Intelligent warehouse inventory dynamic optimization method and system based on supply and demand trend prediction

Abstract

The application discloses an intelligent warehouse inventory dynamic optimization method and system based on supply and demand trend prediction, which comprises the following steps: firstly, obtaining historical warehousing records and delivery records in a warehouse management system, generating warehousing time sequence distribution sequences and delivery time sequence distribution sequences according to time markers, constructing an inventory item flow directed network containing item category nodes and warehousing and delivery directed edges, predicting the flow trend of the network, obtaining item supply and demand trend prediction information, and dynamically adjusting storage locations and replenishment trigger thresholds according to the information to generate inventory dynamic optimization instructions.

Description

Technical Field

[0001] This invention relates to the field of intelligent warehousing technology, and more specifically, to an intelligent warehousing inventory dynamic optimization method and system based on supply and demand trend prediction. Background Technology

[0002] With the rapid development of e-commerce, intelligent manufacturing, and supply chain collaborative management, the types of goods, inventory size, and turnover frequency in warehousing systems are constantly increasing. Traditional warehousing inventory management usually relies on manual experience, fixed replenishment rules, or simple upper and lower inventory thresholds for inventory control, which makes it difficult to reflect in a timely manner the changes in inbound and outbound of different product categories and the trends of supply and demand fluctuations at different time periods.

[0003] While existing warehouse management systems can record the inbound and outbound times, inbound and outbound quantities, and item category information, most systems only use this data for inventory balance statistics, book-to-physical reconciliation, or historical queries. They fail to fully leverage the time-series changes and supply-demand trends implicit in historical inventory turnover records. When demand for certain items continues to rise, using fixed replenishment trigger thresholds can easily lead to untimely replenishment, inventory shortages, or order fulfillment delays. Conversely, when supply increases but demand decreases for certain items, failure to adjust storage locations and inventory strategies in a timely manner can result in unreasonable storage space utilization, increased picking routes, and higher warehousing costs. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for intelligent warehousing inventory dynamic optimization based on supply and demand trend prediction.

[0005] In a first aspect, embodiments of the present invention provide an intelligent warehousing inventory dynamic optimization method based on supply and demand trend prediction, the method comprising:

[0006] Obtain a set of historical inventory turnover records from the warehouse management system. The set of historical inventory turnover records includes a sequence of inbound records with an inbound time stamp and a sequence of outbound records with an outbound time stamp. The inbound record sequence includes an inbound item category identifier and an inbound quantity parameter. The outbound record sequence includes an outbound item category identifier and an outbound quantity parameter.

[0007] Based on the inbound time stamp, the inbound record sequence is arranged along the time axis to generate an inbound time sequence distribution sequence. Based on the outbound time stamp, the outbound record sequence is arranged along the time axis to generate an outbound time sequence distribution sequence.

[0008] A directed network for the flow of inventory items is constructed based on the inbound time sequence distribution sequence and the outbound time sequence distribution sequence. The directed network for the flow of inventory items includes item category nodes, inbound directed edges, and outbound directed edges. The inbound directed edges carry the inbound quantity parameter, and the outbound directed edges carry the outbound quantity parameter.

[0009] The flow trend prediction processing is performed on the directional network of inventory items to generate supply and demand trend prediction information for the items. The supply and demand trend prediction information for the items includes the demand trend direction identifier and the supply trend direction identifier of the item category.

[0010] Based on the predicted supply and demand trends of the goods, the storage locations and replenishment trigger thresholds of the warehouse inventory are dynamically adjusted to generate a set of dynamic inventory optimization instructions, which includes storage location adjustment instructions and replenishment trigger threshold update instructions.

[0011] Secondly, embodiments of the present invention provide an intelligent warehousing and inventory dynamic system based on supply and demand trend prediction, including at least one service node;

[0012] The service node includes a storage unit and a computing unit; the storage unit is used to store program code; the computing unit is used to run the program code to perform the intelligent warehousing inventory dynamic optimization based on supply and demand trend prediction as described in the first aspect.

[0013] Compared to existing technologies, the beneficial effects of this invention include: Employing the intelligent warehousing inventory dynamic optimization method and system based on supply and demand trend prediction disclosed in this invention, historical inventory flow records are acquired, including time-stamped inbound and outbound record sequences, recording item categories and quantity parameters. Inbound and outbound records are sequentially arranged according to time stamps, generating inbound and outbound time-series distribution sequences. A directed network for inventory item flow is constructed based on these two time-series distribution sequences, with item categories as nodes and directed edges carrying inbound and outbound quantity parameters to characterize the flow relationships. Flow trend prediction is performed on this network to generate item supply and demand trend prediction information, including demand trend direction identifiers and supply trend direction identifiers for item categories. Finally, based on the supply and demand trend prediction information, the storage locations and replenishment trigger thresholds for warehouse inventory items are dynamically adjusted, generating storage location adjustment instructions and replenishment trigger threshold update instructions. This invention can adapt to changes in supply and demand, effectively reducing inventory backlog and stockout risks, and improving warehousing dynamic optimization capabilities. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as limiting the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 A flowchart illustrating the steps of the intelligent warehousing inventory dynamic optimization method based on supply and demand trend prediction provided in this embodiment of the invention;

[0016] Figure 2 This is a schematic diagram of the construction of a directed network for the flow of inventory items provided in an embodiment of the present invention;

[0017] Figure 3 A schematic diagram illustrating the adjustment of directed edge weights for inventory item flow based on external supply chain events, as provided in this embodiment of the invention.

[0018] Figure 4 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0020] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0021] In order to solve the technical problems mentioned in the background art Figure 1 This is a flowchart illustrating the intelligent warehousing inventory dynamic optimization method based on supply and demand trend prediction provided in this embodiment. The following is a detailed description of the intelligent warehousing inventory dynamic optimization method based on supply and demand trend prediction.

[0022] Step S201: Obtain a set of historical inventory turnover records in the warehouse management system. The set of historical inventory turnover records includes a sequence of inbound records with an inbound time stamp and a sequence of outbound records with an outbound time stamp. The inbound record sequence includes an inbound item category identifier and an inbound quantity parameter. The outbound record sequence includes an outbound item category identifier and an outbound quantity parameter.

[0023] Step S202: Arrange the inbound record sequence along the time axis according to the inbound time marker to generate an inbound time sequence distribution sequence; and arrange the outbound record sequence along the time axis according to the outbound time marker to generate an outbound time sequence distribution sequence.

[0024] Step S203: Construct a directed network for the flow of inventory items based on the inbound time sequence distribution sequence and the outbound time sequence distribution sequence. The directed network for the flow of inventory items includes item category nodes, inbound directed edges, and outbound directed edges. The inbound directed edges carry the inbound quantity parameter, and the outbound directed edges carry the outbound quantity parameter.

[0025] Step S204: Perform flow trend prediction processing on the directed network of inventory item flow to generate item supply and demand trend prediction information. The item supply and demand trend prediction information includes the demand trend direction identifier and supply trend direction identifier of item category.

[0026] Step S205: Based on the predicted supply and demand trend information of the goods, the storage location and replenishment trigger threshold of the warehouse inventory are dynamically adjusted to generate a set of dynamic inventory optimization instructions, which includes storage location adjustment instructions and replenishment trigger threshold update instructions.

[0027] In this embodiment of the invention, for example, the server acts as the execution entity, connected to the warehouse management system, the order system, the supplier collaboration system, and the warehouse scheduling equipment. The server obtains a set of historical inventory turnover records for the most recent ninety days from the warehouse management system. This set includes a sequence of inbound records with inbound time stamps and a sequence of outbound records with outbound time stamps. The inbound record sequence includes the category identifier of the inbound item and the inbound quantity parameter, such as "Frozen Chicken Breast - Category A, May 1, 2026, 08:10, 120 boxes inbound" and "Pre-prepared Dishes - Category C, May 1, 2026, 11:20, 80 boxes inbound". The outbound record sequence includes the category identifier of the outbound item and the outbound quantity parameter, such as "Frozen Chicken Breast - Category A, May 1, 2026, 15:20, 95 boxes outbound" and "Pre-prepared Dishes - Category C, May 1, 2026, 18:40, 120 boxes outbound". When retrieving records, the server simultaneously reads the batch number, warehouse area number, supplier number, and order source, which are used for subsequent merging and verification of the flow data of the same item category.

[0028] The server arranges the inbound record sequence along the time axis based on the inbound time stamp, generating an inbound time-series distribution sequence; similarly, it arranges the outbound record sequence along the time axis based on the outbound time stamp, generating an outbound time-series distribution sequence. For multiple inbound or outbound transactions of the same item category on the same day, the server retains the time sequence of each transaction and calculates cumulative values ​​across daily, weekly, and replenishment cycle dimensions. For example, in the ready-to-eat meals category C, the outbound quantity increased from 120 boxes per day to 210 boxes per day between May 1st and May 7th, while the inbound quantity only increased from 80 boxes per day to 100 boxes per day. This allows the server to establish a data basis for "demand growth outpacing supply growth" in the time-series distribution sequence. For the frozen chicken breast category A, which experiences a fixed outbound peak on weekends, the server retains its periodic fluctuation characteristics to avoid misjudging normal sales peaks as abnormal changes.

[0029] The server constructs a directed network for the flow of inventory items based on the inbound and outbound time-series distribution sequences. The server establishes item category nodes based on the inbound item category identification, such as "Frozen Chicken Breast - Category A Node," "Rice, Flour, and Oil - Category B Node," and "Pre-prepared Vegetables - Category C Node," forming a first mapping relationship between inbound item category identifications and item category nodes. The server then queries this first mapping relationship based on the outbound item category identification and associates the outbound record with the corresponding node. Subsequently, the server connects item category nodes corresponding to adjacent inbound time markers with directed edges for inbound flow, and connects item category nodes corresponding to adjacent outbound time markers with directed edges for outbound flow, assigning inbound and outbound quantity parameters as edge weight parameters to the corresponding directed edges. The server aggregates the edge weight parameters corresponding to the same item category identification to obtain the total inbound and outbound flow parameters for each item category, and appends them to the corresponding nodes. For example, the pre-cooked food-Category C node records "680 boxes of goods received in the past seven days and 1280 boxes of goods shipped out in the past seven days", thus forming a directed network of inventory flow that reflects the supply replenishment path and the demand consumption path.

[0030] The server performs trend prediction processing on the directed network of inventory flow. It extracts the total inbound and outbound flow parameters for each item category node, forming a pairing sequence of item category flow volumes, and divides it according to three-day windows, seven-day windows, and replenishment cycle windows. The server compares the cumulative inbound flow values ​​within adjacent time windows, generating an inbound growth trend marker or an inbound decline trend marker, and determines the supply trend direction identifier based on the dominant inbound trend marker across multiple consecutive windows; simultaneously, it compares the cumulative outbound flow values ​​within adjacent time windows to determine the demand trend direction identifier. The server also recursively accumulates the edge weight parameters along the connection directions of the directed inbound and outbound flow edges, generating supply path trend direction identifiers and demand path trend direction identifiers, and integrates them with the node's own trend. For the case where frozen dumplings (Category D) and ready-to-eat meals (Category C) are linked for outbound shipment in promotional orders, the server tracks the indirectly related nodes along the directed edge of the outbound flow, and accumulates the edge weight parameters with decreasing tracking steps to obtain the indirect impact weight of outbound shipment. The demand trend direction indicator is then corrected to make the trend prediction more consistent with the actual changes in order combinations.

[0031] The server also integrates external supply chain event data streams. The supplier collaboration system pushes notifications such as "The main supplier of pre-prepared meals (Category C) is undergoing production line maintenance, resulting in a 30% supply decrease over the next five days," and the order system pushes notifications such as "Live-stream promotions are expected to increase pre-prepared meals (Category C) orders by 45%." The server performs semantic parsing on these notifications, extracting the capacity change direction identifier, order fluctuation direction identifier, and corresponding affected item category identifiers, and mapping them to the corresponding item category nodes. Based on this, the server corrects the supply trend direction identifier and demand trend direction identifier, and updates the trend prediction weight coefficients of the inbound and outbound directed edges. The server further performs quantity jump detection on the edge weight parameters. When the outbound quantity of pre-prepared meals (Category C) jumps from 120 boxes per day to 260 boxes, the server marks the outbound abnormal jump edge and queries the external supply chain event data stream based on the time of the change to establish a causal relationship between this jump and the live-stream promotion order fluctuations, thereby correcting the demand trend strength parameter.

[0032] In this embodiment of the invention, when the server performs semantic parsing on the external supply chain event data stream, it adopts a combination of event template matching, keyword direction recognition, and item category dictionary mapping. The server pre-establishes a supply chain event dictionary, which includes at least keywords for increased production capacity, decreased production capacity, increased order volume, decreased order volume, and an item category alias table. Keywords for increased production capacity include "new production line," "resumption of supply," "capacity improvement," "increased supply," and "early arrival"; keywords for decreased production capacity include "production line maintenance," "production stoppage," "delayed delivery," "decreased supply," "out of stock," and "limited supply"; keywords for increased order volume include "promotion," "live streaming sales," "order growth," "increased pre-sales," and "increased demand"; keywords for decreased order volume include "event end," "decreased orders," "decreased demand," and "increased cancellations." The server generates event parsing results according to the following field structure: Event={Event Type, Direction Identifier, Affected Item Category Identifier, Impact Ratio, Effective Start Time, Effective End Time, Confidence Level}; where Event Type includes supply events and demand events; Direction Identifier includes directions of increased production capacity, decreased production capacity, increased order volume, and decreased order volume; Impact Ratio is extracted from percentages, quantity changes, or level terms in the notification text. If the impact ratio is not explicitly stated in the text, a default impact ratio is determined based on a preset level terminology. For example, "Slight Decrease" corresponds to 10%, "Significant Decrease" corresponds to 30%, and "Large Decrease" corresponds to 50%. When the notification text is "The main supplier of pre-prepared dishes - Category C will reduce its supply capacity by 30% in the next five days due to production line maintenance," the server parses it to obtain: Event={Supply Event, Direction of Decrease in Production Capacity, Pre-prepared Dishes - Category C, 30%, Current Day, Current Day + 5 Days, 0.9}; the server adjusts the trend event score based on the event parsing results. For supply events, Sin_event(i) takes a positive value if the direction indicates an increase in capacity, and a negative value if the direction indicates a decrease in capacity. For demand events, Sout_event(i) takes a positive value if the direction indicates an increase in order volume, and a negative value if the direction indicates a decrease in order volume. The event score is determined according to the following formula: Sin_event(i) = Dir_s × Impact × Conf × TimeDecay, Sout_event(i) = Dir_d × Impact × Conf × TimeDecay; where Dir_s and Dir_d are the direction values, with +1 for the growth direction and -1 for the decay direction; Impact is the event's impact ratio, calculated as a decimal, for example, 30% is recorded as 0.30; Conf is the semantic resolution confidence; TimeDecay is the event's effective time decay coefficient, which is 1 during the event's validity period and decreases daily after the validity period, with a value range of 0 to 1.The server updates the trend prediction weight coefficients of the directed edges for inbound and outbound flow based on the event scores: Win_event(e) = Win(e) × max{0, 1 + η × Sin_event(i)}, Wout_event(e) = Wout(e) × max{0, 1 + η × Sout_event(i)}; where Win(e) and Wout(e) are the original inbound and outbound edge weights, respectively, and Win_event(e) and Wout_event(e) are the event-corrected inbound and outbound edge weights, respectively, and η is the event correction coefficient, preferably ranging from 0.2 to 1.0. By constraining the correction coefficients with max{0, ·}, the situation where edge weights are negative can be avoided when there is a significant drop in production capacity or order volume.

[0033] Based on supply and demand trend predictions, the server dynamically allocates storage locations and replenishment trigger thresholds. The server obtains a warehouse inventory layout map, which includes storage location identifiers, location space capacity parameters, and current occupied capacity parameters. For example, Cold Chain Zone 1 is close to the sorting line, has a capacity of 1200 boxes, and currently occupies 920 boxes; Cold Chain Zone 2 is farther from the sorting line, has a capacity of 1800 boxes, and currently occupies 1100 boxes. The server generates a demand ranking queue for item categories based on demand trend direction identifiers and a supply ranking queue for item categories based on supply trend direction identifiers. It also generates a location mapping relationship between item categories and storage location identifiers based on remaining space capacity. For the pre-prepared food category (C) with strong demand growth and high outbound frequency, the server moves it from Cold Chain Zone 2 to Cold Chain Zone 1, closer to the repackaging line, and moves the slower-turnover frozen dumplings category (D) to a later storage location in Cold Chain Zone 2, generating a storage location adjustment instruction.

[0034] Simultaneously, the server determines the adjustment direction of the replenishment trigger threshold based on the demand trend direction indicator and the adjustment range based on the supply trend direction indicator. For the pre-prepared vegetables category (C), due to increased demand and decreased supply, the server raises the replenishment trigger threshold from 300 boxes to 520 boxes; for the rice, flour, and oil category (B), due to decreased demand and stable supply, the server lowers the replenishment trigger threshold from 800 bags to 620 bags. The server also calculates inbound and outbound capacity saturation parameters based on remaining space capacity. When a category's corresponding location approaches its capacity limit, it searches for adjacent item category nodes along the inbound or outbound directed edges, generates a demand transmission plan, and adjusts storage location adjustment instructions and replenishment trigger threshold update instructions accordingly. Finally, the server combines these instructions to form a set of dynamic inventory optimization instructions and sends them to the warehouse management system and warehouse scheduling terminal, achieving dynamic inventory optimization driven by supply and demand trend prediction.

[0035] Please refer to the following: Figure 2 , Figure 2 This is a schematic diagram of the construction of a directed network for the circulation of inventory items provided in an embodiment of the present invention. In this embodiment, the construction of the directed network for the circulation of inventory items based on the inbound time sequence distribution sequence and the outbound time sequence distribution sequence can be implemented through the following example.

[0036] Extract the item category identifier corresponding to each entry time marker in the entry time distribution sequence, and map the item category identifier to an item category node, thus establishing a first mapping relationship between the item category identifier and the item category node;

[0037] Extract the outbound item category identifier corresponding to each outbound time marker in the outbound time sequence distribution sequence, query the first mapping relationship through the outbound item category identifier, and associate the outbound item category identifier with the established item category node;

[0038] The item category nodes corresponding to the item category identifiers of adjacent entry time markers in the entry time distribution sequence are unidirectionally connected by the entry flow directed edge, and the direction of the entry flow directed edge is from the previous entry time marker to the subsequent entry time marker.

[0039] The item category nodes corresponding to the item category identifiers of adjacent outbound time markers in the outbound time sequence are unidirectionally connected by outbound flow directed edges, with the direction of the outbound flow directed edges pointing from the preceding outbound time marker to the following outbound time marker.

[0040] Assign the quantity parameter of each entry record in the entry record sequence to the directed edge of the entry flow corresponding to that entry record, as the edge weight parameter of the directed edge of the entry flow;

[0041] Assign the outbound quantity parameter corresponding to each outbound record in the outbound record sequence to the outbound flow directed edge corresponding to that outbound record, as the edge weight parameter of the outbound flow directed edge;

[0042] Based on the first mapping relationship, the edge weight parameters of all the directed edges of the same item category corresponding to the inbound flow are aggregated to obtain the total inbound flow parameter of the item category. Based on the first mapping relationship, the edge weight parameters of all the directed edges of the same item category corresponding to the outbound flow are aggregated to obtain the total outbound flow parameter of the item category.

[0043] The total inbound and outbound flow parameters of the item category are respectively attached to the corresponding item category node to obtain the item category node carrying the inbound total flow attribute and the outbound total flow attribute.

[0044] The directed network for inventory item circulation is formed by combining the item category nodes, the directed edges for inbound circulation, the directed edges for outbound circulation, the edge weight parameters, the total inbound circulation parameters for the item category, and the total outbound circulation parameters for the item category.

[0045] In an embodiment of the present invention, exemplaryly, in the specific implementation of constructing a directed network for the flow of inventory goods, the server first reads the time-series distribution sequence of inbound goods that has been sorted by time, and extracts the inbound item category identifier corresponding to each inbound time marker. Taking a fresh food central warehouse as an example, the inbound time-series distribution sequence contains records such as "May 1, 2026, 08:10, Frozen Chicken Breast - Category A, 120 boxes inbound", "May 1, 2026, 09:35, Rice, Flour and Oil - Category B, 300 bags inbound", and "May 1, 2026, 11:20, Prepared Food - Category C, 80 boxes inbound". The server maps "Frozen Chicken Breast - Category A", "Rice, Flour and Oil - Category B", and "Prepared Food - Category C" to the corresponding item category nodes, and establishes a first mapping relationship between the inbound item category identifier and the item category node in the memory data table, so that subsequent inbound and outbound data of the same category can be aggregated to the same node.

[0046] Subsequently, the server reads the outbound time-series distribution sequence and extracts the corresponding outbound item category identifier for each outbound time marker. For outbound records such as "May 1, 2026, 15:20, Frozen Chicken Breast - Category A, 95 boxes outbound" and "May 1, 2026, 18:40, Prepared Food - Category C, 120 boxes outbound," the server queries the first mapping relationship through the outbound item category identifier, associating the outbound record with the already established Frozen Chicken Breast - Category A node and Prepared Food - Category C node. Thus, both inbound replenishment and outbound consumption for the same item category fall within the same node system, avoiding inconsistencies in trend judgment caused by scattered storage of inbound and outbound data.

[0047] After node association is completed, the server connects the item category nodes corresponding to adjacent inbound time markers through directed edges, according to the chronological order of the inbound time sequence. The connection direction is from the earlier inbound time marker to the later inbound time marker. For example, the server connects the frozen chicken breast (Category A) node at 08:10 to the rice, flour, and cooking oil (Category B) node at 09:35, and then connects the rice, flour, and cooking oil (Category B) node to the prepared food (Category C) node at 11:20, to express the inbound flow sequence of the central warehouse within that time period. Simultaneously, the server connects the item category nodes corresponding to adjacent outbound time markers through directed edges, according to the chronological order of the outbound time sequence. The connection direction is from the earlier outbound time marker to the later outbound time marker, to express the outbound consumption sequence during order fulfillment.

[0048] The server further assigns the quantity parameter corresponding to each inbound record to the directed edge of the inbound flow corresponding to that record, using it as the edge weight parameter. For example, if 120 boxes of frozen chicken breast (Category A) are inbound, the server writes "120" to the corresponding directed edge of the inbound flow; if 80 boxes of pre-prepared dishes (Category C) are inbound, the server writes "80" to the corresponding edge weight. Similarly, the server assigns the quantity parameter corresponding to each outbound record to the corresponding directed edge of the outbound flow. For example, if 120 boxes of pre-prepared dishes (Category C) are outbound, 160 boxes are outbound the next day, and 260 boxes are outbound during the weekend promotion, these are all stored as edge weight parameters of the directed edge of the outbound flow.

[0049] After assigning edge weights, the server aggregates the edge weight parameters of all directed edges corresponding to the same item category identification for inbound flow, based on the first mapping relationship, to obtain the total inbound flow parameter for the item category. Simultaneously, it aggregates the edge weight parameters of all directed edges corresponding to the same item category identification for outbound flow, to obtain the total outbound flow parameter for the item category. For example, if the total inbound volume of the pre-prepared food-C category is 680 boxes and the total outbound volume is 1280 boxes in the past seven days, the server will append "total inbound volume 680 boxes" and "total outbound volume 1280 boxes" to the pre-prepared food-C category node, forming item category nodes that carry both inbound and outbound volume attributes. Finally, the server combines all item category nodes, directed edges for inbound flow, directed edges for outbound flow, edge weight parameters, and the inbound and outbound volume attributes of each node to form a directed network for inventory item flow, providing a structured data foundation for subsequent supply trends, demand trends, warehouse location adjustments, and replenishment threshold updates.

[0050] In this embodiment of the invention, the process of predicting the flow trend of the inventory goods flow in a directed network to generate supply and demand trend prediction information can be implemented through the following example.

[0051] Extract the total inbound and outbound flow parameters of each item category node in the directed network of inventory item flow to form an item category flow volume pairing sequence.

[0052] The item category turnover pairing sequence is divided into time windows along the time axis to obtain multiple inbound turnover time window sequences and multiple outbound turnover time window sequences. Each inbound turnover time window sequence corresponds to the cumulative value of the total inbound turnover within a time window, and each outbound turnover time window sequence corresponds to the cumulative value of the total outbound turnover within a time window.

[0053] The cumulative total value of the inbound turnover in adjacent time windows is compared with the trend direction. If the cumulative total value of the inbound turnover in the subsequent time window is greater than the cumulative total value of the inbound turnover in the preceding time window, an inbound growth trend mark is generated. If the cumulative total value of the inbound turnover in the subsequent time window is less than the cumulative total value of the inbound turnover in the preceding time window, an inbound decline trend mark is generated.

[0054] The cumulative outbound turnover values ​​of adjacent time windows are compared in terms of trend direction. If the cumulative outbound turnover value of the subsequent time window is greater than that of the preceding time window, an outbound growth trend mark is generated. If the cumulative outbound turnover value of the subsequent time window is less than that of the preceding time window, an outbound decline trend mark is generated.

[0055] The frequency of occurrence of the inbound growth trend marker and the frequency of occurrence of the inbound decline trend marker are counted within multiple consecutive time windows. The inbound trend marker direction with the higher frequency of occurrence is determined as the supply trend direction identifier for this item category.

[0056] The frequency of occurrence of the outbound growth trend marker and the frequency of occurrence of the outbound decline trend marker are counted within multiple consecutive time windows. The outbound trend marker direction with the dominant frequency of occurrence is determined as the demand trend direction identifier for this item category.

[0057] The edge weight parameters of the directed edge of the inbound flow are recursively accumulated along the connection direction of the directed edge of the inbound flow to obtain the cumulative flow sequence of the inbound flow path. The supply path trend direction identifier is generated based on the change direction of the cumulative flow sequence within a continuous time window.

[0058] The edge weight parameters of the outbound flow directed edge are recursively accumulated along the connection direction of the outbound flow directed edge to obtain the cumulative flow sequence of the outbound flow path. The demand path trend direction identifier is generated based on the change direction of the cumulative flow sequence within a continuous time window.

[0059] By combining the supply trend direction identifier and the supply path trend direction identifier, a supply trend direction identifier for the item category is generated; by combining the demand trend direction identifier and the demand path trend identifier, a demand trend direction identifier for the item category is generated.

[0060] The supply trend direction identifier and the demand trend direction identifier of each item category are combined to form the item supply and demand trend prediction information for that item category.

[0061] In an embodiment of the invention, exemplaryly, during the specific implementation of the flow trend prediction process, the server uses the already constructed directed network of inventory flow as the data object, and reads the total inbound and outbound flow parameters of each item category node in the network one by one, and combines the two into an item category flow volume pairing sequence. For example, the pre-cooked food-C category node records a total inbound volume of 680 boxes and a total outbound volume of 1280 boxes in the past seven days; the frozen chicken breast-A category node records a total inbound volume of 950 boxes and a total outbound volume of 910 boxes; and the rice, flour, and cooking oil-B category node records a total inbound volume of 2100 bags and a total outbound volume of 1500 bags. The server writes these paired data into the flow volume pairing sequence according to the corresponding time stamp, so that the supply replenishment and demand consumption of each category can be compared synchronously.

[0062] The server then divides the item category turnover pairing sequence into time windows along the time axis. For example, the server divides the data from the most recent fifteen days into five three-day time windows, calculating the cumulative total inbound and outbound turnover within each window. For the pre-prepared food-C category, the server calculates that the first window saw 240 boxes inbound and 330 boxes outbound, the second window saw 260 boxes inbound and 450 boxes outbound, and the third window saw 300 boxes inbound and 620 boxes outbound. The server compares the cumulative inbound turnover values ​​of adjacent time windows to identify trends, finding that 260 boxes are greater than 240 boxes and 300 boxes are greater than 260 boxes, continuously generating inbound growth trend markers; simultaneously, it compares the cumulative outbound turnover values ​​of adjacent time windows, identifying that 450 boxes are greater than 330 boxes and 620 boxes are greater than 450 boxes, continuously generating outbound growth trend markers. For the rice, flour, grain and oil category B, if the server detects that the outbound volume in the subsequent window is lower than the outbound volume in the previous window, it will generate an outbound decline trend marker.

[0063] After completing trend marking across multiple consecutive time windows, the server counts the frequency of occurrence of the inbound growth trend mark and the inbound decline trend mark for each item category, and determines the inbound trend direction with the most frequent occurrences as the supply trend direction identifier for that item category. The server similarly counts the frequency of occurrence of the outbound growth trend mark and the outbound decline trend mark, and determines the demand trend direction identifier. For example, the ready-to-eat meals (Category C) showed an outbound growth trend mark three times in four consecutive window comparisons, so the server determined its demand trend direction identifier to be a demand growth trend; the frozen chicken breast (Category A) saw both inbound and outbound shipments maintain a slight increase, so the server determined its supply and demand trends to be both stable growth trends; and the rice, flour, and cooking oil (Category B) had a dominant outbound decline mark, so the server determined its demand trend direction identifier to be a demand decline trend.

[0064] The server further refines the process by incorporating path changes in the directed network. It recursively accumulates the edge weights of the inbound flow directed edges along the connection direction of these edges, obtaining a cumulative flow sequence for the inbound flow path. Based on the direction of change of this sequence within a continuous time window, it generates a supply path trend direction indicator. For example, although the upstream inbound flow path for pre-prepared dishes (Category C) shows a slight increase in inbound volume at the node itself, the cumulative flow is reduced due to supplier production line maintenance, resulting in a weakened supply path indicator. Similarly, the server recursively accumulates the edge weights along the outbound flow directed edges, obtaining a cumulative flow sequence for the outbound flow path, and generates a demand path trend direction indicator accordingly. For example, if pre-prepared dishes (Category C) and frozen dumplings (Category D) continuously share outbound orders during live-stream promotions, the server identifies a continuous increase in the cumulative outbound flow, generating a strengthened demand path indicator.

[0065] In this embodiment of the invention, to ensure that the supply trend direction identifier and demand trend direction identifier have deterministic and repeatable calculation, the server converts the trend direction identifier into a trend score for processing. For the i-th item category, within the t-th time window, the server calculates the inbound window change rate and the outbound window change rate as follows: the inbound window change rate is: Rin(i,t)=(Qin(i,t)-Qin(i,t-1)) / max(Qin(i,t-1),ε); the outbound window change rate is: Rout(i,t)=(Qout(i,t)-Qout(i,t-1)) / max(Qout(i,t-1),ε); where Qin(i,t) represents the cumulative inbound turnover of the i-th item category within the t-th time window, Qout(i,t) represents the cumulative outbound turnover of the i-th item category within the t-th time window, and ε is a preset minimum positive number to avoid a denominator of zero, with a value range of 0.01 to 1. When Rin(i,t) is greater than the first growth threshold θ1, the server generates an inbound growth trend marker; when Rin(i,t) is less than the negative first growth threshold -θ1, the server generates an inbound decline trend marker; when Rin(i,t) is between -θ1 and θ1, the server generates an inbound stable trend marker. θ1 can be determined based on the historical fluctuation level of the item category, preferably between 0.05 and 0.15. The outbound growth trend marker, outbound decline trend marker, and outbound stable trend marker are generated by Rout(i,t) in the same way. The server assigns a value of +1 to the inbound growth trend marker, a value of 0 to the inbound stable trend marker, and a value of -1 to the inbound decay trend marker. It then calculates the node supply trend score over N consecutive time windows: Sin_node(i) = Σ[t=1 to N]w_in(t) × M_in(i,t); where M_in(i,t) is the inbound trend marker value corresponding to the t-th time window, and w_in(t) is the time decay weight, with the weight increasing closer to the current time window. Preferably, w_in(t) = ρ (N-t) / Σ[t=1 to N]ρ (N-t) ρ ranges from 0.7 to 0.95. The demand node trend score is calculated using the following formula: Sout_node(i) = Σ[t=1 to N]w_out(t) × M_out(i,t); where M_out(i,t) is the outbound trend marker value corresponding to the t-th time window, and w_out(t) = ρ (N-t) / Σ[t=1 to N]ρ (N-t)By normalizing the time decay weights, the values ​​of Sin_node(i) and Sout_node(i) can be stably kept between -1 and +1, facilitating subsequent comparison with the trend judgment threshold. For path trends, the server recursively accumulates the edge weight parameters along the directed edge connection direction of the inbound flow, generating the inbound path accumulation Pin(i,t), and calculates the inbound path change rate: Rin_path(i,t)=(Pin(i,t)-Pin(i,t-1)) / max(Pin(i,t-1),ε). The server generates the path inbound trend label M_in_path(i,t) based on the comparison result of Rin_path(i,t) and the first growth threshold θ1, and calculates the supply path trend score according to Sin_path(i)=Σ[t=1 to N]w_in(t)×M_in_path(i,t). Similarly, the server recursively accumulates the edge weight parameters along the directed edge connection direction of the outbound flow to generate the cumulative outbound path Pout(i,t), calculates the outbound path change rate: Rout_path(i,t)=(Pout(i,t)-Pout(i,t-1)) / max(Pout(i,t-1),ε), and calculates the demand path trend score according to Sout_path(i)=Σ[t=1 to N]w_out(t)×M_out_path(i,t). The server merges node trends and path trends according to the following formulas: Sin_final(i) = α × Sin_node(i) + β × Sin_path(i) + γ × Sin_event(i); Sout_final(i) = α × Sout_node(i) + β × Sout_path(i) + γ × Sout_event(i); where Sin_final(i) is the final supply trend score, and Sout_final(i) is the final demand trend score; Sin_event(i) and Sout_event(i) are external supply chain event correction scores, which are 0 when no external events are received; α, β, and γ are weighting coefficients, satisfying α + β + γ = 1. Preferably, when making predictions based solely on historical flow data, α is 0.6, β is 0.4, and γ is 0; when accessing external supply chain event data streams, α is 0.5, β is 0.3, and γ is 0.2. When Sin_final(i) ≥ θ2, the server identifies the supply trend direction of the i-th item category as a supply growth trend; when Sin_final(i) ≤ -θ2, it identifies it as a supply decline trend; when Sin_final(i) is between -θ2 and θ2, it identifies it as a stable supply trend. The demand trend direction is determined by comparing Sout_final(i) with θ2. θ2 is preferably set between 0.2 and 0.5.

[0066] When the node trend direction is opposite to the path trend direction, the server does not directly discard either trend, but instead forms the final trend score according to the weighted formula mentioned above. When the absolute value of the final trend score is lower than θ2, it is determined as a stable trend, thereby avoiding incorrect judgments caused by fluctuations in a single window.

[0067] Finally, the server merges the supply trend direction identifier obtained based on time window statistics with the supply path trend direction identifier obtained by recursion based on the inbound path to generate the final supply trend direction identifier for that item category; it also merges the demand trend direction identifier obtained based on time window statistics with the demand path trend direction identifier obtained by recursion based on the outbound path to generate the final demand trend direction identifier. For the pre-prepared vegetables-C category, the server generates a supply and demand trend prediction information of "weak supply growth or tightening, and rapid demand growth"; for the rice, flour, grains, and oils-B category, the server generates a supply and demand trend prediction information of "stable supply and declining demand". The server saves the combined supply and demand trend direction identifiers for each item category as the basis for subsequent warehouse location adjustments, replenishment trigger threshold updates, and abnormal flow corrections.

[0068] In this embodiment of the invention, the dynamic allocation of storage locations and replenishment trigger thresholds of warehouse inventory items based on the predicted supply and demand trend information of the items, and the generation of a set of dynamic inventory optimization instructions, can be implemented through the following examples.

[0069] Obtain a warehouse inventory location layout map, which includes multiple storage location identifiers, location space capacity parameters corresponding to each storage location identifier, and current occupied capacity parameters corresponding to each storage location identifier.

[0070] Based on the demand trend direction indicator in the supply and demand trend prediction information, the item categories are sorted by the urgency of demand. Item categories with a demand trend direction of increasing demand are arranged before item categories with a demand trend of decreasing demand, thus obtaining an item category demand sorting queue.

[0071] Based on the supply trend direction indicator in the supply and demand trend prediction information, the product categories are sorted according to the degree of supply sufficiency. Product categories with a supply trend direction of decreasing supply are arranged before product categories with a supply growth trend, thus obtaining the product category supply sorting queue.

[0072] Based on the order of the item category demand sorting queue and the order of the item category supply sorting queue, and combined with the location space capacity parameter and the current occupied capacity parameter of the storage location identifier, location matching processing is performed to generate a location mapping relationship between each item category and the storage location identifier.

[0073] Based on the location mapping relationship, a storage location adjustment instruction is generated. The storage location adjustment instruction includes the target identifier of the item type to be adjusted, the original storage location identifier, the target storage location identifier, and the adjustment execution sequence.

[0074] Extract the demand trend direction identifier of each item category in the item category demand sorting queue, and determine the adjustment direction of the replenishment trigger threshold according to the demand trend direction identifier. If the demand trend direction identifier is a demand growth trend, the replenishment trigger threshold is adjusted upward, and if the demand trend direction identifier is a demand decline trend, the replenishment trigger threshold is adjusted downward.

[0075] Extract the supply trend direction identifier of each item category in the item category supply sorting queue, and determine the adjustment range of the replenishment trigger threshold according to the supply trend direction identifier. The replenishment trigger threshold with the supply trend direction identifier indicating a supply decline trend is adjusted upward, and the replenishment trigger threshold with the supply trend direction identifier indicating a supply growth trend is adjusted upward.

[0076] The adjustment direction and adjustment range of the replenishment trigger threshold are fused together to generate the replenishment trigger threshold update amount for each item category.

[0077] The replenishment trigger threshold corresponding to each item category is superimposed with the replenishment trigger threshold update amount to generate an updated replenishment trigger threshold. A replenishment trigger threshold update instruction is then generated based on the updated replenishment trigger threshold. The replenishment trigger threshold update instruction includes the item category identification and the updated replenishment trigger threshold.

[0078] The storage location adjustment instruction and the replenishment trigger threshold update instruction are combined to form the inventory dynamic optimization instruction set.

[0079] In an embodiment of the invention, exemplarily, during the specific implementation of dynamic allocation based on predicted supply and demand trends of goods, the server first obtains a warehouse inventory location layout map from the warehouse management system. This layout map records multiple storage location identifiers, such as Cold Chain Zone 1, Cold Chain Zone 2, Ambient Temperature Zone 1, Ambient Temperature Zone 2, and Promotional Pre-positioning Zone, and records the location space capacity parameters and currently occupied capacity parameters for each location. For example, Cold Chain Zone 1 is close to the sorting and verification line, with a location space capacity of 1200 boxes and currently occupying 920 boxes; Cold Chain Zone 2 is farther from the outbound gate, with a location space capacity of 1800 boxes and currently occupying 1100 boxes; the Promotional Pre-positioning Zone has a capacity of 500 boxes and currently occupies 260 boxes. The server calculates the available space for each location based on the above capacity data and associates the available space with the turnover requirements of the item category, replenishment pressure, and outbound frequency.

[0080] The server then reads the demand trend direction indicator from the supply and demand trend prediction information and sorts the product categories according to the urgency of demand. For the ready-to-eat meals (C category), the server identifies its demand trend direction indicator as an increasing trend, and the live-streaming promotional orders are driving a continuous increase in outbound volume, so it is placed at the front of the demand ranking queue. For the frozen chicken breast (A category), the server identifies its weekend outbound volume as steadily increasing, so it is placed in the second tier. For the rice, flour, and cooking oil (B category), the server identifies its demand trend direction indicator as a decreasing trend, so it is placed at the back of the queue. The server also sorts the supply sufficiency according to the supply trend direction indicator, placing product categories with a decreasing supply trend before those with an increasing supply trend. For example, the ready-to-eat meals (C category) are placed in a high-supply-risk position because the main supplier's production line maintenance will lead to a decrease in supply over the next five days; the rice, flour, and cooking oil (B category) have stable supply and sufficient inventory, so they are placed in a lower priority position.

[0081] After forming the demand and supply sorting queues for each item category, the server performs location matching by combining the spatial capacity parameters and current occupied capacity parameters of each area. The server matches the pre-prepared meals (Category C), which has increasing demand, high outbound frequency, and tight supply, to the cold chain zone 1 and promotional pre-positioning area near the review and packaging line. It adjusts the frozen dumplings (Date), with slower turnover, from cold chain zone 1 to a later storage location in cold chain zone 2, and adjusts the rice, flour, and cooking oil (Bate) category, with declining demand, to ambient temperature zone 2. Based on this, the server generates a location mapping relationship and further generates storage location adjustment instructions. These instructions include the item category identifier to be adjusted, the original storage location identifier, the target storage location identifier, and the adjustment execution sequence, such as "Pre-prepared meals (Category C), adjusted from cold chain zone 2 to cold chain zone 1, execution sequence is the off-peak operating period from 22:00 to 23:30 on the same day," to minimize the impact of the adjustment process on order picking.

[0082] The server further extracts the demand trend direction identifier for each category in the product category demand ranking queue to determine the adjustment direction of the replenishment trigger threshold. For the pre-prepared meals (Category C) category, where demand is increasing, the server adjusts the replenishment trigger threshold upwards; for the rice, flour, and cooking oil (Category B) category, where demand is decreasing, the server adjusts the replenishment trigger threshold downwards. The server then extracts the supply trend direction identifier from the supply ranking queue to determine the adjustment magnitude. For the pre-prepared meals (Category C) category, which experiences both demand growth and supply decline, the server increases its threshold by a larger margin, for example, from the original safety stock of 300 boxes to 520 boxes; for the frozen chicken breast (Category A) category, where demand is increasing but supply is increasing simultaneously, the server only increases the threshold from 400 boxes to 450 boxes; for the rice, flour, and cooking oil (Category B) category, where demand is declining and supply is ample, the server decreases the threshold from 800 bags to 620 bags. After merging the adjustment direction and adjustment magnitude, the server generates the updated replenishment trigger threshold for each product category and adds it to the original replenishment trigger threshold to form the updated replenishment trigger threshold.

[0083] In this embodiment of the invention, the server calculates the replenishment trigger threshold update based on both the demand trend strength and the supply risk strength. For the i-th item category, let the original replenishment trigger threshold be T0(i), the updated replenishment trigger threshold be T1(i), the demand trend strength parameter be D(i), and the supply risk strength parameter be G(i), then: T1_raw(i) = T0(i) × [1 + a × D(i) + b × G(i)]; where D(i) is determined based on the demand trend score Sout_final(i), which is positive when demand increases, negative when demand decreases, and 0 when demand is stable; G(i) is determined based on the supply trend score Sin_final(i), which is positive when supply decreases, negative when supply increases, and 0 when supply is stable. Preferably, both D(i) and G(i) are limited to between -1 and +1. 'a' represents the demand adjustment coefficient, preferably between 0.2 and 0.6; 'b' represents the supply risk adjustment coefficient, preferably between 0.1 and 0.5. Further, the server sets upper and lower limits for the updated replenishment trigger threshold: T1(i) = min{Tmax(i), max[Tmin(i), T1_raw(i)]}; where Tmin(i) is the minimum safety stock threshold for this item category, and Tmax(i) is the maximum allowed stock threshold for this item category in the corresponding storage location. If the calculated T1_raw(i) is less than Tmin(i), then Tmin(i) is taken as the updated replenishment trigger threshold; if T1_raw(i) is greater than Tmax(i), then Tmax(i) is taken as the updated replenishment trigger threshold. For example, the original replenishment trigger threshold T0 for the pre-prepared food category C is 300 boxes, the demand growth intensity D is 0.8, the supply risk intensity G is 0.6, the demand adjustment coefficient a is 0.5, and the supply risk adjustment coefficient b is 0.4. Then: T1_raw = 300 × [1 + 0.5 × 0.8 + 0.4 × 0.6] = 492 boxes; if the safety redundancy of 28 boxes in the pre-sale promotion area is considered, then the server will set the updated replenishment trigger threshold to 520 boxes. For location matching, the server calculates a location matching score for each item category: L(i,j)=c1×A(j)+c2×F(i)+c3×D(i)-c4×Dist(j)-c5×Occ(j); where L(i,j) is the matching score between the i-th item category and the j-th storage location; A(j) is the normalized value of the remaining capacity of the j-th location; F(i) is the normalized value of the outbound frequency of the i-th item category; D(i) is the demand trend strength parameter; Dist(j) is the normalized value of the distance from the j-th location to the sorting line or outbound gate; Occ(j) is the current occupancy rate of the j-th location; c1 to c5 are preset weight coefficients, and c1+c2+c3+c4+c5=1.The server determines the location with the largest L(i,j) as the target storage location for that item category; when the remaining capacity of the target location is insufficient, it selects alternative locations in descending order of matching score.

[0084] Finally, the server generates a replenishment trigger threshold update instruction based on the updated replenishment trigger threshold. This instruction includes the item category target identification and the updated replenishment trigger threshold. The server combines the replenishment trigger threshold update instruction with the storage location adjustment instruction to form a set of dynamic inventory optimization instructions, which are then sent to the warehouse management system, warehouse operation terminals, and replenishment task queue. This enables the warehouse system to complete storage location rearrangement, picking route optimization, and replenishment strategy updates based on supply and demand trend predictions.

[0085] In this embodiment of the invention, the method further includes:

[0086] The inbound and outbound time markers are mixed and arranged in chronological order to generate an inventory turnover event timeline.

[0087] Based on the inventory turnover event timeline, calculate the item inventory retention time parameter and item inventory coverage time parameter corresponding to the same item category target identification;

[0088] The average inventory retention time is obtained by aggregating the inventory retention time parameters of the same item category target identification, and the average inventory coverage time is obtained by aggregating the inventory coverage time parameters of the same item category target identification.

[0089] The average inventory retention time is compared with the upper limit threshold of retention time to generate an inventory backlog indicator; the average inventory coverage time is compared with the lower limit threshold of coverage time to generate an inventory shortage indicator.

[0090] Adjust the procurement plan parameters according to the supply trend direction indicator of the item category marked with the inventory backlog indicator, and adjust the safety stock baseline parameters according to the demand trend direction indicator of the item category marked with the inventory shortage indicator.

[0091] The adjusted procurement plan parameters and the adjusted safety stock baseline parameters are combined to generate a set of inventory strategy parameter update instructions.

[0092] In this embodiment of the invention, the method further includes:

[0093] Construct an inbound item category association set based on the inbound item category target identification, and construct an outbound item category association set based on the outbound item category target identification;

[0094] In the set of associated categories of inbound items, the matching of associated item categories is determined based on the frequency of common occurrence within the same inbound time window;

[0095] In the set of associated outbound item categories, the matching of outbound associated item categories is determined based on the frequency of common occurrence within the same outbound time window;

[0096] A consistency comparison is performed on the supply trend direction identifier in the matching of the inbound related item categories to generate an inbound supply trend association direction identifier, and the supply trend direction identifier is corrected based on the inbound supply trend association direction identifier.

[0097] A consistency comparison is performed on the demand trend direction identifier in the matching of outbound related item categories to generate an outbound demand trend association direction identifier, and the demand trend direction identifier is corrected based on the outbound demand trend association direction identifier.

[0098] The replenishment trigger threshold adjustment direction is redefined based on the revised supply trend direction indicator, and the storage location matching priority is redefined based on the revised demand trend direction indicator.

[0099] Based on the redefined replenishment trigger threshold adjustment direction, a collaborative adjustment instruction for replenishment of related item categories is generated; based on the redefined storage location matching priority, a collaborative adjustment instruction for location of related item categories is generated.

[0100] The related item category replenishment coordination adjustment instruction and the related item category location coordination adjustment instruction are incorporated into the inventory dynamic optimization instruction set.

[0101] In this embodiment of the invention, the method further includes:

[0102] Extract the inbound flow directed edges connected to each item category node in the directed network of inventory item flow, and trace the related item category nodes that have indirect inbound association with the item category node along the direction of the inbound flow directed edges to form a set of inbound flow reachable nodes.

[0103] Extract the outbound flow directed edges connected to each item category node in the directed network of inventory item flow, and trace the related item category nodes that have indirect outbound association with the item category node along the direction of the outbound flow directed edges to form a set of outbound flow reachable nodes;

[0104] The edge weight parameter of the directed edge of the inbound flow of each associated item category node in the set of reachable nodes for inbound flow is subjected to attenuation accumulation processing. The attenuation coefficient of the attenuation accumulation processing decreases as the number of tracking steps increases, thereby obtaining the inbound indirect influence weight of the item category node.

[0105] The edge weight parameters of the directed edge of the outbound flow of each associated item category node in the set of reachable outbound flow nodes are attenuated and accumulated. The attenuation coefficient of the attenuation accumulation process decreases as the number of tracking steps increases, so as to obtain the outbound indirect influence weight of the item category node.

[0106] The indirect impact weight of warehousing is used to correct the total warehousing flow parameter of the item category node. The indirect impact weight of warehousing and the total warehousing flow parameter of the item category are weighted and fused to generate the corrected total warehousing flow parameter of the item category.

[0107] Based on the indirect impact weight of outbound shipment, the total outbound flow parameter of the item category node is corrected. The indirect impact weight of outbound shipment and the total outbound flow parameter of the item category are weighted and fused to generate the corrected total outbound flow parameter of the item category.

[0108] The steps of generating the inbound growth trend marker and the inbound decline trend marker are re-executed using the corrected item category inbound total flow parameters to obtain the corrected supply trend direction identifier.

[0109] The steps of generating the outbound growth trend marker and the outbound decline trend marker are re-executed using the corrected item category outbound total flow parameters to obtain the corrected demand trend direction identifier;

[0110] The supply and demand trend prediction information of the item is updated based on the corrected supply trend direction indicator and the corrected demand trend direction indicator to generate updated supply and demand trend prediction information of the item.

[0111] The set of dynamic inventory optimization instructions is regenerated based on the updated supply and demand trend prediction information.

[0112] In this embodiment of the invention, exemplaryly, in the specific implementation of the indirect circulation impact correction process, the server uses the already generated directed network of inventory item circulation as the processing object, extracting the inbound circulation directed edges connected to each item category node one by one. Taking pre-cooked dishes - category C as an example, the server continues to trace along the direction of the inbound circulation directed edges, identifying related item category nodes such as frozen seasoning packets - category E and packaging materials - category F in its preceding inbound path, and forming these nodes into an inbound circulation reachable node set. This set is used to indicate that the supply change of pre-cooked dishes - category C is not only determined by its own inbound quantity, but also affected by indirect supply nodes such as matching seasoning packets and packaging materials. The server also extracts the outbound circulation directed edges connected to pre-cooked dishes - category C, traces the related nodes of subsequent joint consumption or combined outbound along the outbound circulation direction, identifies that frozen dumplings - category D and frozen chicken breast - category A have a continuous outbound relationship with pre-cooked dishes - category C in live promotion orders, and forms an outbound circulation reachable node set.

[0113] The server then performs attenuation and accumulation processing on the edge weight parameters of the set of reachable nodes for inbound flow. For the frozen seasoning packet-E category, which is one step adjacent to the ready-to-eat meals-C category, the server uses a higher attenuation coefficient to calculate its inbound edge weight; for the packaging materials-F category, which is two steps away, the server uses a lower attenuation coefficient to calculate its inbound edge weight, so that the further the tracking distance, the weaker the influence of the node. For example, if the frozen seasoning packet-E category has received 300 boxes in the past three days and the packaging materials-F category has received 600 pieces in the past three days, the server accumulates them according to different attenuation coefficients to obtain the inbound indirect influence weight of the ready-to-eat meals-C category. The server also performs attenuation and accumulation processing on the set of reachable nodes for outbound flow, and includes the outbound edge weights of related nodes such as frozen dumplings-D category and frozen chicken breast-A category in descending order of tracking steps to obtain the outbound indirect influence weight of the ready-to-eat meals-C category.

[0114] In this embodiment of the invention, the server determines the attenuation coefficient of the indirect influence weight based on the number of tracking steps. For the i-th item category node, the set of nodes reachable at the k-th step obtained by the server tracing along the directed edge of the inbound flow is denoted as Vin(i,k), where k is the number of tracking steps starting from the current item category node. The server calculates the inbound indirect influence weight according to the following formula: Iin(i)=Σ[k=1 to K](λ_in k×Σ[v∈Vin(i,k)]C_in(i,v)×S_in(v)×Qin(v)); where Iin(i) is the indirect impact weight of the ith item category; K is the maximum number of tracking steps, preferably 2 to 4; λ_in is the inbound attenuation coefficient, ranging from 0.3 to 0.8; C_in(i,v) is the confidence level of the inbound association between reachable node v and current node i, ranging from 0 to 1; S_in(v) is the supply influence direction of reachable node v, taking +1 when supply increases, -1 when supply decreases, and +1 when no clear direction is identified; Qin(v) is the total inbound flow parameter of reachable node v within the corresponding time window. The outbound indirect impact weight is calculated according to the following formula: Iout(i)=Σ[k=1 to K](λ_out k ×Σ[v∈Vout(i,k)]C_out(i,v)×S_out(v)×Qout(v)); where Iout(i) is the indirect impact weight of the outbound shipment for the i-th item category; Vout(i,k) is the set of reachable nodes at the k-th step obtained by tracing the directed edge along the outbound flow; λ_out is the outbound attenuation coefficient; C_out(i,v) is the outbound association confidence; S_out(v) is the demand influence direction of reachable node v, which is +1 when combined with outbound enhancement or demand enhancement, -1 when demand weakens, and +1 when no clear direction is identified; Qout(v) is the total outbound flow parameter of reachable node v within the corresponding time window. The server adjusts the total turnover of each item category according to the following formulas: Qin_adj(i) = max{0, Qin(i) + μ_in × Iin(i)}; Qout_adj(i) = max{0, Qout(i) + μ_out × Iout(i)}; where Qin_adj(i) is the adjusted total inbound turnover parameter, and Qout_adj(i) is the adjusted total outbound turnover parameter; μ_in and μ_out are indirect influence fusion coefficients, preferably ranging from 0.1 to 0.5. When a supply decrease event occurs at a reachable node, S_in(v) can take a negative value to reduce the effective supply of the current item category; when a combined outbound enhancement relationship exists at a reachable node, S_out(v) can take a positive value to increase the effective demand of the current item category. When the same reachable node is repeatedly tracked via multiple paths, the server retains the path with the largest impact weight after decay, or sums the impact weights of multiple paths and limits them to a preset upper limit to avoid duplicate inclusion and distortion of trend strength.

[0115] After obtaining the indirect impact weight, the server weights the inbound indirect impact weights with the total inbound turnover parameters of the ready-to-eat meals-C category itself, generating a revised total inbound turnover parameter. For example, if the ready-to-eat meals-C category itself received 680 boxes in the past seven days, but the supply of its accompanying seasoning packets decreased, weakening its indirect supply capacity, the server will reduce its effective inbound turnover after correction. Simultaneously, the server weights the outbound indirect impact weights with the total outbound turnover parameters of the category itself, generating a revised total outbound turnover parameter. For example, if the ready-to-eat meals-C category itself outbound 1280 boxes in the past seven days, and the frequency of outbound shipments together with the frozen dumplings-D category continues to increase, the server will increase its effective outbound turnover after correction.

[0116] The server re-executes the comparison of adjacent time windows using the corrected total inbound turnover parameters, regenerates inbound growth trend markers or inbound decline trend markers, and obtains the corrected supply trend direction indicator accordingly. For the pre-prepared meals-C category, after considering the decrease in supply of supporting categories, the server corrects the original weak supply growth to tight supply. The server then re-executes the outbound trend marker generation step using the corrected total outbound turnover parameters to obtain the corrected demand trend direction indicator, further increasing the demand growth intensity of the pre-prepared meals-C category. Finally, the server updates the item supply and demand trend prediction information based on the corrected supply trend direction indicator and demand trend direction indicator, and regenerates the inventory dynamic optimization instruction set. It further raises the replenishment trigger threshold for the pre-prepared meals-C category, arranges it and the frozen dumplings-D category in adjacent cold chain picking areas, and includes the supporting seasoning packets-E category in the synchronous replenishment task, so that the inventory optimization results can reflect the combined impact of direct and indirect related turnover.

[0117] Please refer to the following: Figure 3 , Figure 3 This is a schematic diagram illustrating the adjustment of directed edge weights for inventory item flow based on external supply chain events, as provided in an embodiment of the present invention. In this embodiment, the method further includes:

[0118] Acquire external supply chain event data streams, which include upstream supplier capacity change notifications and downstream demand order fluctuation notifications;

[0119] Semantic parsing is performed on the capacity change notification information from the upstream supplier to extract the capacity change direction identifier and the target identifier of the item category affected by the capacity change. The capacity change direction identifier includes the direction of capacity increase and the direction of capacity decrease.

[0120] The order fluctuation notification information from the downstream demand side is semantically parsed to extract the order fluctuation direction identifier and the item category identifier affected by the order fluctuation. The order fluctuation direction identifier includes the direction of order volume increase and the direction of order volume decrease.

[0121] The target identifier of the item category affected by the capacity change is mapped to the item category node in the directed network of inventory item flow. The supply trend direction identifier of the mapped item category node is corrected according to the capacity change direction identifier. When the capacity change direction identifier is the capacity reduction direction, the supply trend direction identifier is adjusted to the supply decline trend direction.

[0122] The order fluctuation affects the item category identification and is mapped to the item category node in the directed network of inventory item circulation. The demand trend direction identification of the mapped item category node is corrected according to the order fluctuation direction identification. When the order fluctuation direction identification is the direction of order volume increase, the demand trend direction identification is adjusted to the direction of demand growth trend.

[0123] Based on the corrected supply trend direction identifier, update the trend prediction weight coefficient of the edge weight parameter of the directed edge of the inbound flow of the mapped item category node to obtain the directed edge of inbound flow after the supply chain event is corrected.

[0124] Based on the corrected demand trend direction identifier, update the trend prediction weight coefficient of the edge weight parameter of the outbound flow directed edge of the mapped item category node to obtain the outbound flow directed edge after supply chain event correction.

[0125] The directed network for the flow of inventory items is reconstructed using the directed edges for inbound and outbound flows corrected by the supply chain events.

[0126] The reconstructed directed network of inventory item flow is reprocessed to predict the flow trend, generating supply and demand trend prediction information for items that incorporates external supply chain events.

[0127] The set of dynamic inventory optimization instructions is regenerated based on the predicted supply and demand trends of goods derived from the integration of external supply chain events.

[0128] In this embodiment of the invention, exemplaryly, during the specific implementation of integrating external supply chain events, the server obtains external supply chain event data streams from the supplier collaboration system, order forecasting system, and customer order platform. These data streams include upstream supplier capacity change notifications and downstream demand-side order fluctuation notifications. For example, the supplier collaboration system pushes a notification stating that "the main supplier of pre-prepared dishes (Category C) will reduce its supply capacity by 30% over the next five days due to production line maintenance," the order forecasting system pushes a notification stating that "pre-prepared dishes (Category C) orders are expected to increase by 45% during the live-stream promotional event," and another supplier pushes a notification stating that "a new production line for rice, flour, and oil (Category B) will be put into operation, increasing supply capacity by 20% over the next week." The server performs semantic parsing on the capacity change notifications, extracting information such as "direction of capacity reduction," "pre-prepared dishes (Category C)," and "direction of capacity increase," "rice, flour, and oil (Category B)." The server also performs semantic parsing on the order fluctuation notifications, extracting information such as "direction of order volume increase," "pre-prepared dishes (Category C)," and converts the aforementioned event information into structured fields that can be recognized by the directed network for inventory flow.

[0129] The server then maps the item category identifiers affected by capacity changes to the corresponding item category nodes in the directed network of inventory item flow. For example, the server maps "Pre-prepared Vegetables - Category C" to the Pre-prepared Vegetables - Category C node, and adjusts the supply trend direction of this node towards the direction of supply decline based on the direction of capacity reduction; the server maps "Rice, Flour, Grains and Oils - Category B" to the Rice, Flour, Grains and Oils - Category B node, and increases its supply sufficiency based on the direction of capacity increase. The server also maps the item category identifiers affected by order fluctuations to the corresponding nodes. For the Pre-prepared Vegetables - Category C, where order volume is increasing, the server adjusts its demand trend direction towards the direction of demand growth. Thus, the server not only relies on historical inbound and outbound quantities to judge trends, but also incorporates future supply and order changes into the supply and demand forecasting process in advance.

[0130] After correcting the trend direction, the server updates the trend prediction weight coefficient of the directed edges for inbound flow connecting the corresponding item category nodes based on the corrected supply trend direction identifier. For example, a decrease in production capacity for the pre-prepared meals (Category C) reduces the effective supply weight of its directed edges for inbound flow, resulting in a corrected inbound flow edge for the server. Conversely, an increase in production capacity for the rice, flour, and oil (Category B) increases the supply weight of its directed edges for inbound flow. Simultaneously, the server updates the trend prediction weight coefficient of the directed edges for outbound flow based on the corrected demand trend direction identifier. For example, an increase in orders for pre-prepared meals (Category C) increases the demand weight of its directed edges for outbound flow, resulting in a corrected outbound flow edge for the server.

[0131] The server reconstructs the directed network of inventory item flow using the corrected inbound and outbound directed edges, and re-executes the flow trend prediction processing on the reconstructed network. For the pre-prepared vegetables (Category C), the server ultimately generates a supply and demand trend prediction information of "supply decline and demand growth increase"; for the rice, flour, and oil (Category B), the server generates a prediction information of "supply increase and demand decline or remain stable". Based on the supply and demand trend prediction information of items after incorporating external supply chain events, the server regenerates the set of dynamic inventory optimization instructions, raising the replenishment trigger threshold for pre-prepared vegetables (Category C) and adjusting the priority location to cold chain zone 1 or the pre-promotion zone, while reducing the procurement urgency and controlling the expansion of storage space for rice, flour, and oil (Category B), so that the warehouse inventory optimization results can respond in advance to changes in the external supply chain.

[0132] In this embodiment of the invention, the method further includes:

[0133] Extract the inbound quantity parameters carried by each inbound flow directed edge in the directed network of inventory item circulation, perform quantity jump detection processing on the inbound quantity parameters of adjacent inbound flow directed edges along the connection direction of the inbound flow directed edges, and calculate the quantity jump amplitude between the inbound quantity parameters of adjacent inbound flow directed edges.

[0134] Extract the outbound quantity parameters carried by each outbound directed edge in the directed network of inventory flow, perform quantity jump detection processing on the outbound quantity parameters of adjacent outbound directed edges along the connection direction of the outbound directed edges, and calculate the quantity jump amplitude between the outbound quantity parameters of adjacent outbound directed edges.

[0135] The magnitude of the quantity jump between the inbound quantity parameters is compared with the preset inbound jump magnitude threshold. Directed edges of inbound flow with a quantity jump magnitude exceeding the inbound jump magnitude threshold are marked as abnormal inbound jump edges.

[0136] The magnitude of quantity jumps between outbound quantity parameters is compared with a preset outbound jump magnitude threshold. Directed edges of outbound flow with quantity jump magnitudes exceeding the outbound jump magnitude threshold are marked as outbound abnormal jump edges. The item category nodes connected to the abnormal jump edges of inbound flow are extracted, and the continuous change sequence of the total inbound flow parameter of the item category corresponding to the item category node along the time axis is obtained. The mutation point location processing is performed on the continuous change sequence to determine the inbound mutation time point within the time window corresponding to the abnormal jump edge of inbound flow.

[0137] Extract the item category nodes connected to the abnormal jump edge of the outbound process, obtain the continuous change sequence of the total outbound flow parameter of the item category corresponding to the item category node along the time axis, perform mutation point location processing on the continuous change sequence, and determine the outbound mutation time point within the time window corresponding to the abnormal jump edge of the outbound process.

[0138] Based on the entry mutation time point, query the external supply chain event data stream, retrieve the upstream supplier capacity change notification information within a preset time range before and after the entry mutation time point, and match the capacity change direction identifier of the capacity change notification information with the quantity jump direction of the entry abnormal jump edge.

[0139] Based on the outbound change time point, query the external supply chain event data stream, retrieve downstream demand order fluctuation notification information within a preset time range before and after the outbound change time point, and match the order fluctuation direction identifier of the order fluctuation notification information with the quantity jump direction of the outbound abnormal jump edge.

[0140] When the capacity change direction identifier is successfully matched with the quantity change direction of the abnormal jump edge of the inbound, the capacity change notification information is established as a causal relationship with the abnormal jump edge of the inbound. When the order fluctuation direction identifier is successfully matched with the quantity jump direction of the abnormal jump edge of the outbound, the order fluctuation notification information is established as a causal relationship with the abnormal jump edge of the outbound.

[0141] Based on the established causal relationship, the trend strength parameter of the supply trend direction indicator of the item category node corresponding to the abnormal jump edge of the inbound is corrected, and the trend strength parameter of the demand trend direction indicator of the item category node corresponding to the abnormal jump edge of the outbound is corrected.

[0142] The supply trend direction indicator and demand trend direction indicator after correcting the trend strength parameter are reintegrated into the commodity supply and demand trend prediction information to generate commodity supply and demand trend prediction information after jump causality correction.

[0143] The set of dynamic inventory optimization instructions is regenerated based on the predicted supply and demand trends of goods after the jump causality correction.

[0144] In this embodiment of the invention, exemplaryly, during the specific implementation of quantity jump detection and causal correction, the server first extracts the quantity parameters carried by each directed edge of the inventory flow network, and performs quantity jump detection on the quantity parameters of adjacent directed edges of the inventory flow along the connection direction of the directed edges of the inventory flow. For example, if 100 boxes of pre-prepared vegetables (Category C) were received on May 6th and 40 boxes were received on May 7th, the server calculates the jump range between adjacent quantities as 60 boxes; if 120 boxes of frozen chicken breast (Category A) were received on May 8th and 125 boxes were received on May 9th, the server calculates the jump range as 5 boxes. Simultaneously, the server extracts the quantity parameters carried by each directed edge of the outbound flow and performs jump detection on adjacent directed edges of the outbound flow along the outbound flow direction. For example, 120 boxes of pre-cooked meals (Category C) were shipped out on May 6th, and 260 boxes were shipped out on May 7th. The server calculated that the jump between the adjacent shipment quantities was 140 boxes.

[0145] The server compares the magnitude of changes in incoming quantities with preset thresholds. For the pre-prepared meals (Category C), if the incoming quantity drops from 100 boxes to 40 boxes, exceeding the preset threshold of 50 boxes, the server marks the corresponding directed edge of the incoming flow as an abnormal incoming quantity change edge. For the frozen chicken breast (Category A), the incoming quantity change is only 5 boxes, so the server does not mark it as abnormal. Similarly, the server compares the magnitude of changes in outgoing quantities with preset thresholds. If the outgoing quantity for pre-prepared meals (Category C) increases from 120 boxes to 260 boxes, exceeding the preset threshold of 80 boxes, the server marks the corresponding directed edge of the outgoing flow as an abnormal outgoing quantity change edge.

[0146] In this embodiment of the invention, the server performs quantity jump detection based on both the absolute change and the relative change rate of the weight parameters of adjacent edges. For the inbound quantity parameters Qin(t-1) and Qin(t) of two adjacent time windows, the server calculates the inbound quantity jump magnitude: ΔQin(t)=|Qin(t)-Qin(t-1)|; and calculates the inbound relative jump rate: Jin(t)=|Qin(t)-Qin(t-1)| / max(Qin(t-1),ε). When ΔQin(t) is greater than the preset absolute inbound jump threshold and Jin(t) is greater than the preset relative inbound jump threshold, the server marks the corresponding inbound flow directed edge as an abnormal inbound jump edge. For the outbound quantity parameters Qout(t-1) and Qout(t) of two adjacent time windows, the server calculates the outbound quantity jump magnitude: ΔQout(t)=|Qout(t)-Qout(t-1)|; and calculates the outbound relative jump rate: Jout(t)=|Qout(t)-Qout(t-1)| / max(Qout(t-1),ε). When ΔQout(t) is greater than the preset absolute outbound jump threshold and Jout(t) is greater than the preset relative outbound jump threshold, the server marks the corresponding outbound flow directed edge as an abnormal outbound jump edge. Preferably, the absolute threshold for inbound and outbound changes are determined based on the historical standard deviation of the corresponding item categories: Th_in_abs(i) = m × σ_in(i), Th_out_abs(i) = m × σ_out(i); where σ_in(i) is the standard deviation of the inbound turnover of the i-th item category within M historical time windows, and σ_out(i) is the standard deviation of the outbound turnover of the i-th item category within M historical time windows, with m ranging from 1.5 to 3. The relative threshold for inbound and outbound changes is preferably 20% to 50%. When establishing causal relationships, the server must simultaneously meet the following conditions: First, the item category identifier affected by the external supply chain event must be consistent with the item category node connected by the abnormal jump edge, or both must belong to the same item category in the preset item category alias table; Second, the effective time of the external supply chain event must fall within the time range ΔT before and after the abnormal jump time point, with ΔT preferably being 1 to 7 days; Third, the event direction must be consistent with the quantity jump direction, i.e., a decrease in production capacity corresponds to a decrease in the number of items entering the warehouse, an increase in production capacity corresponds to an increase in the number of items entering the warehouse, an increase in order volume corresponds to an increase in the number of items leaving the warehouse, and a decrease in order volume corresponds to a decrease in the number of items leaving the warehouse; Fourth, the semantic parsing confidence level must not be lower than the preset confidence threshold, preferably not lower than 0.6.When all the above conditions are met, the server establishes a causal relationship and corrects the trend strength parameter according to the following formula: Strength_adj(i) = Strength_base(i) × [1 + κ × MatchScore]; where Strength_base(i) is the trend strength parameter before correction, Strength_adj(i) is the trend strength parameter after correction, κ is the causal correction coefficient, preferably ranging from 0.1 to 0.5, and MatchScore is the causal matching score. MatchScore is obtained by weighting the item category matching score, time proximity score, directional consistency score, and semantic parsing confidence: MatchScore = p1 × Score_item + p2 × Score_time + p3 × Score_dir + p4 × Conf; where p1 + p2 + p3 + p4 = 1, and the values ​​of Score_item, Score_time, Score_dir, and Conf are all in the range of 0 to 1. If the directions are inconsistent, Score_dir is set to 0, and the server does not establish a causal relationship; if the directions are consistent and other causal relationship conditions are met, Score_dir is set to 1.

[0147] After marking the abnormal transition edges, the server extracts the item category nodes connected to the abnormal transition edges and obtains the continuous change sequence of the total inbound flow parameter of the corresponding item category along the time axis. The server performs mutation point localization processing on this continuous change sequence, pinpointing the time window of a significant decrease in the inbound volume of the pre-prepared food-C category to the morning of May 7th, and determining the inbound mutation time point within this time window. Similarly, the server extracts the item category nodes connected to the abnormal transition edges of outbound flow, obtains the continuous change sequence of their outbound flow parameter along the time axis, and pinpoints the time window of a significant increase in the outbound volume of the pre-prepared food-C category to the evening of May 7th, determining the outbound mutation time point.

[0148] The server queries external supply chain event data streams based on the inbound abrupt change time point, retrieving upstream supplier capacity change notifications within a preset time range before and after that time point. The server retrieves "Pre-prepared Food - Category C main supplier's production line underwent maintenance, resulting in a 30% decrease in supply capacity starting May 7th," and matches the direction of capacity reduction with the direction of quantity decrease in the inbound abrupt change edge. Upon successful matching, a causal relationship is established between the capacity change notification information and the inbound abrupt change edge. The server then queries downstream demand-side order fluctuation notifications based on the outbound abrupt change time point, retrieving "The live-stream promotional activity on the evening of May 7th led to a 45% increase in pre-prepared food - Category C orders," and matches the direction of the order increase with the direction of the quantity increase in the outbound abrupt change edge. Upon successful matching, a causal relationship is established between the order fluctuation notification information and the outbound abrupt change edge.

[0149] Based on the aforementioned causal relationships, the server corrects the trend strength parameters of the item category nodes corresponding to the abnormal jump edges. For the pre-prepared food-C category, the abnormal jump in inventory is causally related to the reduction in supplier capacity, so the server increases the strength of its supply decline trend; the abnormal jump in inventory is causally related to the increase in live-stream promotional orders, so the server increases the strength of its demand growth trend. The server reintegrates the corrected supply trend direction identifier and demand trend direction identifier into the item supply and demand trend prediction information, generating item supply and demand trend prediction information after jump causality correction. Finally, the server regenerates the set of dynamic inventory optimization instructions based on this information, raising the replenishment trigger threshold for the pre-prepared food-C category to a higher level, adjusting it to the cold chain zone 1 and the pre-promotion zone, and simultaneously generating supplier alternative replenishment reminders and peak picking priority adjustment instructions.

[0150] In this embodiment of the invention, the method further includes:

[0151] Extract the change sequence of the edge weight parameters of the inbound flow directed edges of each item category node in the directed network of inventory item circulation along the time axis, and convert the fluctuation characteristics of the edge weight parameters along the time axis into a description of the fluctuation frequency of the inbound flow edge weights.

[0152] Extract the change sequence of the edge weight parameters of the outbound flow directed edges of each item category node in the directed network of inventory item circulation along the time axis, and convert the fluctuation characteristics of the edge weight parameters along the time axis into a description of the fluctuation frequency of the outbound flow edge weights.

[0153] The frequency consistency comparison process is performed between the description of the fluctuation frequency of the edge weight of the inbound flow and the inbound periodic fluctuation feature. The inbound fluctuation frequency component that is consistent with the frequency of the inbound periodic fluctuation feature is identified in the fluctuation of the edge weight parameter of the directed edge of the inbound flow. The inbound fluctuation frequency component with consistent frequency is marked as the inbound habitual fluctuation component, and the inbound fluctuation frequency component with inconsistent frequency is marked as the inbound abnormal fluctuation component.

[0154] The frequency of outbound flow edge weight fluctuation is compared with the frequency consistency of outbound periodic fluctuation characteristics. The outbound fluctuation frequency components that are consistent with the frequency of outbound periodic fluctuation characteristics in the fluctuation of edge weight parameters of the outbound flow directed edge are identified. The outbound fluctuation frequency components with consistent frequencies are marked as outbound habitual fluctuation components, and the outbound fluctuation frequency components with inconsistent frequencies are marked as outbound abnormal fluctuation components.

[0155] The occurrence time of the abnormal fluctuation components of the inbound goods is located along the time axis. The start time, duration and fluctuation amplitude parameters of the abnormal fluctuation components of the inbound goods are extracted to form an inbound goods abnormal event descriptor.

[0156] The occurrence time of the outbound anomaly fluctuation component is located along the time axis, and the start time, duration and fluctuation amplitude parameters of the outbound anomaly fluctuation component are extracted to form an outbound anomaly event descriptor;

[0157] Based on the inbound anomaly event descriptor, query the external supply chain event data stream, retrieve upstream supplier capacity change notification information within the time range covered by the inbound anomaly event descriptor, and perform event feature matching processing between the capacity change notification information and the inbound anomaly event descriptor;

[0158] Based on the outbound anomaly event descriptor, query the external supply chain event data stream, retrieve downstream demand order fluctuation notification information within the time range covered by the outbound anomaly event descriptor, and perform event feature matching processing between the order fluctuation notification information and the outbound anomaly event descriptor;

[0159] When the capacity change notification information matches the descriptor of the inbound anomaly event, the inbound anomaly fluctuation component is separated from the inbound flow edge weight fluctuation frequency description, and the edge weight parameter change sequence of the inbound flow directed edge is reconstructed using the inbound habitual fluctuation component. When the match fails, the inbound anomaly fluctuation component is marked as an unexplained inbound anomaly event.

[0160] When the order fluctuation notification information matches the outbound anomaly event descriptor, the outbound anomaly fluctuation component is separated from the outbound flow edge weight fluctuation frequency description, and the outbound flow directed edge edge weight parameter change sequence is reconstructed using the outbound habitual fluctuation component. When the match fails, the outbound anomaly fluctuation component is marked as an unexplained outbound anomaly event.

[0161] The supply trend direction identifier of the item category node is regenerated based on the change sequence of edge weight parameters of the reconstructed inbound flow directed edges, and the demand trend direction identifier of the item category node is regenerated based on the change sequence of edge weight parameters of the reconstructed outbound flow directed edges.

[0162] The reconstructed supply trend direction identifier and demand trend direction identifier are reintegrated into the commodity supply and demand trend prediction information to generate commodity supply and demand trend prediction information after removing the influence of external events.

[0163] The set of dynamic inventory optimization instructions is regenerated based on the predicted supply and demand trends of goods after the impact of external events has been removed.

[0164] In an embodiment of the invention, exemplaryly, in the specific implementation process of stripping away the influence of external events, the server first extracts the inbound directed edges connected to each item category node in the directed network of inventory flow, and reads the change sequence of the edge weight parameters of each inbound directed edge along the time axis. For example, frozen chicken breast-A category shows 120 boxes inbound on Wednesday, 180 boxes on Friday, and 60 boxes on Sunday over four consecutive weeks. The server converts this fluctuation into an inbound flow edge weight fluctuation frequency description to express its fixed replenishment rhythm. Simultaneously, the server extracts the outbound directed edges connected to each item category node, and converts the change sequence of the outbound quantity parameters along the time axis into an outbound flow edge weight fluctuation frequency description. For example, ready-to-eat meals-C category has 90 to 120 boxes outbound on ordinary weekdays, with a significant increase on weekends and promotional days. The server converts its outbound fluctuations into corresponding fluctuation frequency descriptions.

[0165] The server compares the frequency description of the inbound flow edge weight fluctuation with the aforementioned inbound periodic fluctuation characteristics for frequency consistency. For the frozen chicken breast-A category, the Friday inbound peak and Sunday inbound trough occurred repeatedly in multiple periods, and the server marked this frequency-consistent fluctuation component as the inbound habitual fluctuation component. For the ready-to-eat meals-C category, the inbound volume suddenly dropped from 100 boxes to 40 boxes on May 7, and this change did not conform to its normal replenishment cycle, so the server marked it as the inbound abnormal fluctuation component. Similarly, the server compares the frequency description of the outbound flow edge weight fluctuation with the outbound periodic fluctuation characteristics, marking the stable weekend outbound peak as the outbound habitual fluctuation component, and the abnormal peak of outbound volume jumping from 120 boxes to 260 boxes on the night of the live promotion as the outbound abnormal fluctuation component.

[0166] In this embodiment of the invention, the server represents the change sequence of the edge weight parameters along the time axis as a time series X(t), and generates a fluctuation frequency description by combining moving average decomposition and frequency component identification. Specifically, the server first performs mean-removal processing on X(t) to obtain a centered sequence: Xc(t) = X(t) - mean(X); the server then performs a discrete Fourier transform on Xc(t) to obtain the amplitude spectrum A(f) corresponding to different frequencies. For inventory turnover data with a sampling period of 1 day, the server maps the frequency f to the period P = 1 / f, and extracts frequencies with amplitudes greater than the amplitude threshold Th_amp as candidate fluctuation frequency components. Th_amp can be the sum of the mean and standard deviation of all amplitudes, or 20% to 40% of the largest amplitude. The server performs a consistency comparison between the candidate fluctuation frequency components and the periodic fluctuation characteristics of inbound or outbound inventory. When the period P corresponding to a candidate frequency meets the following condition with the preset period P0, the server marks the candidate frequency component as a normal fluctuation component: |P-P0| / P0≤δ; where δ is the period consistency tolerance, preferably between 0.1 and 0.25. For example, if the preset period is 7 days, and the period corresponding to the candidate frequency is between 5.6 and 8.4 days, the server considers it to be consistent with the period fluctuation. For frequency components that do not meet the period consistency condition and whose amplitude is greater than the abnormal amplitude threshold, the server marks them as abnormal fluctuation components. The server further locates the start time, duration, and fluctuation amplitude of the abnormal fluctuation component on the time axis. The start time is the time point when X(t) first deviates from the normal reconstruction sequence by more than Th_dev, the duration is the length of time that it continuously exceeds Th_dev, and the fluctuation amplitude is the maximum difference between X(t) and the normal reconstruction sequence within the abnormal time period. When the server confirms that the abnormal fluctuation component matches the capacity change notification information or order fluctuation notification information successfully based on the external supply chain event data stream, the server removes the abnormal fluctuation component from the original edge weight parameter change sequence. The stripping method is as follows: The habitual frequency components consistent with the periodic fluctuation characteristics are retained, while the abnormal frequency components that have matched external events are filtered out. An inverse Fourier transform is then performed to obtain the reconstructed edge weight parameter change sequence Xrec(t). In another embodiment, when the number of time series samples is insufficient for frequency domain transformation, the server uses the moving median method to construct the habitual sequence: Xbase(t) = median{X(tr),...,X(t+r)}; where r is the sliding window radius, preferably 2 to 7. The server treats the portion of X(t) - Xbase(t) exceeding the deviation threshold as the abnormal fluctuation component; if this abnormal fluctuation component successfully matches an external event, Xbase(t) replaces X(t) within the abnormal interval, resulting in the reconstructed edge weight parameter change sequence.The server recalculates the supply trend direction indicator based on the reconstructed sequence of changes in the weight parameters of the inbound side, and recalculates the demand trend direction indicator based on the reconstructed sequence of changes in the weight parameters of the outbound side, thereby avoiding misjudgment of long-term inventory trends caused by short-term anomalies that have been interpreted by external events.

[0167] After identifying the abnormal components, the server locates the occurrence time of the inbound abnormal fluctuation components along the timeline, extracts their start time, duration, and fluctuation amplitude parameters, and constructs an inbound abnormal event descriptor. For example, the inbound abnormal event descriptor for the pre-prepared food-C category records "starting time is May 7th, 08:00, lasting for two days, with a decrease of approximately 60 to 80 boxes." The server performs the same processing on the outbound abnormal fluctuation components, forming an outbound abnormal event descriptor, such as "starting at 20:00 on May 7th, lasting overnight until the next morning, with an increase of approximately 140 boxes in outbound shipments." Subsequently, the server queries the external supply chain event data stream based on the inbound abnormal event descriptor, retrieves upstream supplier capacity change notification information within this time range, and identifies "the main supplier of pre-prepared food-C category is undergoing production line maintenance, resulting in a 30% decrease in supply capacity." The server performs event feature matching with the inbound abnormal event descriptor for the direction of capacity reduction, the affected categories, and the duration of the notification, confirming a successful match.

[0168] The server queries downstream demand-side order fluctuation notification information based on the outbound anomaly event descriptor and finds "Live-stream promotion caused a 45% increase in pre-prepared food - Category C order volume". The server matches the direction of the order volume increase, the affected category, and the promotion time range with the outbound anomaly event descriptor, confirming that the outbound anomaly was triggered by an external promotional event. For successfully matched inbound anomaly fluctuation components, the server removes them from the inbound flow edge weight fluctuation frequency description and reconstructs the edge weight parameter change sequence of the inbound flow directed edge using the inbound habitual fluctuation components; for inbound anomalies that do not match supply chain events, the server marks them as unexplained inbound anomaly events and reserves them for the anomaly monitoring module. The server performs the same processing on outbound anomalies: successfully matched promotional outbound peaks are removed, and unmatched outbound anomalies are marked as unexplained outbound anomaly events.

[0169] Finally, based on the reconstructed sequence of changes in the weight parameters of the directed edges in the inbound flow, the server regenerates the supply trend direction identifier for the item category node; based on the reconstructed sequence of changes in the weight parameters of the directed edges in the outbound flow, it regenerates the demand trend direction identifier. For example, after removing the peak of live-stream promotions, the long-term demand for the pre-prepared meals-C category is still growing but at a reduced intensity; the server corrects its demand trend from strong growth to stable growth. After removing the impact of supplier maintenance, the server corrects its long-term supply trend from significant decline to short-term tightness. The server reintegrates the reconstructed supply trend direction identifiers and demand trend direction identifiers into the item supply and demand trend prediction information, generating item supply and demand trend prediction information after removing the impact of external events. Based on this, it regenerates the set of dynamic inventory optimization instructions, ensuring that replenishment thresholds, warehouse location adjustments, and procurement plans can respond to external events without being misled by one-off abnormal fluctuations in the long term.

[0170] In this embodiment of the invention, the method further includes:

[0171] Extract the storage location identifier corresponding to each item category node in the directed network of inventory item circulation, and calculate the remaining space capacity parameter of the storage location identifier based on the location space capacity parameter and the current occupied capacity parameter.

[0172] The total inbound flow parameter of the item category node is mapped to the remaining space capacity parameter of the storage location identifier corresponding to the item category node to generate the inbound capacity saturation parameter of the storage location identifier.

[0173] The total outbound flow parameter of the item category node is mapped to the remaining space capacity parameter of the storage location identifier corresponding to the item category node to generate the outbound capacity saturation parameter of the storage location identifier.

[0174] Extract item category nodes whose inbound capacity saturation parameter exceeds the upper limit of inbound capacity, and mark these item category nodes as inbound capacity limit item categories. Extract item category nodes whose outbound capacity saturation parameter exceeds the upper limit of outbound capacity, and mark these item category nodes as outbound capacity limit item categories.

[0175] Extract the supply trend direction identifier of the category of goods with limited storage capacity. When the supply trend direction identifier is a supply growth trend, search for adjacent item category nodes that have storage association with the category of goods with limited storage capacity along the direction of the storage flow, and obtain the remaining space capacity parameters of the adjacent item category nodes.

[0176] The excess inbound demand of the inbound capacity of the item category is transmitted to the adjacent item category node in the opposite direction of the inbound flow direction. The inbound quantity parameter corresponding to the excess inbound demand is allocated to the remaining space capacity parameter of the adjacent item category node to generate an inbound demand transmission scheme.

[0177] Extract the demand trend direction identifier of the category of goods with limited warehouse capacity. When the demand trend direction identifier is a demand growth trend, search for adjacent item category nodes that are associated with the outbound flow along the direction of the outbound flow direction, and obtain the remaining space capacity parameters of the adjacent item category nodes.

[0178] The excess outbound demand of the outbound capacity limit item category is transmitted to the adjacent item category node in the opposite direction of the outbound flow directed edge. The outbound quantity parameter corresponding to the excess outbound demand is allocated to the remaining space capacity parameter of the adjacent item category node to generate an outbound demand transmission scheme.

[0179] Based on the inbound demand transmission plan, adjust the replenishment trigger threshold update instructions and storage location adjustment instructions for inbound goods categories with limited capacity and adjacent goods category nodes; based on the outbound demand transmission plan, adjust the replenishment trigger threshold update instructions and storage location adjustment instructions for outbound goods categories with limited capacity and adjacent goods category nodes.

[0180] The adjusted replenishment trigger threshold update instruction and the adjusted storage location adjustment instruction will be incorporated into the inventory dynamic optimization instruction set.

[0181] In this embodiment of the invention, exemplaryly, during the specific implementation of capacity saturation transmission, the server first extracts the storage location identifier corresponding to each item category node from the directed network of inventory item flow, and reads the location space capacity parameter and the currently occupied capacity parameter of that location, and calculates the remaining space capacity parameter. For example, pre-cooked dishes - category C are currently stored in cold chain zone 1, which has a total capacity of 1200 boxes and currently occupies 1100 boxes; the server calculates the remaining space capacity as 100 boxes. Frozen dumplings - category D are stored in cold chain zone 2, which has a total capacity of 1800 boxes and currently occupies 1050 boxes; the server calculates the remaining space capacity as 750 boxes. The server then binds the above remaining space capacity with the total inbound and outbound flow parameters of the item category.

[0182] In this embodiment of the invention, the server calculates the capacity saturation based on the remaining space capacity parameter of the storage location. For the i-th item category, the location space capacity parameter of its corresponding storage location is Cap(i), and the currently occupied capacity parameter is Occ(i). Then, the remaining space capacity parameter is: Remain(i) = Cap(i) - Occ(i). The server calculates the inbound capacity saturation parameter based on the predicted inbound quantity Qin_pred(i): Sat_in(i) = Qin_pred(i) / max(Remain(i), ε). The server calculates the outbound operation quantity Qout_pred(i). The outbound capacity saturation parameter is calculated based on the outbound capacity Pout(i) of the location per unit time: Sat_out(i) = Qout_pred(i) / max(Pout(i), ε). When Sat_in(i) is greater than the inbound saturation threshold Th_sat_in, the server records the item category as an inbound capacity limit item category; when Sat_out(i) is greater than the outbound saturation threshold Th_sat_out, the server records the item category as an outbound capacity limit item category. Th_sat_in and Th_sat_out are preferably set to values ​​between 0.8 and 1.2. For inbound capacity limit item categories, the server calculates excess inbound demand: Excess_in(i) = max(0, Qin_pred(i) - Remain(i)). The server searches for adjacent item category nodes along the inbound flow directed edge and allocates excess inbound demand according to the remaining capacity, spatial distance, and category compatibility of adjacent nodes. For adjacent node j, the allocation ratio is: Ratio(i,j)=[Remain(j)×Compat(i,j) / Dist(i,j)] / Σj[Remain(j)×Compat(i,j) / Dist(i,j)]; where Remain(j) is the remaining capacity of the corresponding location of the adjacent node, Compat(i,j) is the item category compatibility coefficient, which takes a value of 0 to 1, and Dist(i,j) is the spatial distance parameter. The server determines the amount of data transferred to node j in the following way: Transfer_in(i,j)=min(Excess_in(i)×Ratio(i,j),Remain(j)); for item categories with extremely limited outbound capacity, the server calculates the excess outbound demand: Excess_out(i)=max(0,Qout_pred(i)-Pout(i)). The server searches for adjacent item category nodes along the directed edge of the outbound flow or in the opposite direction, allocates excess outbound demand to adjacent locations that are closer to the outbound outlet or have higher outbound processing capacity, and generates an outbound demand propagation plan.The server adjusts the storage location adjustment command based on Transfer_in(i,j) or Transfer_out(i,j) to move the transferred inventory quantity to an adjacent location; at the same time, it recalculates the replenishment trigger threshold based on the actual remaining capacity after the transfer to avoid triggering large-scale replenishment when the original location is already saturated.

[0183] The server then maps the total inbound turnover parameter of each item category to the remaining space capacity parameter of the corresponding location, generating an inbound capacity saturation parameter. For example, if the total inbound turnover of the pre-prepared meals-C category is 260 boxes in the future replenishment window, but the remaining space in cold chain zone 1 is only 100 boxes, the server calculates that its inbound capacity saturation exceeds the upper limit and marks the pre-prepared meals-C category as an item category with an inbound capacity limit. Simultaneously, the server maps the total outbound turnover parameter to the remaining space capacity parameter, generating an outbound capacity saturation parameter. For example, if the frozen chicken breast-A category requires frequent replenishment, picking, and temporary turnover in the weekend outbound window, and the current front-end picking location has insufficient remaining capacity, the server marks it as an item category with an outbound capacity limit.

[0184] The server extracts the supply trend direction identifier for the category of goods with limited storage capacity. When the supply trend direction identifier for the pre-prepared meals (C category) is an increasing supply trend, the server searches for adjacent item category nodes with storage associations along the direction of the storage flow, identifying that the locations of the frozen seasoning packets (E category) and frozen dumplings (D category) still have remaining space. The server further obtains the remaining space capacity parameters of these adjacent nodes and performs demand transmission processing for the excess storage demand of the pre-prepared meals (C category) along the reverse direction of the storage flow. For example, if the pre-prepared meals (C category) exceeds the carrying capacity of cold chain zone 1 by 160 boxes, the server will keep 100 boxes in cold chain zone 1 and allocate 60 boxes to available storage locations in the adjacent cold chain zone 2, generating a storage demand transmission plan.

[0185] The server also extracts the demand trend direction identifiers for categories with limited warehouse capacity. When the demand trend direction identifier for frozen chicken breast (Category A) shows a demand growth trend, the server searches for adjacent nodes with outbound associations along the directed edge of the outbound flow. It identifies that prepared meals (Category C) and frozen seasoning packets (Category E) are being outbound together in a combined order, and that there is still some spare capacity in the adjacent front-end area. The server propagates the excess outbound demand for frozen chicken breast (Category A) in the reverse direction of the outbound flow, allocating some high-frequency picking stock to adjacent front-end areas, generating an outbound demand propagation plan to reduce picking congestion in a single location.

[0186] Finally, the server adjusts the replenishment trigger threshold update instructions and storage location adjustment instructions for the pre-prepared food-C category and its adjacent nodes according to the inbound demand transmission scheme. For example, it increases the replenishment threshold for the pre-prepared food-C category and adds temporary receiving locations in the cold chain zone 2. The server adjusts the storage location priority for the frozen chicken breast-A category and related categories according to the outbound demand transmission scheme, placing frequently occurring outbound categories on adjacent picking paths. The server incorporates the adjusted replenishment trigger threshold update instructions and storage location adjustment instructions into the inventory dynamic optimization instruction set, enabling the system to maintain inventory buffering and outbound support through related location transmission even when locations are close to capacity limits.

[0187] This invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the processor executes the computer instructions, the computer device 100 performs the aforementioned actions provided in this invention. Figure 4 As shown, Figure 4 This is a structural block diagram of a computer device 100 provided in an embodiment of the present invention. The computer device 100 includes a memory 111, a processor 112, and a communication unit 113. To enable data transmission or interaction, the memory 111, the processor 112, and the communication unit 113 are electrically connected to each other directly or indirectly.

[0188] For illustrative purposes, the foregoing description has been made with reference to specific embodiments. However, the foregoing illustrative discussions are not intended to be exhaustive or to limit the present disclosure to the precise forms disclosed. Numerous modifications and variations are possible in accordance with the foregoing teachings. These embodiments were chosen and described in order to best illustrate the principles of the present disclosure and its practical application, thereby enabling those skilled in the art to best utilize the disclosure and to employ various embodiments with different modifications to suit a particular intended application.

Claims

1. A method for dynamic optimization of intelligent warehousing inventory based on supply and demand trend prediction, characterized in that, The method includes: Obtain a set of historical inventory turnover records from the warehouse management system. The set of historical inventory turnover records includes a sequence of inbound records with an inbound time stamp and a sequence of outbound records with an outbound time stamp. The inbound record sequence includes an inbound item category identifier and an inbound quantity parameter. The outbound record sequence includes an outbound item category identifier and an outbound quantity parameter. Based on the inbound time stamp, the inbound record sequence is arranged along the time axis to generate an inbound time sequence distribution sequence. Based on the outbound time stamp, the outbound record sequence is arranged along the time axis to generate an outbound time sequence distribution sequence. A directed network for the flow of inventory items is constructed based on the inbound time sequence distribution sequence and the outbound time sequence distribution sequence. The directed network for the flow of inventory items includes item category nodes, inbound directed edges, and outbound directed edges. The inbound directed edges carry the inbound quantity parameter, and the outbound directed edges carry the outbound quantity parameter. The flow trend prediction processing is performed on the directional network of inventory items to generate supply and demand trend prediction information for the items. The supply and demand trend prediction information for the items includes the demand trend direction identifier and the supply trend direction identifier of the item category. Based on the predicted supply and demand trends of the goods, the storage locations and replenishment trigger thresholds of the warehouse inventory are dynamically adjusted to generate a set of dynamic inventory optimization instructions, which includes storage location adjustment instructions and replenishment trigger threshold update instructions.

2. The method according to claim 1, characterized in that, The construction of a directed network for the flow of inventory items based on the inbound time-series distribution sequence and the outbound time-series distribution sequence includes: Based on the inbound item category identification in the inbound time sequence distribution sequence, an item category node is established, and a first mapping relationship is formed between the inbound item category identification and the item category node; Based on the outbound item category identification in the outbound time sequence distribution sequence, query the first mapping relationship and associate the outbound item category identification with the corresponding item category node; According to the chronological order, the item category nodes corresponding to adjacent inbound time markers are connected by inbound flow directed edges, and the item category nodes corresponding to adjacent outbound time markers are connected by outbound flow directed edges. The inbound quantity parameter is assigned to the corresponding inbound flow directed edge as an edge weight parameter, and the outbound quantity parameter is assigned to the corresponding outbound flow directed edge as an edge weight parameter. Based on the first mapping relationship, the edge weight parameters corresponding to the same item category are aggregated to obtain the total inbound flow parameters and the total outbound flow parameters of the item category. The total inbound and outbound flow parameters of the item category are attached to the corresponding item category node to form the directed network of the inventory item flow.

3. The method according to claim 2, characterized in that, The process of predicting the flow trend of the directed network of inventory items to generate supply and demand trend prediction information includes: Extract the total inbound and outbound flow parameters of the item category corresponding to the item category node to form an item category flow volume pairing sequence; The time window is divided into the item category turnover pairing sequence to obtain the inbound turnover time window sequence and the outbound turnover time window sequence; Compare the cumulative total value of inbound turnover within adjacent time windows to generate inbound growth trend markers or inbound decline trend markers, and determine the supply trend direction indicator based on the inbound trend marker that appears most frequently within multiple consecutive time windows. Compare the cumulative total outbound turnover within adjacent time windows to generate outbound growth trend markers or outbound decline trend markers, and determine the demand trend direction indicator based on the outbound trend marker that appears most frequently within multiple consecutive time windows. Along the connection direction of the directed edge of the inbound flow and the directed edge of the outbound flow, the edge weight parameters are recursively accumulated to generate supply path trend direction identifiers and demand path trend direction identifiers respectively. By integrating the supply trend direction identifier and the supply path trend direction identifier, a supply trend direction identifier for the corresponding item category is generated; by integrating the demand trend direction identifier and the demand path trend direction identifier, a demand trend direction identifier for the corresponding item category is generated. The supply trend direction identifier and the demand trend direction identifier of each item category are combined to form the item supply and demand trend prediction information.

4. The method according to claim 1, characterized in that, The method involves dynamically allocating the storage locations and replenishment trigger thresholds of warehouse inventory items based on the predicted supply and demand trends of the items, generating a set of dynamic inventory optimization instructions, including: Obtain a warehouse inventory location layout map, which includes storage location identifiers, location space capacity parameters, and current occupied capacity parameters; A demand sorting queue for item categories is generated based on the demand trend direction identifier, and a supply sorting queue for item categories is generated based on the supply trend direction identifier. Based on the item category demand sorting queue, the item category supply sorting queue, the location space capacity parameter, and the current occupied capacity parameter, a location mapping relationship between item categories and storage location identifiers is generated. Generate storage location adjustment instructions based on the location mapping relationship; The adjustment direction of the replenishment trigger threshold is determined based on the demand trend direction indicator, and the adjustment range of the replenishment trigger threshold is determined based on the supply trend direction indicator. By combining the adjustment direction and the adjustment magnitude, a replenishment trigger threshold update amount is generated, and a replenishment trigger threshold update instruction is generated based on the replenishment trigger threshold update amount; The storage location adjustment instruction and the replenishment trigger threshold update instruction are combined to form the inventory dynamic optimization instruction set.

5. The method according to claim 2, characterized in that, The method further includes: Tracing the associated item category nodes that have an indirect inbound association with the item category node along the direction of the directed edge of the inbound flow, a set of reachable nodes for inbound flow is formed; Tracing the associated item category nodes that have an indirect outbound association with the item category node along the direction of the outbound flow, a set of outbound flow reachable nodes is formed; The edge weight parameters in the set of reachable nodes for the inbound flow are subjected to decay accumulation processing that decreases with the number of tracking steps to obtain the inbound indirect influence weight; The edge weight parameters in the set of reachable nodes for outbound flow are subjected to decay accumulation processing that decreases with the number of tracking steps to obtain the outbound indirect impact weight; The total inbound turnover parameter of the item category is adjusted according to the inbound indirect impact weight, and the total outbound turnover parameter of the item category is adjusted according to the outbound indirect impact weight. The supply trend direction indicator is regenerated based on the revised total inbound circulation parameters of the item category, and the demand trend direction indicator is regenerated based on the revised total outbound circulation parameters of the item category. Based on the regenerated supply trend direction identifier and demand trend direction identifier, update the commodity supply and demand trend prediction information, and regenerate the inventory dynamic optimization instruction set.

6. The method according to claim 1, characterized in that, The method further includes: Acquire external supply chain event data streams, which include upstream supplier capacity change notifications and downstream demand order fluctuation notifications; Semantic parsing is performed on the capacity change notification information to extract the capacity change direction identifier and the target identifier of the item category affected by the capacity change; Semantic parsing is performed on the order fluctuation notification information to extract the order fluctuation direction identifier and the target identifier of the item category affected by the order fluctuation; Map the target identifier of the item category affected by the capacity change to the item category node, and correct the supply trend direction identifier according to the capacity change direction identifier; Map the item category identifier affected by the order fluctuation to the item category node, and correct the demand trend direction identifier according to the order fluctuation direction identifier; The trend prediction weight coefficient of the inbound flow directed edge is updated according to the corrected supply trend direction identifier, and the trend prediction weight coefficient of the outbound flow directed edge is updated according to the corrected demand trend direction identifier. The directed network for the flow of inventory items is reconstructed using the updated directed edges for inbound and outbound flow. Based on the reconstructed directed network for the flow of inventory items, the supply and demand trend prediction information of items, which incorporates external supply chain events, is regenerated, and the set of dynamic inventory optimization instructions is regenerated.

7. The method according to claim 6, characterized in that, The method further includes: Extract the inbound quantity parameters carried by the directed edge of the inbound flow, and perform quantity jump detection along the connection direction of the directed edge of the inbound flow to obtain the inbound quantity jump amplitude. Extract the outbound quantity parameters carried by the directed edge of the outbound flow, and perform quantity jump detection along the connection direction of the directed edge of the outbound flow to obtain the outbound quantity jump amplitude; The jump amplitude of the inbound quantity and the jump amplitude of the outbound quantity are compared with the corresponding jump amplitude thresholds, and abnormal jump edges of inbound and outbound are marked. Based on the total turnover parameters of the item category nodes connected by the abnormal jump edges of the inbound and outbound processes, the abnormal jump times of the inbound and outbound processes are located respectively. Based on the inbound and outbound abrupt change time points, query the external supply chain event data stream, match the capacity change direction identifier and the order fluctuation direction identifier with the corresponding quantity jump direction, and establish a causal relationship based on the matching results; Based on the causal relationship, the trend strength parameters of the supply trend direction indicator and the demand trend direction indicator are corrected to generate the commodity supply and demand trend prediction information after jump causality correction; The set of dynamic inventory optimization instructions is regenerated based on the predicted supply and demand trends of the goods after the jump causality correction.

8. The method according to claim 1, characterized in that, The method further includes: Extract the change sequence of the edge weight parameters of the inbound flow directed edges of each item category node in the directed network of inventory item circulation along the time axis, and convert the fluctuation characteristics of the edge weight parameters along the time axis into a description of the fluctuation frequency of the inbound flow edge weights. Extract the change sequence of the edge weight parameters of the outbound flow directed edges of each item category node in the directed network of inventory item circulation along the time axis, and convert the fluctuation characteristics of the edge weight parameters along the time axis into a description of the fluctuation frequency of the outbound flow edge weights. The frequency consistency of the description of the fluctuation frequency of the edge weight of the inbound flow is compared with the frequency consistency of the periodic fluctuation characteristics of the inbound flow. The frequency components of the inbound flow fluctuation that are consistent with the frequency of the periodic fluctuation characteristics of the inbound flow are identified. The frequency components of the inbound flow fluctuation that are consistent with the frequency of the periodic fluctuation characteristics of the inbound flow are marked as the inbound normal fluctuation components, and the frequency components of the inbound flow fluctuation that are inconsistent with the frequency of the inbound flow are marked as the inbound abnormal fluctuation components. The frequency consistency of the outbound flow edge weight fluctuation frequency description and the outbound periodic fluctuation characteristics is compared. The outbound fluctuation frequency components that are consistent with the outbound periodic fluctuation characteristics in the edge weight parameter fluctuation of the outbound flow directed edge are identified. The outbound fluctuation frequency components with consistent frequencies are marked as outbound habitual fluctuation components, and the outbound fluctuation frequency components with inconsistent frequencies are marked as outbound abnormal fluctuation components. The occurrence time of the abnormal fluctuation components of the inbound goods is located along the time axis. The start time, duration and fluctuation amplitude parameters of the abnormal fluctuation components of the inbound goods are extracted to form an inbound goods abnormal event descriptor. The occurrence time of the outbound anomaly fluctuation component is located along the time axis, and the start time, duration and fluctuation amplitude parameters of the outbound anomaly fluctuation component are extracted to form an outbound anomaly event descriptor; Based on the inbound anomaly event descriptor, query the external supply chain event data stream, retrieve upstream supplier capacity change notification information within the time range covered by the inbound anomaly event descriptor, and perform event feature matching processing between the capacity change notification information and the inbound anomaly event descriptor; Based on the outbound anomaly event descriptor, query the external supply chain event data stream, retrieve downstream demand order fluctuation notification information within the time range covered by the outbound anomaly event descriptor, and perform event feature matching processing between the order fluctuation notification information and the outbound anomaly event descriptor; When the capacity change notification information matches the descriptor of the inbound anomaly event, the inbound anomaly fluctuation component is separated from the inbound flow edge weight fluctuation frequency description, and the edge weight parameter change sequence of the inbound flow directed edge is reconstructed using the inbound habitual fluctuation component. When the match fails, the inbound anomaly fluctuation component is marked as an unexplained inbound anomaly event. When the order fluctuation notification information matches the outbound anomaly event descriptor, the outbound anomaly fluctuation component is separated from the outbound flow edge weight fluctuation frequency description, and the outbound flow directed edge edge weight parameter change sequence is reconstructed using the outbound habitual fluctuation component. When the match fails, the outbound anomaly fluctuation component is marked as an unexplained outbound anomaly event. The supply trend direction identifier of the item category node is regenerated based on the change sequence of edge weight parameters of the reconstructed inbound flow directed edges, and the demand trend direction identifier of the item category node is regenerated based on the change sequence of edge weight parameters of the reconstructed outbound flow directed edges. The reconstructed supply trend direction identifier and demand trend direction identifier are reintegrated into the commodity supply and demand trend prediction information to generate commodity supply and demand trend prediction information after removing the influence of external events. The set of dynamic inventory optimization instructions is regenerated based on the predicted supply and demand trends of goods after the impact of external events has been removed.

9. The method according to claim 4, characterized in that, The method further includes: Calculate the remaining space capacity parameter based on the location space capacity parameter corresponding to the storage location identifier and the currently occupied capacity parameter; An inbound capacity saturation parameter is generated based on the total inbound turnover parameter of the item category and the remaining space capacity parameter; an outbound capacity saturation parameter is generated based on the total outbound turnover parameter of the item category and the remaining space capacity parameter. The categories of goods with extremely limited inbound capacity are identified based on the inbound capacity saturation parameter, and the categories of goods with extremely limited outbound capacity are identified based on the outbound capacity saturation parameter. When the supply trend direction of the category of goods with limited inbound capacity is identified as a supply growth trend, the adjacent category nodes are retrieved along the directed edge of the inbound flow, and the excess inbound demand is transmitted to the adjacent category nodes to generate an inbound demand transmission scheme. When the demand trend direction of the outbound capacity limit item category is identified as a demand growth trend, the adjacent item category nodes are retrieved along the outbound flow directed edge, and the excess outbound demand is transmitted to the adjacent item category nodes to generate an outbound demand transmission scheme. Based on the inbound demand transmission scheme and the outbound demand transmission scheme, adjust the replenishment trigger threshold update instruction and the storage location adjustment instruction; The adjusted replenishment trigger threshold update instruction and the adjusted storage location adjustment instruction are incorporated into the inventory dynamic optimization instruction set.

10. A smart warehousing and inventory dynamic optimization system based on supply and demand trend prediction, characterized in that, Includes at least one service node; The service node includes a storage unit and a computing unit; the storage unit is used to store program code. The computing unit is used to run the program code to execute the intelligent warehousing inventory dynamic optimization method based on supply and demand trend prediction as described in any one of claims 1 to 9.

Images