Big data-based library age management system

Abstract

The invention relates to the technical field of storage age management, and particularly discloses a storage age management system based on big data, which is characterized in that storage age distribution of a group of commodities is calculated by extracting stock dates and arrival dates of each commodity in a specific commodity set; and further introducing a data analysis technology based on deep learning to perform periodic time sequence pattern recognition and global pattern context perception on the library age distribution data of the group of commodities to predict the future inventory demand of the commodities, and further generating corresponding purchase suggestions based on comparison between the prediction result and the current inventory. In this way, the enterprise inventory can be managed more accurately, so that the inventory turnover rate is increased, the inventory cost is reduced, and the market competitiveness of enterprises is enhanced.

Description

Technical Field

[0001] The present application relates to the technical field of inventory age management, and more specifically, to an inventory age management system based on big data. Background Art

[0002] In the modern business environment, inventory management is a key link in supply chain optimization. Effective inventory management can not only ensure sufficient supply of goods to meet market demand, but also reduce holding costs, reduce capital occupation, and improve the operational efficiency of the enterprise. In inventory management, warehouse managers, operations personnel, and financial personnel usually need to regularly check the age of the company's existing inventory and evaluate the overall health of the existing inventory in order to make decisions on warehousing planning, operations planning, and cost planning.

[0003] Inventory age refers to the time from the time each item of inventory is put into storage to the current date. By calculating the inventory age of goods, companies can understand the age of their inventory, assess inventory liquidity, and make more effective warehousing, sales, and financial plans accordingly.

[0004] Traditional inventory age management methods are usually based on manual records and simple spreadsheets to track inventory dates and arrival dates. This method has obvious limitations, such as untimely data updates, prone to errors, and difficulty in processing large amounts of data. With the development of information technology, more efficient and accurate technical tools are provided for the inventory age records of goods. However, although some existing inventory age management systems can monitor the inventory level of goods in real time, most of them are limited to reflecting the current inventory status, and lack the ability to deeply analyze historical inventory data and predict future inventory demand, making it difficult for enterprises to plan in advance and can only respond passively to market changes.

[0005] Therefore, an optimized inventory age management system based on big data is expected. Summary of the invention

[0006] The present application provides an inventory age management system based on big data, which can manage the enterprise inventory more accurately, thereby improving inventory turnover, reducing inventory costs, and enhancing the market competitiveness of the enterprise.

[0007] In a first aspect, a big data-based inventory age management system is provided, comprising:

[0008] A storage age related data extraction module is used to extract the storage age related data of all first SKUs from the background database, wherein the storage age related data includes the SKU inventory date and the SKU arrival date;

[0009] A stock age calculation module, used to calculate the stock age value of each first SKU based on the stock age related data of each SKU to obtain the stock age distribution of the first SKU;

[0010] A future inventory demand prediction module, configured to perform future inventory demand prediction based on the inventory age distribution of the first SKU to obtain a future inventory demand prediction value;

[0011] A purchasing suggestion generating module, used for generating purchasing suggestions based on the comparison between the future inventory demand forecast value and the current inventory quantity;

[0012] Among them, the future inventory demand prediction module includes: an inventory age distribution pattern feature extraction unit, which is used to perform pattern feature extraction on the inventory age distribution of the first SKU based on singular spectrum decomposition to obtain multiple inventory age cycle pattern coding feature vectors; a context-aware aggregation unit, which is used to perform context-aware autocorrelation gated aggregation analysis on the multiple inventory age cycle pattern coding feature vectors to obtain a inventory age multi-time series period significant aggregation coding vector; an inventory demand prediction unit, which is used to generate the future inventory demand prediction value based on the inventory age multi-time series period significant aggregation coding vector.

[0013] The application provides a big data-based inventory management system, which calculates the inventory age distribution of a group of commodities by extracting the inventory date and arrival date of each commodity in a specific commodity set, and then further introduces a data analysis technology based on deep learning to perform periodic time series pattern recognition and global pattern context perception on the inventory age distribution data of the group of commodities to predict the future inventory demand of the commodity, and then generates corresponding procurement recommendations based on the comparison between the prediction results and the current inventory. In this way, the enterprise inventory can be managed more accurately, thereby improving inventory turnover, reducing inventory costs, and enhancing the market competitiveness of the enterprise. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application are briefly introduced below. Obviously, the drawings described below only relate to some embodiments of the present application, and are not intended to limit the present application.

[0015] Figure 1 This is a schematic block diagram of an inventory age management system based on big data according to an embodiment of the present application.

[0016] Figure 2 This is a data flow diagram of the big data-based inventory age management system according to an embodiment of the present application.

[0017] Figure 3 This is a schematic block diagram of a future inventory demand prediction module in a big data-based inventory age management system according to an embodiment of the present application.

[0018] Figure 4 This is a schematic block diagram of a feature extraction unit for inventory age distribution pattern in an inventory age management system based on big data according to an embodiment of the present application.

[0019] Figure 5 This is a schematic block diagram of a context-aware aggregation unit in a big data-based inventory age management system according to an embodiment of the present application. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present application.

[0021] In response to the above technical problems, the technical concept of this application is: by extracting the inventory date and arrival date of each commodity in a specific commodity set, the inventory age distribution of the group of commodities is calculated, and then, the data analysis technology based on deep learning is further introduced to perform periodic time series pattern recognition and global pattern context perception on the inventory age distribution data of the group of commodities to predict the future inventory demand of the commodity, and then generate corresponding procurement recommendations based on the comparison between the prediction results and the current inventory. In this way, the enterprise inventory can be managed more accurately, thereby improving inventory turnover, reducing inventory costs, and enhancing the market competitiveness of the enterprise.

[0022] like Figure 1 and Figure 2 As shown, the inventory age management system based on big data includes: an inventory age related data extraction module 10, which is used to extract the inventory age related data of all first SKUs from the background database, and the inventory age related data includes the SKU inventory date and the SKU arrival date; an inventory age calculation module 20, which is used to calculate the inventory age value of each first SKU based on the inventory age related data of each SKU to obtain the inventory age distribution of the first SKU; a future inventory demand prediction module 30, which is used to predict the future inventory demand based on the inventory age distribution of the first SKU to obtain the future inventory demand prediction value; a procurement suggestion generation module 40, which is used to generate procurement suggestions based on the comparison between the future inventory demand prediction value and the current inventory amount.

[0023] Exemplarily, in the inventory age related data extraction module 10, the inventory age related data of all first SKUs are extracted from the background database, and the inventory age related data includes the SKU inventory date and the SKU arrival date. It should be understood that the first SKU (Stock Keeping Unit) refers to a unique identifier of a specific group or series of goods. By extracting the inventory age related data of each SKU, it is helpful to understand the inventory status and liquidity of the group of goods, thereby achieving more effective inventory management.

[0024] For example, in the inventory age calculation module 20, based on the inventory age related data of each SKU, the inventory age value of each first SKU is calculated to obtain the inventory age distribution of the first SKU. It should be understood that the SKU inventory date refers to the last inventory record date of each SKU in the warehouse, and the SKU arrival date refers to the date when the goods arrive at the warehouse. Based on the time difference between the SKU inventory date and the SKU arrival date, the inventory time length (i.e., inventory age value) of each SKU can be calculated to obtain the inventory age distribution data of the first SKU, which is helpful to understand the inventory turnover speed of the group of goods and provide effective data support for subsequent inventory optimization.

[0025] Exemplarily, in the future inventory demand prediction module 30, based on the inventory age distribution of the first SKU, future inventory demand prediction is performed to obtain the future inventory demand prediction value. It should be understood that by analyzing the newness and oldness of existing inventory goods (i.e., inventory age), the enterprise can better understand the sales speed of different goods. This analysis helps to adjust the procurement strategy, ensure that fast-moving goods have sufficient inventory to meet customer needs, and avoid unsalable goods occupying too much capital and storage space, thereby improving inventory turnover and reducing holding costs. Accurate demand forecasting also enables enterprises to respond to market changes more flexibly, plan production and procurement activities in advance, ensure sufficient product supply when needed, improve customer satisfaction and service levels, and enhance the competitive advantage of enterprises in the market. In addition, using the periodic patterns and other contextual information in historical data to make scientific and reasonable predictions of future inventory demand provides strong data support for management to help them make more informed business decisions. Therefore, a deep learning algorithm is further introduced to conduct an in-depth analysis of the inventory age distribution of the first SKU to predict future inventory demand and guide the inventory management and business decisions of the enterprise, so as to achieve the purpose of optimizing resource allocation and enhancing market competitiveness.

[0026] In one embodiment, Figure 3 As shown, the future inventory demand prediction module 30 includes: an inventory age distribution pattern feature extraction unit 31, which is used to perform pattern feature extraction on the inventory age distribution of the first SKU based on singular spectrum decomposition to obtain multiple inventory age cycle pattern coding feature vectors; a context-aware aggregation unit 32, which is used to perform context-aware autocorrelation gated aggregation analysis on the multiple inventory age cycle pattern coding feature vectors to obtain an inventory age multi-time series period significant aggregation coding vector; an inventory demand prediction unit 33, which is used to generate the future inventory demand prediction value based on the inventory age multi-time series period significant aggregation coding vector.

[0027] In one embodiment, Figure 4As shown, the inventory age distribution pattern feature extraction unit 31 includes: a singular spectrum decomposition subunit 311, which is used to input the inventory age distribution of the first SKU into a time series encoder based on singular spectrum decomposition to obtain multiple inventory age distribution subsequences; a time series pattern feature extraction subunit 312, which is used to perform time series pattern feature extraction on each inventory age distribution subsequence in the multiple inventory age distribution subsequences to obtain the multiple inventory age cycle pattern encoding feature vectors.

[0028] Exemplarily, in the singular spectrum decomposition subunit 311, the inventory age distribution of the first SKU is input into a time series encoder based on singular spectrum decomposition to obtain multiple inventory age distribution subsequences. It should be understood that considering that the inventory age distribution data usually contains a variety of periodic patterns, such as seasonal fluctuations, long-term growth or recession trends, etc. Therefore, in order to more accurately capture the hidden patterns in the inventory age distribution data, the present application uses the singular spectrum decomposition technology to perform time series encoding on the inventory age distribution of the first SKU, so as to decompose the complex inventory age distribution data into several subsequences with different periodic patterns. Singular Spectrum Analysis (SSA) is a powerful time series analysis tool, which can decompose the original inventory age distribution data into multiple independent subsequences by constructing a trajectory matrix of the inventory age distribution data, and through singular value decomposition (SVD), data reorganization and sequence reconstruction, wherein each inventory age distribution subsequence corresponds to inventory age distribution information on different time scales, for example, some subsequences may reflect annual sales cycles, while others may show quarterly or monthly changes. In this way, the periodic characteristics in the inventory age distribution data can be more effectively identified and extracted, providing more accurate input for subsequent predictive analysis.

[0029] Exemplarily, in the time series pattern feature extraction subunit 312, time series pattern feature extraction is performed on each of the multiple inventory age distribution subsequences to obtain the multiple inventory age period pattern encoding feature vectors. It should be understood that in order to further integrate and optimize the inventory age periodic pattern features of different time scales, so as to more comprehensively understand the dynamic change law of the inventory status of the goods and improve the accuracy of future inventory demand forecasting, the present application proposes a feature significance aggregation method, which identifies and strengthens the periodic pattern features that are significantly associated with inventory status changes by performing global pattern context perception based on graph walking and feature fusion strategy based on gating mechanism on the multiple inventory age period pattern encoding feature vectors, while suppressing unimportant noise information, thereby providing more comprehensive and accurate data support for the forecast of future inventory demand.

[0030] In one embodiment, the time series pattern feature extraction subunit 312 is used to: input each of the multiple storage age distribution subsequences into the time series pattern feature extractor based on the LSTM-RNN model to obtain the multiple storage age cycle pattern encoding feature vectors. That is, in order to further analyze and capture the time series characteristics contained in each storage age distribution subsequence, the present application uses the time series pattern feature extractor based on the LSTM-RNN model to extract features from each storage age distribution subsequence to obtain multiple storage age cycle pattern encoding feature vectors. It should be known to those skilled in the art that LSTM (Long Short-Term Memory Network) is an improved version of RNN (Recurrent Neural Network) specially designed for processing and predicting time series data with long-term dependencies. It effectively solves the gradient vanishing or exploding problem encountered by traditional RNN when processing long time interval information by introducing special structures - memory unit, input gate, forget gate and output gate. Specifically, each of the plurality of storage age distribution subsequences is input into a temporal pattern feature extractor based on the LSTM-RNN model, including that each storage age distribution subsequence is fed into the network one by one as an input sequence of the LSTM-RNN model. When a storage age distribution subsequence enters the model, the LSTM unit updates its internal memory state according to the current input and the state at the previous moment. Specifically, the forget gate determines which information should be discarded from the memory unit; the input gate controls how new information is added to the memory unit; and the output gate determines how much of the state of the memory unit will be output. This mechanism allows the LSTM to selectively remember or forget past information and can capture patterns and trends on different time scales. As the storage age distribution subsequences are continuously input, the LSTM-RNN model gradually learns the temporal features in these sequences, including but not limited to seasonal fluctuations, trend changes and other potential laws. For each input storage age distribution subsequence, after being processed by multiple layers of LSTM units, a compact and expressive feature representation, the so-called storage age cycle pattern encoding feature vector, is finally generated in the last layer of the network. This feature vector condenses the key information of the original inventory age distribution subsequence and can be used to describe the inventory behavior pattern during this period, such as sales frequency, inventory turnover speed, etc. Since the LSTM-RNN model has powerful nonlinear modeling capabilities and excellent memory characteristics, it can not only adapt well to complex and changeable inventory data, but also effectively extract the hidden cyclical and non-cyclical components. Therefore, the eigenvector encoding the inventory age cycle pattern obtained in this way not only retains the important attributes of the original data, but also provides a solid foundation for further advanced analysis, such as accurate prediction of future inventory demand.

[0031] In one embodiment, Figure 5As shown, the context-aware aggregation unit 32 includes: a context-aware enhancement subunit 321, which is used to perform graph walking context-aware enhancement on the multiple inventory age cycle pattern encoding feature vectors to obtain multiple inventory age cycle pattern context-aware enhanced feature vectors; a gated aggregation subunit 322, which is used to perform significance aggregation on the multiple inventory age cycle pattern encoding feature vectors based on the feature correlation between the multiple inventory age cycle pattern context-aware enhanced feature vectors and the multiple inventory age cycle pattern encoding feature vectors to obtain the inventory age multi-time series period significance aggregation encoding vector.

[0032] In one embodiment, the context-aware enhancement subunit includes: a topological structure feature extraction secondary subunit, which is used to extract the topological structure features of the multiple inventory age cycle pattern encoding feature vectors to obtain a inventory age cycle pattern graph walk topological feature matrix; a global context walk secondary subunit, which is used to input the inventory age cycle pattern graph walk topological feature matrix and the multiple hyperbolic space mapped inventory age cycle pattern encoding feature vectors into a global context walk encoder based on a graph convolutional neural network model to obtain the multiple inventory age cycle pattern context-aware enhancement feature vectors.

[0033] In one embodiment, the topological structure feature extraction secondary subunit is used to: input each of the plurality of storage age cycle pattern coding feature vectors into a hyperbolic space mapper to obtain a plurality of storage age cycle pattern coding feature vectors after hyperbolic space mapping. Specifically, the process can be expressed as follows:

[0034]

[0035]

[0036] in, represents the plurality of inventory age cycle pattern encoding feature vectors, , , and Respectively represent the first, second, and third of the plurality of inventory age cycle pattern encoding feature vectors and The time series encoding feature vector of the storage age cycle pattern, is the number of the encoding feature vectors of the inventory age cycle pattern, and Represent the first linear mapping matrix and the second linear mapping matrix respectively, represents the first of the encoding feature vectors of the storage age cycle pattern after multiple hyperbolic space mappings The time-series encoding feature vector of the storage age cycle pattern after hyperbolic space mapping;

[0037] The Poincare distance between any two of the multiple hyperbolic space mapped storage age cycle pattern encoding feature vectors is calculated to obtain the storage age cycle pattern graph walk topology matrix. Specifically, the process can be expressed as follows:

[0038]

[0039] in, , Respectively represent the first and After hyperbolic space mapping, the time series encoding feature vector of the storage age cycle pattern is represents the calculation of the norm of the vector, represents the arccosine function, represents the Poincare distance metric function, It represents the first The element value at the position;

[0040] The walk topology matrix of the storage age cycle pattern graph is subjected to dilated convolution coding to obtain the walk topology feature matrix of the storage age cycle pattern graph. Specifically, the process can be expressed by the formula:

[0041]

[0042] in, Represents the topological matrix of the storage age cycle pattern graph. represents the dilated convolution operation, Represents the wandering topological characteristic matrix of the storage age cycle pattern graph.

[0043] Specifically, considering that the volume of hyperbolic space grows exponentially compared to Euclidean space, it is more suitable for embedding tree-like or tree-like structures to represent hierarchical data sets. Therefore, in order to better represent the correlation between the characteristics of each storage age cycle pattern, the present application further maps the multiple storage age cycle pattern encoding feature vectors to the hyperbolic space, so as to utilize the hyperbolic geometric characteristics to maintain the hierarchical structure and long-tail distribution in high-dimensional data, and enhance the relative distance representation capability between storage age cycle pattern characteristics.

[0044] Next, the Poincare distance between the time-series encoding feature vectors of each storage age cycle pattern after hyperbolic space mapping is calculated to represent the relative proximity between each storage age cycle pattern, and a walk topology matrix of the storage age cycle pattern graph is constructed to reveal the global topological structure of the multiple storage age cycle pattern encoding feature vectors. Subsequently, the dilated convolution coding technology is applied to the walk topology matrix of the storage age cycle pattern graph to utilize the sparse connection characteristics of the dilated convolution coding technology to further capture the dependencies between storage age cycle patterns of different time scales over a long distance, thereby enhancing the feature expression capability.

[0045] Exemplarily, in the global context walk secondary subunit, the topological feature matrix of the storage age cycle pattern graph walk and the plurality of hyperbolic space mapped storage age cycle pattern encoding feature vectors are input into the global context walk encoder based on the graph convolutional neural network model to obtain the plurality of storage age cycle pattern context-aware enhanced feature vectors. Specifically, the process can be expressed as follows:

[0046]

[0047] in, represents a graph convolutional neural network, Indicates the The corresponding context-aware enhanced feature vector of the inventory age cycle pattern.

[0048] That is, the temporal encoding feature vectors of each storage age cycle pattern are used as nodes in the graph structure, and the topological feature matrix of the storage age cycle pattern graph walk is used as the edge of the graph walk. The walk is performed on the entire graph structure through the graph convolution operation to transfer and accumulate information between each storage age cycle pattern, so that each node can access the information of its multi-hop neighbors to enhance its feature representation by using contextual information, thereby extracting multiple storage age cycle pattern context-aware enhanced feature vectors with rich contextual information.

[0049] In one embodiment, the gated aggregation subunit is used to: input each group of corresponding storage age cycle pattern context-aware enhanced feature vectors and storage age cycle pattern time-series encoded feature vectors in the multiple storage age cycle pattern context-aware enhanced feature vectors and the multiple storage age cycle pattern encoded feature vectors into an autocorrelation gating unit to obtain multiple storage age cycle pattern autocorrelation gated significant confidence factors. Specifically, the process can be expressed by the formula:

[0050]

[0051] in, Indicates the The corresponding autocorrelation gated significant confidence factor of the storage age cycle pattern, represents the autocorrelation gating unit, It means subtracting by position. Represents a classifier based on the Softmax function.

[0052] The multiple storage age cycle pattern autocorrelation gated significant confidence factors are input into a normalization unit based on a Softmax function to obtain multiple storage age cycle pattern autocorrelation gated significant confidence weight factors. Specifically, the process can be expressed by the formula:

[0053]

[0054] in, represents an exponential function with a natural constant as base, Indicates the The corresponding autocorrelation gated significant confidence weight factor of the storage age cycle pattern.

[0055] Based on the autocorrelation gated significant confidence weight factors of the multiple storage age cycle patterns, the position-weighted sum of the multiple storage age cycle pattern encoding feature vectors is calculated to obtain the storage age multi-time series period significant aggregation encoding vector.

[0056]

[0057] in, It means point multiplication by position. Represents the significant aggregation encoding vector of multiple time series periods of the storage age.

[0058] It should be understood that the context-aware enhanced inventory cycle pattern features and the original inventory cycle pattern features are sent to the autocorrelation gating unit to measure the consistency and correlation between the two. If the original inventory cycle pattern features and the context-aware enhanced feature representation have a high degree of consistency and correlation, it can be considered that the inventory cycle pattern plays a leading role in the characterization of the current inventory status time series change pattern. In this way, the relative importance of each inventory cycle pattern can be effectively revealed, and multiple inventory cycle pattern autocorrelation gating significant confidence factors can be generated to guide the subsequent feature aggregation process. Then, the multiple inventory cycle pattern autocorrelation gating significant confidence factors are normalized to ensure that in the subsequent feature aggregation process, the sum of the weight factors of each inventory cycle pattern feature is 1. Finally, the multiple inventory cycle pattern encoding feature vectors are weighted and aggregated to achieve the emphasis on key periodic pattern features and the suppression of noise information, and the inventory age multi-time series period significant aggregation encoding vector is obtained. In this way, not only the characteristics of the periodic patterns of storage age at various time scales are comprehensively considered, but also the quality of the characteristics is further improved by learning the interactions and influences between different periodic patterns.

[0059] Exemplarily, in the inventory demand prediction unit 33, the future inventory demand forecast value is generated based on the inventory age multi-time series period significant aggregation coding vector. In one embodiment, the inventory demand prediction unit is used to: input the inventory age multi-time series period significant aggregation coding vector into a decoder-based inventory demand prediction module to obtain the future inventory demand forecast value. Specifically, the decoder trains the optimal mapping relationship between the inventory age periodic pattern characteristics and future inventory demand based on historical data, so that in practical applications, it can effectively map the potential information in the inventory age multi-time series period significant aggregation coding vector to the future inventory demand, and convert the encoded inventory age periodic pattern characteristics into specific inventory demand forecast values, thereby providing scientific decision support for inventory management.

[0060] In a specific embodiment, the decoder-based inventory demand forecasting module uses a multi-layer perceptron (MLP) as a core component, aiming to convert the key features extracted from the multi-time-series-period significant aggregation coding vector of inventory age into specific future inventory demand forecast values. First, the multi-time-series-period significant aggregation coding vector of inventory age is input into the multi-layer perceptron. The multi-time-series-period significant aggregation coding vector of inventory age condenses the periodic pattern characteristics of inventory age at different time scales, providing the decoder with a rich description of past inventory behavior. The multi-layer perceptron consists of several fully connected layers, each of which contains a certain number of neurons, each of which is connected to all neurons in the previous layer and is nonlinearly transformed by an activation function. As data flows through the first hidden layer of the multi-layer perceptron, it begins to undergo a series of complex computational operations. At this stage, the input features are weighted summed and combined with the bias term, and then passed to the activation function. Common activation functions such as ReLU (Rectified Linear Unit) can introduce nonlinear factors, allowing the model to capture more complex relationships. The output after activation becomes the input of the next layer, and so on. Each layer tries to learn different aspects of the input data, thereby gradually abstracting higher-level representations. This layered processing allows the model to not only understand information at a single point in time, but also to grasp trends and patterns across multiple time steps. After passing through several hidden layers, the data reaches the output layer. The design of the output layer depends on the specific requirements of the forecasting task, in this case, to generate forecasts for future inventory demand. Therefore, the output layer usually has only one or a few neurons to give continuous numerical forecasts. In order to ensure that the forecast results have good statistical properties, the last layer may apply additional activation functions, such as linear activation functions to maintain the original range of the output, or Sigmoid / Softmax functions to limit the output value range. The specific choice depends on the actual application scenario. Finally, the multi-layer perceptron outputs the results representing the forecast values ​​of future inventory demand.

[0061] Taking into account that each of the multiple inventory age cycle pattern coding feature vectors respectively represents the inventory age time series correlation characteristics of the first SKU in different time periods, when performing feature sequence aggregation analysis based on graph walk autocorrelation gating, insufficient time scale correspondence of inventory age time series characteristics in different time periods will lead to the lack of decision-making of the aggregation feature instance of the inventory age multi-time series period significant aggregation coding vector, thereby affecting the accuracy of the future inventory demand forecast value obtained by the decoder-based inventory demand forecast module.

[0062] Based on this, before the inventory age multi-time series period significant aggregation coding vector is input into the decoder-based inventory demand prediction module, the inventory age multi-time series period significant aggregation coding vector is first optimized, including the steps of:

[0063] Calculate the distance between each pair of eigenvalues ​​of the multi-time series period significant aggregation coding vector of the storage age, such as the L2 distance, and take the square root of the distance to obtain the multi-time series period significant aggregation coding distance representation matrix of the storage age, that is,

[0064]

[0065] in, represents the significant aggregation coding vector of multiple time series periods of the storage age, and The first one represents the significant aggregation coding vector of the multi-time series period of the storage age and Eigenvalues, The first one represents the significant aggregation coding vector of the multi-time series period of the storage age and The distance between eigenvalues, Indicates the significant aggregation of multiple time series periods of the storage age in the encoding distance representation matrix The eigenvalue of the position;

[0066] The autocorrelation matrix of the multi-time-series-period-significant-aggregation-coding vector of the age of storage is obtained as a row vector, that is, ,in, represents the significant aggregation coding vector of multiple time series periods of the storage age, represents the transpose of a vector, represents matrix multiplication, Represents the significant aggregated coded autocorrelation matrix of multiple time series periods of storage age;

[0067] The multi-time-series-period significant aggregation coding vector of the storage age is matrix-multiplied with the multi-time-series-period significant aggregation coding distance representation matrix of the storage age to obtain the first-level mapping vector of the multi-time-series-period significant aggregation coding of the storage age, that is, ,in, represents the significant aggregation coding vector of multiple time series periods of the storage age, The matrix representing the significant aggregation coding distance of multiple time series periods of the storage age is represented. represents matrix multiplication, Indicates the first-level mapping vector of significant aggregation encoding of multiple time series periods of storage age;

[0068] The first-level mapping vector of the significant aggregation coding of the multi-series period of the age of the warehouse is matrix-multiplied by the matrix product of the significant aggregation coding distance representation matrix of the multi-series period of the age of the warehouse and the self-correlation matrix of the significant aggregation coding of the multi-series period of the age of the warehouse to obtain the multi-level mapping vector of the significant aggregation coding of the multi-series period of the age of the warehouse ,in, Indicates the first-level mapping vector of significant aggregation encoding of multiple time series periods of storage age, The matrix representing the significant aggregation coding distance of multiple time series periods of the storage age is represented. represents matrix multiplication, Represents the significant aggregation coded autocorrelation matrix of multiple time series periods of storage age, Represents the multi-level mapping vector of significant aggregation encoding of multiple time series periods of storage age;

[0069] The multi-level mapping vector of the multi-time period significant aggregation code of the storage age is interpolated with the associated eigenvector of the multi-time period significant aggregation code of the storage age composed of the eigenvalues ​​of the auto-correlation matrix of the multi-time period significant aggregation code of the storage age to obtain an optimized multi-time period significant aggregation code vector of the storage age, wherein interpolation or zero filling is performed when the eigenvalue is insufficient.

[0070] That is, through the linear target mapping representation based on the similarity distance representation matrix of the inventory age multi-time series period significantly aggregated coding vector, the self-correlation of the complete similarity instantiation of the inventory age multi-time series period significantly aggregated coding vector is instantiated by a secondary target mapping representation based on a multi-level distribution hierarchy, and the association mismatch negative impact factor is compensated by the association fusion kernel bias to improve the instance decision degree of the eigenvalue of the inventory age multi-time series period significantly aggregated coding vector under similarity constraints, that is, the significance degree of the eigenvalue as an instance for the decoding regression decision, so as to improve the accuracy of the future inventory demand forecast value obtained by the inventory demand forecast module based on the decoder input of the inventory age multi-time series period significantly aggregated coding vector.

[0071] Exemplarily, in the purchase suggestion generation module 40, a purchase suggestion is generated based on the comparison between the future inventory demand forecast value and the current inventory quantity. For example, if the forecasted inventory demand value is higher than the current inventory quantity, the system will recommend increasing the purchase quantity to avoid potential inventory shortages; conversely, if the forecasted value is lower than the current inventory quantity, the system will recommend reducing or postponing purchases to avoid excess inventory and related costs. In this way, enterprises can make more accurate and timely inventory management decisions based on the forecast results, thereby avoiding capital occupation and storage cost increases caused by excessive stockpiling, and also preventing the loss of sales opportunities due to insufficient inventory, thereby reducing business risks.

[0072] In summary, according to the embodiment of the present application, the inventory age management system based on big data is explained, which calculates the inventory age distribution of the group of goods by extracting the inventory date and arrival date of each commodity in a specific commodity set, and then further introduces the data analysis technology based on deep learning to perform periodic time series pattern recognition and global pattern context perception on the inventory age distribution data of the group of goods to predict the future inventory demand of the goods, and then generates corresponding procurement recommendations based on the comparison between the prediction results and the current inventory. In this way, the enterprise inventory can be managed more accurately, thereby improving inventory turnover, reducing inventory costs, and enhancing the market competitiveness of the enterprise.

[0073] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0074] It should be understood that the specific examples in this article are only intended to help those skilled in the art better understand the embodiments of the present application, rather than to limit the scope of the embodiments of the present application.

[0075] It should also be understood that in the various embodiments of the present application, the size of the serial number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0076] It should also be understood that the various implementation methods described in this specification can be implemented individually or in combination, and the embodiments of the present application are not limited to this.

[0077] Unless otherwise stated, all technical and scientific terms used in the embodiments of the present application are the same as the meanings generally understood by those skilled in the art of the technical field of the present application. The terms used in the present application are only for the purpose of describing specific embodiments and are not intended to limit the scope of the present application. The term "and / or" used in the present application includes any and all combinations of one or more related listed items. The singular forms of "a kind of", "above" and "the" used in the embodiments of the present application and the appended claims are also intended to include majority forms, unless the context clearly indicates other meanings. In addition, the terms "first", "second" etc. are only used for description purposes and cannot be understood as indicating or suggesting relative importance.

[0078] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0079] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0080] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0081] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0082] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A big data-based inventory age management system, characterized in that: include: A storage age related data extraction module is used to extract the storage age related data of all first SKUs from the background database, wherein the storage age related data includes the SKU inventory date and the SKU arrival date; A stock age calculation module, used to calculate the stock age value of each first SKU based on the stock age related data of each SKU to obtain the stock age distribution of the first SKU; A future inventory demand prediction module, configured to perform future inventory demand prediction based on the inventory age distribution of the first SKU to obtain a future inventory demand prediction value; A purchasing suggestion generating module, used for generating purchasing suggestions based on the comparison between the future inventory demand forecast value and the current inventory quantity; Among them, the future inventory demand prediction module includes: an inventory age distribution pattern feature extraction unit, which is used to perform pattern feature extraction on the inventory age distribution of the first SKU based on singular spectrum decomposition to obtain multiple inventory age cycle pattern coding feature vectors; a context-aware aggregation unit, which is used to perform context-aware autocorrelation gated aggregation analysis on the multiple inventory age cycle pattern coding feature vectors to obtain a inventory age multi-time series period significant aggregation coding vector; an inventory demand prediction unit, which is used to generate the future inventory demand prediction value based on the inventory age multi-time series period significant aggregation coding vector.

2. The big data-based inventory management system according to claim 1 is characterized in that: The storage age distribution pattern feature extraction unit comprises: A singular spectrum decomposition subunit, used for inputting the inventory age distribution of the first SKU into a time series encoder based on singular spectrum decomposition to obtain a plurality of inventory age distribution subsequences; The time series pattern feature extraction subunit is used to extract the time series pattern features of each of the multiple storage age distribution subsequences to obtain the multiple storage age cycle pattern encoding feature vectors.

3. The inventory age management system based on big data according to claim 2 is characterized in that: The temporal pattern feature extraction subunit is used to: Each of the multiple storage age distribution subsequences is input into a time series pattern feature extractor based on the LSTM-RNN model to obtain the multiple storage age cycle pattern encoding feature vectors.

4. The big data-based inventory management system according to claim 3 is characterized in that: The context-aware aggregation unit comprises: A context-aware enhancement subunit, configured to perform graph walk context-aware enhancement on the plurality of storage age cycle pattern encoding feature vectors to obtain a plurality of storage age cycle pattern context-aware enhanced feature vectors; The gated aggregation subunit is used to perform significance aggregation on the multiple storage age cycle pattern encoding feature vectors based on the feature correlation between the multiple storage age cycle pattern context-aware enhanced feature vectors and the multiple storage age cycle pattern encoding feature vectors to obtain the storage age multi-time series period significant aggregation encoding vector.

5. The big data-based inventory age management system according to claim 4 is characterized in that: The context-aware reinforcement subunit comprises: A topological structure feature extraction secondary subunit is used to extract the topological structure features of the plurality of storage age cycle pattern encoding feature vectors to obtain a storage age cycle pattern graph wandering topological feature matrix; The global context walk secondary sub-unit is used to input the warehouse age cycle pattern graph walk topological feature matrix and the multiple warehouse age cycle pattern encoding feature vectors after hyperbolic space mapping into the global context walk encoder based on the graph convolutional neural network model to obtain the multiple warehouse age cycle pattern context-aware enhanced feature vectors.

6. The big data-based inventory age management system according to claim 5 is characterized in that: The topological structure feature extraction secondary subunit is used for: Inputting each of the plurality of storage age cycle pattern time sequence coding feature vectors into a hyperbolic space mapper to obtain a plurality of storage age cycle pattern coding feature vectors after hyperbolic space mapping; Calculating the Poincare distance between any two of the plurality of hyperbolic space mapped storage age cycle pattern encoding feature vectors to obtain a storage age cycle pattern graph walk topology matrix; The walk topology matrix of the storage age cycle pattern diagram is subjected to dilated convolution coding to obtain the walk topology feature matrix of the storage age cycle pattern diagram.

7. The big data-based inventory age management system according to claim 6 is characterized in that: The gated polymerization subunit is used for: Inputting each group of corresponding storage age cycle pattern context-aware enhanced feature vectors and storage age cycle pattern time-series encoded feature vectors in the multiple storage age cycle pattern context-aware enhanced feature vectors and the multiple storage age cycle pattern encoded feature vectors into an autocorrelation gating unit to obtain multiple storage age cycle pattern autocorrelation gating significant confidence factors; Inputting the plurality of storage age cycle pattern autocorrelation gated significant confidence factors into a normalization unit based on a Softmax function to obtain a plurality of storage age cycle pattern autocorrelation gated significant confidence weight factors; Based on the autocorrelation gated significant confidence weight factors of the multiple storage age cycle patterns, the position-weighted sum of the multiple storage age cycle pattern encoding feature vectors is calculated to obtain the storage age multi-time series period significant aggregation encoding vector.

8. The big data-based inventory age management system according to claim 7 is characterized in that: The inventory demand forecasting unit is used to: The inventory age multi-time period significant aggregation coding vector is input into a decoder-based inventory demand prediction module to obtain the future inventory demand prediction value.

Images