Search

CN-121981394-A - Low-profit order cause tracing method and system based on multi-source data

CN121981394ACN 121981394 ACN121981394 ACN 121981394ACN-121981394-A

Abstract

The invention discloses a low-profit order cause tracing method and system based on multi-source data, which belong to the technical field of data processing and business intelligence, and comprise the steps of obtaining an order accompanying shadow file; detecting microscopic association by using a causal discovery algorithm with mixed constraint and score, constructing a primary causal probe network, mapping to generate a multi-layer nested causal forest, completing forest recombination evaluation based on a periodic trigger signal, optimizing a controllable variable and developing node state evolution reasoning through multi-path backtracking locking collaborative cause combination when a low-profit order is identified, and finally outputting a strategy simulation result with profit prediction and risk annotation. The invention adopts the technical scheme of constructing a dynamic order shadow file, generating a multi-layer nested causal forest and carrying out collaborative backtracking and strategy simulation, can deeply reveal the low-profit composite cause and provides quantized optimization decision support.

Inventors

  • ZHU RONGFENG

Assignees

  • 深圳智汇创想科技有限责任公司

Dates

Publication Date
20260505
Application Date
20260126

Claims (10)

  1. 1.A method for tracing low profit order causes based on multi-source data, the method comprising: Acquiring full-period multi-source heterogeneous data of an order to be analyzed, and obtaining a dynamically updated order accompanying shadow file through time sequence alignment and associated archiving; based on the order accompanying shadow files of the historical orders, performing microcosmic association strength detection by using a causal discovery algorithm with mixed constraint and score to obtain a primary causal probe network; Hierarchical topology mapping is carried out on the primary causal probe network, a root layer network representing direct causal relation, a dry layer network representing conditional modulation relation and a canopy network representing environmental association relation are respectively extracted, and a multi-layer nested causal forest is generated; acquiring a periodic trigger signal, performing periodic recombination evaluation on the multi-layer nested causal forest, and generating an updated multi-layer nested causal forest; when a low-profit order is identified, carrying out multi-path causal backtracking on an order accompanying shadow file of the low-profit order by utilizing the updated multi-layer nested causal forest to obtain a cooperative cause combination; And carrying out numerical adjustment on the controllable variables in the collaborative cause combination, inputting the adjusted parameters into the updated multi-layer nested causal forest to carry out node state evolution reasoning so as to obtain a strategy simulation result comprising a profit predicted value and a risk labeling position.
  2. 2. The method for tracing back a low profit order cause based on multi-source data according to claim 1, the method is characterized in that the dynamically updated order accompanying shadow archive comprises the following steps: acquiring cost data, production data, logistics data and market data of an order to be analyzed from a service system interface in real time to obtain original multi-source data; Performing field mapping based on order identification and synchronizing with a time stamp on the original multi-source data to generate a unified data view; and carrying out structural hierarchical organization on the unified data view according to the life cycle stage of the order to obtain a dynamically updated order-accompanied shadow file.
  3. 3. A low-profit order cause tracing method based on multi-source data according to claim 2, wherein said deriving a primary causal probe network comprises: performing data preprocessing on the order-accompanying shadow files of the historical orders, and extracting candidate feature variables; calculating the condition independence and causal confidence coefficient between the candidate feature variables by adopting a causal discovery algorithm based on constraint and score mixing, and obtaining a causal directed graph representing microscopic association between data; and storing the node relation of the causal directed graph to obtain a primary causal probe network.
  4. 4. A method of low profit order cause and trace back based on multi-source data according to claim 3, wherein said generating a multi-layer nested causal forest comprises: Screening association paths with direct statistical correlation with profit indexes according to microscopic association relations in the primary causal probe network, and constructing a root layer network; Acquiring an algorithm for finding a condition factor, identifying the condition factor affecting the intensity or direction of an associated path in the root layer network in the primary causal probe network, and constructing a dry layer network describing the modulation relation between the condition factor and the associated path; acquiring an external environment data stream, analyzing the association between the external environment data stream and the structural stability changes of the root layer network and the dry layer network, and constructing a canopy network; And carrying out multidimensional topology fusion and hierarchical nested mapping on the root layer network, the dry layer network and the canopy network to generate a multi-layer nested causal forest.
  5. 5. The method for tracing back a low profit order cause based on multi-source data according to claim 4, the method is characterized in that the generating the updated multi-layer nested causal forest comprises the following steps: after receiving the periodic trigger signal, acquiring an order accompanying shadow file corresponding to the newly added order to obtain incremental sample data; Substituting the incremental sample data into the multi-layer nested causal forest, and calculating the saliency probability of each connecting edge in the root layer network, the dry layer network and the canopy network to obtain a relationship strength evaluation value; and rejecting or integrating new connection into the existing connection according to the relation strength evaluation value to obtain an updated multi-layer nested causal forest.
  6. 6. A method of low profit order cause tracing over multi-source data according to claim 5, comprising: calculating sensitivity weight of data acquisition according to the newly added node association relation in the updated multi-layer nested causal forest, and generating a data acquisition optimization instruction; and adjusting the acquisition frequency of the front-end system on the original multi-source data by utilizing the data acquisition optimization instruction.
  7. 7. The method for tracing back a low profit order cause based on multi-source data according to claim 5, the method is characterized in that the obtaining of the synergistic cause combination comprises the following steps: tracing back to obtain a direct cause affecting the profit index along the root layer network of the updated multi-layer nested causal forest; Positioning a dry layer network with a modulation relation with the direct cause, extracting a condition variable with the limited direct cause, and obtaining a preliminary cause combination; and retrieving environment data of a period corresponding to the low-profit order, mapping the environment data to the canopy network, extracting environment factors which generate amplification effects on the primary cause combination, and carrying out vector combination with the primary cause combination to obtain a synergistic cause combination.
  8. 8. The method of claim 7, wherein obtaining the policy simulation result including the profit prediction value and the risk tagging bit comprises: Setting a change step length for the controllable variables in the collaborative cause combination, and generating an adjustment strategy containing a parameter correction value; mapping the parameter correction value in the adjustment strategy to a root layer node of the updated multi-layer nested causal forest, and calculating a state update value of the root layer node; And calculating chain reaction and risk probability triggered by the state update value by using the constraint relation between the dry layer network and the canopy network, and generating a strategy simulation result comprising predicted profit and risk identification.
  9. 9. The multi-source data based low profit order cause tracing method of claim 1, further comprising: Receiving a profit target value input by a user to obtain a target constraint condition; in the updated multi-layer nested causal forest, reverse path optimization is carried out by taking the target constraint condition as a boundary, and a feasible adjustment parameter set meeting the condition is generated; And carrying out multi-objective pareto analysis on the cost and profit of the feasible adjustment parameter set, and generating and recommending an optimization strategy scheme.
  10. 10. A low profit order cause tracing system based on multi-source data, the system comprising: The shadow file construction module is used for acquiring full-period multi-source heterogeneous data of an order to be analyzed, and dynamically updated order accompanying shadow files are obtained through time sequence alignment and associated archiving; the causal probe generation module is used for detecting microcosmic association strength by using a causal discovery algorithm mixed with constraint and scoring based on the order syndrome shadow file of the historical order to obtain a primary causal probe network; the nested forest construction module is used for carrying out hierarchical topology mapping on the primary causal probe network, respectively extracting a root layer network representing a direct causal relation, a dry layer network representing a conditional modulation relation and a canopy network representing an environmental association relation, and generating a multi-layer nested causal forest; the network dynamic reorganization module is used for acquiring periodic trigger signals, carrying out periodic reorganization evaluation on the multi-layer nested causal forest, and generating an updated multi-layer nested causal forest; The cause cooperation backtracking module is used for carrying out multipath causal backtracking on the order accompanying shadow files of the low-profit orders by utilizing the updated multi-layer nested causal forest when the low-profit orders are identified, so as to obtain cooperation cause combinations; And the strategy simulation module is used for carrying out numerical adjustment on the controllable variables in the collaborative cause combination, inputting the adjusted parameters into the updated multi-layer nested causal forest to carry out node state evolution reasoning so as to obtain a strategy simulation result comprising a profit prediction value and a risk labeling bit.

Description

Low-profit order cause tracing method and system based on multi-source data Technical Field The invention relates to the technical field of data processing and business intelligence, in particular to a low-profit order cause tracing method and system based on multi-source data. Background In modern enterprise management, especially in industries relying on order driving such as manufacturing industry, retail industry and the like, the fine analysis and management of the profitability of each order are key to improving the overall operation efficiency and core competitiveness. The source tracing of the low profit order is a core link in the management activity, and aims to identify key factors which lead to lower profit than expected by analyzing the whole process data of the order from quotation, production to delivery, and provide decision support for subsequent process optimization, cost control and pricing strategy adjustment. This process typically involves integrating and analyzing the large amounts of data generated by multiple business systems within an enterprise, such as enterprise resource planning systems, production execution systems, supply chain management systems, and the like. In the related art, the Chinese patent publication No. CN120875676A discloses an operation decision intelligent analysis method and system based on cross-domain data fusion. The method comprises the steps of carrying out entity identification and relation mapping on heterogeneous data from different business systems through a multi-source heterogeneous data dynamic fusion algorithm based on semantic mapping based on an enterprise multi-domain ontology knowledge base, establishing a unified data model, carrying out analysis processing on the model through a mixed intelligent decision engine fusing causal reasoning and deep learning, utilizing an analysis result of the mixed intelligent decision engine to establish a multi-level causal relation network among business variables through a causal relation discovery algorithm, using a self-adaptive business scene analysis model based on reinforcement learning, dynamically adjusting an analysis strategy according to business environment changes, generating a pareto optimal decision scheme set through a multi-objective optimization algorithm, and outputting operation decision suggestions. With respect to the related art as described above, the inventors consider that it has technical drawbacks in practical applications. The method is characterized in that a unified data model and a multi-level causal network based on semantic mapping are established, but the method is still in a static and flattened organization mode, so that a dynamic track of profit fluctuation cannot be accurately restored when a strong time sequence service is processed, and in causal logic, only the transverse division of service dimension is remained, so that direct inducement and environment interference cannot be effectively distinguished, and the tracing analysis precision is insufficient. Disclosure of Invention In order to solve the problems, the invention provides a low-profit order cause tracing method and system based on multi-source data, which adopts the technical scheme of constructing a dynamic order shadow file, generating a multi-layer nested causal forest and carrying out collaborative backtracking and strategy simulation, can deeply reveal the low-profit composite cause and provide quantized optimization decision support. The above object can be achieved by the following scheme: A low-profit order cause tracing method based on multi-source data comprises the steps of obtaining full-period multi-source heterogeneous data of an order to be analyzed, obtaining a dynamically updated order-accompanying shadow file through time sequence alignment and associated archiving, detecting microcosmic association strength by using a causal discovery algorithm mixed with constraint and score based on the order-accompanying shadow file of a historical order to obtain a primary causal probe network, carrying out hierarchical topological mapping on the primary causal probe network, respectively extracting a root layer network representing a direct causal relationship, a dry layer network representing a conditional modulation relationship and a canopy network representing an environmental association relationship to generate a multi-layer causal nested forest, obtaining a periodic trigger signal, carrying out periodic recombination evaluation on the multi-layer nested causal forest to generate an updated multi-layer nested forest, carrying out multi-path causal trace on the order-accompanying file of the low profit by using the updated multi-layer nested causal nested forest when the low-profit order is identified to obtain a collaborative causal combination, carrying out numerical adjustment on controllable variables in the collaborative causal probe network, and inputting adjusted parameters into the mult