An intelligent prediction and dynamic scheduling method and system for industrial goods inventory demand

Abstract

The present application relates to the technical field of industrial internet of things and intelligent warehousing, in particular to an industrial goods inventory demand intelligent prediction and dynamic scheduling method and system; the method comprises: time alignment and behavior intensity driven event processing of multi-source heterogeneous data on the terminal side, compression into memory fragments after consistency verification and uploading to the cloud; the cloud fuses to build a continuously updated inventory behavior memory, identifies the demand disturbance source and generates a structured propagation chain; then, an inventory tension state field is built, the demand evolution path is deduced and the risk level is identified; finally, a terminal-cloud collaborative dynamic scheduling mechanism is established, the scheduling strategy is corrected in combination with the real-time state on the terminal side, and the model is updated. The present application realizes inventory intelligent closed-loop regulation and control based on terminal-cloud collaboration and pressure evolution constraints, and changes passive response into active risk prediction and adaptive scheduling.

Description

Technical Field

[0001] This invention relates to the fields of industrial Internet of Things and smart warehousing technology, specifically to a method and system for intelligent forecasting and dynamic scheduling of industrial product inventory demand. Background Technology

[0002] With intensifying global manufacturing competition and increasingly complex supply chain systems, inventory management for industrial products has evolved from a simple function of material storage to a crucial collaborative link spanning the entire chain of production, logistics, procurement, and sales. Modern industrial sites are characterized by multi-variety, small-batch production, rapid fluctuations in customer demand, diverse equipment maintenance and spare parts, and frequent cross-regional warehousing and transfers. This means that inventory changes no longer follow a single, fixed pattern, but exhibit strong dynamism and multi-factor coupling. At the same time, the widespread deployment of IoT sensing technology, edge computing nodes, and cloud computing platforms provides solid technical support for the real-time acquisition of multi-source heterogeneous data such as warehousing operations, production consumption, equipment operating status, and order triggering. This also makes the detailed characterization of inventory status and global visual monitoring feasible.

[0003] Chinese invention patent CN116720811B discloses a warehouse management method and system based on the Internet of Things. It establishes a predictive multi-path adaptive scheduling model, constructs an initial solution set through encoding, calculates fitness, and performs roulette wheel selection and cross-mutation iterative optimization of the number of picking carts and paths. At the same time, it establishes an adaptive inventory prediction model, determines the optimal hyperplane based on historical inventory data, and uses momentum gradient to update parameters to predict future inventory changes. Then, it guides the optimization and adjustment of the storage location of warehouse items based on the prediction results, effectively improving the efficiency of warehouse operations and the scientific nature of inventory management.

[0004] However, within the supply chain network, inventory changes between different warehousing nodes, production units, and logistics links are often interconnected. An abnormal consumption or replenishment delay in one link may be transmitted and amplified along the business path, thus having a chain reaction on the overall inventory stability. This places higher demands on the overall coordination capability and response flexibility of inventory management. In the face of the new opportunities brought about by the aforementioned industrial and technological developments, how to fully integrate distributed sensing data, accurately understand the dynamic evolution of inventory pressure, and generate reasonable resource allocation decisions in a timely manner has become an important direction that urgently needs to be addressed in the field of industrial inventory management. Summary of the Invention

[0005] The purpose of this invention is to address the problems existing in the background technology by proposing a method and system for intelligent prediction and dynamic scheduling of industrial product inventory demand.

[0006] The technical solution of this invention: a method for intelligent forecasting and dynamic scheduling of industrial supplies inventory demand, characterized by the following specific implementation steps: S1. At the edge computing node on the edge side, perform synchronous perception and time alignment processing on multi-source heterogeneous data related to industrial supplies inventory. Then, through the behavior intensity driving mechanism, transform continuous inventory data into discrete behavior event units. At the same time, introduce behavior consistency constraints to achieve event credibility screening and vectorized expression. Finally, perform time-weighted compression on the event sequence to form inventory behavior memory fragments and upload them to the cloud. S2. In the cloud, the inventory behavior of different warehousing nodes, production nodes and logistics nodes is unified and integrated to build an inventory behavior memory that can be continuously updated. By analyzing the continuous evolution characteristics of inventory behavior, the source of demand disturbance is identified. Furthermore, the demand disturbance propagation chain is established by combining the relationship between inventory business. The credibility of the propagation path is corrected and dynamically modified, and finally a structured demand disturbance propagation chain is output. S3. By integrating the persistence of inventory behavior, the degree of impact of demand propagation, and inventory recovery capability, tension is calculated for each inventory node, and the global inventory pressure distribution is formed by combining the propagation relationship between nodes. By analyzing the trend of inventory tension changes, the demand evolution path is deduced, potential areas of concentrated inventory pressure are identified, and risk levels are generated. S4. Construct a dynamic scheduling mechanism for end-to-cloud collaboration based on the constraints of inventory pressure evolution. Determine scheduling priorities by analyzing the urgency of inventory risk areas, generate dynamic scheduling schemes by combining demand propagation direction, inventory resource status and on-site execution conditions, and use the real-time status of the end side to correct the cloud scheduling strategy. At the same time, collect scheduling execution results and update the inventory behavior memory, demand propagation relationship and inventory tension status.

[0007] Preferably, step S1 specifically includes: Multi-source heterogeneous data such as inventory inbound and outbound, equipment status, production consumption and order triggering are collected on the edge. Time drift correction is performed on each data source through a unified time base, and local smoothing is performed on the data in combination with a sliding window to eliminate the impact of sampling frequency differences and instantaneous noise. A comprehensive behavior intensity function is constructed to characterize the overall change range of the inventory system through weighted fusion of multiple types of data, and a dynamic threshold triggering mechanism is constructed using the behavior intensity difference between adjacent time points. When the change exceeds the threshold, an inventory behavior event unit is generated. The generated inventory behavior event units are introduced with expected action values ​​inferred from historical behavior patterns. A behavior consistency scoring function is constructed to measure the degree of deviation between the actual execution behavior and historical patterns. Event triggering sources, execution actions, result feedback and consistency scores are mapped together into a unified low-dimensional vector representation. The vectors of inventory behavior events that are continuous in time are aggregated into behavior memory segments by time segment, and time decay weights are introduced to compress the events within the segments to generate a compact inventory behavior memory representation. Only the compressed result is uploaded to the cloud.

[0008] Preferably, the specific process of correcting time drift for each data source using a unified time base and combining this with local smoothing of the data using a sliding window is as follows: A unified time reference unit is established in the edge computing node, using the local high-precision clock or the industrial gateway master clock as a reference, and drift correction is performed on each data source through periodic time calibration packages; after time alignment, a sliding stable window is introduced for local smoothing, and the window length is determined based on the average response cycle of inventory operations.

[0009] Preferably, the specific process of constructing a comprehensive behavioral intensity function and characterizing the overall change magnitude of the inventory system through weighted fusion of multiple types of data is as follows: A comprehensive behavior intensity function is constructed, and the overall activity level of the inventory system at the current moment is calculated by weighted summation of multi-source data. The behavior contribution weight of each data type is obtained by statistical analysis of historical inventory consumption and business correlation. The dynamic event triggering threshold used in the dynamic threshold triggering mechanism is adaptively determined by performing quantile statistics on the distribution of historical inventory behavior intensity.

[0010] Preferably, the specific process of constructing a behavioral consistency scoring function to measure the degree of deviation between actual performance behavior and historical patterns is as follows: Reference actions are established using historical similar inventory behaviors, and the current inventory behavior is compared with the reference actions. The consistency score is calculated based on the ratio of the execution intensity corresponding to the actual action executed in the current event to the reference execution intensity obtained from the statistics of historical similar inventory behaviors. The inventory behavior events are further encoded to form an inventory behavior vector, which consists of the event source code, the action code, and the result code.

[0011] Preferably, the specific process of introducing time decay weight to weighted compress events within a segment is as follows: construct an inventory behavior memory segment for inventory behavior event vectors that occur consecutively within the same business cycle, and calculate the memory segment expression using a time decay method, wherein the time decay coefficient is set according to the average turnover cycle of industrial supplies inventory; edge nodes only upload the compressed inventory behavior memory segment and necessary index information to the cloud.

[0012] Preferably, step S2 specifically includes: The cloud receives inventory behavior memory fragments uploaded by each edge node, collects and sorts them uniformly according to industrial product identification, business type, time sequence and inventory flow relationship, and performs weighted fusion based on the relevance of inventory behavior, consistency of behavior and business continuity to construct an inventory behavior memory that fully records the historical evolution of industrial product inventory. Based on the memory of inventory behavior, the direction, magnitude and duration of changes of adjacent inventory behaviors are analyzed according to the order of occurrence of inventory behaviors. The continuity index of inventory behavior is calculated, and the continuous evolution trend is identified accordingly. Furthermore, the initial event that triggered the continuous inventory change is located in reverse and identified as the source of demand disturbance. Starting from the source of demand disturbance, we establish inventory behavior relationships by combining inventory flow relationships, production work order relationships, warehousing and allocation relationships and process matching relationships, calculate the propagation intensity of demand disturbance between each inventory behavior node, and dynamically generate the demand disturbance propagation chain according to the actual evolution sequence of inventory business. By combining the consistency score of inventory behavior, the intensity of demand disturbance propagation, and the historical propagation verification results, the credibility of each propagation path in the demand disturbance propagation chain is calculated and corrected. Credible propagation relationships are preserved and the paths to be verified are continuously tracked and updated. Finally, a structured demand disturbance propagation chain containing information on propagation source, propagation direction, propagation level, propagation intensity, and credibility is output.

[0013] Preferably, step S3 specifically includes: Based on the intensity and credibility of demand disturbance propagation and the continuity of inventory behavior indicators, combined with the replenishment response capability, allocation capability and alternative supply capability of inventory nodes, an inventory node tension calculation mechanism is established to quantify the supply and demand pressure of each industrial product inventory node and form a node tension value that reflects the current inventory risk level. By leveraging the node relationships and propagation credibility in the demand disturbance propagation chain, the tension of each inventory node is correlated and integrated, so that each inventory node is jointly affected by its own inventory status and the demand changes of related nodes, thus constructing an inventory tension state field covering the entire inventory network. Based on the inventory tension state field analysis, the pressure change trend of each inventory node is analyzed. By calculating the direction and growth rate of tension change between adjacent nodes, the possible future diffusion direction of demand disturbance is identified, and the inventory demand evolution path is generated according to the propagation chain sequence. Based on the node pressure level, growth trend, and demand propagation stability in the inventory tension state field, continuous high-pressure inventory nodes are regionalized, the inventory risk index is calculated and the risk level is generated, and the industrial products involved and the scope of impact are output.

[0014] Preferably, step S4 specifically includes: Based on abnormal inventory tension areas, risk indices, and demand evolution trends, a comprehensive evaluation of different inventory areas is conducted. By integrating inventory pressure levels, tension growth rates, demand propagation ranges, and the criticality of industrial products, scheduling priorities are calculated to determine the inventory nodes requiring priority intervention and resource scheduling targets. Based on the evolution path of inventory demand and the resource status of inventory nodes, the resource supply capacity and scheduling adaptability between different inventory nodes are analyzed. Combining the reliability of demand propagation, inventory adjustability and scheduling execution conditions, a dynamic scheduling scheme including the outgoing node, the incoming node, the allocation quantity and the execution time is generated. The dynamic scheduling scheme is executed by using an edge-cloud collaboration mechanism. The cloud is responsible for scheduling planning based on the global inventory status, while the edge combines the real-time inventory status, equipment status and operation constraints collected from the warehouse site to make local adjustments to the scheduling scheme. By correcting the scheduling path, resource source and execution order in real time, the adaptability of the scheduling scheme in the actual industrial environment is improved. Collect actual execution data during the inventory scheduling process, including allocation results, completion time, inventory changes and demand response, and calculate scheduling execution deviation. When there is a large deviation between the actual results and the plan, update the inventory behavior memory weight, demand propagation credibility and inventory tension status to complete closed-loop optimization.

[0015] The technical solution of this invention: An intelligent forecasting and dynamic scheduling system for industrial supplies inventory demand, used to execute the aforementioned intelligent forecasting and dynamic scheduling method for industrial supplies inventory demand, comprising: The edge-side multi-source inventory behavior perception and event structured modeling module is used to collect and preprocess multi-source data on inventory changes, material requisition, replenishment, transfer, inventory count and production consumption in the entire process of industrial supplies inventory business. It constructs inventory behavior events and forms inventory behavior memory fragments, providing unified and standardized structured input data for subsequent inventory behavior analysis. The Inventory Behavior Memory Fusion and Demand Disturbance Propagation Analysis Module is used to fuse inventory behavior memory fragments, construct an inventory behavior memory body, generate a demand disturbance propagation chain based on the temporal and business relationships between inventory behaviors, calculate the demand propagation strength and propagation credibility, and identify the inventory demand formation process and its propagation relationship. The module for constructing the inventory tension state field and extrapolating the demand evolution is used to calculate the inventory node tension based on the inventory behavior memory and the demand disturbance propagation chain, construct the inventory tension state field, extrapolate the inventory demand evolution path, identify abnormal areas of inventory tension, and output the inventory risk level and demand development trend, providing a basis for inventory scheduling decisions. The edge-cloud collaborative dynamic scheduling and closed-loop optimization module is used to generate dynamic inventory scheduling strategies based on the inventory tension state field, demand evolution path and inventory risk level. It combines the real-time warehouse status on the edge side to complete the scheduling execution and local correction, and updates the inventory behavior memory, demand propagation relationship and inventory tension state based on the execution feedback, so as to realize the closed-loop optimization of inventory demand forecasting and dynamic scheduling.

[0016] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects: This invention designs an intelligent forecasting and dynamic scheduling method and system for industrial supplies inventory demand. Through a collaborative architecture that integrates event-driven compression based on edge behavior intensity with global memory fusion in the cloud, it effectively solves the practical problems of difficult synchronization of multi-source heterogeneous data and high communication load in industrial scenarios, significantly improving the real-time performance and lightweight nature of inventory status perception. Based on a demand disturbance propagation chain constructed using behavioral consistency constraints and credibility correction, it can accurately trace the source of demand changes and clearly depict their diffusion path in the supply chain. This breaks through the limitations of traditional inventory management that only focuses on quantity fluctuations, fundamentally enhancing the system's understanding of demand uncertainty. Adaptability; By introducing the concept of inventory tension state field, the system quantifies and expresses the pressure on nodes themselves and the impact of related propagation, achieving a leap from single-point risk identification to regional situational awareness. Combined with evolution path deduction, the system possesses forward-looking early warning capabilities. The dynamic scheduling mechanism of edge-cloud collaboration fully leverages the complementary advantages of global optimization in the cloud and rapid response at the edge. At the same time, through feedback loops, it continuously corrects model parameters, enabling prediction and scheduling capabilities to adaptively evolve with changes in the production environment. Ultimately, it realizes the transformation of inventory resources from a passive response to an active defensive control mode, providing solid technical support for lean inventory management in complex industrial environments. Attached Figure Description

[0017] Figure 1 The flowchart shows a method for intelligent forecasting and dynamic scheduling of industrial supplies inventory demand proposed in this invention. Figure 2 This is a system architecture diagram of an intelligent forecasting and dynamic scheduling system for industrial supplies inventory demand proposed in this invention. Detailed Implementation

[0018] Example 1, as Figure 1 As shown, the present invention proposes an intelligent forecasting and dynamic scheduling method for industrial supplies inventory demand, which includes the following specific implementation steps: S1. At the edge computing node on the edge side, perform synchronous perception and time alignment processing on multi-source heterogeneous data related to industrial supplies inventory. Then, through the behavior intensity driving mechanism, transform continuous inventory data into discrete behavior event units. At the same time, introduce behavior consistency constraints to achieve event credibility screening and vectorized expression. Finally, perform time-weighted compression on the event sequence to form inventory behavior memory fragments and upload them to the cloud. S2. Based on the inventory behavior memory fragment uploaded in step S1, the inventory behavior of different warehousing nodes, production nodes and logistics nodes is uniformly integrated in the cloud to build an inventory behavior memory that can be continuously updated. By analyzing the continuous evolution characteristics of inventory behavior, the source of demand disturbance is identified. Furthermore, the demand disturbance propagation chain is established by combining the relationship between inventory business. The credibility of the propagation path is corrected and dynamically modified. Finally, a structured demand disturbance propagation chain is output. S3. Based on the inventory behavior memory and demand disturbance propagation chain output in step S2, construct the inventory tension state field. By integrating the persistence of inventory behavior, the degree of impact of demand propagation, and inventory recovery capability, calculate the tension of each inventory node and form a global inventory pressure distribution state by combining the propagation relationship between nodes. By analyzing the trend of inventory tension change, deduce the demand evolution path, identify potential areas of concentrated inventory pressure, and generate risk levels. S4. Based on the inventory tension state field, demand evolution path, and inventory anomaly area information output in step S3, construct an edge-cloud collaborative dynamic scheduling mechanism based on inventory pressure evolution constraints. Determine scheduling priority by analyzing the urgency of inventory risk areas, generate dynamic scheduling schemes by combining demand propagation direction, inventory resource status, and on-site execution conditions, and modify cloud scheduling strategies using real-time edge status. At the same time, collect scheduling execution results and update the inventory behavior memory, demand propagation relationship, and inventory tension state.

[0019] In a specific embodiment, step S1 involves performing unified perception, time alignment, behavior extraction, and structured representation on multi-source heterogeneous data from warehousing operations, production consumption, and equipment operation in the end-side node of the industrial supplies inventory system, forming inventory behavior event units and their compressed memory representations that can be used for subsequent cloud modeling. The specific implementation steps include: S11. Collect multi-source heterogeneous data such as inventory inflows and outflows, equipment status, production consumption, and order triggering on the edge. Correct time drift for each data source using a unified time base, and perform local smoothing of the data using a sliding window to eliminate the impact of sampling frequency differences between different systems and instantaneous noise. This enables the multi-source data to form a time-consistent basic data stream for inventory behavior. The specific implementation process is as follows: In industrial settings, different types of inventory-related data come from different devices and systems, including but not limited to RFID readers, PLC controllers, MES system interfaces, warehouse scanning terminals, and equipment status sensors. Each data source has differences in sampling frequency, transmission delay, and clock accuracy. To ensure that multi-source data has a unified temporal semantic basis on the edge, a unified time reference unit is first established in the edge computing node, with the local high-precision clock or the industrial gateway master clock as a reference, and drift correction of each data source is performed through periodic time calibration packages. For the i-th type of inventory behavior data, its original observation sequence is denoted as The aligned behavior sequence is obtained through time offset correction. ; After time alignment, a sliding stabilizing window is further introduced for local smoothing: ; In the formula, W represents the length of the sliding window, and its value is determined based on the average response cycle of inventory operations, such as the time interval between batch inbound and outbound operations. This represents a smoothed sequence of inventory behavior observations, used to reduce the impact of transient noise on subsequent event identification. S12. Construct a comprehensive behavior intensity function to characterize the overall change magnitude of the inventory system through weighted fusion of multiple types of data, and use the behavior intensity difference between adjacent time points to construct a dynamic threshold triggering mechanism. When the change exceeds the threshold, an inventory behavior event unit is generated. The specific implementation process is as follows: After completing the multi-source data alignment and smoothing process, the edge nodes perform intensity-driven modeling of inventory behavior, that is, dynamically identify event boundaries based on the intensity of changes in inventory behavior, and construct a comprehensive inventory behavior intensity function: ; In the formula, The overall inventory behavior intensity at time t is used to reflect the current overall activity level of the inventory system; n represents the number of inventory behavior data types involved in the modeling. This represents the behavioral contribution weight of the i-th type of data. This weight is not fixed, but is obtained through the statistics of historical inventory consumption and business correlation. For example, the weight of outbound behavior is usually higher than the weight of equipment status. Based on this, an event triggering mechanism is constructed by differentiating the intensity of behavior at adjacent time points: ; In the formula, This represents the dynamic event trigger threshold, which is adaptively determined by performing quantile statistics on the intensity distribution of historical inventory behavior (e.g., the 90%~97% quantile range) to adapt to different warehousing business rhythms. When the triggering conditions are met, the current moment is defined as the point where the inventory behavior event occurs, and an inventory behavior event unit is constructed. : ; In the formula, Indicates the timestamp of the event; Indicates the type of event trigger source, such as order-driven, production consumption-driven, equipment malfunction-driven, or allocation-driven; Indicates the type of inventory action performed, such as inbound, outbound, transfer, or replenishment; This indicates the feedback quantity of the behavioral outcome, which typically includes inventory changes, execution delay time, or response deviation value; S13. Introduce expected action values ​​inferred from historical behavior patterns to the generated inventory behavior event units, construct a behavior consistency scoring function to measure the degree of deviation between actual execution behavior and historical patterns, and map the event trigger source, execution action, result feedback, and consistency score together into a unified low-dimensional vector representation. The specific implementation process is as follows: To ensure that the data uploaded to the cloud accurately reflects the actual operating status of the inventory, the edge nodes further perform consistency verification on inventory behavior events. For similar inventory behaviors, there are usually relatively stable execution patterns in long-term operation. For example, the variation range of material requisition quantity corresponding to the same production work order is usually small, and the spare parts consumption corresponding to the same equipment maintenance cycle has a relatively stable trend. Therefore, this embodiment uses historical similar inventory behaviors to establish reference actions and compares the current inventory behavior with the reference actions. The consistency evaluation is calculated using the following formula: ; In the formula, Indicates the consistency score of inventory behavior; This indicates the execution intensity corresponding to the actual action performed in the current event; This indicates the reference execution intensity obtained based on statistics of similar historical inventory behavior; The closer the consistency score is to 1, the more consistent the current inventory behavior is with historical patterns; the lower the consistency score, the more likely the current inventory behavior may be affected by unforeseen events, and it needs to be closely monitored in subsequent demand analysis. Further encode inventory behavior events to form inventory behavior vectors: ; In the formula, The feature mapping function representing the edge nodes; Indicates the source code of the event; Indicates the code for the action to be performed; Indicates the encoding of the execution result; S14. Aggregate the time-continuous inventory behavior event vectors by time segments to form behavior memory segments, and introduce time decay weights to weight and compress the events within the segments to generate a compact inventory behavior memory representation. Only the compressed results are uploaded to the cloud to reduce communication load and retain key behavior evolution information, achieving a balance between lightweight memory storage on the edge and efficient modeling input in the cloud. The specific implementation process is as follows: Edge nodes continuously receive inventory behavior vectors and construct inventory behavior memory fragments according to the continuous inventory business process; For consecutive inventory behavior events occurring within the same business cycle, establish inventory behavior memory fragments: ; In the formula, This represents the memory segment of the j-th inventory behavior; Represents an inventory behavior event vector; Considering that recent inventory behavior typically has a greater impact on inventory demand forecasting than earlier inventory behavior, a time decay method is used to calculate the memory segment representation: ; ; In the formula, This represents a compressed memory fragment of inventory behavior; m represents the number of inventory behavior events contained in the fragment. Indicates the time decay weight; This represents the time decay factor, which is set based on the average inventory turnover period of industrial products. After the memory fragments are constructed, the edge nodes only upload the inventory behavior memory fragments and necessary index information to the cloud; the cloud then aggregates the inventory behavior memory fragments from multiple edge nodes.

[0020] In a specific embodiment, step S2 reconstructs the long-term behavioral evolution of industrial goods across different warehousing, production, and logistics nodes, and establishes a continuously updated inventory behavior memory. Based on this memory, it analyzes the evolution of inventory demand formation, identifies the location, direction, and scope of demand disturbances, and forms a demand disturbance propagation chain with business logic constraints. The specific implementation steps include: S21. The cloud receives inventory behavior memory fragments uploaded by each edge node, and uniformly collects and sorts them according to industrial product identification, business type, time sequence, and inventory flow relationship. It then performs weighted fusion based on the recentity of inventory behavior, behavioral consistency, and business continuity to construct an inventory behavior memory that can completely record the historical evolution of industrial product inventory. The specific implementation process is as follows: The cloud first receives inventory behavior memory fragments uploaded from various edge computing nodes. Since different edge nodes correspond to different warehousing areas, different production units, or different logistics nodes, the same industrial product usually corresponds to multiple inventory behavior memory fragments in the actual circulation process. If each memory fragment is analyzed separately, the inventory behavior evolution process is easily artificially fragmented, and the true inventory change pattern of industrial products in the entire supply chain cannot be reflected. Therefore, this embodiment first collects all memory fragments of inventory behavior according to the unique identifier of industrial products, batch information, warehousing node, production work order number and the time of inventory behavior occurrence, and reorders them according to the actual flow order of inventory business to restore the evolution sequence of inventory behavior of industrial products from warehousing, storage, transfer, material requisition, consumption to replenishment. The j-th inventory behavior memory fragment uploaded in step S1 is represented as follows: The inventory behavior memory is then represented as: ; ; ; ; ; In the formula, B represents the inventory behavior memory; N represents the number of inventory behavior memory fragments participating in the fusion. Indicates the weight of the fusion of memory fragments representing inventory behavior; Indicates the memory validity period coefficient; Indicates the reliability coefficient of inventory behavior; Indicates the continuity coefficient of inventory operations; This represents the time decay coefficient, which is determined based on the average inventory turnover period of industrial products. This represents the time interval between the current moment and the memory segment representing inventory behavior. This indicates the number of events belonging to the same inventory business process within the memory segment of inventory behavior; This represents the total number of events in the inventory behavior memory segment; For example, if a certain inventory behavior segment includes: production material requisition → workshop delivery → production consumption → replenishment request, all belonging to the same work order, then... If it includes: production material requisition, manual inventory, temporary material borrowing, and return to inventory, then business continuity decreases. Correspondingly reduced; After the integration is completed, each industrial product forms a unique corresponding inventory behavior memory. The inventory behavior memory not only stores changes in inventory quantity, but also stores the order of inventory behavior, duration of behavior, consistency of behavior, and continuous relationship between different inventory behaviors, thus being able to fully reflect the long-term evolution of the inventory status of industrial products. S22. Based on the memory of inventory behavior, analyze the direction, magnitude, and duration of changes in adjacent inventory behaviors according to their order of occurrence. Calculate the continuity index of inventory behavior and identify whether inventory behavior forms a continuous evolution trend. Further reverse-engineer the initial event that triggered the continuous inventory change, identify it as the source of demand disturbance, and determine the propagation starting point for demand propagation analysis. The specific implementation process is as follows: Once the inventory behavior memory is established, the cloud further analyzes whether the inventory behavior has gradually evolved from random fluctuations to a continuous changing trend; Industrial product inventory changes gradually form under the continuous effect of multiple inventory behaviors. For example, an increase in production tasks will first manifest as an increase in production material requisition, which will then lead to a decrease in safety stock, and further trigger replenishment requests and warehouse transfers. Therefore, this embodiment identifies whether a continuous evolution process is formed between inventory behaviors based on the inventory behavior memory. Based on the order of inventory behavior, the consecutive inventory behaviors within the inventory behavior memory are compared one by one to analyze the consistency of adjacent inventory behaviors in terms of the direction, magnitude and duration of change. Inventory continuity is represented by the following indicators: ; In the formula, L represents the inventory behavior continuity index; K represents the number of inventory behavior fragments contained in the inventory behavior memory. This indicates the degree of continuous change between the (k-1)th inventory behavior segment and the kth inventory behavior segment; It should be noted that the degree of continuous change is considered in conjunction with three factors: whether the direction of inventory behavior changes is consistent, that is, whether the inventory changes are all showing a continuous increase or a continuous decrease; whether the magnitude of adjacent inventory behavior changes maintains a stable increase or a stable decrease, so as to avoid misjudgment of continuity due to occasional inventory adjustments; and whether the time interval between adjacent inventory behaviors is within the normal inventory business cycle. When the time span is too large, it is considered that there is no continuous evolution relationship between inventory behaviors. When the inventory behavior continuity indicator reaches the set requirements, it indicates that the inventory behavior has formed a continuous evolution process; Subsequently, a reverse search was conducted along the direction of inventory behavior evolution to locate the position where continuous evolution began, and the inventory event that first caused continuous changes in inventory behavior was identified as the source of demand disturbance. Demand disturbances include not only increased orders, but also actual business events such as production plan adjustments, increased demand for spare parts due to equipment failures, delayed replenishment caused by supply delays, and changes in regional inventory due to warehousing and allocation. These factors can more accurately reflect the reasons for the formation of industrial supplies inventory demand. S23. Starting from the source of demand disturbance, establish inventory behavior relationships by combining inventory flow relationships, production work order relationships, warehousing and allocation relationships, and process matching relationships. Calculate the propagation intensity of demand disturbances between each inventory behavior node, and dynamically generate a demand disturbance propagation chain according to the actual evolution sequence of inventory business. The specific implementation process is as follows: After identifying the sources of demand disturbances, the cloud platform further analyzes the diffusion process of inventory demand throughout the entire inventory network. Because industrial products typically have relationships such as process compatibility, assembly, spare parts substitution, and warehousing and allocation, a change in the inventory of one industrial product often triggers a chain reaction of changes in the inventory of other related industrial products. This embodiment restores the actual propagation path of inventory demand within the supply chain based on inventory behavior memory. Inventory behavior relationships are established based on the inventory flow information recorded in the inventory behavior memory. These relationships include shared material requisition for the same production work order, shared consumption for the same equipment, shared allocation for the same storage area, and continuous material supply for upstream and downstream processes. For any two inventory behavior nodes, define the demand disturbance propagation strength (the greater the propagation strength, the more stable the inventory demand propagation relationship between the nodes): ; In the formula, This indicates the propagation strength of the demand disturbance from node i to node j; Indicator representing the continuity of inventory behavior at node i; This represents the correlation coefficient of inventory operations, which is obtained based on the historical number of inventory transfers, the number of times they jointly participated in production tasks, and the number of times they jointly consumed inventory. This represents the time propagation preservation coefficient, used to reflect whether the propagation of inventory behavior has temporal continuity; Taking the source of demand disturbance as the starting point of propagation, propagation connections are established sequentially for all inventory behavior nodes that meet the propagation conditions, and a demand disturbance propagation chain is generated according to the actual order in which inventory behavior occurs. It should be noted that the propagation chain established in this embodiment is automatically adjusted as the inventory behavior memory is continuously updated. When new changes occur in the inventory behavior, the propagation chain can automatically add new propagation nodes, adjust the propagation order, or delete invalid propagation relationships, so that the propagation chain can always truly reflect the evolution of industrial product inventory demand in the supply chain. S24. Based on the consistency score of comprehensive inventory behavior, the intensity of demand disturbance propagation, and historical propagation verification results, the credibility of each propagation path in the demand disturbance propagation chain is calculated and corrected. Credible propagation relationships are retained, and the paths to be verified are continuously tracked and updated. Finally, a structured demand disturbance propagation chain containing information on propagation source, propagation direction, propagation level, propagation intensity, and credibility is output. The specific implementation process is as follows: To ensure that the demand disturbance propagation chain can accurately reflect the inventory demand propagation process, the credibility of the demand disturbance propagation chain is further corrected. For each propagation path in the propagation chain, the propagation credibility is obtained by combining the consistency score of inventory behavior, propagation strength, and historical propagation verification results, and then normalizing the results. ; In the formula, Indicates the credibility of the dissemination; This represents the historical propagation verification success rate, which is obtained by statistically analyzing whether node j experiences a corresponding inventory change within a reasonable time after node i undergoes an inventory change during the historical operation process. , and These represent the weight coefficients corresponding to inventory behavior consistency, propagation strength, and historical propagation verification results, respectively. Each weight is calibrated offline using historical inventory scheduling data, and the sum of the weights is 1. Regarding the credibility of dissemination Propagation paths that meet the set requirements will be retained; propagation paths with low credibility will not be deleted directly, but will be temporarily marked as pending verification; when new inventory behavior memory fragments are uploaded to the cloud, they will be re-verified in conjunction with the new inventory behavior to avoid frequent changes in the propagation chain due to short-term inventory anomalies. Output a structured demand disturbance propagation chain. Each propagation chain includes the demand disturbance source, propagation node sequence, propagation direction, propagation level, propagation intensity, propagation credibility, and corresponding industrial product identification information.

[0021] In a specific embodiment, step S3, based on the output of step S2, further combines the operational status of the inventory nodes themselves, the impact of demand disturbance propagation, and inventory recovery capabilities to establish an inventory tension state expression mechanism. This mechanism quantifies and integrates the demand pressures dispersed across different inventory nodes, forming an inventory tension state field that can describe the distribution of inventory pressure. Furthermore, it infers the future direction of demand evolution based on the trend of inventory tension changes. The specific implementation steps include: S31. Based on the demand disturbance propagation strength, propagation credibility, and inventory behavior continuity indicators output from step S2, and combined with the replenishment response capability, allocation capability, and alternative supply capability of inventory nodes, an inventory node tension calculation mechanism is established. Supply and demand pressure is quantified for each industrial product inventory node, forming a node tension value reflecting the current inventory risk level. This realizes the transformation from inventory quantity state to inventory pressure state. The specific implementation process is as follows: After generating the demand disturbance propagation chain in step S2, the cloud first calculates the inventory pressure status for each inventory node. It should be noted that the inventory status of industrial supplies cannot be judged solely by the current inventory quantity. For example, for a certain type of critical equipment maintenance spare parts, even if the current inventory quantity meets production needs, if the failure frequency of the corresponding equipment increases, the consumption trend continues to rise, and the supply cycle is long, then that inventory point may still quickly enter a risky state. Therefore, this embodiment comprehensively considers demand sustainability, the impact of demand propagation, and inventory recovery capabilities to construct an inventory node tension value; For any inventory node i, define the inventory node tension value: ; In the formula, This represents the current inventory tension value corresponding to inventory node i; This indicates the combined impact strength of demand disturbances, used to characterize the degree to which the current node is affected by changes in demand from other inventory nodes; The inventory recovery capability coefficient describes the ability of an inventory node to return to a normal state through replenishment, allocation, or replacement of inventory. It should be noted that the overall impact intensity of demand disturbances... Based on the demand perturbation propagation chain generated in step S2, the following is calculated: ; In the formula, This represents the number of associated nodes that can influence demand for inventory node i. This indicates the strength of the demand disturbance propagation from inventory node j to inventory node i; This indicates the strength of the demand disturbance propagation from inventory node j to inventory node i; It should be further explained that the inventory recovery capability coefficient Determined based on actual inventory business data, including but not limited to: historical average replenishment response time, supplier delivery cycle, similar inventory substitution capability, and cross-warehouse transfer capability; Specifically, when inventory replenishment is fast and alternative resources are plentiful, An increase indicates that the inventory node has strong recovery capabilities and its inventory tension is reduced; when the inventory node has a long supply cycle and weak substitution ability, ... As it decreases, its inventory tension increases; S32. Utilizing the node relationships and propagation credibility in the demand disturbance propagation chain, the tension of each inventory node is correlated and integrated, so that each inventory node is jointly affected by its own inventory status and the demand changes of related nodes, thus constructing an inventory tension state field covering the entire inventory network. The specific implementation process is as follows: Because there are production, supply and logistics links among industrial supplies inventory, a change in demand at one inventory node will spread to other nodes along the demand disturbance propagation chain formed in step S2. Therefore, this embodiment further utilizes the demand disturbance propagation chain to model the tension influence relationship between each inventory node, so that a dynamically related inventory tension state field is formed between the inventory nodes. For inventory node i, define its overall inventory tension state: ; In the formula, This represents the comprehensive tension state value of inventory node i in the inventory tension state field; This indicates the underlying tension generated by the inventory node itself; This represents the demand disturbance propagation attenuation coefficient, used to describe the residual strength of the demand after it has been affected by distance, time, and business relevance during the propagation process from node j to node i. It should be noted that the demand disturbance propagation attenuation coefficient The following factors are considered: the logistical distance between the two inventory nodes; the business hierarchy between the two inventory nodes; and the time interval after demand propagation occurs. For example, material demand propagation distances between adjacent workstations on the same production line are short and business connections are strong. The ratio is relatively high; while there are only occasional transfers between cross-regional warehouses, and their... Lower; In this way, the system not only considers the pressure of the inventory node itself, but also the impact of the pressure of the surrounding inventory nodes, so that the inventory tension can be dynamically transmitted along the demand disturbance propagation chain. Overall inventory tension status across all inventory nodes Together they constitute the inventory tension state field: ; In the formula, This represents the inventory tension state field corresponding to the entire inventory network; This indicates the number of inventory nodes in the inventory network; It should be noted that the inventory tension field can reflect: the current location of concentrated inventory pressure; the direction of inventory pressure propagation; and high-risk inventory-related areas. S33. Based on the inventory tension state field analysis, the pressure change trend of each inventory node is analyzed. By calculating the direction and growth rate of tension changes between adjacent nodes, the possible future diffusion direction of demand disturbances is identified, and the inventory demand evolution path is generated according to the propagation chain sequence to predict the development trend of potential inventory pressure. The specific implementation process is as follows: After completing the construction of the inventory tension state field, further analysis is conducted on the development trend of inventory pressure. This embodiment uses the dynamic changes in the inventory tension state field to predict the future development direction of inventory demand; Specifically, the system analyzes the changes in inventory tension between adjacent inventory nodes according to the demand disturbance propagation chain generated in step S2; Define the change in inventory tension: ; In the formula, This represents the change in tension during the transfer of inventory pressure from node i to node j; and These represent the current overall inventory tension status of the two related inventory nodes; when At that time, the current inventory pressure at node j is higher than that at node i, and the demand disturbance may continue to spread towards node j; Further combining the tension change trends over multiple consecutive time periods, the inventory tension growth trend is calculated: ; In the formula, This represents the rate of increase in tension at inventory node i; Indicates the current inventory tension; This indicates the inventory tension from the previous analysis period; Indicates the analysis period; If a certain inventory node meets the following conditions: tension value continues to increase, tension growth rate remains positive, and it is located downstream of the demand disturbance propagation chain, then the node is determined to be a potential demand diffusion target. Based on the above judgment results, the continuously affected inventory nodes are connected in the order of propagation to generate a demand evolution path; The demand evolution path includes: the starting point of demand disturbance, intermediate propagation points, potential pressure concentration points, and the expected time range of impact; This path can anticipate the potential development of future inventory pressure, enabling the scheduling system to intervene before actual inventory shortages occur. S34. Based on the node pressure levels, growth trends, and demand propagation stability in the inventory tension state field, regionalize continuously high-pressure inventory nodes, calculate the inventory risk index and generate risk levels, and output the industrial products involved and the scope of impact. The specific implementation process is as follows: This embodiment further identifies inventory areas that need to be prioritized based on the inventory tension state field and demand evolution path; For each inventory node in the inventory tension state field, perform regional aggregation. When multiple adjacent inventory nodes simultaneously meet the following conditions: the overall inventory tension is higher than the preset threshold, the tension growth trend continues to increase, and there is a stable demand propagation path connecting them, then the above nodes are divided into the same inventory tension abnormal region. For each abnormal region, calculate the regional risk index: ; In the formula, This indicates the risk index for an abnormal inventory area; num represents the number of inventory nodes included in this abnormal area. This represents the average credibility of demand dissemination in the region. Based on the risk index and preset thresholds, abnormal inventory areas are classified into different levels: High-risk areas: Inventory pressure continues to increase while demand transmission remains stable; Medium-risk areas: Inventory pressure is increasing, but there is room for recovery; Low-risk areas: Inventory pressure fluctuates in the short term but has not formed a sustained trend; Final output: Location of the abnormal area, types of industrial products involved, risk level, and direction of pressure propagation.

[0022] In a specific embodiment, step S4, based on the inventory tension state field, demand evolution path, and inventory anomaly area information output in step S3, combined with the real-time warehousing execution status, inventory availability status, and logistics operation status collected from the edge, performs global scheduling decisions in the cloud and performs on-site constraint correction through the edge terminal. The specific implementation steps include: S41. Based on the abnormal inventory tension areas, risk index, and demand evolution trend output in step S3, a comprehensive evaluation is conducted on different inventory areas. By integrating inventory pressure levels, tension growth rates, demand propagation range, and the criticality of industrial products, scheduling priorities are calculated to determine the inventory nodes requiring priority intervention and resource scheduling targets. The specific implementation process is as follows: The abnormal inventory area output in step S3 includes multiple industrial product nodes with an accumulation trend of inventory pressure. However, different areas have different degrees of impact on the production process. Therefore, it is necessary to first determine the scheduling objects that need to be prioritized based on the evolution of inventory tension. In actual industrial production environments, some inventory nodes may have low current inventory levels, but their demand is stable and their replenishment cycle is short, so they will not immediately affect production. On the other hand, some inventory nodes may not have reached the point of being out of stock, but their associated production tasks are intensive, their demand is growing rapidly, and their supply recovery cycle is long. Once inventory declines occur, they will directly affect the continuity of production. Therefore, this embodiment comprehensively considers the inventory pressure level, the rate of pressure growth, and the scope of the impact of demand propagation to dynamically prioritize areas with abnormal inventory. For the i-th inventory anomaly region, define a scheduling priority evaluation value: ; In the formula, This represents the scheduling priority evaluation value of the i-th inventory anomaly region; The demand propagation impact range coefficient is used to indicate the number of production nodes that may be affected by this inventory area; The criticality coefficient of industrial supplies indicates the importance of the corresponding industrial supplies in this inventory area to the production task. Among them, the influence range coefficient of demand propagation Based on the demand disturbance propagation chain formed in step S2, the following is determined: In the formula, This indicates the number of associated inventory nodes that can be affected by the abnormal region; This indicates the total number of nodes in the inventory network; Keyness coefficient of industrial products Determined based on the historical consumption frequency of industrial products, the degree of production linkage, and the difficulty of substitution; For example, for ordinary consumable materials, there are many alternative sources, so the degree of criticality is low; for spare parts for special equipment, the replacement cycle is long and affects the operation of key production equipment, so the degree of criticality is high. Based on the above calculations, the following inventory areas are prioritized for scheduling: those with high current inventory pressure, rapid risk growth, large impact range, and significant impact on production. S42. Based on the evolution path of inventory demand and the resource status of inventory nodes, analyze the resource supply capacity and scheduling adaptability between different inventory nodes. Combining the reliability of demand propagation, inventory adjustability, and scheduling execution conditions, generate a dynamic scheduling scheme that includes the outgoing node, the incoming node, the allocation quantity, and the execution time. This achieves proactive matching of inventory resources with demand pressure areas. The specific implementation process is as follows: After determining the scheduling target, the system further dynamically matches the inventory resources according to the demand evolution path generated in step S3; Traditional inventory scheduling typically selects allocation sources based on fixed rules such as proximity and maximum inventory quantity. For example, even if a nearby inventory node currently has sufficient inventory, if it is also in a phase of rapid inventory strain, allocating from that node may lead to new inventory risks. This embodiment selects scheduling resources that can reduce overall inventory pressure based on the evolution trend of inventory tension and the direction of demand propagation. When a candidate inventory node j allocates resources to a risk node i, a scheduling matching evaluation value is defined as follows: ; ; In the formula, This represents the inventory schedulable capacity coefficient of inventory node j, which is the proportion of inventory resources that the current inventory node can actually use for allocation. This indicates the actual amount of inventory that can be allocated, i.e., the amount of available inventory after deducting locked inventory, safety stock, and allocated inventory; This indicates the current total inventory quantity at the inventory node; This indicates the feasibility of scheduling between nodes, and is calculated based on: the logistics distance between nodes, the current status of transportation resources, warehousing capacity, and the estimated allocation time. This indicates the degree of scheduling matching between inventory node j and inventory risk node i; when The closer the value is to 1, the more abundant the inventory resources and the stronger the ability to allocate them; when... The closer it is to 0, the more likely it is that most of the inventory at the inventory node has been occupied or reserved, and it is no longer suitable to continue as a node to be transferred out. Meanwhile, to avoid transferring inventory pressure from one region to another, the system performs tension constraint judgment on candidate transfer nodes: after the transfer is completed: If so, then the node is considered to have the ability to allocate resources; In the formula, This indicates the predicted inventory tension at node j after the allocation is completed; This indicates the maximum allowable inventory tension threshold; Based on the scheduling matching evaluation value, select inventory resources that meet the constraints and generate a dynamic scheduling scheme; The scheduling plan should include at least: the outgoing inventory node, the incoming inventory node, the category of industrial supplies, the quantity to be transferred, the transfer time window, and the execution priority level; S43. Utilize an edge-cloud collaborative mechanism to execute a dynamic scheduling scheme. The cloud is responsible for scheduling planning based on the global inventory status, while the edge, combined with real-time data collected from the warehouse site on inventory status, equipment status, and operational constraints, makes local adjustments to the scheduling scheme. By correcting the scheduling path, resource source, and execution order in real time, the adaptability of the scheduling scheme in actual industrial environments is improved. The specific implementation process is as follows: Because industrial warehousing sites are characterized by dynamic changes, such as: temporary occupation of AGV equipment, changes in warehouse workers, changes in storage location status, and insertion of temporary production tasks; Therefore, the scheduling strategy generated in the cloud cannot be directly used as the final execution instruction and needs to be adjusted by the end side according to the real-time status of the site. This embodiment adopts an end-to-cloud collaborative execution method; The cloud primarily accomplishes: determining the scheduling direction based on the inventory tension state field, determining the resource flow direction based on the demand evolution path, and generating an initial scheduling plan based on the global inventory status. The edge terminal mainly performs the following tasks: obtaining real-time inventory status, obtaining the operating status of field equipment, determining whether the current scheduled task meets the execution conditions, and adjusting the local execution order. For cloud-generated scheduling schemes The endpoint generates the final execution plan based on the real-time status: ; Correction amount This can occur under the following circumstances: the actual inventory quantity is inconsistent with the cloud record, the target storage location cannot currently perform storage or retrieval operations, the transportation equipment is unavailable, or a new production task changes the demand priority. For example: The cloud plans to transfer a certain industrial product from warehouse A to production line B, but the terminal detects that the equipment in warehouse A is currently under maintenance and cannot be shipped out. In this case, the terminal readjusts the transfer source based on the backup inventory node and feeds the adjustment result back to the cloud. By making real-time corrections at the endpoint, the dynamic scheduling strategy can adapt to real industrial environments. Output the final execution scheduling instruction and on-site execution status information; S44. Collect actual execution data during the inventory scheduling process, including allocation results, completion time, inventory changes, and demand response status. Calculate the scheduling execution deviation. When there is a significant deviation between the actual results and the plan, update the inventory behavior memory weight, demand propagation reliability, and inventory tension status. This allows the inventory forecasting and dynamic scheduling model to be continuously corrected as the industrial production process changes, achieving closed-loop optimization. The specific implementation process is as follows: After the scheduling is completed, the system needs to determine whether the scheduling strategy is effective based on the actual execution results and update the inventory status model using the feedback results. This embodiment establishes a scheduling feedback mechanism to create a closed loop between inventory forecasting, risk identification, and scheduling processes; Continuous data collection on the device side includes: actual allocation quantity, actual completion time, actual inventory changes, production demand response status, and reasons for scheduling deviations, which are then uploaded to the cloud. Deviation between planned and actual results of cloud computing: ; when This indicates a discrepancy between the current inventory behavior model or demand propagation model and the actual operating status. In the formula, Indicates the deviation in scheduling execution; Indicates the planned scheduling results; This indicates the actual scheduling result; Indicates the preset deviation threshold; At this point, the cloud-based execution model is corrected: the weights for fusion of inventory behavior memory fragments in step S2 are adjusted. Correct the reliability of demand disturbance propagation; update the inventory tension state field in step S3; For example, if the actual consumption of a certain industrial product exceeds the forecast result multiple times in a row, the importance of the corresponding inventory behavior memory segment will be increased, so that subsequent inventory tension calculations will pay more attention to such behavior changes. Ultimately, this enables the inventory scheduling system to continuously optimize as industrial production processes change.

[0023] Example 2, as Figure 2As shown, the present invention proposes an intelligent forecasting and dynamic scheduling system for industrial supplies inventory demand, which is used to execute the intelligent forecasting and dynamic scheduling method for industrial supplies inventory demand proposed in Embodiment 1. It includes: a multi-source inventory behavior perception and event structured modeling module on the edge side, an inventory behavior memory fusion and demand disturbance propagation analysis module, an inventory tension state field construction and demand evolution deduction module, and an edge-cloud collaborative dynamic scheduling and closed-loop optimization module.

[0024] The edge-side multi-source inventory behavior perception and event structured modeling module is used to collect and preprocess multi-source data such as inventory changes, material requisition, replenishment, transfer, inventory count and production consumption in the entire process of industrial supplies inventory business. It constructs inventory behavior events and forms inventory behavior memory fragments, providing unified and standardized structured input data for subsequent inventory behavior analysis. The Inventory Behavior Memory Fusion and Demand Disturbance Propagation Analysis Module is used to fuse inventory behavior memory fragments, construct an inventory behavior memory body, generate a demand disturbance propagation chain based on the temporal and business relationships between inventory behaviors, calculate the demand propagation strength and propagation credibility, and identify the inventory demand formation process and its propagation relationship. The module for constructing the inventory tension state field and extrapolating the demand evolution is used to calculate the inventory node tension based on the inventory behavior memory and the demand disturbance propagation chain, construct the inventory tension state field, extrapolate the inventory demand evolution path, identify abnormal areas of inventory tension, and output the inventory risk level and demand development trend, providing a basis for inventory scheduling decisions. The edge-cloud collaborative dynamic scheduling and closed-loop optimization module is used to generate dynamic inventory scheduling strategies based on the inventory tension state field, demand evolution path and inventory risk level. It combines the real-time warehouse status on the edge side to complete the scheduling execution and local correction, and updates the inventory behavior memory, demand propagation relationship and inventory tension state based on the execution feedback, so as to realize the closed-loop optimization of inventory demand forecasting and dynamic scheduling.

[0025] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method for intelligent forecasting and dynamic scheduling of industrial goods inventory demand, characterized in that, The specific implementation steps include the following: S1. At the edge computing node on the edge side, perform synchronous perception and time alignment processing on multi-source heterogeneous data related to industrial supplies inventory. Then, through the behavior intensity driving mechanism, transform continuous inventory data into discrete behavior event units. At the same time, introduce behavior consistency constraints to achieve event credibility screening and vectorized expression. Finally, perform time-weighted compression on the event sequence to form inventory behavior memory fragments and upload them to the cloud. S2. In the cloud, the inventory behavior of different warehousing nodes, production nodes and logistics nodes is unified and integrated to build an inventory behavior memory that can be continuously updated. By analyzing the continuous evolution characteristics of inventory behavior, the source of demand disturbance is identified. Furthermore, the demand disturbance propagation chain is established by combining the relationship between inventory business. The credibility of the propagation path is corrected and dynamically modified, and finally a structured demand disturbance propagation chain is output. S3. By integrating the persistence of inventory behavior, the degree of impact of demand propagation, and inventory recovery capability, tension is calculated for each inventory node, and the global inventory pressure distribution is formed by combining the propagation relationship between nodes. By analyzing the trend of inventory tension changes, the demand evolution path is deduced, potential areas of concentrated inventory pressure are identified, and risk levels are generated. S4. Construct a dynamic scheduling mechanism for end-to-cloud collaboration based on the constraints of inventory pressure evolution. Determine scheduling priorities by analyzing the urgency of inventory risk areas, generate dynamic scheduling schemes by combining demand propagation direction, inventory resource status and on-site execution conditions, and use the real-time status of the end side to correct the cloud scheduling strategy. At the same time, collect scheduling execution results and update the inventory behavior memory, demand propagation relationship and inventory tension status.

2. The intelligent forecasting and dynamic scheduling method for industrial supplies inventory demand according to claim 1, characterized in that, Step S1 specifically includes: Multi-source heterogeneous data such as inventory inbound and outbound, equipment status, production consumption and order triggering are collected on the edge. Time drift correction is performed on each data source through a unified time base, and local smoothing is performed on the data in combination with a sliding window to eliminate the impact of sampling frequency differences and instantaneous noise. A comprehensive behavior intensity function is constructed to characterize the overall change range of the inventory system through weighted fusion of multiple types of data, and a dynamic threshold triggering mechanism is constructed using the behavior intensity difference between adjacent time points. When the change exceeds the threshold, an inventory behavior event unit is generated. The generated inventory behavior event units are introduced with expected action values ​​inferred from historical behavior patterns. A behavior consistency scoring function is constructed to measure the degree of deviation between the actual execution behavior and historical patterns. Event triggering sources, execution actions, result feedback and consistency scores are mapped together into a unified low-dimensional vector representation. The vectors of inventory behavior events that are continuous in time are aggregated into behavior memory segments by time segment, and time decay weights are introduced to compress the events within the segments to generate a compact inventory behavior memory representation. Only the compressed result is uploaded to the cloud.

3. The intelligent forecasting and dynamic scheduling method for industrial supplies inventory demand according to claim 2, characterized in that, The specific process of correcting time drift for each data source using a unified time base and performing local smoothing of the data using a sliding window is as follows: A unified time reference unit is established in the edge computing node, using the local high-precision clock or the industrial gateway master clock as a reference, and drift correction is performed on each data source through periodic time calibration packages; after time alignment, a sliding stable window is introduced for local smoothing, and the window length is determined based on the average response cycle of inventory operations.

4. The intelligent forecasting and dynamic scheduling method for industrial supplies inventory demand according to claim 3, characterized in that, The specific process of constructing a comprehensive behavioral intensity function and characterizing the overall change magnitude of the inventory system through weighted fusion of multiple types of data is as follows: Construct a comprehensive behavior intensity function, and use multi-source data weighted summation to calculate the overall activity level of the inventory system at the current moment. The behavioral contribution weight of each data type is obtained by statistical analysis of historical inventory consumption and business correlation. The dynamic event triggering threshold used in the dynamic threshold triggering mechanism is adaptively determined by performing quantile statistics on the intensity distribution of historical inventory behavior.

5. The intelligent forecasting and dynamic scheduling method for industrial supplies inventory demand according to claim 4, characterized in that, The specific process of constructing a behavioral consistency scoring function to measure the degree of deviation between actual performance behavior and historical patterns is as follows: Reference actions are established using historical similar inventory behaviors, and the current inventory behavior is compared with the reference actions. The consistency score is calculated based on the ratio of the execution intensity corresponding to the actual action executed in the current event to the reference execution intensity obtained from the statistics of historical similar inventory behaviors. The inventory behavior events are further encoded to form an inventory behavior vector, which consists of the event source code, the action code, and the result code.

6. The intelligent forecasting and dynamic scheduling method for industrial supplies inventory demand according to claim 5, characterized in that, The specific process of introducing time decay weight to weight and compress events within a segment is as follows: construct an inventory behavior memory segment for inventory behavior event vectors that occur consecutively within the same business cycle, and calculate the memory segment expression using the time decay method, wherein the time decay coefficient is set according to the average turnover cycle of industrial supplies inventory; Edge nodes only upload compressed inventory behavior memory fragments and necessary index information to the cloud.

7. The intelligent forecasting and dynamic scheduling method for industrial supplies inventory demand according to claim 6, characterized in that, Step S2 specifically includes: The cloud receives inventory behavior memory fragments uploaded by each edge node, collects and sorts them uniformly according to industrial product identification, business type, time sequence and inventory flow relationship, and performs weighted fusion based on the relevance of inventory behavior, consistency of behavior and business continuity to construct an inventory behavior memory that fully records the historical evolution of industrial product inventory. Based on the memory of inventory behavior, the direction, magnitude and duration of changes of adjacent inventory behaviors are analyzed according to the order of occurrence of inventory behaviors. The continuity index of inventory behavior is calculated, and the continuous evolution trend is identified accordingly. Furthermore, the initial event that triggered the continuous inventory change is located in reverse and identified as the source of demand disturbance. Starting from the source of demand disturbance, we establish inventory behavior relationships by combining inventory flow relationships, production work order relationships, warehousing and allocation relationships and process matching relationships, calculate the propagation intensity of demand disturbance between each inventory behavior node, and dynamically generate the demand disturbance propagation chain according to the actual evolution sequence of inventory business. By combining the consistency score of inventory behavior, the intensity of demand disturbance propagation, and the historical propagation verification results, the credibility of each propagation path in the demand disturbance propagation chain is calculated and corrected. Credible propagation relationships are preserved and the paths to be verified are continuously tracked and updated. Finally, a structured demand disturbance propagation chain containing information on propagation source, propagation direction, propagation level, propagation intensity, and credibility is output.

8. The intelligent forecasting and dynamic scheduling method for industrial supplies inventory demand according to claim 7, characterized in that, Step S3 specifically includes: Based on the intensity and credibility of demand disturbance propagation and the continuity of inventory behavior indicators, combined with the replenishment response capability, allocation capability and alternative supply capability of inventory nodes, an inventory node tension calculation mechanism is established to quantify the supply and demand pressure of each industrial product inventory node and form a node tension value that reflects the current inventory risk level. By leveraging the node relationships and propagation credibility in the demand disturbance propagation chain, the tension of each inventory node is correlated and integrated, so that each inventory node is jointly affected by its own inventory status and the demand changes of related nodes, thus constructing an inventory tension state field covering the entire inventory network. Based on the inventory tension state field analysis, the pressure change trend of each inventory node is analyzed. By calculating the direction and growth rate of tension change between adjacent nodes, the possible future diffusion direction of demand disturbance is identified, and the inventory demand evolution path is generated according to the propagation chain sequence. Based on the node pressure level, growth trend, and demand propagation stability in the inventory tension state field, continuous high-pressure inventory nodes are regionalized, the inventory risk index is calculated and the risk level is generated, and the industrial products involved and the scope of impact are output.

9. The intelligent forecasting and dynamic scheduling method for industrial supplies inventory demand according to claim 8, characterized in that, Step S4 specifically includes: Based on abnormal inventory tension areas, risk indices, and demand evolution trends, a comprehensive evaluation of different inventory areas is conducted. By integrating inventory pressure levels, tension growth rates, demand propagation ranges, and the criticality of industrial products, scheduling priorities are calculated to determine the inventory nodes requiring priority intervention and resource scheduling targets. Based on the evolution path of inventory demand and the resource status of inventory nodes, the resource supply capacity and scheduling adaptability between different inventory nodes are analyzed. Combining the reliability of demand propagation, inventory adjustability and scheduling execution conditions, a dynamic scheduling scheme including the outgoing node, the incoming node, the allocation quantity and the execution time is generated. The dynamic scheduling scheme is executed by using an edge-cloud collaboration mechanism. The cloud is responsible for scheduling planning based on the global inventory status, while the edge combines the real-time inventory status, equipment status and operation constraints collected from the warehouse site to make local adjustments to the scheduling scheme. By correcting the scheduling path, resource source and execution order in real time, the adaptability of the scheduling scheme in the actual industrial environment is improved. Collect actual execution data during the inventory scheduling process, including allocation results, completion time, inventory changes and demand response, and calculate scheduling execution deviation. When there is a large deviation between the actual results and the plan, update the inventory behavior memory weight, demand propagation credibility and inventory tension status to complete closed-loop optimization.

10. An intelligent forecasting and dynamic scheduling system for industrial supplies inventory demand, used to execute the intelligent forecasting and dynamic scheduling method for industrial supplies inventory demand as described in any one of claims 1 to 9, characterized in that, include: The edge-side multi-source inventory behavior perception and event structured modeling module is used to collect and preprocess multi-source data on inventory changes, material requisition, replenishment, transfer, inventory count and production consumption in the entire process of industrial supplies inventory business. It constructs inventory behavior events and forms inventory behavior memory fragments, providing unified and standardized structured input data for subsequent inventory behavior analysis. The Inventory Behavior Memory Fusion and Demand Disturbance Propagation Analysis Module is used to fuse inventory behavior memory fragments, construct an inventory behavior memory body, generate a demand disturbance propagation chain based on the temporal and business relationships between inventory behaviors, calculate the demand propagation strength and propagation credibility, and identify the inventory demand formation process and its propagation relationship. The module for constructing the inventory tension state field and extrapolating the demand evolution is used to calculate the inventory node tension based on the inventory behavior memory and the demand disturbance propagation chain, construct the inventory tension state field, extrapolate the inventory demand evolution path, identify abnormal areas of inventory tension, and output the inventory risk level and demand development trend, providing a basis for inventory scheduling decisions. The edge-cloud collaborative dynamic scheduling and closed-loop optimization module is used to generate dynamic inventory scheduling strategies based on the inventory tension state field, demand evolution path and inventory risk level. It combines the real-time warehouse status on the edge side to complete the scheduling execution and local correction, and updates the inventory behavior memory, demand propagation relationship and inventory tension state based on the execution feedback, so as to realize the closed-loop optimization of inventory demand forecasting and dynamic scheduling.

Images