inventory data; inputting the historical optimal order quantity and the target feature data into the multi-product inventory optimization neural network model to obtain the target optimal order quantity, wherein the target feature data includes at least one of the following: inventory feature data, commodity feature data, historical commodity demand data, promotion feature data, and derivative feature data, wherein the multi-product inventory optimization neural network model integrates at least one of the following: a multi-layer perceptron module, a long short-term memory network algorithm module, a graph learning module, a graph convolutional network module, and an attention mechanism module.

[0146] In an optional implementation: the initial inventory data and actual order quantity of each period are determined based on historical inventory data; and the order demand information and current period inventory data of two adjacent periods among multiple periods are determined; the historical optimal order quantity is determined based on the initial inventory data, the actual order quantity, the order demand information, and the current period inventory data.

[0147] In an optional implementation: cycle holding cost, cycle out-of-stock cost, where the current commodity cost is the minimum cost generated from the delivery time of the commodity order within the target period to all subsequent periods; the minimum total cost of the commodity in each period is determined based on the current commodity cost; the minimum total cost of the commodity is calculated by a reverse algorithm to obtain the historical optimal order quantity.

[0148] In an optional implementation: determining the association information between each commodity according to the graph learning module; inputting the association information and the historical optimal order quantity into the graph convolutional network module to obtain the target optimal order quantity.

[0149] In an optional implementation: determine the product node corresponding to each product; calculate the embedding vector of each product node, wherein the embedding vector is used to represent the product node feature; calculate the adjacency matrix between each product based on the embedding vector; and determine the association information based on the adjacency matrix.

[0150] In an optional implementation: historical commodity demand data is input into a long short-term memory network algorithm module to obtain a commodity time series vector, wherein the commodity time series vector is used to capture the correlation between commodities in the time dimension; and promotion feature data is input into a multi-layer perceptron module to obtain a promotion feature vector; the commodity time series vector and the promotion feature vector are input into an attention mechanism module to obtain attention weight information; and the target optimal order quantity is determined based on the attention weight information.

[0151] In an optional implementation: the attention weight information is normalized to obtain processed attention weight information, the hidden state of the product time series vector is weighted based on the processed attention weight information to obtain a target product time series vector; the target product time series vector is input into a graph convolutional network module to obtain a target optimal order quantity.

[0152] The storage unit 520 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 5201 and / or a cache storage unit 5202 , and may further include a read-only storage unit (ROM) 5203 .

[0153] The storage unit 520 may also include a program / utility 5204 having a set (at least one) of program modules 5205, such program modules 5205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0154] Bus 530 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0155] The electronic device 500 may also communicate with one or more external devices 600 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 500, and / or may communicate with any device that enables the electronic device 500 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 550. Furthermore, the electronic device 500 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 560. As shown, the network adapter 560 communicates with other modules of the electronic device 500 via a bus 530. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 500, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0156] Through the description of the above implementation methods, it is easy for those skilled in the art to understand that the example implementation methods described here can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the implementation methods of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the implementation methods of the present application.

[0157] In addition, the above-mentioned figures are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present application, and are not intended to be limiting. It is easy to understand that the processes shown in the above-mentioned figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0158] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any variations, uses or adaptations of the present application, which follow the general principles of the present application and include common knowledge or customary technical means in the art that are not disclosed in the present application. The specification and embodiments are to be regarded as exemplary only, and the true scope and spirit of the present application are indicated by the claims.

Claims

1. A method for optimizing inventory management, characterized in that: include: Get historical inventory data for multiple periods; Determine the historical optimal order quantity for each of the multiple periods according to the historical inventory data; The historical optimal order quantity and target feature data are input into a multi-product inventory optimization neural network model to obtain a target optimal order quantity, wherein the target feature data includes at least one of the following: inventory feature data, commodity feature data, historical commodity demand data, promotion feature data, and derivative feature data; wherein the multi-product inventory optimization neural network model integrates at least one of the following: a multi-layer perceptron module, a long short-term memory network algorithm module, a graph learning module, a graph convolutional network module, and an attention mechanism module.

2. The method according to claim 1, characterized in that Determining the historical optimal order quantity for each of the multiple periods according to the historical inventory data includes: Determining the initial inventory data and the actual order quantity of each period respectively based on the historical inventory data; and, Determining order demand information of two adjacent cycles among the multiple cycles and current cycle inventory data; The historical optimal order quantity is determined according to the initial inventory data, the actual order quantity, the order demand information, and the current cycle inventory data.

3. The method according to claim 2, characterized in that Determining the historical optimal order quantity according to the initial inventory data, the actual order quantity, the order demand information, and the current cycle inventory data includes: Calculate the current commodity cost between orders according to the initial inventory data, the actual order quantity, the order demand information, and the current cycle inventory data, wherein the current commodity cost includes: cycle holding cost and cycle shortage cost, wherein the current commodity cost is the minimum cost generated from the delivery time of the commodity order within the target cycle to all subsequent cycles; Determine the minimum total cost of the goods in each period according to the current cost of the goods; The minimum total cost of the commodity is calculated by a reverse algorithm to obtain the historical optimal order quantity.

4. The method according to claim 1, characterized in that: The multi-product inventory optimization neural network model integrates the graph learning module and the graph convolutional network module. The target feature data is the commodity feature data. The historical optimal order quantity and the target feature data are input into the multi-product inventory optimization neural network model to obtain the target optimal order quantity, including: Determine association information between each product according to the graph learning module; The association information and the historical optimal order quantity are input into a graph convolutional network module to obtain the target optimal order quantity.

5. The method according to claim 4, characterized in that Determining association information between each product according to the graph learning module includes: Determine the commodity node corresponding to each commodity; Calculating an embedding vector of each commodity node, wherein the embedding vector is used to represent a feature of the commodity node; Calculate the adjacency matrix between each of the commodities based on the embedding vector; The association information is determined based on the adjacency matrix.

6. The method according to claim 1, characterized in that The multi-product inventory optimization neural network model integrates the long short-term memory network algorithm module, the multi-layer perceptron module, and the attention mechanism module. The target feature data is the historical commodity demand data and the promotion feature data. The historical optimal order quantity and the target feature data are input into the multi-product inventory optimization neural network model to obtain the target optimal order quantity, including: Inputting the historical commodity demand data into the long short-term memory network algorithm module to obtain a commodity time series vector, wherein the commodity time series vector is used to capture the relevance of commodities in the time dimension; and, Inputting the promotion feature data into the multi-layer perceptron module to obtain a promotion feature vector; Inputting the product time series vector and the promotion feature vector into the attention mechanism module to obtain attention weight information; The target optimal order quantity is determined according to the attention weight information.

7. The method according to claim 6, characterized in that Determining the target optimal order quantity according to the attention weight information includes: Normalizing the attention weight information to obtain processed attention weight information, and weighting the hidden state of the product time series vector based on the processed attention weight information to obtain a target product time series vector; The target commodity time series vector is input into the graph convolutional network module to obtain the target optimal order quantity.

8. An inventory management optimization device, characterized in that: include: An acquisition unit, used to acquire historical inventory data of multiple periods; A first determining unit, configured to determine a historical optimal order quantity for each of the multiple periods according to the historical inventory data; The second determination unit is used to input the historical optimal order quantity and the target feature data into the multi-product inventory optimization neural network model to obtain the target optimal order quantity, wherein the target feature data includes at least one of the following: inventory feature data, commodity feature data, historical commodity demand data, promotion feature data, and derivative feature data, wherein the multi-product inventory optimization neural network model integrates at least one of the following: a multi-layer perceptron module, a long short-term memory network algorithm module, a graph learning module, a graph convolutional network module, and an attention mechanism module.

9. A computer-readable storage medium, characterized in that: The storage medium includes a stored program, wherein the program executes the inventory management optimization method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: include: One or more processors, a memory, a display device, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include a method for executing the inventory management optimization method described in any one of claims 1 to 7.

Images