CN-122022965-A - Intelligent decision method and system for purchasing demand

CN122022965ACN 122022965 ACN122022965 ACN 122022965ACN-122022965-A

Abstract

The invention discloses an intelligent decision-making method and system for purchasing demands, belongs to the technical field of demand prediction, and aims to solve the technical problems that the conventional purchasing demand prediction method is single in data dimension, poor in real-time adaptability and low in effective decision-making support, so that prediction accuracy is low, and suppliers cannot accurately predict market demands. The method comprises the steps of extracting multi-mode purchasing data based on a historical purchasing record and preprocessing the data, extracting features of the preprocessed multi-mode purchasing data, selecting the extracted features based on a multi-dimensional analysis method to obtain purchasing demand features, constructing a purchasing demand prediction model, training the purchasing demand prediction model based on edge calculation and a streaming real-time learning technology, inputting the multi-mode purchasing data and the purchasing demand features into the purchasing demand prediction model to obtain purchasing demand prediction results, and generating a visual decision report according to the purchasing demand prediction results.

Inventors

ZHAO GUANG
GUO JIAN
REN XUESHEN
DIAO JUNMING
LI HOUMING

Assignees

山东土地数字科技集团有限公司

Dates

Publication Date: 20260512
Application Date: 20260123

Claims (10)

1. An intelligent decision-making method for purchasing demands, which is characterized by comprising the following steps: based on the historical purchasing record, extracting multi-mode purchasing data and preprocessing the data; Extracting features of the preprocessed multi-mode purchasing data, and selecting the extracted features based on a multidimensional analysis method to obtain purchasing demand features; constructing a purchasing demand prediction model, and training the purchasing demand prediction model based on edge calculation and a streaming real-time learning technology; inputting the multi-mode purchasing data and the purchasing demand characteristics into the purchasing demand prediction model to obtain a purchasing demand prediction result; and generating a visual decision report according to the purchasing demand prediction result, wherein the visual decision report at least comprises a time dimension, a space dimension and a strategy dimension.
2. The intelligent purchasing demand decision-making method according to claim 1, wherein the multi-mode purchasing data is extracted based on the historical purchasing record and data preprocessing is performed, and the method specifically comprises the steps of: Extracting structured data, semi-structured data and unstructured data from the historical purchasing record; Converting the structured data and the semi-structured data into structural feature vectors through a multi-mode fusion framework fused with an attention mechanism, extracting semantic feature vectors from text data in unstructured data, and generating visual feature vectors from image data in unstructured data; According to the structural feature vector, the semantic feature vector, the visual feature vector and the preset modal weight parameter, a fusion feature vector is obtained so as to perform data fusion on the structured data, the semi-structured data and the unstructured data, and the multi-modal purchasing data is obtained; and performing data cleaning, normalization processing and time sequence alignment processing on the multi-mode purchase data to obtain the preprocessed multi-mode purchase data.
3. The intelligent purchasing demand decision-making method according to claim 1, wherein the feature extraction is performed on the preprocessed multi-mode purchasing data, and the method specifically comprises: extracting time features and market features from the multi-mode purchasing data based on a sliding window technology, wherein the market features at least comprise a demand gap index, a market liveness index and a price elasticity index; And extracting text keywords from the multi-mode purchase data to obtain text features, wherein the text keywords at least comprise commodity attribute keywords, provider feature keywords and market dynamic keywords.
4. The intelligent purchasing demand decision-making method according to claim 3, wherein the feature selection is performed on the extracted features based on a multidimensional analysis method to obtain purchasing demand features, and the method specifically comprises: Calculating the correlation value of each extracted feature and a target variable based on a mutual information method, and carrying out feature screening according to the correlation value to obtain a first screening feature, wherein the target variable is a purchase quantity; Obtaining importance scores of each feature in the first screening features through a random forest model; Sorting the first screening features according to importance scores, and iteratively removing redundant features in the first screening features through a recursive feature elimination algorithm to obtain second screening features; Determining the causal relation among all the features in the second screening feature through a causal inference model to obtain a causal feature pair; extracting a causal chain in the causal graph, and only reserving core features on each causal chain to obtain a third screening feature; And performing feature dimension reduction on the high-dimensional features in the third screening features, performing visual cluster analysis to obtain a dimension reduction diagram, deleting one of the features if the distance between the two features in the dimension reduction diagram is lower than a preset distance threshold value, deleting the feature if a certain feature deviates from a cluster area, and finishing the final feature screening to obtain the final purchasing demand feature.
5. The intelligent purchasing demand decision-making method according to claim 1, wherein the building of the purchasing demand prediction model specifically comprises: Constructing a basic model layer based on the LSTM neural network; Adding a multi-head self-attention module at the output end of the basic model layer, and distributing dynamic weights to different time step characteristics; An online learning optimizer based on a sliding window updating mechanism is constructed at the output end of the multi-head self-attention module, and the purchasing demand prediction model is obtained; carrying out light weight treatment on the purchasing demand prediction model to obtain a light weight sub-model; the lightweight sub-model is deployed on a plurality of edge nodes, and the purchasing demand prediction model is deployed on a cloud, wherein the edge nodes at least comprise a provider local system, a purchasing platform edge server and a logistics node terminal.
6. The intelligent purchasing demand decision-making method according to claim 5, wherein training the purchasing demand prediction model based on edge calculation and streaming real-time learning technology specifically comprises: Carrying out global model parameter initialization on the purchase demand prediction model deployed at the cloud end, and issuing the global model parameter initialization to a plurality of edge nodes; the edge node trains the lightweight sub-model based on the local real-time data stream and the received global model parameters, and calculates the local parameter updating quantity; uploading the local parameter updating quantity and the local data quantity of each edge node to a cloud end, and updating global model parameters of the purchase demand prediction model; issuing the updated global model parameters to edge nodes to replace the model parameters of the lightweight sub-model, and completing a round of collaborative training; In the edge node, based on the streaming computing framework, the local real-time data stream is processed, and the lightweight sub-model is trained and updated in real time.
7. The intelligent purchasing demand decision method of claim 6, wherein after training the purchasing demand prediction model based on edge calculation and streaming real-time learning techniques, the method further comprises: optimizing a parameter space of the purchase demand prediction model based on a Bayesian optimization framework and a proxy model; constructing a Stacking integrated model, and carrying out integrated optimization on the purchase demand prediction model; and stopping optimizing after the loss value of the verification set is not reduced and the preset times are continued, so as to save the optimal model parameters.
8. The intelligent purchasing demand decision method according to claim 1, wherein after inputting the multi-modal purchasing data and the purchasing demand feature into the purchasing demand prediction model to obtain a purchasing demand prediction result, the method further comprises: dynamically updating the feature importance weight of the purchasing demand feature according to the market fluctuation rate; Calculating the deviation value of the actual transaction amount and the purchasing demand prediction result, and converting the deviation value into a model parameter correction amount; And optimizing and updating the purchasing demand prediction model according to the updated feature importance weight and the model parameter correction quantity.
9. The intelligent decision-making method for purchasing demands according to claim 1, wherein generating a visual decision report according to the purchasing demand prediction result specifically comprises: In the time dimension, generating a purchasing quantity prediction curve of the current year and the next year according to the purchasing demand prediction result; In the space dimension, constructing a demand thermodynamic diagram of a purchasing area and purchasing products according to the purchasing demand prediction result, wherein the deeper the color in the demand thermodynamic diagram is, the larger the demand gap is; in the strategy dimension, generating capacity adjustment suggestions of suppliers according to the purchasing demand prediction result; Typesetting the purchase quantity prediction curve, the demand thermodynamic diagram and the capacity adjustment suggestion according to a preset template to obtain the visual decision report.
10. An intelligent purchasing demand decision-making system, the system comprising: the characteristic engineering module is used for extracting multi-mode purchasing data based on a historical purchasing record and preprocessing the data, extracting characteristics of the preprocessed multi-mode purchasing data, and selecting the extracted characteristics based on a multi-dimensional analysis method to obtain purchasing demand characteristics; The model prediction module is used for constructing a purchasing demand prediction model, and training the purchasing demand prediction model based on edge calculation and a streaming real-time learning technology; the decision module is used for generating a visual decision report according to the purchasing demand prediction result, wherein the visual decision report at least comprises a time dimension, a space dimension and a strategy dimension.

Description

Intelligent decision method and system for purchasing demand Technical Field The invention relates to the technical field of demand prediction, in particular to an intelligent purchasing demand decision method and system. Background In the current bidding purchasing field, purchasing demand prediction is a key link for a provider to formulate an operation strategy. At present, in the field of bidding purchasing, some purchasing demand prediction methods exist, and mainly include methods of simple statistical methods based on historical data, manual prediction methods driven by experience, traditional time sequence models and the like. However, these existing purchasing demand prediction techniques still have a number of drawbacks that need to be addressed. Firstly, the data dimension is single, the existing prediction method depends on single dimension data such as historical purchasing quantity, other multidimensional key information is ignored, and the real condition and potential gap of market demands cannot be comprehensively reflected. And secondly, the adaptability of the prediction model is insufficient, the nonlinear relation and the multivariate interaction are difficult to process by the traditional method, the response capability to sudden market events is weak, real-time market data cannot be integrated quickly, the prediction model is updated, the prediction result is delayed from market change, and the prediction precision is low. And thirdly, part of prediction methods rely on experience of purchasing personnel or suppliers to carry out subjective judgment, and lack of data support, so that the stability of a prediction result is poor. Finally, the prediction results are presented in the form of original data, visual analysis and targeted decision suggestions are lacked, and the actual operation behaviors of suppliers such as capacity planning, bidding strategy formulation and the like are difficult to direct. These problems lead to the dilemma that suppliers often face surplus or insufficient capacity, high bidding flow rate, unreasonable resource allocation and the like, seriously affect the market competitiveness and the operation benefit of the suppliers, and restrict the overall trading efficiency of bidding purchasing markets. Disclosure of Invention The embodiment of the invention provides an intelligent decision-making method and system for purchasing demands, which are used for solving the technical problems that the traditional purchasing demand forecasting method is single in data dimension, poor in real-time adaptability and low in effective decision-making support, so that forecasting precision is low, and suppliers cannot accurately forecast market demands. The embodiment of the invention adopts the following technical scheme: in one aspect, an embodiment of the present invention provides a method for intelligently deciding purchasing requirements, where the method includes: based on the historical purchasing record, extracting multi-mode purchasing data and preprocessing the data; Extracting features of the preprocessed multi-mode purchasing data, and selecting the extracted features based on a multidimensional analysis method to obtain purchasing demand features; constructing a purchasing demand prediction model, and training the purchasing demand prediction model based on edge calculation and a streaming real-time learning technology; inputting the multi-mode purchasing data and the purchasing demand characteristics into the purchasing demand prediction model to obtain a purchasing demand prediction result; and generating a visual decision report according to the purchasing demand prediction result, wherein the visual decision report at least comprises a time dimension, a space dimension and a strategy dimension. In one possible implementation, based on the historical purchase record, multi-mode purchase data is extracted and data preprocessing is performed, and specifically includes: Extracting structured data, semi-structured data and unstructured data from the historical purchasing record; Converting the structured data and the semi-structured data into structural feature vectors through a multi-mode fusion framework fused with an attention mechanism, extracting semantic feature vectors from text data in unstructured data, and generating visual feature vectors from image data in unstructured data; According to the structural feature vector, the semantic feature vector, the visual feature vector and the preset modal weight parameter, a fusion feature vector is obtained so as to perform data fusion on the structured data, the semi-structured data and the unstructured data, and the multi-modal purchasing data is obtained; and performing data cleaning, normalization processing and time sequence alignment processing on the multi-mode purchase data to obtain the preprocessed multi-mode purchase data. In a possible implementation manner, the feature extraction is performed on the