Supply chain supply and demand dynamic matching method based on multi-agent game

Abstract

The application discloses a supply chain supply and demand dynamic matching method based on multi-agent game, relates to the technical field of supply chain management, and comprises the following steps: constructing a multi-agent system, dividing agents, carrying an edge module, interacting with the MQTT protocol through a block chain, and encrypting sensitive data; real-time sensing of supply and demand information, parameter acquisition according to a dynamic frequency, synchronization to a shared middle station after filtering and cleaning; generation of a game strategy, construction of a model with Nash equilibrium as a target, and iterative generation of an initial strategy; execution of dynamic matching, demand allocation according to priority, monitoring of deviation and negotiation adjustment; multi-dimensional evaluation of matching effect, comprehensive score calculation through an analytic hierarchy process; iterative optimization of the strategy, parameter adjustment according to the evaluation result, and updating of a strategy library to adapt to dynamic changes. The application breaks the information island of the supply chain, balances the interests of all participants to achieve overall optimization, improves the supply and demand matching precision and dynamic adaptation capability, and enhances the efficiency and stability of the supply chain.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of supply chain management, and particularly relates to a supply chain supply-demand dynamic matching method based on multi-agent game. BACKGROUND

[0002] With the scale development of manufacturing industry, retail industry and the like, the supply chain system gradually presents the characteristics of multi-node, cross-region and high dynamics, and the accuracy and timeliness of supply-demand matching become the core factors affecting the efficiency of the supply chain. At present, most of the supply chains still rely on the traditional "hierarchical" information transmission mode, and there are obvious information islands among the suppliers, manufacturers, distributors and demand sides - the suppliers are difficult to obtain the actual demand fluctuation of the downstream in real time, and can only make production capacity plan according to the historical order; the manufacturers often appear the situation that the production scheduling is out of sync with the raw material supply due to the lag of the raw material inventory data; the distributors lack real-time collaboration with the demand side, which easily leads to regional inventory overstock or shortage. Such information asymmetry directly leads to the mismatch between supply and demand, for example, during the retail industry promotion, the demand increases sharply, and the suppliers cannot quickly adjust the production capacity due to the failure to timely perceive the demand change, which eventually leads to the terminal shortage; on the contrary, during the off-season of the manufacturing industry, the demand decreases, and the suppliers still maintain high production capacity, which causes the overstock of raw materials and increases the warehouse and capital costs.

[0003] The existing supply chain supply-demand matching method mostly adopts a static planning model, relies on the manually preset parameters and fixed rules to perform matching, and is difficult to adapt to the dynamically changing market environment. The traditional method usually updates the supply-demand data in "monthly" or "weekly" cycles, and cannot respond to sudden situations such as order quantity fluctuation, equipment failure and transportation delay in real time; and lacks a game mechanism for multi-link collaboration, and each participant usually only aims to maximize its own interests - the suppliers tend to increase the price to increase profits, and the demand side pursues low price and fast delivery, which leads to the imbalance of the supply-demand game and makes it difficult to achieve the overall optimal solution. For example, the manufacturer selects the supplier with the lowest price to reduce costs, but the supplier may not be able to deliver on time due to insufficient production capacity, which eventually leads to the production stagnation of the manufacturer; the demand side excessively compresses the supplier cycle to shorten the delivery time, which forces the supplier to rush production, and instead increases the product defect rate, forming a dilemma of "local optimum, global inefficiency".

[0004] In addition, the data security and trust problem in the supply chain collaboration process, the lack of strategy optimization ability and other defects further restrict the supply and demand matching effect. When each link interacts in data, it is easy to face the risk of leakage of sensitive information such as quotation and cost, which leads to the concealment of key data by some participants due to concerns about information security, affecting the matching accuracy; at the same time, the existing method lacks a systematic evaluation and strategy iteration mechanism for the matching effect, so that even if the supply and demand mismatch is found, it is difficult to quickly locate the problem source and adjust the strategy. For example, when there is a regional demand surge, the traditional method cannot quickly coordinate the resource scheduling of surrounding suppliers and distributors, and can only passively bear the high cost of inventory shortage or emergency procurement; and lacks an effective dispute resolution mechanism, when the supply and demand parties have differences on the matching result or income distribution, it is difficult to reach a consensus through a traceable evidence chain, further weakening the stability of supply chain collaboration. SUMMARY

[0005] The supply chain supply and demand dynamic matching method based on multi-agent game provided by the application solves the problems mentioned in the prior art.

[0006] In order to achieve the above purpose, the application adopts the following technical scheme:

[0007] A supply chain supply and demand dynamic matching method based on multi-agent game, comprising the following steps:

[0008] Multi-agent system construction step: divide the supply chain agent type, including supplier, manufacturer, distributor, demand end agent; each agent carries an edge computing module, adopts block chain P2P communication protocol and MQTT Internet of Things protocol for data interaction; establish an agent interaction standard, define JSON format data, and use AES-256 to encrypt sensitive data;

[0009] Supply and demand information real-time sensing step: each agent collects core parameters through sensors and system interfaces, the collection frequency is dynamically adjusted according to parameters, the data is cleaned by Kalman filtering algorithm, and after removing outliers, it is synchronized to the supply chain shared data center;

[0010] Game strategy generation step: each agent as a game participant, sets the objective function, generates the initial strategy by using iterative game algorithm, the supplier agent quotes based on the cost curve, the demand end agent adjusts the order distribution ratio based on demand elasticity, and the manufacturer agent balances raw material procurement and production rhythm;

[0011] Dynamic matching execution step: the shared data center starts the matching process according to the game strategy, and allocates demand according to the priority of emergency order> high profit order> regular order; real-time monitoring of supply and demand deviation, when the supply and demand deviation> 8%, trigger dynamic negotiation; the matching result is fed back to each agent, and the matching list is generated;

[0012] The matching effect multi-dimensional evaluation step: an evaluation index system is constructed, the first level index is supply-demand matching degree, cost control rate, response timeliness and cooperation stability; supply-demand matching degree = actual matching amount / theoretical optimal matching amount x 100%, cost control rate = (target cost-actual cost) / target cost x 100%, response timeliness = (promised delivery time-actual delivery time) / promised delivery time x 100%; the analytic hierarchy process is used to calculate the comprehensive score, and three levels of "excellent", "qualified" and "unqualified" are divided;

[0013] Strategy iteration optimization step: adjust the game parameters according to the evaluation results, the "excellent" strategy is included in the optimal strategy library, the "qualified" scene is fine-tuned the constraint condition, and the "unqualified" scene is regenerated strategy; update the strategy library every month, eliminate invalid strategies, and supplement new scene strategies.

[0014] Further, it further includes a supply-demand gap prediction module, which predicts the supply-demand gap in the future period through historical data and real-time parameters, provides a prediction basis for game strategy generation, and the prediction formula is: , wherein, is is the cumulative supply-demand gap in the period, is the prediction start time, is the prediction end time, is the real-time demand rate at time t, is the real-time supply rate at time t; through the prediction, the supply-demand imbalance risk in the next 12-24 hours is identified in advance, when is greater than the preset threshold, the game strategy is triggered to adjust in advance, and the matching lag is reduced.

[0015] Further, it further includes a game profit distribution mechanism, which distributes the collaborative income according to the contribution of each agent in the matching process, and the income distribution formula is: , wherein, is the income of the i-th agent, is the total income of the supply chain cooperation, is the contribution of the i-th agent, is the total contribution of all game agents, K is the cooperation coefficient, P i is the cooperation satisfaction of the i-th agent.

[0016] Further, the communication security mechanism is added in the multi-agent system construction: each agent adopts a double-key system, and identity authentication is performed through RSA-2048 algorithm before data interaction, and a temporary encryption channel is established after the authentication; for critical data, the "sharded storage + multi-signature access" strategy is adopted, the data is split into 3-5 pieces and stored in different agent nodes, and the digital signature of ≥2 nodes is required to read; the communication key is updated regularly, the abnormal communication behavior is detected, and when the abnormality is triggered, the communication is temporarily blocked and the alarm is pushed to the supply chain management end.

[0017] Further, the priority rule is optimized in the dynamic matching execution: a multi-dimensional priority evaluation model is constructed, in addition to the basic "order urgency", a "supply chain collaboration value" index is added; the urgent order is classified according to "delivery time difference", the time difference < 24 hours is first level, 24-48 hours is second level, > 48 hours is third level, the collaboration value is classified according to "weighted score", the score ≥ 80 points is A level, 60-79 points is B level, < 60 points is C level; the matching priority is determined according to "emergency level + collaboration value level" combination, and the overall income of the supply chain is improved.

[0018] Further, the abnormal data self-healing mechanism is added in the real-time sensing of supply and demand information: when the parameters collected by an agent are interrupted or deviate, and the deviation > 5σ, the self-healing process is started; short-term interruption, < 5 minutes, "historical trend prediction + adjacent agent data completion" is adopted; long-term interruption, ≥ 5 minutes, trigger the backup collection channel; after data recovery, the self-healing accuracy is optimized through "deviation correction algorithm", the continuity of sensing data is ensured, and the game strategy is prevented from being invalid due to data loss.

[0019] Further, the strategy adjustment rate model is introduced in the strategy iteration optimization, the optimal adjustment amplitude of the strategy parameter is calculated through the derivative, and the formula is: wherein, v is the strategy adjustment rate, k is the adjustment coefficient, E is the comprehensive score of the matching effect, is the core parameter of the game strategy, is the derivative of the comprehensive score to the strategy parameter; when is positive, the same direction adjustment is taken to improve E; when is negative, the reverse adjustment is taken to avoid the matching fluctuation caused by the too large adjustment amplitude of the strategy.

[0020] Further, it also includes demand fluctuation coping mechanism: the demand side agent monitors the demand fluctuation coefficient in real time, and when the fluctuation coefficient is greater than 1.5, it is determined as "large fluctuation", triggering fluctuation warning; after receiving the warning, the supplier and the manufacturer agent start "capacity elastic reserve", the supplier reserves 10-15% of the capacity, and the manufacturer adjusts the production shift to "three shifts"; the "fluctuation coping constraint" is added when the game strategy is generated, which allows the supplier to increase the price by no more than 8% when the demand increases, and the delivery time is promised to be unchanged; the "batch delivery" mode is adopted in dynamic matching, and large orders are split into 3-5 batches, which are delivered according to the demand rhythm, reducing the risk of inventory backlog or shortage in the supply chain.

[0021] Further, the index calculation logic is refined in the matching effect evaluation: supply-demand matching degree = (actual matching amount-supply-demand deviation amount) / theoretical optimal matching amount x 100%, wherein the supply-demand deviation amount includes excess matching and shortage matching, and the weight of shortage matching is 1.5 times of that of excess matching; cost control rate = (target cost-actual cost-emergency cost) / target cost x 100%, emergency cost refers to the adjustment cost when the matching is abnormal; response timeliness = (promised delivery time-actual delivery time-unavoidable delay) / promised delivery time x 100%, the unavoidable delay needs to be proved by a third party; through the refined calculation, the evaluation result is more in line with the actual operation scene.

[0022] Further, it also includes a blockchain storage and audit module: the supply and demand data, game strategy parameters, matching results and key information of income distribution records of each agent are uploaded to the alliance chain in the format of "time stamp + agent signature", and the block generation interval is less than or equal to 5 minutes; the audit node queries the on-chain data in real time, and verifies the compliance of the matching process; when there is a dispute, the on-chain storage data is called for tracing, and the tracing result is used as the basis for dispute resolution; through the blockchain storage, the whole matching process is traceable and tamper-proof, and the trust of supply chain collaboration is improved.

[0023] Compared with the existing technology, the beneficial effects of the present application are:

[0024] In the aspect of information interaction, the multi-agent system breaks the hierarchical information transmission mode of the traditional supply chain, and each link agent realizes real-time data sharing through an efficient communication protocol, while the encryption technology and security mechanism ensure that sensitive information is not leaked, which not only solves the problem of information asymmetry, but also eliminates the participants' concerns about data security, making the supply and demand data transmission more timely and more reliable.

[0025] In terms of supply-demand matching accuracy and collaborative efficiency, the introduction of game strategy realizes the balance of the interests of each participant and the overall optimization. Unlike the limitations of each link in the traditional method, the present application takes Nash equilibrium and Pareto optimality as the goal, so that the suppliers, manufacturers, demanders and other agents can consider their own interests and the overall efficiency of the supply chain in the game process, avoiding the global inefficiency caused by local maximum; the dynamic matching mechanism can monitor the supply-demand deviation in real time, quickly trigger negotiation adjustment, reduce the mismatch caused by demand fluctuations or supply changes, and the supply-demand gap prediction function can identify potential imbalance risks in advance, reducing the probability of supply-demand mismatch.

[0026] The matching effect evaluation system can comprehensively measure supply-demand matching, cost control, response timeliness and other multi-dimensional performances, provide accurate basis for strategy iteration, and make the system continuously optimize according to market changes; the strategy adjustment mechanism ensures that the game parameters or constraint conditions can be adjusted scientifically when the matching effect is not good, improving the fault tolerance and evolution ability of the system. In addition, the blockchain storage and audit module makes the matching process traceable and tamper-proof, provides reliable basis for dispute resolution, greatly enhances the trust between each link and reduces the collaborative friction. Overall, the present application makes the supply chain supply-demand matching more accurate, the collaboration more efficient, and the operation more stable, providing strong support for the sustainable development of multi-node supply chain. BRIEF DESCRIPTION OF DRAWINGS

[0027] Fig. 1 The schematic diagram of the supply chain supply-demand dynamic matching method based on multi-agent game proposed by the present application is shown in the figure;

[0028] Fig. 2 The figure for the income distribution of each agent and the contribution degree;

[0029] Fig. 3 The comparison figure of order priority and matching response time. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0031] In the description of the present application, it should be understood that the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application.

[0032] In addition, the terms "first", "second" are only for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited. In addition, the terms "mounting", "connection", "connection" should be broadly understood, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be the communication between two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances, and the present application will be further described in detail below with reference to the drawings.

[0033] Referring to Figs. 1 to 3 A supply chain supply and demand dynamic matching method based on multi-agent game, comprising the following steps:

[0034] The multi-agent system is constructed, first dividing the types of agents at each link of the supply chain: the supplier agent is mainly responsible for capacity reporting and offer generation, the manufacturer agent undertakes production plan control and raw material demand coordination, the distributor agent carries out inventory level statistics and distribution demand analysis, and the demand side agent carries out market order collection and demand forecasting. Each agent is equipped with an edge computing module, the module computing power is ≥2TOPS, and a blockchain P2P communication protocol and an MQTT Internet of Things protocol are used to realize cross-agent data interaction, ensuring that the communication delay is ≤100ms. In addition, an agent interaction standard is established, and the data format is a JSON structure, which includes parameter identification, timestamp and data precision; sensitive data such as offers and costs are encrypted using AES-256 to ensure information transmission security.

[0035] The supply and demand information is perceived in real time, and each agent collects core parameters through sensors and system interfaces: the supplier agent collects capacity utilization rate (value range 0-100%), raw material inventory (measured in units), and delivery period (measured in days); the manufacturer agent collects work-in-process quantity, equipment failure rate (value range 0-5%), and production energy consumption; the distributor agent collects warehouse turnover rate and regional demand density; and the demand-side agent collects order quantity (measured in units / hour), demand fluctuation coefficient (value range 0-2), and customer satisfaction (value range 0-100 points). The parameter collection frequency is dynamically adjusted according to the type, with high-frequency parameters such as order quantity collected at a frequency of 1 time / minute, and low-frequency parameters such as capacity collected at a frequency of 1 time / hour; after the collected data is cleaned by Kalman filtering algorithm and outliers with deviation > 3σ are removed, it is synchronized to the supply chain shared data platform.

[0036] Game strategy generation, a game model is constructed with Nash equilibrium and Pareto optimality as the goal: each agent is an independent game participant and sets a target function, where the supplier agent aims to minimize cost and the demand-side agent aims to maximize satisfaction; at the same time, the game constraints are clarified, including capacity upper limit, inventory threshold, and transportation time limit (requirement ≤ 72 hours). An iterative game algorithm is used to generate the initial strategy, with the number of iterations ≤ 50 times; during the generation process, the supplier agent formulates a quote based on the cost curve (including raw material cost and production marginal cost), the demand-side agent adjusts the order allocation ratio based on demand elasticity, and the manufacturer agent balances raw material procurement rhythm and production progress; the final initial strategy must meet the requirement of "total supply and demand matching deviation ≤ 5%".

[0037] Dynamic matching execution, the supply chain shared data platform starts the matching process according to the generated game strategy: first, allocate demand according to "priority rules", with the priority order being emergency orders > high-profit orders > regular orders; then the supplier agent responds to raw material supply requests, the manufacturer agent adjusts production scheduling synchronously, and the distributor agent optimizes inventory scheduling. During the matching process, the supply and demand deviation is monitored in real time, and when the deviation > 8%, the agent triggers dynamic negotiation, including adjusting the supply amount or demand allocation ratio, with a single adjustment response time ≤ 30 seconds; after the matching is completed, the results are fed back to each agent in real time, and a matching list is generated, which contains the participants, matching quantity, delivery time, and cost.

[0038] Multi-dimensional evaluation of matching effect, and construction of a hierarchical evaluation index system: the first-level indexes include supply-demand matching degree (weight 0.4), cost control rate (weight 0.3), response timeliness (weight 0.2) and cooperation stability (weight 0.1). The calculation formulas of the first-level indexes are as follows: supply-demand matching degree = actual matching amount / theoretical optimal matching amount * 100%, cost control rate = (target cost - actual cost) / target cost * 100%, response timeliness = (promised delivery time - actual delivery time) / promised delivery time * 100%. The analytic hierarchy process is used to calculate the comprehensive score, and the score range is 0-100 points, wherein 85 points or more are determined as “excellent”, 60-84 points are determined as “qualified”, and 60 points or less are determined as “unqualified”.

[0039] Strategy iteration optimization, and adjustment of game parameters according to the matching effect evaluation results: for the “excellent” scene, the current game strategy is retained and is included in the optimal strategy library; for the “qualified” scene, the game constraint conditions are fine-tuned, for example, the production capacity fluctuation threshold is widened by 1-2%; for the “unqualified” scene, the game strategy is regenerated, and the upper limit of the iteration number is increased to 80 times. The optimal strategy library is updated every month, and the invalid strategies that are not called for 3 times in a row are excluded, and new scene strategies such as sudden demand surge and supplier disruption are supplemented, so as to adapt to the dynamic change demand of the supply chain.

[0040] In the present application, a supply-demand gap prediction module is further included, the future period supply-demand gap is predicted through historical data and real-time parameters, and a prediction formula is provided for the game strategy generation, and the prediction formula is: wherein, is to the cumulative supply-demand gap (unit: pieces) in the period, is the prediction start time (unit: hour), is the prediction end time (unit: hour), is the real-time demand rate at t time (unit: pieces / hour, fitted by the demand end agent based on order quantity), is the real-time supply rate at t time (unit: pieces / hour, calculated by the supplier and the manufacturer agent in cooperation); through the prediction, the supply-demand imbalance risk in the future 12-24 hours can be identified in advance, when > preset threshold (such as 10% of the total supply amount), the game strategy is triggered to be adjusted in advance, and the matching lag is reduced.

[0041] In the present application, a game income distribution mechanism is further included, and the collaborative income is distributed according to the contribution of each agent in the matching process, and the income distribution formula is: wherein, is the income of the i th agent (unit: yuan), The total revenue of the supply chain cooperation (unit: yuan, which is the sum of cost savings and profit increases after matching), The contribution degree of the i-th agent (0-1, based on capacity input, response speed, and cost control calculation), The total contribution degree of all participating agents (j=1, 2,..., n, n is the number of agents), K is the cooperation coefficient (0.1-0.3, encouraging deep cooperation of multiple agents), P i The cooperation satisfaction of the i-th agent (0-1, obtained by mutual evaluation of other agents); through the distribution mechanism, the "free riding" phenomenon can be avoided, and the enthusiasm of each agent participating in the game can be improved.

[0042] In the present application, a communication security mechanism is added in the construction of the multi-agent system: each agent adopts a double key system, the private key is stored in the local security chip, and the public key is uploaded to the blockchain. Before data interaction, identity authentication is completed through the RSA-2048 algorithm, and after authentication, a temporary encrypted channel is established to ensure the safety of data transmission. For key data such as cost details and demand prediction models, the "sharded storage + multi-signature access" strategy is adopted, the data is first split into 3-5 pieces and stored in different agent nodes, and when reading, the digital signature of ≥2 nodes needs to be obtained to operate. The system will update the communication key regularly, the update cycle is every 24 hours, and at the same time, it will detect abnormal communication behavior in real time, such as high-frequency invalid requests, data format abnormalities, etc., when the abnormality is triggered, the communication will be temporarily blocked and the alarm will be pushed to the supply chain management end.

[0043] In the present application, the priority rule is optimized in dynamic matching execution: a multi-dimensional priority evaluation model is constructed, in addition to the basic "order urgency" index, a "supply chain cooperation value" index is added, which includes three dimensions of long-term cooperation time, historical fulfillment rate and emergency response ability. Among them, the urgent order is classified according to "delivery time difference", the delivery time difference < 24 hours is classified as level one, 24-48 hours is classified as level two, and > 48 hours is classified as level three; the supply chain cooperation value is classified according to "weighted score", the weighted score ≥ 80 points is classified as A level, 60-79 points is classified as B level, and < 60 points is classified as C level. The order matching priority is determined by the combination of "urgency level + cooperation value level", for example, the priority of level one + A level is higher than that of level one + B level, and the priority of level one + B level is higher than that of level two + A level, through the determination rule, the overall revenue of the supply chain is improved.

[0044] In the present application, abnormal data self-recovery mechanism is added in real-time sensing of supply and demand information: when the parameters collected by an intelligent agent are interrupted (such as sensor failure) or deviation exceeds the standard (deviation> 5σ), the self-recovery process is started; short-term interruption (<5 minutes) adopts "historical trend prediction + adjacent intelligent agent data completion" (such as interruption of manufacturer raw material demand, use the average demand of the previous 1 hour + supplier inventory data to complete); long-term interruption (≥5 minutes) triggers the standby collection channel (such as switching to the standby sensor when the main sensor fails, and enabling the manual input interface if the standby sensor is not deployed); after data recovery, the self-recovery accuracy is optimized through "deviation correction algorithm" (compare the self-recovery data with the actual recovery data, and correct the prediction model) to ensure the continuity of the sensing data and avoid the invalidation of the game strategy due to data loss.

[0045] In the present application, a policy adjustment rate model is introduced in policy iteration optimization, and the optimal adjustment amplitude of the policy parameter is calculated through derivative, and the formula is: Wherein, v is the policy adjustment rate (0-1, the larger the value, the larger the adjustment amplitude), k is the adjustment coefficient, the value is 0.05-0.2, and k takes a small value according to the stability setting of the supply chain, E is the comprehensive score of the matching effect (0-100 points), which is the core parameter of the game strategy, such as the supplier bidding coefficient, the demand allocation proportion coefficient, which is the derivative of the comprehensive score to the strategy parameter, indicating the sensitivity of the parameter change to the score; when is positive, the same direction adjustment is taken to improve E; when is negative, the reverse adjustment is taken, which avoids the matching fluctuation caused by the too large adjustment amplitude of the strategy and improves the stability of the iteration optimization.

[0046] In the present application, it also includes a demand fluctuation response mechanism: the demand end intelligent agent monitors the demand fluctuation coefficient in real time, and when the fluctuation coefficient> 1.5, it is determined as "large fluctuation", and the fluctuation warning is triggered immediately. After receiving the warning, the supplier intelligent agent and the manufacturer intelligent agent start the "capacity elastic reserve" measure immediately - the supplier reserves 10-15% of the capacity to respond to potential demand changes, and the manufacturer adjusts the production shift to "three shifts" to improve the production capacity. In the game strategy generation link, the "fluctuation response constraint" condition is added, for example, when the demand increases sharply, the supplier is allowed to increase the price but the amplitude is ≤8%, and at the same time the supplier needs to promise that the delivery time remains unchanged. In the dynamic matching process, the "batch delivery" mode is adopted, and large orders are split into 3-5 batches, and delivered in turn according to the actual demand rhythm, which reduces the risk of inventory accumulation or material shortage in the supply chain.

[0047] In the present application, the matching effect evaluation is refined by the index calculation logic: the supply-demand matching degree calculation formula is (actual matching amount-supply-demand deviation amount) / theoretical optimal matching amount*100%, wherein the supply-demand deviation amount includes "excessive matching" (supply is greater than demand) and "shortage matching" (supply is less than demand), and the weight of shortage matching is set to 1.5 times of that of excessive matching. The cost control rate calculation formula is (target cost-actual cost-emergency cost) / target cost*100%, wherein the emergency cost refers to the adjustment cost generated when the supply-demand matching is abnormal, such as urgent transportation cost, temporary capacity leasing cost, etc. The response timeliness calculation formula is (promised delivery time-actual delivery time-uncontrollable delay) / promised delivery time*100%, wherein the uncontrollable delay needs to provide third-party proof materials, such as weather warning notice, traffic control file, etc. Through the refined calculation logic of each index, the evaluation result is more suitable for the actual operation scene of the supply chain, and provides accurate basis for the iteration optimization of subsequent game strategy.

[0048] In the present application, the blockchain storage and audit module is also included: the key information of each agent, such as supply-demand data, game strategy parameters, matching results, and income distribution records, is uploaded to the consortium chain (such as Fabric blockchain) in the format of "time stamp+agent signature", and the block generation interval is ≤5 minutes. The audit node is jointly assumed by the core enterprises of the supply chain and the third-party institutions, which can query the on-chain data in real time and verify the compliance of the matching process, such as checking whether the matching is according to the priority rules and whether the income distribution is consistent with the preset formula. When there is a dispute (such as the agent is not satisfied with the income distribution result), the on-chain storage data is called to trace the source, and the result obtained by tracing the source is directly used as the basis for dispute resolution. Through the way of blockchain storage, the entire supply-demand matching process is traceable and the data is tamper-proof, thereby improving the trust degree in the supply chain collaboration process.

[0049] The specific implementation of the system is further illustrated by two embodiments as follows:

[0050] Embodiment one: supply-demand dynamic matching of automobile parts manufacturing industry supply chain

[0051] This embodiment is aimed at the automobile parts supply chain (covering engine suppliers, tire suppliers, automobile manufacturers, regional distributors, and demand side of automobile manufacturers), realizes the supply-demand collaboration of core parts such as engines and tires, solves the problem of "capacity mismatch and delivery delay" in traditional supply chain, and the specific execution process is as follows:

[0052] Multi-agent system construction

[0053] Four types of agents are classified and hardware and communication modules are configured: the supplier agent includes 3 engine suppliers (A1, A2, A3) and 2 tire suppliers (B1, B2), each equipped with an NVIDIA Jetson Xavier NX edge computing module (21 TOPS of computing power, 8 GB of memory), connected to the production PLC system through a PCIe interface; the manufacturer agent is the vehicle manufacturer (C1), with an edge module using Huawei Atlas300I (8 TOPS of computing power), connected to the MES production execution system; the distributor agent is the North China and East China regional warehouse center (D1, D2), with a module of Raspberry Pi 4B (0.5 TOPS of computing power, with a 4G module); the demand-side agent is the vehicle manufacturer's sales department (E1), with a module integrated into the ERP system (2 TOPS of computing power). The communication uses Hyperledger Fabric consortium chain (8 nodes, block generation interval of 5 minutes) and MQTT v3.1.1 protocol (QoS level 2, ensuring message delivery), and sensitive data (such as A1's cost details) are encrypted by AES-256, with the key stored by the agent's local security chip (using the national standard SM4 algorithm).

[0054] Real-time perception of supply and demand information

[0055] Each agent collects parameters at a dynamic frequency: engine supplier A1 collects order quantity every 1 minute (read from the ERP system, units: units / hour), capacity utilization every 1 hour (real-time transmission from PLC, range 0-100%, current 85%), raw material inventory (warehouses management system, aluminum alloy inventory 200 tons), delivery period (calculated by production scheduling system, current 3 days); vehicle manufacturer C1 collects work-in-process quantity every 5 minutes (MES system, engine work-in-process 150 units), equipment failure rate (equipment management system, punch press failure rate 1.2%); distributor D1 collects warehouse turnover rate every 30 minutes (inventory management system, current 3.2 times / month), regional demand density (based on order distribution in the past 24 hours, North China region 120 units / day); demand-side E1 collects market order quantity every 1 minute (sales system, current 180 units / hour), demand fluctuation coefficient (based on data in the past 12 hours, 1.2), customer satisfaction (after-sales system, 92 points). The collected data is cleaned by Kalman filtering (process noise covariance Q=0.01, measurement noise covariance R=0.1), and after removing an abnormal capacity data of A1 (120%, deviation 3.2σ), it is synchronized to the supply chain shared data platform (deployed on an Ali Cloud ECS server with 8 cores and 16 GB of memory).

[0056] Game strategy generation and dynamic matching

[0057] A game model is constructed with "Nash equilibrium" as the goal: the objective function of A1 is to minimize the cost (C = 600x + 3x², x is the daily engine production capacity), and the constraint condition is that the production capacity is upper limited to 500 units / day and the raw material inventory is greater than or equal to 50 tons; the objective function of E1 is to maximize the satisfaction (S = 95 - 0.02y, y is the delivery delay hours), and the constraint condition is that the daily demand is greater than or equal to 450 units. An iterative game algorithm is adopted: in the first iteration, A1 offers 1300 yuan / unit, E1 allocates 300 units, and the supply-demand deviation is 6.7%; in the second iteration, A1 adjusts the offer based on the cost curve to 1280 yuan / unit, E1 allocates 350 units, the deviation is 4.4% (≤5%), and the initial strategy is generated. In dynamic matching, E1 initiates an emergency order (delivery time limit 20 hours, first-class emergency degree, A-class collaborative value, long-term cooperation 5 years, and fulfillment rate 98%), and the middle station allocates according to the "first-class + A-class" priority: A1 supplies 200 units, A2 supplies 180 units, D1 dispatches 70 units from the inventory, the total supply is 450 units, the demand is 450 units, and the matching deviation is 0. When A2 has a sudden equipment failure, the actual supply decreases to 160 units, and the deviation is 8.9% (>8%), triggering negotiation: A1 temporarily increases the production capacity by 20 units (starts a backup production line, response time 25 seconds), and A2 repairs the equipment, and the supply is restored after 3 hours, and the matching returns to balance.

[0058] Formula application and effect evaluation

[0059] Perform supply-demand gap prediction: t1=0, t2=24 hours, D(t)=180+30sin(πt / 12) (simulate daily demand fluctuation), S(t)=170+25sin(πt / 12), substitute into the formula G(0,24)=∫0² 4 [180+30sin(πt / 12)-170-25sin(πt / 12)]dt=∫0² 4 [10+5sin(πt / 12)]dt=240+(-60 / π)×(cos2π-cos0)=240 units, predict that the cumulative gap in 24 hours is 240 units, and adjust A1 production capacity to 480 units / day in advance. In the revenue distribution, R tot =60 million yuan (cost savings 4 million yuan + profit increment 2 million yuan), A1's C i =0.3 (production capacity input 480 units, response time 25 seconds), ΣC j =1.2 (A1=0.3, A2=0.25, C1=0.35, D1=0.2, E1=0.1), K=0.2, Pᵢ=0.9 (other intelligent agents mutual evaluation), substitute Rᵢ=60×(0.3 / 1.2)×(1+0.2×0.9)=60×0.25×1.18=17.7 million yuan, A1's revenue is reasonable, and the cooperation enthusiasm is improved.

[0060] Table 1: Comparison of matching effect of example one and traditional supply chain

[0061]

[0062] Table 1 data reflects the advantages of the present application: the supply-demand matching degree is improved due to the real-time collaboration and dynamic negotiation of multiple agents, avoiding the mismatch caused by "information lag" in traditional methods; the cost control rate is improved due to the balance of bidding and demand through game strategy, reducing redundant procurement; the response to emergency orders is fast due to the priority rules and fast negotiation mechanism, while the traditional method needs manual coordination of multiple levels; the prediction accuracy rate is 92%, which can avoid gaps in advance, while the traditional method has no prediction ability and often responds passively to shortages. The overall adaptation to the "multi-variety, high timeliness" demand of automobile parts supply chain reduces the risk of production stagnation.

[0063] Example two: dynamic matching of supply and demand in fresh e-commerce supply chain

[0064] This example is aimed at the fresh e-commerce supply chain (farmer suppliers, cold chain logistics, regional warehouse distributors, platform demand side), solves the problem of "easy loss and large demand fluctuation" of fresh food, and realizes accurate matching of supply and demand and loss reduction.

[0065] Multi-agent system and information perception

[0066] Intelligent agent division: farmer suppliers (F1-F5, vegetable growers), cold chain logistics (L1-L3), regional warehouses (W1-W4), e-commerce platforms (P1). F1-F5 are equipped with Huawei Atlas200IDKA2 edge modules (8TOPS of computing power), which collect field harvest quantities through LoRa sensors; L1-L3 modules are integrated into cold chain vehicles (5TOPS of computing power), which collect GPS positions and vehicle compartment temperatures (0-4℃) in real time; W1-W4 modules are Hikvision AI boxes (4TOPS of computing power), which access warehouse temperature and humidity systems; P1 module is deployed on e-commerce cloud servers (32TOPS of computing power). Communication uses FISCOBCOS alliance chain (6 nodes) and MQTT QoS level 1, and sensitive data (F1's purchase price) is encrypted by AES-256.

[0067] Information collection frequency: F1 collects yield (unit: kg / hour), capacity utilization rate (80%, based on planting area), loss rate (5%) every 30 minutes; L1 collects transportation timeliness (current 4 hours / 100 kilometers), temperature deviation (±0.5℃) every 10 minutes; W1 collects inventory turnover rate (2.5 times / day), regional demand density (5000 kg / day in South China) every 20 minutes; P1 collects order quantity (current 800 kg / hour), demand fluctuation coefficient (1.6, peak during holidays), customer satisfaction (88 points) every 5 minutes. After Kalman filtering (Q=0.02, R=0.15) and removing one abnormal temperature data (8℃, deviation 4σ) of L2, the data is synchronized to the middle platform.

[0068] Game strategy and dynamic matching optimization

[0069] Game objective: F1's cost function C=8y+2y² (y is yield, unit: 100 kg), with constraint condition loss rate ≤8%; P1's satisfaction function S=90-0.05z (z is out-of-stock rate%), with constraint condition daily demand ≥12000 kg. Iterative game: F1 offers 5 yuan / kg for the first time, P1 allocates 1500 kg of orders, with a deviation of 8%; F1 adjusts the offer to 4.8 yuan / kg for the second time, allocates 1800 kg, with a deviation of 3.3%, and generates a strategy. During dynamic matching, P1's demand fluctuation coefficient rises to 1.8 (>1.5) due to holidays, triggering a warning: F1-F5 reserve 15% capacity (F1 increases 200 kg of yield), L1-L3 adjust to "24-hour cold chain transportation", and the game strategy adds a new constraint (F1's offer increases by 7%, but promises 24-hour delivery); W1 splits large orders (3000 kg) into 3 batches (1000 kg / batch, with 6-hour intervals) to avoid inventory accumulation.

[0070] When F3 sensor fails (long-term interruption, ≥5 minutes), self-healing is started: use the average yield in the previous 1 hour (800 kg / hour) + adjacent data of F2 (F2 yield 850 kg / hour, F3 and F2 have similar planting areas) to complete, and after data recovery, modify the prediction model (self-healing data deviates from actual data by 3%, adjust model parameters). During strategy iteration, E=78 points (pass), θ=F1 offer coefficient 0.9 (current offer 4.8 yuan=5×0.96, θ=0.96 here), dE / dθ=0.6 (offer coefficient increases by 0.1, E increases by 6 points), k=0.1 (supply chain is relatively stable), substitute v=0.1×0.6=0.06, fine-tune θ to 0.98, F1 offer 4.9 yuan / kg, E rises to 83 points.

[0071] Blockchain storage and effect evaluation

[0072] Blockchain record: F1's harvest amount (800 kg / hour, timestamp 2024-10-0108:00), L1's transportation record (GPS coordinates 113.2°E, 23.1°N, temperature 2°C), matching result (F1 supply 1800 kg, L1 transportation 1500 kg), and income distribution (R tot = 300,000 yuan, F1's Rᵢ= 30×(0.25 / 1.0)×(1+0.2×0.85)= 8.1,000 yuan) are uploaded on the chain, and audit nodes (e-commerce platform and agricultural bureau) can query. When F1 has objections to the income, the on-chain Cᵢ calculation record (capacity input 1800 kg, response time 30 seconds) is retrieved, the matching rules are confirmed, and the dispute is resolved within 1 hour.

[0073] Table 2: Comparison of matching effects between Example Two and traditional fresh food supply chain

[0074]

[0075] Table 2 data highlights the adaptability of fresh food scenarios: loss rate reduction due to dynamic matching and batch delivery, avoiding traditional "overstocking"; fast response to fluctuations due to early warning and flexible capacity, traditional methods require manual coordination between farmers and logistics; fast dispute resolution due to blockchain record, traditional methods rely on paper records for traceability; strategy iteration improves satisfaction, traditional methods have no optimization mechanism and long-term low satisfaction. This method solves the "time-sensitive and high loss" pain points of fresh food supply chain, ensuring food freshness and supply stability.

[0076] Reference Fig. 2 , the chart is based on the income distribution formula (K=0.2) calculation, =0.32+0.35+0.21+0.18=1.06. Manufacturer B contributes the most (0.35, responsible for production coordination and scheduling adjustment), with an income of 21.3 million yuan; demand side D has the lowest contribution (0.18), with an income of 11.2 million yuan, consistent with the "more effort, more income" logic. Traditional equal distribution gives each 20 million yuan, but ignores contribution differences, which may lead to a decrease in supplier A's (high contribution) enthusiasm. The distribution mechanism of the invention avoids the "free rider" phenomenon, such as distributor C, who has a high level of satisfaction (0.95), with an income of 0.8 million yuan more than simply calculating based on contribution, effectively improving the willingness of each agent to cooperate deeply.

[0077] Reference Fig. 3, The chart embodies the advantages of the multi-dimensional priority rule of the application. The traditional single priority does not distinguish the order urgency and the collaborative value, the response time of all orders is 120 seconds, the completion rate is only 85.2%, and the high-value urgent order cannot be processed in priority. In the application, the "first + A" order responds the fastest, the completion rate is 100%, and the customer satisfaction is 96 points; even if the "first + B" order has the same urgency, due to the slightly lower collaborative value, the response time is extended to 28 seconds, ensuring that resources are tilted to high-value demand. The rule balances urgency and long-term supply chain interests, avoids the one-sidedness of traditional methods "heavy emergency and light value" or "heavy value and light emergency", and improves overall collaborative benefits.

[0078] The above is only a preferred embodiment of the application, but the protection scope of the application is not limited thereto, any person skilled in the art can make equivalent replacement or change within the technical range disclosed by the application according to the technical scheme and the inventive concept of the application, which should be covered within the protection scope of the application.

Claims

1. A method for dynamic matching of supply and demand in a supply chain based on multi-agent game theory, characterized in that, Includes the following steps: The steps for building a multi-agent system are as follows:

1. Divide the supply chain agents into types, including suppliers, manufacturers, distributors, and demand-side agents; 2. Each agent is equipped with an edge computing module and uses the blockchain P2P communication protocol and the MQTT IoT protocol for data interaction; 3. Establish agent interaction standards, define JSON format data, and use AES-256 to encrypt sensitive data. Real-time supply and demand information sensing steps: Each intelligent agent collects core parameters through sensors and system interfaces. The collection frequency is dynamically adjusted according to the parameters. The data is cleaned by the Kalman filter algorithm to remove outliers and then synchronized to the supply chain shared data platform. Game strategy generation steps: Each agent, as a participant in the game, sets an objective function and uses an iterative game algorithm to generate an initial strategy. The supplier agent quotes prices based on the cost curve, the demand-side agent adjusts the order allocation ratio based on demand elasticity, and the manufacturer agent balances raw material procurement and production rhythm. Dynamic matching execution steps: The shared data platform initiates the matching process according to the game strategy, allocating demand according to the priority of urgent orders > high-profit orders > regular orders; it monitors the supply and demand deviation in real time, and triggers dynamic negotiation when the supply and demand deviation is greater than 8%; The matching results are fed back to each agent, generating a matching list; Multi-dimensional evaluation steps for matching effect: Construct an evaluation indicator system, with primary indicators being supply and demand matching degree, cost control rate, response timeliness, and collaboration stability; Supply and demand matching degree = actual matching quantity / theoretical optimal matching quantity × 100%; cost control rate = (target cost - actual cost) / target cost × 100%; response timeliness = (promised delivery time - actual delivery time) / promised delivery time × 100%; the comprehensive score is calculated using the analytic hierarchy process and divided into three levels: "excellent", "qualified" and "unqualified". Strategy iteration and optimization steps: Adjust game parameters based on evaluation results, include "excellent" strategies into the optimal strategy library, fine-tune constraints for "qualified" scenarios, and regenerate strategies for "unqualified" scenarios; update the strategy library monthly, remove ineffective strategies, and add strategies for new scenarios.

2. The supply chain dynamic matching method based on multi-agent game theory as described in claim 1, characterized in that, It also includes a supply-demand gap prediction module, which predicts the supply-demand gap in future periods using historical data and real-time parameters, providing a predictive basis for generating game strategies. The prediction formula is as follows: ,in, for to The cumulative supply and demand gap over the period. To predict the start time, To predict the end time, Let be the real-time demand rate at time t. Let t be the real-time supply rate at time t; this forecast can be used to identify the risk of supply-demand imbalance in the next 12-24 hours in advance. >When a preset threshold is reached, the game strategy is adjusted in advance to reduce matching lag.

3. The supply chain dynamic matching method based on multi-agent game theory as described in claim 1, characterized in that, It also includes a game-theoretic payout mechanism, which distributes collaborative payouts based on the contributions of each agent in the matching process. The payout distribution formula is as follows: ,in, For the i-th agent, For the total benefits of supply chain collaboration, The contribution of the i-th agent. The sum of contributions from all participating agents in the game, where K is the cooperation coefficient and P is the total contribution. i Let be the collaboration satisfaction of the i-th agent.

4. The supply chain dynamic matching method based on multi-agent game theory according to claim 1, characterized in that, A communication security mechanism is incorporated into the construction of the multi-agent system: each agent adopts a dual-key system, and identity is authenticated through the RSA-2048 algorithm before data interaction. After successful authentication, a temporary encrypted channel is established. For critical data, a "sharded storage + multi-signature access" strategy is adopted, which splits the data into 3-5 pieces and stores them on different agent nodes. Digital signatures from ≥2 nodes are required for reading. The communication key is updated regularly, abnormal communication behavior is detected, and communication is temporarily blocked and an alarm is pushed to the supply chain management end when an anomaly is triggered.

5. The supply chain dynamic matching method based on multi-agent game theory according to claim 1, characterized in that, Dynamic matching optimizes priority rules during execution: A multi-dimensional priority evaluation model is constructed, adding a "supply chain collaboration value" indicator in addition to the basic "order urgency"; urgent orders are graded according to "delivery time difference", with a time difference of <24 hours as level 1, 24-48 hours as level 2, and >48 hours as level 3; collaboration value is graded according to "weighted score", with a score ≥80 as level A, 60-79 as level B, and <60 as level C; matching priority is determined by a combination of "urgency level + collaboration value level" to improve the overall benefits of the supply chain.

6. The supply chain dynamic matching method based on multi-agent game theory according to claim 1, characterized in that, An abnormal data self-healing mechanism is added to the real-time perception of supply and demand information: when the parameters collected by a certain agent are interrupted or the deviation exceeds the standard (deviation > 5σ), the self-healing process is initiated; for short-term interruptions (< 5 minutes), "historical trend prediction + data completion by neighboring agents" is adopted; for long-term interruptions (≥ 5 minutes), the backup acquisition channel is triggered; after the data is recovered, the self-healing accuracy is optimized through "deviation correction algorithm" to ensure the continuity of perception data and avoid the failure of game strategy due to data loss.

7. The supply chain dynamic matching method based on multi-agent game theory according to claim 1, characterized in that, In the iterative optimization of the strategy, a policy adjustment rate model is introduced. The optimal adjustment range of the policy parameters is calculated using the derivative, and the formula is as follows: Where v is the strategy adjustment rate, k is the adjustment coefficient, and E is the overall matching performance score. As a core parameter of the game strategy, The derivative of the overall score with respect to the strategy parameters; when When the value is positive, adjust in the same direction. To increase E; when E is negative, adjust in the opposite direction. This avoids matching fluctuations caused by excessively large strategy adjustments.

8. The supply chain dynamic matching method based on multi-agent game theory according to claim 1, characterized in that, It also includes a demand fluctuation response mechanism: the demand-side intelligent agent monitors the demand fluctuation coefficient in real time, and when the fluctuation coefficient is greater than 1.5, it is judged as "significant fluctuation" and triggers a fluctuation warning; after receiving the warning, the supplier and manufacturer intelligent agents activate "capacity elastic reserve", with suppliers reserving 10-15% of capacity and manufacturers adjusting production shifts to "three shifts"; when the game strategy is generated, a "fluctuation response constraint" is added, allowing suppliers to raise their prices by ≤8% when demand surges, and they must promise that the delivery time remains unchanged; in dynamic matching, a "batch delivery" mode is adopted, splitting large orders into 3-5 batches and delivering them according to the demand rhythm to reduce the risk of supply chain inventory backlog or shortage.

9. The supply chain dynamic matching method based on multi-agent game theory according to claim 1, characterized in that, The detailed calculation logic of indicators in the matching effect evaluation is as follows: Supply and demand matching degree = (actual matching amount - supply and demand deviation amount) / theoretical optimal matching amount × 100%, where the supply and demand deviation amount includes over-matching and shortage matching, and the weight of shortage matching is 1.5 times that of over-matching; Cost control rate = (target cost - actual cost - emergency cost) / target cost × 100%, where emergency cost refers to the adjustment cost when matching is abnormal; Response timeliness = (promised delivery time - actual delivery time - force majeure delay) / promised delivery time × 100%, where force majeure delay requires third-party proof; Through detailed calculation, the evaluation results are made more consistent with the actual operation scenario.

10. The supply chain dynamic matching method based on multi-agent game theory according to claim 1, characterized in that, It also includes a blockchain evidence storage and auditing module: key information such as supply and demand data, game strategy parameters, matching results, and profit distribution records of each intelligent agent are uploaded to the consortium blockchain in the format of "timestamp + intelligent agent signature", with a block generation interval of ≤5 minutes; audit nodes query on-chain data in real time to verify the compliance of the matching process; when disputes arise, the on-chain evidence storage data is retrieved for tracing, and the tracing results serve as the basis for dispute adjudication; blockchain evidence storage makes the entire matching process traceable and tamper-proof, improving the trust in supply chain collaboration.

Images