Search

CN-122022373-A - Intelligent purchasing source-seeking method and system based on multi-objective dynamic optimization

CN122022373ACN 122022373 ACN122022373 ACN 122022373ACN-122022373-A

Abstract

The invention discloses an intelligent purchasing source searching method and system based on multi-objective dynamic optimization, which relate to the technical field of intelligent supply chain management and comprise the following steps of S1, constructing a multi-objective scene tree and strategy plan library, enumerating typical conflict modes possibly occurring among a plurality of targets of cost, quality, delivery and risk in purchasing process based on historical data and industry knowledge, setting a plurality of weight proportioning intervals corresponding to each conflict mode, and adopting a simulation deduction algorithm. According to the intelligent purchasing source searching method and system based on multi-objective dynamic optimization, the complex multi-objective dynamic optimization problem is decomposed into two stages of off-line pre-calculation and on-line quick matching, so that when the state of a supply chain changes, the system can quickly lock a feasible strategy set through causal inference according to key signal indexes, and self-adaptive fine adjustment is performed through an on-line learning algorithm, and the response capability of the system to market fluctuation and emergencies is enhanced while the decision timeliness is ensured.

Inventors

  • SUN XUJIE
  • WU XIYUN

Assignees

  • 济南行者网络科技有限公司

Dates

Publication Date
20260512
Application Date
20260227

Claims (10)

  1. 1. An intelligent purchasing source-seeking method based on multi-target dynamic optimization is characterized by comprising the following steps: S1, constructing a multi-target scene tree and a strategy plan library, namely enumerating typical conflict modes possibly occurring among a plurality of targets of cost, quality, delivery and risk in the purchasing process based on historical data and industry knowledge, setting a plurality of corresponding weight proportion intervals for each conflict mode, adopting a simulation deduction algorithm, and performing off-line simulation calculation on different weight proportion intervals under each conflict mode to generate the strategy plan library comprising strategy schemes and applicable boundary conditions and expected effects thereof; S2, real-time monitoring and scene matching, namely, monitoring a group of key signal indexes reflecting the state of a supply chain in real time, triggering a scene matching engine when the monitored key signal index changes beyond a preset threshold value, analyzing a target conflict mode implied by the current key signal index combination based on a preset causal inference model by the scene matching engine, and matching a strategy candidate set associated with the current target conflict mode from the strategy plan library; S3, performing self-adaptive strategy execution and fine adjustment, namely taking the matched strategy candidate set as a strategy set to be selected, performing strategy selection and execution in the strategy set to be selected by adopting an online learning algorithm based on real-time business context information of a current purchasing task, and performing instant evaluation and parameter adjustment on the applicability of the selected strategy according to feedback information generated in the strategy execution process; and S4, generating and displaying an interpretable decision map, namely recording and storing the whole process data and the logic relation from key signal index monitoring, target conflict mode inference, strategy matching and selection to strategy execution and feedback in a correlated manner, forming a decision map taking a causal chain as a core, and performing visual display.
  2. 2. The intelligent purchasing source searching method based on the multi-objective dynamic optimization of claim 1, wherein in step S1, the construction of the multi-objective scene tree and strategy plan library specifically comprises the following steps: S11, defining conflict mode types among a plurality of targets based on supply chain historical abnormal events and business rules, wherein each conflict mode is triggered by at least constraint condition changes among two targets; S12, setting a weight proportioning interval, namely presetting a plurality of different target weight distribution intervals according to each defined conflict mode based on expert experience or historical optimization results, wherein each interval represents one possible target priority trend in the conflict mode; s13, offline simulation deduction, namely simulating the influence of each weight proportion interval set in the step S12 on the final supplier selection and order distribution scheme under different market environment parameters and supply chain events in an offline environment by adopting a Monte Carlo tree search algorithm and combining a game theory equilibrium solving method; And S14, generating a plan library, namely integrating the simulation deduction result of the step S13 to form a structured plan library, wherein each plan record at least comprises an associated conflict mode identification, an applicable weight proportioning interval, a recommended supplier combination or order allocation strategy, an expected cost range of the strategy, an expected risk level and a boundary condition for the strategy to take effect.
  3. 3. The intelligent purchasing source searching method based on multi-objective dynamic optimization of claim 2, wherein in step S2, the key signal indexes comprise at least two of price indexes of core raw materials, congestion state indexes of main logistics channels and preset public opinion alarm levels of a plurality of key suppliers, the causal inference model is a Bayesian network or a structural causal model trained based on historical data, nodes of the causal inference model represent key signal indexes and potential objective conflict modes, and edges represent causal relation strength among variables.
  4. 4. The intelligent purchasing source searching method based on multi-objective dynamic optimization according to claim 3, wherein in the step S3, the online learning algorithm is a contextual multi-arm slot machine algorithm, the real-time business context information comprises the amount of a current purchasing order, the criticality level of a product type and the emergency degree of a demand, the feedback information generated in the policy execution process comprises at least one of the quotation response speed, the order confirmation state and the preliminary quality sampling result of a provider, the instant evaluation and the parameter adjustment are specifically that each purchasing decision is regarded as one slot machine rocker arm selection, the feedback information obtained after policy execution is converted into an instant rewarding value, and the instant rewarding value is used for updating the expected rewarding estimation of the selected policy under the current business context, and further influencing the subsequent policy selection probability.
  5. 5. The intelligent purchasing source searching method based on multi-objective dynamic optimization according to claim 4, wherein the method is characterized by further comprising the step S5 of collecting final completion result data of an actual purchasing task and comparing and analyzing with expected effects in a strategy plan library after the step S4, calibrating causal relation parameters in a definition of a related conflict mode, setting of a weight proportioning interval or a causal relation parameter in a causal deducing model based on a comparison analysis result, recording manual correction operation of a system recommended strategy by a user, and taking the corrected success strategy as new sample data for supplementing or updating the strategy plan library.
  6. 6. The intelligent purchasing source searching method based on the multi-objective dynamic optimization of claim 5, wherein the comparison analysis in the step S5 comprises a counterfactual reasoning analysis, specifically, for the completed purchasing task, based on a decision graph and a causal reasoning model, simulating a result possibly generated if another strategy which is not selected in the strategy candidate set is executed, and comparing the simulation result with an actual result to generate a strategy comparison analysis report, wherein the strategy comparison analysis report is used for guiding the updating direction of the strategy plan library.
  7. 7. An intelligent procurement sourcing system based on multi-objective dynamic optimization for implementing the method of any of claims 1-6, comprising: the scene construction and plan management module is used for executing the step S1, constructing a multi-target scene tree, and generating and managing a strategy plan library through offline simulation deduction; The signal monitoring and intelligent matching module is used for executing the step S2, monitoring preset key signal indexes in real time, starting scene matching through a built-in causal inference model when the indexes are abnormal, and outputting a strategy candidate set from a strategy plan library; The strategy execution and self-adaptive optimization module is used for executing step S3, receiving a strategy candidate set and real-time service context information, and completing dynamic selection, execution and fine adjustment of the strategy through an online learning algorithm; The decision visualization and interpretation module is used for executing step S4, collecting and correlating the operation data of each module, and constructing and displaying an interpretable decision map; and the knowledge precipitation and closed-loop learning module is used for executing the step S5, collecting the final execution result, carrying out effect comparison and inverse fact analysis, and carrying out iterative updating on the strategy plan library and the causal inference model according to the result.
  8. 8. The intelligent purchasing source-seeking system based on multi-objective dynamic optimization of claim 7, wherein the scenario construction and planning management module comprises: a conflict pattern library for storing predefined multi-objective conflict patterns and description rules thereof; the weight interval configuration unit is used for configuring a plurality of weight proportion intervals for each mode in the conflict mode library; the offline simulation engine is integrated with a Monte Carlo tree search algorithm and a game theory solver and is used for carrying out multi-round simulation deduction according to the configured weight interval; and the plan storage unit is used for storing the structured strategy plans output by the offline simulation engine and supporting retrieval based on conflict mode identification and boundary conditions.
  9. 9. The intelligent purchasing source-seeking system based on multi-objective dynamic optimization of claim 7, wherein the signal monitoring and intelligent matching module comprises: The signal acquisition interface is used for acquiring preset key signal index data from an external data source in real time; The threshold judging unit is used for comparing the acquired index data with a preset threshold and judging whether the matching is triggered or not; The causal matching engine is internally provided with a Bayesian network model, the priori structure and parameters of the network model are obtained based on historical data training, and the causal matching engine is used for deducing the most probable collision mode according to the input real-time signal index data when triggering matching and accessing the plan storage unit to obtain a strategy candidate set according to the most probable collision mode.
  10. 10. The intelligent purchasing source-seeking system based on multi-objective dynamic optimization of claim 7, wherein the policy enforcement and adaptive optimization module comprises: the context feature extraction unit is used for extracting a real-time business context feature vector from the current purchase order information; The strategy exploration and utilization unit is used for realizing a context multi-arm slot machine algorithm, and the unit is used for receiving a strategy candidate set and a context feature vector, calculating the selection probability of each candidate strategy in the current context and outputting a selection result; And the feedback processing unit is used for collecting the feedback information of the suppliers in the strategy execution process, quantifying the feedback information into an instant rewarding value and feeding the instant rewarding value back to the strategy exploration and utilization unit so as to update the value estimation of the strategy.

Description

Intelligent purchasing source-seeking method and system based on multi-objective dynamic optimization Technical Field The invention relates to the technical field of intelligent supply chain management, in particular to an intelligent purchasing source searching method and system based on multi-objective dynamic optimization. Background With the increasing complexity of the global supply chain and the increasing competition in the market, the enterprise purchasing and sourcing activities are not limited to single cost control, but multiple targets such as quality, lead time, risk management and the like need to be comprehensively considered. The traditional purchasing source searching method is mainly based on static decision, relies on manual experience or a fixed weight model to select and evaluate suppliers, and is difficult to deal with dynamic change environments such as market demand fluctuation and supply chain emergencies. Especially when facing the multi-objective optimization problem, the traditional method often causes poor adaptability and insufficient risk resistance of purchasing strategies due to fixed objective weight and lack of dynamic adjustment mechanism, and cannot realize intelligent configuration and collaborative optimization of all-link resources. In the prior art, some intelligent purchasing systems try to introduce an optimization algorithm, for example, an intelligent purchasing source-seeking method and system based on a multi-objective dynamic optimization and real-time feedback mechanism described in an authority publication number CN120782512A, which realize dynamic allocation of multi-objective weights and simulation prediction of supplier behaviors through reinforcement learning and dynamic game theory, and improve adaptability and accuracy of purchasing decisions. However, the system generally relies on a complex real-time feedback mechanism and multi-source heterogeneous data fusion, and in actual deployment, challenges such as high data integration cost, system response delay, weak model interpretability and the like are faced, and when a sudden supply chain interruption or multi-target conflict severe scene is handled, the problems of strategy adjustment lag, incomplete optimization dimension and the like still exist. Therefore, on the basis of ensuring the real-time performance and reliability of decision making, the method and the system for intelligent purchasing source searching, which are simple in structure, rapid in response and capable of supporting multi-objective dynamic collaborative optimization, are constructed, and are the technical problems to be solved in the current supply chain management field. Disclosure of Invention The invention aims to provide an intelligent purchasing source searching method and system based on multi-objective dynamic optimization, which are used for solving the problems in the background technology. In order to solve the technical problems, the invention provides the following technical scheme that the intelligent purchasing source searching method and system based on multi-objective dynamic optimization comprises the following steps: S1, constructing a multi-target scene tree and a strategy plan library, namely enumerating typical conflict modes possibly occurring among a plurality of targets of cost, quality, delivery and risk in the purchasing process based on historical data and industry knowledge, setting a plurality of corresponding weight proportion intervals for each conflict mode, adopting a simulation deduction algorithm, and performing off-line simulation calculation on different weight proportion intervals under each conflict mode to generate the strategy plan library comprising strategy schemes and applicable boundary conditions and expected effects thereof; S2, real-time monitoring and scene matching, namely, monitoring a group of key signal indexes reflecting the state of a supply chain in real time, triggering a scene matching engine when the monitored key signal index changes beyond a preset threshold value, analyzing a target conflict mode implied by the current key signal index combination based on a preset causal inference model by the scene matching engine, and matching a strategy candidate set associated with the current target conflict mode from the strategy plan library; S3, performing self-adaptive strategy execution and fine adjustment, namely taking the matched strategy candidate set as a strategy set to be selected, performing strategy selection and execution in the strategy set to be selected by adopting an online learning algorithm based on real-time business context information of a current purchasing task, and performing instant evaluation and parameter adjustment on the applicability of the selected strategy according to feedback information generated in the strategy execution process; and S4, generating and displaying an interpretable decision map, namely recording and storing the whole process