An industry-level multi-agent-based manufacturing channel coordination decision-making method

Abstract

This invention discloses a manufacturing channel collaborative decision-making method based on industrial-grade multi-agent systems, relating to the fields of intelligent decision-making and multi-agent collaborative control in manufacturing. The method includes collecting internal data from manufacturing enterprises, performing time alignment and consistency checks to generate a collaborative basic dataset; constructing a collaborative semantic structure reflecting the relationships between products, channels, contracts, and fulfillment, and outputting channel collaborative state description data; multiple business agents extracting demand fluctuations, inventory pressure, and fulfillment risk states based on the channel collaborative state description data, and converting them into channel constraints and production / supply constraints to form a collaborative strategy input set; performing parallel inference and evaluation of the collaborative strategy input set in a digital twin environment, selecting strategy results that meet preset stability and consistency conditions, and outputting channel allocation and production scheduling decisions. This invention significantly reduces actual execution risks and improves order fulfillment rates, inventory turnover efficiency, and fulfillment stability.

Description

Technical Field

[0001] This invention relates to the field of intelligent decision-making and multi-agent collaborative control technology in manufacturing, and in particular to a manufacturing channel collaborative decision-making method based on industrial-level multi-agent systems. Background Technology

[0002] With the continuous advancement of digital transformation and intelligent upgrading in the manufacturing industry, the business operations of enterprises, including production, supply, sales, and fulfillment, are exhibiting highly networked, dynamic, and complexly coupled characteristics. Especially under multi-channel sales models and personalized order-driven production systems, manufacturing enterprises urgently need to use intelligent decision-making methods to collaboratively optimize the allocation of product, channel, inventory, and fulfillment resources. In recent years, multi-agent technology, digital twin technology, and semantic modeling methods have been gradually applied to the field of business collaboration and decision support in the manufacturing industry. This enables enterprises to deduce and evaluate complex business strategies in a virtual environment, thereby improving the rationality of production scheduling and the efficiency of channel resource allocation, becoming an important development direction for intelligent decision-making systems in the manufacturing industry.

[0003] CN114610889B discloses a method for constructing a semantic network for a multi-value chain collaborative data space in the manufacturing industry. By constructing a semantic network for this data space, it enables semantic representation and structured correlation analysis of multi-source manufacturing data. The key focus of this technology is on ontology modeling, semantic similarity calculation, and semantic network structure optimization for cross-value chain data resources, thereby identifying key nodes and evaluating the consistency and completeness of the multi-value chain semantic network. While this method is beneficial for improving the organization and knowledge discovery capabilities of multi-source data in the manufacturing industry, it primarily emphasizes the construction of static semantic networks and indicator analysis. It lacks a closed-loop decision-making mechanism for actual production scheduling and channel allocation, and has not yet formed a dynamic decision-making process that directly transforms semantic analysis results into executable collaborative strategies. This makes it difficult to support real-time simulation and optimization of multi-channel supply, production, and sales collaboration in complex scenarios.

[0004] CN120764548A discloses a knowledge retrieval-enhanced method for collaborative knowledge extraction across multiple value chains in the manufacturing industry. It uses a multi-agent dynamic execution graph and a large language model to parse, reason, and express business knowledge across multiple stages using natural language. While this technology emphasizes intelligent collaboration and knowledge generation capabilities for knowledge retrieval and business analysis scenarios, its core functions are concentrated on cross-stage knowledge extraction, logical verification, and natural language summarization. This makes it difficult to support manufacturing enterprises in quantitatively evaluating and updating their production, supply, and sales collaboration strategies in a closed-loop manner.

[0005] In summary, existing technologies for constructing collaborative semantic networks across multiple value chains in the manufacturing industry and for extracting knowledge from multiple agents generally suffer from an overemphasis on knowledge modeling and analysis and a neglect of decision-making deduction and execution loops. These technologies are insufficient to meet the collaborative scheduling and resource allocation decision-making needs under multiple channels, constraints, and objectives. Summary of the Invention

[0006] In view of the problems existing in the construction technology of multi-value chain collaborative semantic network in manufacturing and multi-agent knowledge extraction technology, this invention is proposed.

[0007] Therefore, the problem to be solved by this invention is how to achieve stability screening and closed-loop updating of channel allocation and production scheduling strategies.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0009] In a first aspect, the present invention provides a manufacturing channel collaborative decision-making method based on industrial-level multi-agent systems, comprising: S1: collecting production, inventory, and order data within the manufacturing enterprise, as well as fulfillment and demand data from various sales channels, and performing time alignment and consistency verification to generate a collaborative basic dataset; S2: based on the collaborative basic dataset, constructing a collaborative semantic structure reflecting the relationship between products, channels, contracts, and fulfillment, and outputting channel collaborative status description data; S3: multiple business agents extracting demand fluctuations, inventory pressure, and fulfillment risk status based on the channel collaborative status description data, and converting the corresponding business rules into channel constraints and production and supply constraints to form a collaborative strategy input set; S4: performing parallel inference and evaluation on the collaborative strategy input set in a digital twin environment, screening strategy results that meet preset conditions for stability and consistency, outputting channel allocation and production scheduling decisions, and feeding the strategy results back to the collaborative basic dataset to update subsequent decision-making basis.

[0010] As a preferred embodiment of the manufacturing channel collaborative decision-making method based on industrial-level multi-agent as described in this invention, step S1 includes: acquiring production work order records, warehouse inbound and outbound flow data, sales order details, and fulfillment status logs from various channel portals in the manufacturing information system, and using these as source data nodes; attaching an event timestamp accurate to the second and a globally unique data identifier to each source data node; based on the event timestamp, uniformly resampling the inventory level data with continuous numerical attributes to a time grid with a period of n minutes using cubic spline interpolation; for discrete event-driven production completion data, accumulating and statistically analyzing the number of completion events within each time grid using event aggregation; retaining all records for sales order detail data; and specifying the globally unique data identifier. The sequence is: Entity Type - Entity ID - Operation Sequence; For data located within the same time grid, the inventory level data and production completion data of the same product material are associated based on the Entity ID field; Simultaneously, for sales order details and fulfillment status logs, association matching is performed based on the order number and Entity ID, and the timestamp of the fulfillment status log is required to be within the time grid where the sales order details are located or within m consecutive time grids thereafter; If both types of association are successful, they are merged and marked as consistent records; If the inventory level data, production completion data, and sales order details are successfully associated, but the corresponding fulfillment status log cannot be matched within m consecutive time grids thereafter, the sales order details are merged with the inventory level data and production completion data, and a consistent status marker with fulfillment lag is generated.

[0011] As a preferred embodiment of the manufacturing channel collaborative decision-making method based on industrial-level multi-agent as described in this invention, step S1 further includes: merging the consistent record and the record with a consistency status marker indicating performance lag according to the time grid and the entity ID to form an intermediate record with the time grid-entity ID as the primary key; for each primary key, if the corresponding intermediate record simultaneously contains inventory level data, production completion data, sales order details, and performance status logs, then a decision identifier with full-link visibility is generated for the primary key; if it only contains sales order details and a consistency status marker indicating performance lag, then a decision identifier requiring performance tracking is generated; counting data entries that fail association verification within K consecutive hours, if the data entries corresponding to the entity ID accumulate to more than U times, then the entity ID is included in a dynamic removal list; based on the dynamic removal list, filtering the intermediate records and retaining records not included in the dynamic removal list; organizing all filtered records according to the primary key and decision identifier in a columnar storage format to generate a collaborative basic dataset.

[0012] As a preferred embodiment of the manufacturing channel collaborative decision-making method based on industrial-grade multi-agent as described in this invention, step S2 includes: extracting all successfully associated intermediate records containing sales order details from the collaborative basic dataset; parsing the channel code from each sales order detail; extracting the historical demand sequence and associated entity ID set corresponding to each channel code; simultaneously calculating the historical order fulfillment rate of each channel; constructing a directed channel dependency network based on the overlap of entity ID sets between channels and the Pearson correlation coefficient of historical demand sequences; wherein the dependency direction points from channels with low historical order fulfillment rates to channels with high rates, and the edge weight of the dependency edge is determined by the product of the overlap and the correlation coefficient; for each channel node in the channel dependency network, the sum of the weights of the in-degree dependency edges is defined as the initial supply constraint pressure value of the channel; simultaneously, according to the decision identifier, if more than P% of the entity IDs under a channel are identified as requiring fulfillment tracking, then a fulfillment anchor point attribute is added to the channel node; the historical order fulfillment rate is the number of fulfilled orders in the past 24 hours divided by the total number of orders received during the same period.

[0013] As a preferred embodiment of the manufacturing channel collaborative decision-making method based on industrial-level multi-agent as described in this invention, step S2 further includes: performing constraint pressure iterative propagation on the channel dependency network: starting from the channel node with the performance anchor attribute, the initial supply constraint pressure value is distributed and transmitted to neighboring nodes along the out-degree dependency edge according to the edge weight ratio; after multiple rounds of iterative propagation, all the distributed values ​​received by each channel node are aggregated and added to its own initial supply constraint pressure value to obtain the final constraint pressure value of the channel; each channel code, the set of all associated entity IDs in the channel dependency network, and the final constraint pressure value are integrated into a record and output as structured channel collaborative state description data.

[0014] As a preferred embodiment of the manufacturing channel collaborative decision-making method based on industrial-level multi-agent as described in this invention, the business agents include a demand fluctuation identification agent, an inventory pressure assessment agent, and a fulfillment risk diagnosis agent; the demand fluctuation identification agent traverses each channel code in the channel collaborative state description data, extracts the demand sequence of each entity ID within multiple fixed time grids of the past, decomposes each demand sequence using multi-scale discrete wavelet transform, and extracts the energy value of high-frequency components and the trend slope of low-frequency components; the high-frequency energy values ​​of all entities under the same channel are normalized and averaged to obtain the instantaneous demand fluctuation intensity; the consistency of the trend slope sign of more than L% of entities under the same channel is used as the demand trend consistency indicator of the channel; the inventory pressure assessment agent... The force assessment agent parses the set of entity IDs associated with each channel in the channel collaboration status description data, obtains the inventory level data of each entity in the current time grid and the previous multiple time grids from the collaboration basic dataset, and performs the following operations: calculates the ratio of the standard deviation to the mean of the inventory level data of each entity in the most recent five time grids, as the inventory variation coefficient of the entity ID; if the current inventory level of an entity ID is lower than the safety stock threshold, it is marked as a supply warning state; if an entity ID is marked as a supply warning, the inventory variation coefficients of all associated channels are multiplied by a preset penalty factor; the weighted average of the inventory variation coefficients of all entities associated with each channel is defined as the comprehensive inventory pressure index of the channel, where the weight is determined by the historical demand ratio of each entity ID in the channel.

[0015] As a preferred embodiment of the manufacturing channel collaborative decision-making method based on industrial-level multi-agent as described in this invention, the following steps are taken: the performance risk diagnosis agent performs joint diagnosis based on the final constraint pressure value of each channel in the channel collaborative status description data and the performance status in the collaborative basic dataset: for a channel, the proportion of the number of intermediate records in the corresponding intermediate records that contain at least one sales order detail with a performance lag consistency status marker to the total number of intermediate records in the channel is used as the explicit delay rate; the final constraint pressure values ​​of all channels are normalized by min-max and mapped to the [0,1] interval; a composite performance risk score is calculated by weighting the final constraint pressure value and the explicit delay rate; if a channel meets any of the following conditions, it is determined to enter the performance circuit breaker state: the composite performance risk score is greater than Risk scoring threshold; explicit delay rate greater than delay rate threshold and final constraint pressure value greater than 1.5 times the average constraint pressure value; receive output from three agents and generate a set of machine-readable constraints for each channel: channel constraints: including instantaneous demand fluctuation intensity, demand trend consistency identifier, composite performance risk score and performance circuit breaker status marker; supply and production constraints include: comprehensive inventory pressure index, high inventory pressure entity ID list and key capacity constraint entity ID list; the high inventory pressure entity ID list refers to entity IDs whose inventory variation coefficient is higher than the variation threshold or are in a supply warning state; the key capacity constraint entity ID list refers to entity IDs that are currently occupying bottleneck production resources; align and associate these two sets of constraints for all channels according to channel codes and package them into a collaborative strategy input set.

[0016] As a preferred embodiment of the manufacturing channel collaborative decision-making method based on industrial-grade multi-agent as described in this invention, step S4 includes: creating multiple identical virtual manufacturing system twin copies based on the current collaborative basic dataset, each twin copy loading the same initial inventory status, in-production work orders, and unfulfilled order pool; parsing the collaborative strategy input set according to channel codes, generating at least three different resource allocation trial strategies for each set of channel constraints and production supply constraints, including: a first resource allocation trial strategy, prioritizing the fulfillment of order demands from channels with low instantaneous demand fluctuation intensity; in production scheduling, prioritizing production for entity IDs in the key capacity constraint entity ID list; a second resource allocation trial strategy, prioritizing the fulfillment of order demands from channels with low composite fulfillment risk scores; and in inventory allocation... The system prioritizes supplying entity IDs in the list of entity IDs with high inventory pressure that are also associated with channels under fulfillment circuit breaker status. A third resource allocation trial strategy, based on the comprehensive inventory pressure index of each channel, prioritizes allocating materials corresponding to entity IDs in the list of entity IDs with high inventory pressure from channels with low inventory pressure. Production resources are tilted towards key capacity-constrained entity IDs associated with the channel with the highest inventory pressure. The generated strategies are injected into independent twin copies. In each twin copy, the operation process for a specified future period is simulated at a preset speedup ratio. During the simulation, simulated order events are dynamically injected based on historical demand patterns, and the production, inventory consumption, and fulfillment processes are fully deduced. After the simulation, performance indicators for each channel are collected from each twin copy, including order fulfillment rate, inventory turnover days, and average order delay time.

[0017] As a preferred embodiment of the manufacturing channel collaborative decision-making method based on industrial-level multi-agent as described in this invention, step S4 further includes: performing strategy screening based on performance indicators collected through simulation; calculating the coefficient of variation of the order fulfillment rate of all channels under each strategy as an indicator for evaluating the horizontal stability of the strategy; the coefficient of variation is the ratio of the standard deviation to the average value of the order fulfillment rate; for each channel, calculating the maximum difference between the order fulfillment rates under all strategies, if it is less than a preset consistency tolerance threshold, then determining that the channel has consistent results in the current strategy set; selecting strategies that simultaneously meet the following conditions from all strategy results: the horizontal stability index of the strategy is lower than the average value, and it can enable channels in a fulfillment circuit breaker state exceeding a preset proportion to obtain consistency; outputting the selected strategy results as executable channel allocation and production scheduling decision instructions.

[0018] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the manufacturing channel collaborative decision-making method based on industrial-level multi-agent as described in the first aspect of the present invention are implemented.

[0019] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the manufacturing channel collaborative decision-making method based on industrial-level multi-agent as described in the first aspect of the present invention are implemented.

[0020] The beneficial effects of this invention are as follows: This invention realizes a substantial upgrade of multi-channel supply and demand collaborative decision-making in the manufacturing industry from "experience-driven" to "model-driven and simulation verification-driven" through industrial-grade multi-agent collaborative reasoning and digital twin parallel simulation mechanism; by performing unified time gridding, correlation consistency verification and semantic structured modeling on multi-source business data such as production, inventory, orders and fulfillment, it effectively eliminates decision-making biases caused by scattered data scope and time sequence misalignment in traditional systems.

[0021] By utilizing three types of intelligent agents—demand fluctuations, inventory pressure, and fulfillment risks—to perform parallel diagnosis of channel status and automatically generate machine-readable collaborative constraints, channel allocation and production scheduling decisions become interpretable and reproducible. Simultaneously, by leveraging a digital twin environment to accelerate simulation and stability and consistency screening of multiple strategies, actual execution risks are significantly reduced, and order fulfillment rates, inventory turnover efficiency, and fulfillment stability are improved. This, in turn, enhances the overall accuracy, robustness, and intelligence of cross-channel resource allocation for manufacturing enterprises. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating a collaborative decision-making method for manufacturing channels based on industrial-level multi-agent systems.

[0024] Figure 2 This is a flowchart of a manufacturing channel collaborative decision-making method based on industrial-level multi-agent systems.

[0025] Figure 3 A schematic diagram of the structure of an electronic device for implementing the industrial-grade multi-agent collaborative decision-making method for manufacturing channels in this embodiment of the invention. Detailed Implementation

[0026] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0027] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0028] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0029] like Figures 1-2 As shown, the manufacturing channel collaborative decision-making method based on industrial-level multi-agent systems includes:

[0030] S1: Collect production, inventory, and order data from within the manufacturing enterprise, as well as fulfillment and demand data from various sales channels, and perform time alignment and consistency verification to generate a collaborative basic dataset.

[0031] S1.1: Obtain production work order records, warehouse inbound and outbound flow records, sales order details, and fulfillment status logs from various channel portals in the manufacturing information system, and use them as source data nodes.

[0032] Specifically, from the manufacturing enterprise's manufacturing execution system or enterprise resource planning system, obtain production work order records organized by production task. Each record includes information such as work order number, product material code (entity ID), planned quantity, start and completion timestamps, and the production unit it belongs to. From the warehouse management system, obtain warehouse inbound and outbound flow records organized by material movement. Each flow record includes material code, changed quantity, remaining quantity after the change, warehouse location, operation timestamp, and associated document number. From the order management system, obtain sales order details, including order number, channel code, customer information, product material code, ordered quantity, order status, creation time, and required delivery date. Through the application programming interfaces (APIs) provided by various sales channels (such as e-commerce platforms, direct sales portals, and distributor systems), periodically pull or receive fulfillment status logs in real time. The logs must include order number, tracking number, status update timestamp, and specific status enumeration values ​​(such as shipped, in transit, and signed for).

[0033] The above four types of data are uniformly abstracted into source data nodes with time attributes.

[0034] S1.2: Attach an event timestamp accurate to the second and a globally unique data identifier to each of the source data nodes.

[0035] Based on event timestamps, inventory level data with continuous numerical attributes is uniformly resampled to a time grid with a period of n minutes (where n is the quantile of the average order arrival interval) using cubic spline interpolation. This approach ensures high-fidelity fitting of the continuous trend of inventory changes over time while maintaining curve smoothness. Compared to linear interpolation, it better reflects the actual inventory decay or replenishment process, providing more accurate time-series gradient information for subsequent stress assessment. Specifically, the process involves: first, sorting all inventory flows for a given material by time to form a discrete time-inventory point sequence; then, using this sequence as control points, constructing a cubic spline interpolation function; and finally, calculating the value at the end of each target time grid as the representative inventory level within that grid.

[0036] For production completion data driven by discrete events, an event aggregation method is used to cumulatively count the number of completion events within each time grid. This aggregation operation transforms discrete events into cumulative output within each grid, facilitating synchronous comparison with demand consumption during the same period.

[0037] For sales order details, retain all records to preserve the details of each transaction request and fulfillment action, ensuring the granularity of subsequent correlation verification.

[0038] The globally unique data identifier is entity type-entity ID-operation sequence, such as "MATERIAL-A001-STOCKIN-000123".

[0039] S1.3: For data located within the same time grid, the inventory level data and production completion data of the same product material are associated based on the entity ID field. Simultaneously, for sales order details and fulfillment status logs, association matching is performed based on the order number and entity ID, requiring the timestamp of the fulfillment status log to be within the time grid where the sales order details are located or within m consecutive time grids thereafter. If both types of association are successful, they are merged and marked as consistent records. If the inventory level data, production completion data, and sales order details are successfully associated, but the corresponding fulfillment status log cannot be matched within m consecutive time grids thereafter, the sales order details are merged with the inventory level data and production completion data, and a consistent status marker with fulfillment lag is generated. Other cases are considered as missing data, and no intermediate records are generated.

[0040] Specifically, if, within the current time grid, for a given entity ID, inventory level data and production completion data exist, and sales order details exist, and a fulfillment status log matching both the order number and entity ID of the sales order details can be found within this time grid or within the subsequent m consecutive time grids, then the four types of data—inventory, production, order, and fulfillment—are merged and marked as a consistent record. If, within the current time grid, for a given entity ID, inventory level data, production completion data, and sales order details exist, but within the subsequent m consecutive time grids (taking the number of grids corresponding to the maximum latency tolerance window of the channel fulfillment log), a fulfillment status log matching both the order number and entity ID cannot be found, then the sales order details are merged with the inventory and production data, and a consistent status marker with fulfillment lag is generated for the corresponding record. Other cases (such as missing inventory, production, or order data) are considered data missing, and no intermediate record is generated.

[0041] S1.4: The consistent record and the record with the consistency status mark of performance lag are merged according to the time grid and the entity ID to form an intermediate record with the time grid-entity ID as the primary key.

[0042] S1.5: For each primary key, if the corresponding intermediate record contains inventory level data, production completion data, sales order details and fulfillment status logs, then a decision identifier with full-link visibility is generated for the primary key; if it only contains sales order details and a consistency status flag with fulfillment lag, then a decision identifier that needs to be tracked for fulfillment is generated.

[0043] S1.6: Count the data entries that fail the association check (i.e., isolated data entries that are neither consistent records nor merely lagging records, but completely unrelated) within K consecutive hours (e.g., K=24). If the number of data entries corresponding to an entity ID exceeds U times (e.g., U=5), the data source quality of the corresponding entity ID is considered to be continuously unreliable, and the entity ID is added to the dynamic removal list.

[0044] As can be seen, this mechanism achieves adaptive management of data quality, automatically filters out continuously abnormal data sources, prevents "dirty data" from polluting the overall dataset, and improves system robustness.

[0045] Based on the dynamic removal list, intermediate records are filtered, retaining those not included in the list. All filtered records are then organized in a columnar storage format according to the primary key and decision identifier to generate a collaborative basic dataset. Subsequent agent analysis often involves batch calculations based on specific columns (such as the inventory level column for all records). Columnar storage provides extremely high compression ratios and query performance, significantly accelerating large-scale data analysis and meeting the data I / O efficiency requirements of real-time decision-making.

[0046] S2: Based on the aforementioned collaborative basic dataset, construct a collaborative semantic structure that reflects the relationship between products, channels, contracts, and performance, and output channel collaborative status description data.

[0047] S2.1: Extract all successfully associated intermediate records containing sales order details from the collaborative basic dataset, and parse the channel code from each sales order detail.

[0048] S2.2: Collect and extract the historical demand sequence and associated entity ID set corresponding to each channel code in the past 24 hours, and calculate the historical order fulfillment rate of each channel.

[0049] The historical order fulfillment rate is the number of orders fulfilled in the past 24 hours divided by the total number of orders received during the same period.

[0050] The historical demand sequence is a demand time series of length (24×60) / n, which is calculated by counting the total number of channel requests for all entity IDs in each time grid of the past 24 hours at set time grids (e.g., 15 minutes).

[0051] S2.3: Based on the overlap of entity ID sets among various channels (the overlap of entity ID sets can be calculated using methods such as the Jaccard similarity coefficient, which is not uniquely limited in this embodiment of the invention) and the Pearson correlation coefficient of historical demand sequences (a high positive correlation indicates that the demand of both often rises and falls together, and is likely to generate supply pressure at the same time), a directed channel dependency network is constructed.

[0052] The dependence shifts from channels with low historical order fulfillment rates to channels with high rates. The logic behind this is that when supply is tight, the unmet demand pressure from channels with lower fulfillment rates (which may be due to lower priority or slower response) is more likely to be transmitted to channels with higher fulfillment rates and more efficient operations.

[0053] The edge weight of a dependent edge is determined by the product of the overlap and the correlation coefficient. Here, the positive part of the correlation coefficient is taken to capture the resonance pressure generated by "co-directional fluctuations" and ignore the offsetting effect of opposite fluctuations.

[0054] The weights comprehensively reflect the dependence intensity of those vying for the same batch of resources (high overlap) and simultaneously reaching out for them (high positive correlation).

[0055] S2.4: For each channel node in the channel dependency network, the sum of the weights of the in-degree dependent edges is defined as the initial supply constraint pressure value of the channel; simultaneously, based on the decision identifier, if more than P% (e.g., based on the 75th quantile of historical statistics) of entity IDs under a channel are identified as requiring performance tracking, then a performance anchor attribute is attached to the channel node, which refers to assigning a risk marker to a specific node (i.e., the channel) in the channel dependency network graph. These anchors will serve as active sources of pressure propagation.

[0056] S2.5: Perform constraint pressure iterative propagation on the channel-dependent network:

[0057] Starting from the channel node with the attribute of fulfillment anchor, the initial supply constraint pressure value is distributed and transmitted to neighboring nodes along the out-degree dependent edge according to the edge weight ratio.

[0058] After multiple rounds of iterative propagation, all allocation values ​​received by each channel node are aggregated and added to its own initial supply constraint pressure value to obtain the final constraint pressure value of the channel.

[0059] For example, suppose the first After round of iterations, the node The supply constraint pressure value For each node with the performance anchor attribute Or the previous round of pressure value The node will determine its current total pressure. Assign along all out-degree edges. Assign to neighboring nodes. The pressure is:

[0060] ;

[0061] in, For the source node The set of terminal channel nodes for all out-degree dependent edges from which a path originates. For set One of the downstream channel nodes, For the node Pointing to channel node The edge weights of the dependent edges. For the node Pointing to node The edge weights of the dependent edges; this allocation rule is based on the strength of the dependency proportionally.

[0062] Each channel node In this iteration, the pressure allocation received from all in-degree neighbors was updated to the pressure value. for:

[0063] ;

[0064] in, The attenuation factor (can be set according to actual needs; in this embodiment of the invention, it is set to 0.3); This represents the initial supply constraint pressure value. To point to channel nodes The set of starting channel nodes for all in-degree dependent edges is obtained directly from the constructed directed channel dependency network; For the current node In the In the round of iteration, from all in-degree neighbor nodes The sum of the received pressure allocation values.

[0065] The above operations can ensure the initial supply constraint pressure value of the channel nodes. A portion of the influence is always retained to prevent the propagation process from completely obscuring the original structure; at the same time, the natural dissipation of pressure during the transmission process is simulated, making the model more consistent with physical reality and ensuring the convergence of the algorithm.

[0066] Furthermore, the Euclidean norm change of the pressure values ​​between adjacent rounds for all channel nodes is calculated. The iteration terminates when the change is less than a preset convergence threshold (which can be set to a very small positive number, such as 1e-6, depending on actual needs). At this point, the pressure value of each channel node is the final constraint pressure value. It integrates the static structural pressure of the channel with the diffuse risk pressure dynamically transmitted from the network's performance anchor points.

[0067] It should be noted that this invention applies graph algorithms to supply chain risk analysis, which can reveal how local fulfillment problems (anchor points) trigger systemic, chain-reaction supply constraints through complex channel dependency networks, thereby identifying those hidden vulnerable channels that appear healthy in their own data but are highly susceptible to risks from other parties due to their network location sensitivity.

[0068] S2.6: Integrate each channel code, the set of all associated entity IDs in the channel dependency network, and the final constraint pressure value into a single record, and output it as structured channel collaboration status description data.

[0069] The channel collaboration status description data set combines the micro-resource composition (entity set) and macro-network potential (pressure value) of the channel, completing the construction from basic data to collaborative semantics, and providing a unified and insightful status perception input for downstream professional intelligent agents.

[0070] S3: Multiple business intelligence agents extract demand fluctuations, inventory pressure, and performance risk status based on the channel collaboration status description data, and transform the corresponding business rules into channel constraints and supply and demand constraints to form a collaborative strategy input set.

[0071] The business intelligence agents include a demand fluctuation identification agent, an inventory pressure assessment agent, and a performance risk diagnosis agent.

[0072] S3.1: The demand fluctuation identification agent traverses each channel code in the channel collaboration status description data and extracts the demand sequence of each entity ID in the past multiple (e.g., 72) time grids.

[0073] S3.1.1: Use multi-scale discrete wavelet transform to decompose each demand sequence and extract the energy value of the high-frequency component and the trend slope of the low-frequency component.

[0074] Specifically, the Daubechies 4 (db4) wavelet is selected as the mother wavelet for a three-level decomposition. The decomposition process is achieved through convolution and downsampling: the sequence is first passed through a pair of low-pass and high-pass filters to obtain approximation coefficients (low frequency) and detail coefficients (high frequency), respectively. This process is then repeated for the approximation coefficients, iterating three times. Finally, the original sequence is decomposed into: a third-level approximation coefficient (representing the long-term trend) and three levels of detail coefficients (representing high-frequency fluctuations at different time scales).

[0075] S3.1.2: Normalize the high-frequency energy values ​​of all entities under the same channel and then calculate the average value to obtain the instantaneous demand fluctuation intensity.

[0076] Specifically, calculate the first level of detail coefficients. The energy, as the instantaneous fluctuation energy of an entity within a channel, is specifically calculated by squaring all values ​​in the detailed coefficient sequence and then summing these squared values ​​to obtain the average. This energy value captures the intensity of random fluctuations in demand at the finest time scale. To perform cross-entity aggregation, the instantaneous fluctuation energy of all entities is normalized using a minimum-maximum normalization method. The instantaneous demand fluctuation intensity of a channel is defined as the sum of the normalized energy; a higher value indicates greater instability in the overall channel demand in the near term.

[0077] S3.1.3: The consistency of the trend slope sign (positive / negative) of low-frequency components of more than L% of entities under the same channel shall be used as the demand trend consistency indicator of the channel.

[0078] Specifically, a linear least squares fit is performed on the third-level approximation coefficients to obtain the trend slope. The proportion of entities in the statistical channel with the same trend slope sign (positive for an upward trend, negative for a downward trend) is then calculated. If this proportion exceeds a preset threshold L% (which can be set according to actual needs, with a default of 60%), a demand trend consistency indicator is generated, and the dominant trend direction (positive or negative) is recorded; otherwise, a demand trend divergence indicator is generated. This indicator helps determine whether the channel as a whole is in an expansionary, contractionary, or chaotic state, providing a forward-looking signal for long-term capacity allocation.

[0079] It can be seen that, through discrete wavelet transform, the agent achieves time-frequency domain decoupling of demand signals, enabling independent and accurate quantification of short-term random fluctuations and long-term deterministic trends. This avoids the volatility misjudgment caused by trend interference in traditional time-domain statistical methods (such as moving standard deviation). The trend consistency indicator provides collective momentum information on the evolution of channel demand, enhancing the decision-making system's forward-looking adaptability to market trends.

[0080] S3.2: The inventory pressure assessment agent parses the set of entity IDs associated with each channel in the channel collaboration status description data, obtains the inventory level data of each entity in the current time grid and the previous multiple time grids from the collaboration basic dataset, and performs the following operations:

[0081] (1) Calculate the ratio of the standard deviation to the mean of the inventory level data of the most recent five time grids for each inventory level. Use this as the inventory variation coefficient of the entity ID. It reflects the degree of dispersion of the inventory level relative to the mean. The higher the value, the more unstable the inventory management and the greater the difficulty of prediction.

[0082] (2) If the current inventory level of an entity ID is lower than the safety stock threshold, it is marked as a supply warning state, indicating that the available inventory of materials is lower than the safety buffer calculated based on historical consumption, and there is an immediate risk of stockout.

[0083] First, historical average daily demand is calculated based on the entity's total demand sequence across all channels over the past 72 hours (default). The safety stock threshold is determined by the product of historical average daily demand and a safety factor (default 1.5).

[0084] (3) If an entity ID is marked as a supply warning, the inventory variation coefficient of all associated channels will be multiplied by a preset penalty factor (set according to actual needs, the default value of which is 1.3 in this embodiment of the invention).

[0085] (4) The weighted average of the inventory variation coefficients of all entities associated with each channel is defined as the comprehensive inventory pressure index of the channel, where the weight is determined by the historical demand share of each entity ID in the channel. For example, the weight can be the channel's demand for the entity in the past 24 hours / the total demand of the channel in the past 24 hours. The weighted design ensures that core materials with high demand contribute more to the channel inventory pressure index, making the index more closely reflect the actual importance of the business.

[0086] Furthermore, the intelligent agent simultaneously outputs two key material lists: a list of entity IDs with high inventory pressure, containing all entity IDs that meet the criteria of having an inventory coefficient of variation greater than the variation threshold (a configurable system parameter that can be dynamically set based on a certain quantile, such as the 75th quantile, of the historical inventory coefficient of variation distribution of all materials) or are in a "supply warning" state. This identifies materials with abnormal inventory conditions or those on the verge of shortage, requiring priority monitoring or replenishment.

[0087] List of key capacity constraint entity IDs: By parsing the production work order records in the collaborative basic dataset that are currently in "production", we can identify which work orders occupy production resources (such as production lines, molds, and key equipment) that belong to the preset bottleneck resource set.

[0088] The entity IDs corresponding to these work orders are extracted and listed, indicating the materials that are currently occupying key capacity and may restrict the overall output elasticity. These materials are the key targets for capacity allocation.

[0089] S3.3: The performance risk diagnosis agent performs a joint diagnosis based on the final constraint pressure value of each channel in the channel collaboration status description data and the performance status in the collaboration basic dataset:

[0090] For a channel, the percentage of intermediate records containing at least one sales order detail with a fulfillment lag consistency status marker, relative to the total number of intermediate records in that channel, is taken as the explicit delay rate.

[0091] The final constraint pressure values ​​of all channels are min-max normalized and mapped to the [0,1] interval.

[0092] The final constraint pressure value and explicit delay rate are weighted to calculate the composite performance risk score. That is, the normalized network pressure and explicit delay rate are integrated by using a convex combination method to form the composite performance risk score.

[0093] If a channel meets any of the following conditions, it is determined to enter the performance circuit breaker state: the composite performance risk score is greater than the risk score threshold (set according to actual needs, and can be set to 0.7 in this embodiment of the invention); the explicit delay rate is greater than the delay rate threshold (set according to actual needs, and can be set to 0.3 in this embodiment of the invention); and the final constraint pressure value is greater than 1.5 times the average constraint pressure value.

[0094] In other words, condition 1 is used to capture high-risk channels under comprehensive assessment; condition 2 serves as a fast track to capture extreme cases where severe delays have occurred and the network is under significant pressure. Channels deemed to be subject to circuit breakers indicate that their fulfillment capabilities are nearing or have already failed, requiring isolation measures in subsequent decision-making, such as suspending the acceptance of new orders, prioritizing the clearing of backlogged orders, or initiating emergency supply processes.

[0095] It can be seen that the dual-condition circuit breaker mechanism combines the robustness of comprehensive judgment with the ability to respond quickly to extreme situations. It prevents both misjudgment and omission, significantly improving the sensitivity and accuracy of risk management and providing a key decision sentinel for ensuring the overall supply chain's reliability.

[0096] S3.4: Receive outputs from the three agents and generate a set of machine-readable constraints for each channel:

[0097] Channel constraints include instantaneous demand fluctuation intensity, demand trend consistency indicator, composite performance risk score, and performance circuit breaker status marker.

[0098] Supply and demand constraints include: a comprehensive inventory pressure index, a list of entity IDs with high inventory pressure, and a list of entity IDs with key capacity constraints.

[0099] S3.5: Align and link these two sets of constraints from all channels according to their channel codes, and package them into a collaborative strategy input set. Using channels as the basic unit, each unit contains seven key constraint attributes, comprehensively and without redundancy depicting the overall business status from demand to supply to fulfillment.

[0100] S4: Perform parallel inference and evaluation on the input set of the collaborative strategy in the digital twin environment, screen the strategy results that meet the preset conditions of stability and consistency, output channel allocation and production scheduling decisions, and feed the strategy results back to the collaborative basic dataset to update the basis for subsequent decisions.

[0101] S4.1: Based on the current collaborative base dataset, create multiple identical virtual manufacturing system twins, each twin loading the same initial inventory status, in-production work orders, and unfulfilled order pool.

[0102] S4.2: Parse the collaborative strategy input set according to channel coding, and generate at least three different resource allocation probing strategies for each set of channel constraints and supply-demand constraints, including:

[0103] The first resource allocation trial strategy prioritizes meeting the order demands of channels with low instantaneous demand fluctuations; in terms of production scheduling, it prioritizes arranging production for entity IDs in the list of key capacity-constrained entity IDs.

[0104] For all channels, sort them in ascending order of the intensity of instantaneous demand fluctuation. The lower the fluctuation intensity, the more stable and predictable the channel demand. The strategy prioritizes fulfilling order demands from the more stable channels that are ranked higher.

[0105] The second resource allocation trial strategy prioritizes meeting the order needs of channels with low composite fulfillment risk scores; in terms of inventory allocation, priority is given to ensuring the supply of entity IDs associated with channels in the list of entity IDs with high inventory pressure and which are also in a fulfillment circuit breaker state.

[0106] In other words, when transferring inventory between channels, priority is given to materials that are both on the list of entities with high inventory pressure (i.e., those with tight or abnormal inventory) and associated with at least one channel that is under fulfillment suspension. The aim is to ensure precise transportation to the most vulnerable links and attempt to quickly alleviate the shortage of core materials in channels that have been suspended.

[0107] The third resource allocation trial strategy is to prioritize the allocation of materials corresponding to entities with high inventory pressure from channels with low inventory pressure, based on the comprehensive inventory pressure index of each channel; and to tilt production resources toward key capacity-constrained entities associated with the channel with the highest inventory pressure.

[0108] S4.4: Inject the generated strategies into independent twin copies. In each twin copy, simulate the operation process for a specified future period at a preset speedup ratio. During the simulation, simulated order events are dynamically injected based on historical demand patterns, and the production, inventory consumption and fulfillment processes are fully deduced.

[0109] Specifically, the generated strategies are loaded into different, independent twin copies. Within each copy, the simulation engine advances the virtual clock at a preset speedup ratio (which can be set according to actual needs; in this embodiment, it is set to 10x speed). The future period of the simulation is a specified T hours (e.g., T=72). To simulate the uncertainty of demand, during the simulation process, based on recent historical demand patterns in the collaborative basic dataset (e.g., the average demand rate and distribution of each channel and each material during the same period of the past week), time series simulation algorithms (e.g., autoregressive models or random sampling based on historical distributions) are used to dynamically inject simulated order arrival events.

[0110] Within each simulation step of each time grid, the engine strictly executes the following logic chain in sequence: Receive newly arrived simulated orders and unfulfilled historical orders; allocate materials to orders from available inventory according to the strategy carried by this replica (e.g., the first resource allocation trial strategy) and defined rules. If inventory is insufficient, generate an order delay record. Advance the progress of work orders in production. Based on the production scheduling logic in the resource allocation trial strategy (e.g., prioritizing bottleneck materials), trigger an event upon work order completion to put the products into inventory. Reduce inventory based on order allocation results, and update the fulfillment status for orders that have been allocated.

[0111] This cycle continues until the simulation ends, fully demonstrating the end-to-end process of demand injection, strategy decision-making, production response, inventory consumption, and fulfillment.

[0112] S4.5: After the simulation, collect the performance indicators of each channel from each twin copy, including order fulfillment rate (the proportion of the number of order lines that are successfully fulfilled to the total number of order lines under the corresponding channel), inventory turnover days (calculated based on the average inventory cost and sales cost during the simulation period, reflecting inventory efficiency) and average order delay time (the average difference between the actual completion time and the required time for all delayed orders).

[0113] S4.6: Strategies are selected based on performance indicators collected through simulation.

[0114] Calculate the coefficient of variation (i.e., the ratio of the standard deviation to the mean of the order fulfillment rate) for all channels under each strategy, as an indicator to evaluate the horizontal stability of the strategy.

[0115] For each channel, calculate the maximum difference between the order fulfillment rates under all strategies. If it is less than the preset consistency tolerance threshold (which can be set according to actual conditions, and can be set to 0.05 in this embodiment of the invention), then the channel is determined to have consistent results in the current strategy set.

[0116] From all strategy results, strategies that simultaneously meet the following conditions are selected: the strategy's horizontal stability index is lower than the average coefficient of variation of all strategies, and it can achieve consistency for channels that exceed a preset proportion (e.g., 80%) and are in a fulfillment circuit breaker state. The selected strategy results are then output as executable channel allocation and production scheduling decision instructions.

[0117] It can be seen that the first condition ensures that the selected strategy itself is robust; the second condition requires that the strategy can effectively stabilize most high-risk channels (regardless of whether they are improved, at least not causing the results to fluctuate drastically due to different strategies). If multiple strategies meet this condition, the one with the lowest horizontal stability index (i.e., the most robust) is selected.

[0118] S4.7: Decode the selected optimal strategy results into specific channel allocation suggestions decomposed by time grid and entity ID, as well as corresponding production scheduling adjustment instructions, and output them as the final decision.

[0119] At the same time, this decision is recorded as a "verified strategy" and, together with the key process indicators generated in the simulation, is associated with the corresponding "time grid-entity ID" primary key record in the collaborative basic dataset. When the next S2 step constructs collaborative semantics, these historical experiences can be used to optimize model parameters (e.g., correct the edge weight calculation method of the channel dependency network), so that the entire system becomes more and more accurate in its decision-making and more and more adapted to the unique operating environment of the enterprise as the running time increases.

[0120] This embodiment also provides a computer device applicable to the manufacturing channel collaborative decision-making method based on industrial-level multi-agent systems, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the manufacturing channel collaborative decision-making method based on industrial-level multi-agent systems proposed in the above embodiment.

[0121] like Figure 3 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0122] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0123] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, central processing unit (CPU), graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as an industrial-grade multi-agent manufacturing channel collaborative decision-making method.

[0124] In some embodiments, the industry-grade multi-agent-based manufacturing channel collaborative decision-making method can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the industry-grade multi-agent-based manufacturing channel collaborative decision-making method described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to execute the industry-grade multi-agent-based manufacturing channel collaborative decision-making method by any other suitable means (e.g., by means of firmware).

[0125] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0126] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the manufacturing channel collaborative decision-making method based on industrial-level multi-agent as proposed in the above embodiments.

[0127] In summary, the beneficial effects of this invention are as follows: This invention achieves a substantial upgrade from "experience-driven" to "model-driven and simulation verification-driven" multi-channel supply and demand collaborative decision-making in the manufacturing industry through industrial-grade multi-agent collaborative reasoning and digital twin parallel simulation mechanisms; by performing unified time gridding, correlation consistency verification, and semantic structured modeling on multi-source business data such as production, inventory, orders, and fulfillment, it effectively eliminates decision-making biases caused by scattered data scope and time sequence misalignment in traditional systems.

[0128] By utilizing three types of intelligent agents—demand fluctuations, inventory pressure, and fulfillment risks—to perform parallel diagnosis of channel status and automatically generate machine-readable collaborative constraints, channel allocation and production scheduling decisions become interpretable and reproducible. Simultaneously, by leveraging a digital twin environment to accelerate simulation and stability and consistency screening of multiple strategies, actual execution risks are significantly reduced, and order fulfillment rates, inventory turnover efficiency, and fulfillment stability are improved. This, in turn, enhances the overall accuracy, robustness, and intelligence of cross-channel resource allocation for manufacturing enterprises.

[0129] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A collaborative decision-making method for manufacturing channels based on industrial-level multi-agent systems, characterized in that: include: S1: Collect production, inventory, and order data from within manufacturing enterprises, as well as fulfillment and demand data from various sales channels, and perform time alignment and consistency verification to generate a collaborative basic dataset; S2: Based on the aforementioned collaborative basic dataset, construct a collaborative semantic structure that reflects the relationship between products, channels, contracts, and performance, and output channel collaborative status description data; S3: Multiple business intelligence agents extract demand fluctuations, inventory pressures, and fulfillment risk statuses based on the channel collaboration status description data, and transform the corresponding business rules into channel constraints and supply and demand constraints to form a collaborative strategy input set. S4: Perform parallel inference and evaluation on the input set of the collaborative strategy in the digital twin environment, screen the strategy results that meet the preset conditions of stability and consistency, output channel allocation and production scheduling decisions, and feed the strategy results back to the collaborative basic dataset to update the basis for subsequent decisions. Constructing a collaborative semantic structure includes: extracting all successfully associated intermediate records containing sales order details from the collaborative basic dataset; parsing the channel code from each sales order detail; extracting the historical demand sequence and associated entity ID set corresponding to each channel code; simultaneously calculating the historical order fulfillment rate for each channel; and constructing a directed channel dependency network based on the overlap of entity ID sets between channels and the Pearson correlation coefficient of historical demand sequences; wherein the dependency direction points from channels with low historical order fulfillment rates to channels with high rates, and the edge weight of the dependency edge is determined by the product of the overlap and the positive value of the correlation coefficient; for the channel dependency network... For each channel node, the sum of the weights of the in-degree dependent edges is defined as the initial supply constraint pressure value of the channel. Simultaneously, based on the decision identifier, if more than P% of the entity IDs under a channel are identified as requiring performance tracking, a performance anchor point attribute is added to the channel node. Iterative propagation of constraint pressure is performed on the channel dependency network: starting from the channel node with the performance anchor point attribute, the initial supply constraint pressure value is distributed and transmitted to neighboring nodes along the out-degree dependent edges according to the edge weight ratio. After multiple rounds of iterative propagation, all the distributed values ​​received by each channel node are aggregated and added to its own initial supply constraint pressure value to obtain the final constraint pressure value of the channel. Based on performance indicators collected through simulation, strategy selection is performed: the coefficient of variation of order fulfillment rate for all channels under each strategy is calculated as an indicator to evaluate the horizontal stability of the strategy; the coefficient of variation is the ratio of the standard deviation to the mean of the order fulfillment rate; for each channel, the maximum difference between the order fulfillment rates under all strategies is calculated, and if it is less than a preset consistency tolerance threshold, the channel is determined to have consistent results in the current strategy set; strategies that meet the following conditions are selected from all strategy results: the horizontal stability index of the strategy is lower than the average value, and it can enable channels in a fulfillment circuit breaker state exceeding a preset proportion to achieve consistency; the selected strategy results are output as executable channel allocation and production scheduling decision instructions.

2. The manufacturing channel collaborative decision-making method based on industrial-level multi-agent as described in claim 1, characterized in that: S1 includes: Acquire production work order records, warehouse inbound and outbound flow records, sales order details, and fulfillment status logs from various channel portals in the manufacturing information system, and use them as source data nodes; Each source data node is appended with an event timestamp accurate to the second and a globally unique data identifier. Based on the event timestamp, inventory level data with continuous numerical attributes is uniformly resampled to a time grid with an n-minute period using cubic spline interpolation. For discrete event-driven production completion data, event aggregation is used to cumulatively count the number of completion events in each time grid. For sales order detail data, all records are retained. The globally unique data identifier is entity type-entity ID-operation sequence; For data located within the same time grid, the inventory level data and production completion data of the same product material are associated based on the entity ID field. Simultaneously, for sales order details and fulfillment status logs, association matching is performed based on the order number and entity ID, requiring the timestamp of the fulfillment status log to be within the time grid where the sales order details are located or within m consecutive time grids thereafter. If both associations are successful, they are merged and marked as consistent records. If the inventory level data, production completion data, and sales order details are successfully associated, but no corresponding fulfillment status log can be matched within m consecutive time grids thereafter, the sales order details are merged with the inventory level data and production completion data, and a consistent status marker with fulfillment lag is generated.

3. The manufacturing channel collaborative decision-making method based on industrial-level multi-agent as described in claim 2, characterized in that: S1 further includes: The consistent record and the record with the consistency status mark of performance lag are merged according to the time grid and the entity ID to form an intermediate record with the time grid-entity ID as the primary key. For each primary key, if the corresponding intermediate record contains inventory level data, production completion data, sales order details, and fulfillment status logs, then a decision identifier with full-link visibility is generated for the primary key; if it only contains sales order details and a consistency status flag with fulfillment lag, then a decision identifier requiring fulfillment tracking is generated. If data entries that fail the association verification are counted within K consecutive hours, and the number of data entries corresponding to an entity ID exceeds U, then the entity ID is added to a dynamic removal list. Based on the dynamic removal list, the intermediate records are filtered, and records not included in the dynamic removal list are retained. All filtered records are organized in a columnar storage format based on the primary key and decision identifier to generate a collaborative basic dataset.

4. The manufacturing channel collaborative decision-making method based on industrial-level multi-agent as described in claim 3, characterized in that: The historical order fulfillment rate is the number of orders fulfilled in the past 24 hours divided by the total number of orders received during the same period.

5. The manufacturing channel collaborative decision-making method based on industrial-level multi-agent as described in claim 4, characterized in that: Each channel code, the set of all associated entity IDs in the channel dependency network, and the final constraint pressure value are integrated into a single record, outputting structured channel collaboration status description data.

6. The manufacturing channel collaborative decision-making method based on industrial-level multi-agent as described in claim 5, characterized in that: The business intelligence agents include a demand fluctuation identification intelligence agent, an inventory pressure assessment intelligence agent, and a performance risk diagnosis intelligence agent. The demand fluctuation identification agent traverses each channel code in the channel collaborative state description data, extracts the demand sequence of each entity ID within multiple fixed time grids of the past, decomposes each demand sequence using multi-scale discrete wavelet transform, and extracts the energy value of high-frequency components and the trend slope of low-frequency components; normalizes the high-frequency energy values ​​of all entities under the same channel and calculates the average to obtain the instantaneous demand fluctuation intensity; the consistency of the trend slope sign of the low-frequency components of more than L% of entities under the same channel is used as the demand trend consistency indicator of the channel. The inventory pressure assessment agent parses the set of entity IDs associated with each channel in the channel collaboration status description data, obtains the inventory level data of each entity in the current time grid and the previous multiple time grids from the collaboration basic dataset, and performs the following operations: Calculate the ratio of the standard deviation to the mean of the inventory level data for the most recent five time grids for each inventory level, and use it as the inventory variation coefficient for the entity ID; If the current inventory level of an entity ID is lower than the safety stock threshold, it will be marked as a supply warning state. If an entity ID is marked as a supply warning, the inventory variation coefficient of all associated channels will be multiplied by a preset penalty factor. The weighted average of the inventory variation coefficients of all entities associated with each channel is defined as the comprehensive inventory pressure index of that channel, where the weights are determined by the historical demand share of each entity ID in that channel.

7. The manufacturing channel collaborative decision-making method based on industrial-level multi-agent as described in claim 6, characterized in that: The performance risk diagnosis agent performs a joint diagnosis based on the final constraint pressure values ​​of each channel in the channel collaboration status description data and the performance status in the collaboration base dataset: For a channel, the proportion of the number of intermediate records that contain at least one sales order detail with a fulfillment lag consistency status marker to the total number of intermediate records in the channel is used as the explicit delay rate; the final constraint pressure values ​​of all channels are min-max normalized and mapped to the [0,1] interval. A composite performance risk score is calculated by weighting the final constraint pressure value and the explicit delay rate. If a channel meets any of the following conditions, it will be determined to enter the performance circuit breaker state: the composite performance risk score is greater than the risk score threshold; the explicit delay rate is greater than the delay rate threshold and the final constraint pressure value is greater than 1.5 times the average constraint pressure value; Receive outputs from three agents and generate a set of machine-readable constraints for each channel: Channel constraints include: intensity of instantaneous demand fluctuations, consistency indicator of demand trends, composite performance risk score, and performance circuit breaker status marker; Supply and demand constraints include: a comprehensive inventory pressure index, a list of entity IDs with high inventory pressure, and a list of entity IDs with key capacity constraints. The list of entity IDs with high inventory pressure refers to entity IDs whose inventory variation coefficient is higher than the variation threshold or who are in a supply warning state; the list of entity IDs with critical capacity constraints refers to entity IDs that are currently occupying bottleneck production resources. All two sets of constraints for all channels are associated and aligned according to the channel code, and packaged into a collaborative strategy input set.

8. The manufacturing channel collaborative decision-making method based on industrial-level multi-agent as described in claim 7, characterized in that: S4 includes: Based on the current collaborative foundation dataset, create multiple identical virtual manufacturing system twins, each twin loading the same initial inventory status, in-production work orders, and unfulfilled order pool; The input set of collaborative strategies is parsed according to channel codes, generating at least three different resource allocation probing strategies for each set of channel constraints and supply-demand constraints, including: The first resource allocation trial strategy prioritizes meeting the order demands of channels with low instantaneous demand fluctuations; in terms of production scheduling, production is prioritized for entity IDs in the list of key capacity-constrained entity IDs. The second resource allocation trial strategy prioritizes meeting the order needs of channels with low composite fulfillment risk scores; in terms of inventory allocation, priority is given to ensuring the supply of entity IDs associated with channels in the list of entity IDs with high inventory pressure and which are simultaneously in a fulfillment circuit breaker state. The third resource allocation trial strategy is to prioritize the allocation of materials corresponding to the entity ID list with high inventory pressure from channels with low inventory pressure, based on the comprehensive inventory pressure index of each channel. Production resources are being shifted toward key capacity-constrained entities (IDs) associated with the channels experiencing the highest inventory pressure. The generated strategies are injected into separate twin copies; In each twin copy, the operation process for a specified future period is simulated at a preset speedup ratio. During the simulation, simulated order events are dynamically injected based on historical demand patterns, and the production, inventory consumption and fulfillment processes are fully deduced. After the simulation is completed, performance metrics for each channel are collected from each twin copy, including order fulfillment rate, inventory turnover days, and average order delay time.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the manufacturing channel collaborative decision-making method based on industrial-level multi-agent as described in any one of claims 1 to 8.

Images