CN-122022427-A - Material purchasing flow optimization method and system based on deep learning

CN122022427ACN 122022427 ACN122022427 ACN 122022427ACN-122022427-A

Abstract

The invention relates to the technical field of material purchasing management, and particularly discloses a material purchasing flow optimization method and system based on deep learning, wherein a unified data set containing time sequence characteristics and static characteristics is constructed through multi-source data acquisition; then decomposing the time sequence features into trend, period and fluctuation components by adopting a multi-scale decomposition technology, and dynamically adjusting the feature contribution degree by combining with static features to realize accurate demand prediction; the method comprises the steps of establishing a multidimensional evaluation matrix, quantifying the backorder influence, combining a provider capacity evaluation system, dividing purchasing urgency into three levels through a decision tree algorithm, further establishing a task association network, calculating priority scores through feature propagation and depth fusion to form an optimized task sequence, recommending an optimal approval path through multidimensional similarity matching based on historical approval data, and realizing continuous update of approval features.

Inventors

HOU XIAOHU
GUO XIANGZHENG
ZHANG XIAOHUI
ZHANG JIN
YAN JIANLI
YU JIA
LI YILIN
CHENG HUIHUI

Assignees

国网山西省电力有限公司物资分公司
国网山西招标有限公司

Dates

Publication Date: 20260512
Application Date: 20260120

Claims (10)

1. The material purchasing process optimization method based on deep learning is characterized by comprising the following steps of: s1, acquiring purchase records, inventory data and provider information from each business system of an enterprise, and constructing a unified data set containing time sequence characteristics and static characteristics after data cleaning; S2, inputting a unified data set into a demand prediction process, decomposing the time sequence features into trend components, periodic components and fluctuation components, dynamically adjusting the contribution degree of the features at each time point by combining the material attributes and the supplier information in the static features, and outputting material demand prediction results of a plurality of future periods by analyzing the internal correlation among the multi-dimensional features; S3, calculating the number of gaps of the materials according to the demand prediction result and combining the current actual stock quantity, automatically creating a purchasing application when the gaps exist, and dividing purchasing emergency into three levels; s4, constructing a hierarchical purchasing task as a task association network, wherein each node represents one purchasing task, and node characteristics comprise price information, emergency degree and supplier reliability of the tasks; And S5, according to the priority task sequence, automatically matching corresponding approval flows, namely adopting a synchronous approval mode for the high-priority task, setting a processing time limit, automatically switching approval paths after overtime, and optimizing approval path selection and reducing the whole approval time by analyzing historical approval data.
2. The deep learning-based material purchasing process optimization method of claim 1, wherein the prediction process specifically comprises: Carrying out multi-scale decomposition on the time sequence characteristics, adopting sliding window analysis to extract trend components of different time spans, identifying periodic components through spectrum analysis, and adopting a residual error separation method to obtain fluctuation components; Constructing multidimensional feature tensors by using the decomposed components, material classification characteristics in static features and supplier stability information, establishing association mapping relations among features, and dynamically adjusting importance coefficients of feature dimensions according to material use scenes; Reconstructing the characteristic representation of each time point based on the importance coefficient, and mining the cooperative influence of each characteristic under different time spans through characteristic interactive learning to generate a material demand prediction result of a plurality of future periods.
3. The optimization method for material purchasing flow based on deep learning according to claim 2, wherein the calculation process of the importance coefficient is as follows: Constructing scene feature vectors based on the criticality of the materials in the production line and the emergency degree of supply, and quantizing the material usage scene into feature combinations with different weights; For each feature dimension, calculating interaction strength between the feature dimension and a scene feature vector, and determining initial importance scores of the feature dimensions according to historical performance data of the features in different scenes; The final value of the feature importance coefficient is adjusted according to the variation amplitude and the stability index of the feature in the current time window by dynamically correcting the initial importance score; The adjusted importance coefficients are applied to the reconstruction process of the feature tensor.
4. The deep learning-based material purchasing process optimization method of claim 1, wherein the classifying purchasing urgency into three levels specifically comprises: calculating expected inventory gaps of each period according to the future multiple period demand in the demand prediction result and combining the current actual inventory and the safety inventory standard; establishing a multidimensional evaluation matrix comprising gap duration, gap influence range and production dependency degree, and quantifying potential influence caused by lack of stock by analyzing a critical path dependency relationship of materials in a production link; constructing a supplier response capability assessment system based on historical supply data, and integrating the supplier delivery timing rate, the quality qualification rate and the emergency supply capability to form a supply stability score; The inventory gap degree, the backorder influence degree and the supply stability score are input into a three-layer decision tree, and the purchasing emergency degree is accurately divided into three grades of routine, emergency and urgent through a preset grading threshold.
5. The deep learning-based material purchasing process optimization method of claim 4, wherein the multi-dimensional evaluation matrix construction process is as follows: identifying key path nodes of materials in a production link by analyzing a topological structure of a production flow, and determining a linkage influence range caused by material shortage; constructing an influence propagation network based on production plan data, quantifying the conduction effect of the backorder event in different production links, and calculating the number and importance degree of production procedures covered by the influence range; And combining the notch duration and the influence range data, and fusing the duration index, the influence range index and the production dependency degree index into a unified comprehensive influence value by adopting a multidimensional normalization processing method.
6. The method for optimizing a material purchasing process based on deep learning according to claim 1, wherein the step of calculating the priority score of each task by analyzing the association relation between the tasks and forming task sequences with different priorities specifically comprises: Establishing a connection edge between purchasing tasks based on the cooperative use relation of materials in the production link and the overlap ratio of suppliers to form a task association network; The emergency degree of adjacent tasks and the reliability characteristics of suppliers are transmitted in a two-way mode, and the characteristic representation of each node is updated; Integrating the characteristics of the node and the characteristics of the neighbor nodes, and calculating the comprehensive influence score of each task; and sequencing the purchasing tasks according to the comprehensive influence score to generate a priority task sequence with a sequential execution sequence.
7. The deep learning-based material purchasing process optimization method of claim 6, wherein the comprehensive node self-feature and neighbor node feature calculate a comprehensive influence score of each task, and specifically comprises: Carrying out multi-scale feature extraction on the self features of the nodes to obtain multi-level feature representation comprising local features and global features; capturing potential association modes between the characteristics of the node and the characteristics of the neighbor nodes, and establishing nonlinear mapping relations among the characteristics; Deep fusion is carried out on the feature representations of different levels, and node comprehensive feature representations with distinguishing degrees are generated; Based on the node comprehensive feature representation, calculating the comprehensive influence score of each task through feature importance discrimination.
8. The deep learning-based material purchasing process optimization method according to claim 1, wherein optimizing approval path selection by analyzing historical approval data specifically comprises: extracting approval duration, an approver decision mode and task feature data from the historical approval record, and constructing an approval feature library; Based on task features in the current priority task sequence, multidimensional similarity matching is carried out in an approval feature library, and an optimal approval path of a historical similar case is found out; According to the approval effect evaluation result of the similar case, the configuration parameters of the approval path are adjusted by combining the workload state of the current approver; and continuously recording the newly generated approval data, and updating the approval feature library in real time to form a closed-loop improvement process of approval path optimization.
9. The method for optimizing a material purchasing process based on deep learning according to claim 8, wherein the step of performing multidimensional similarity matching in an approval feature library to find an optimal approval path of a historical similar case comprises the following steps: performing multidimensional projection on the feature vector of the current task and the historical cases in the approval feature library, and constructing a feature similarity comparison space; setting a dynamic matching threshold based on the task emergency degree and the approval complexity, and screening out historical approval cases meeting the similarity requirement; Verifying validity of the approval path of the screened historical cases, and eliminating cases with approval failure records or overtime records; And selecting the path with highest approval efficiency from the rest cases as the optimal approval path of the current task.
10. A deep learning-based material purchasing process optimization system, characterized by being configured to perform the deep learning-based material purchasing process optimization method of any one of claims 1-9, including: The data acquisition and preprocessing module is used for acquiring purchasing records, inventory data and supplier information from each business system of an enterprise, and constructing a unified data set containing time sequence characteristics and static characteristics after data cleaning; The demand prediction analysis module inputs the unified data set into a demand prediction process, decomposes the time sequence characteristics into trend components, period components and fluctuation components, dynamically adjusts the contribution degree of the characteristics of each time point by combining the material attribute and the supplier information in the static characteristics, and outputs the material demand prediction results of a plurality of future periods by analyzing the internal correlation among the multidimensional characteristics; the purchasing demand generation module is used for calculating the number of the gaps of the materials according to the demand prediction result and the current actual stock quantity, automatically creating purchasing applications when the gaps exist, and dividing purchasing emergency into three grades; The task priority ordering module is used for constructing the classified purchasing tasks into a task association network, wherein each node represents one purchasing task, and the node characteristics comprise the price information, the emergency degree and the reliability of suppliers of the tasks; the intelligent approval optimizing module automatically matches corresponding approval flows according to the priority task sequence, adopts a synchronous approval mode for the high-priority task, sets a processing time limit, automatically switches the approval paths after overtime, optimizes the approval path selection by analyzing historical approval data, and reduces the whole approval time.

Description

Material purchasing flow optimization method and system based on deep learning Technical Field The invention relates to the technical field of material purchase management, in particular to a material purchase flow optimization method and system based on deep learning. Background In the current enterprise material purchasing management practice, a mode of combining a statistical prediction method based on historical consumption data with a manual experience decision is generally adopted. Such conventional methods typically rely on simple moving average or exponential smoothing methods for demand prediction, determine purchase priorities via fixed ABC classification rules, and process purchase applications using a linear approval process. Especially in industries with high requirements on material supply continuity, such as electric power energy, large-scale manufacturing and the like, purchasing management relates to multiple links, such as multi-source data integration, complex demand prediction, dynamic decision and the like. In the prior art, although the basic purchasing management function can be completed, when complex scenes such as sudden demand change, multi-factor association decision and the like are dealt with, the rigid processing mode and the single-dimension decision basis of the system are difficult to meet the requirements of modern enterprise fine management. The prior art has the following defects: The nonlinear change characteristics of material demands in a complex market environment cannot be accurately captured by the traditional prediction model, and the purchasing priority level is judged to lack comprehensive consideration of multidimensional association factors, so that the prediction deviation rate is higher when sudden demands are handled, and the normal production operation and emergency response capability of enterprises are affected. The core of the problem is that the prior art treats the demand prediction, priority judgment and approval flows as independent links, and lacks a full-flow collaborative optimization mechanism based on data driving. Disclosure of Invention The invention aims to provide a material purchasing flow optimization method and a material purchasing flow optimization system based on deep learning, so as to solve the problems in the background. The aim of the invention can be achieved by the following technical scheme: The material purchasing process optimization method based on deep learning comprises the following steps: s1, acquiring purchase records, inventory data and provider information from each business system of an enterprise, and constructing a unified data set containing time sequence characteristics and static characteristics after data cleaning; S2, inputting a unified data set into a demand prediction process, decomposing the time sequence features into trend components, periodic components and fluctuation components, dynamically adjusting the contribution degree of the features at each time point by combining the material attributes and the supplier information in the static features, and outputting material demand prediction results of a plurality of future periods by analyzing the internal correlation among the multi-dimensional features; S3, calculating the number of gaps of the materials according to the demand prediction result and combining the current actual stock quantity, automatically creating a purchasing application when the gaps exist, and dividing purchasing emergency into three levels; s4, constructing a hierarchical purchasing task as a task association network, wherein each node represents one purchasing task, and node characteristics comprise price information, emergency degree and supplier reliability of the tasks; And S5, according to the priority task sequence, automatically matching corresponding approval flows, namely adopting a synchronous approval mode for the high-priority task, setting a processing time limit, automatically switching approval paths after overtime, and optimizing approval path selection and reducing the whole approval time by analyzing historical approval data. The prediction process specifically comprises the following steps: Carrying out multi-scale decomposition on the time sequence characteristics, adopting sliding window analysis to extract trend components of different time spans, identifying periodic components through spectrum analysis, and adopting a residual error separation method to obtain fluctuation components; Constructing multidimensional feature tensors by using the decomposed components, material classification characteristics in static features and supplier stability information, establishing association mapping relations among features, and dynamically adjusting importance coefficients of feature dimensions according to material use scenes; Reconstructing the characteristic representation of each time point based on the importance coefficient, and mining the cooperative influen