Material price prediction method and material price prediction system

Abstract

This disclosure provides a material price forecasting method and system. The method includes the following steps: Receiving a specification description text of a material to be forecasted. Obtaining specification data of the material to be forecasted from the specification description text using a large language model based on a specified data format. Selecting at least one material with similar specifications from multiple historical purchased materials based on the specification data of the material to be forecasted. Establishing a regression model based on multiple actual purchase prices of the at least one material with similar specifications, and determining the predicted purchase price of the material to be forecasted based on the regression model. Outputting the predicted purchase price of the material to be forecasted.

Claims

1. A material price forecasting method, applicable to a material price forecasting system including a processor, the method comprising: Receive a specification description text of a material to be predicted; use a large language model to obtain the specification data of the material to be predicted from the specification description text based on a specified data format; select at least one material with similar specifications from multiple historical purchased materials based on the specification data of the material to be predicted; establish a regression model based on multiple actual purchase prices of the at least one material with similar specifications, and determine the predicted purchase price of the material to be predicted based on the regression model; and output the predicted purchase price of the material to be predicted.

2. The material price forecasting method as described in claim 1, wherein the step of obtaining the specification data of the material to be forecasted from the specification description text using the large language model based on the specified data format includes: The large language model is used to extract a key specification parameter of the material to be predicted from the specification description text, and the key specification parameter is output as specification data of the material to be predicted in accordance with the specified data format.

3. The material price forecasting method as described in claim 2, wherein the step of obtaining the specification data of the material to be forecasted from the specification description text using the large language model based on the specified data format further includes: A regular expression is used to perform a validity check on the key specification parameters in the specification data of the material to be predicted, so as to revise the key specification parameters that fail the validity check.

4. The material price forecasting method as described in claim 2, wherein the step of using the regular expression to perform the validity check on the key specification parameters in the specification data of the material to be forecasted, in order to revise the key specification parameters that fail the validity check, includes: If the key specification parameters in the specification data of the material to be predicted fail the validity check, the key specification parameters of the material to be predicted are extracted again from the specification description text using the large language model to update the key specification parameters.

5. The material price forecasting method as described in claim 4, wherein the step of using the regular expression to perform the validity check on the key specification parameters in the specification data of the material to be forecasted, in order to revise the key specification parameters that fail the validity check, further includes: If the number of times the key specification parameter in the specification data of the material to be predicted fails the validity test reaches a threshold value, the key specification parameter is determined to be a default value.

6. The material price forecasting method as described in claim 1, wherein the step of selecting at least one material with similar specifications from the plurality of historically purchased materials based on the specification data of the material to be forecasted includes: Reflecting that the specifications of the material to be predicted differ from the historical specifications of each of the plurality of historically purchased materials, multiple material clusters and multiple sub-clusters of each of the plurality of material clusters are obtained by cluster analysis based on the historical purchasing information of the plurality of historically purchased materials; and based on the specifications of the material to be predicted, it is determined that the material to be predicted belongs to a target material cluster among the plurality of material clusters and to a target sub-cluster among the plurality of sub-clusters of the target material cluster, wherein some of the plurality of historically purchased materials in the target sub-cluster are the at least one material with similar specifications.

7. The material price forecasting method as described in claim 6, wherein the steps of establishing the regression model based on the multiple actual purchase prices of the at least one material with similar specifications, and determining the predicted purchase price of the material to be forecasted based on the regression model, include: Based on the multiple actual purchase prices of the at least one similar material, the specification data of the at least one similar material, and a purchase variable, a first regression model is established; and the predicted purchase price of the material to be predicted is determined based on the first regression model.

8. The material price forecasting method as described in claim 6, wherein the step of selecting at least one material with similar specifications from the plurality of historically purchased materials based on the specification data of the material to be forecasted includes: If the specification data of the material to be predicted is the same as the historical specification data of one of the plurality of historically purchased materials, then the plurality of historically purchased materials are selected, wherein one of them is the at least one material with similar specifications.

9. The material price forecasting method as described in claim 8, wherein the steps of establishing the regression model based on the multiple actual purchase prices of the at least one material with similar specifications, and determining the predicted purchase price of the material to be forecasted based on the regression model, include: A second regression model is established based on the multiple actual purchase prices of the at least one similar material and a purchase variable, wherein the specification data of the at least one similar material is the same as the specification data of the material to be predicted; and the predicted purchase price of the material to be predicted is determined based on the second regression model.

10. The material price forecasting method as described in claim 6 further includes: Using the large language model based on the specified data format, the historical specification data of each of the multiple historical purchased materials is obtained from the historical purchase information of the multiple historical purchased materials; a first cluster analysis is performed based on the historical specification data of each of the multiple historical purchased materials to group the multiple historical purchased materials into multiple material clusters; and a second cluster analysis is performed based on the historical specification data of the multiple historical purchased materials in one of the multiple material clusters to group the multiple historical purchased materials in one of the multiple material clusters into multiple sub-clusters.

11. The material price forecasting method as described in claim 6 further includes: Analyze the specification parameter distribution information of the historical specification data of the historical purchased materials in the target sub-cluster; The large language model is used to compare the specification data of the material to be predicted with the specification parameter distribution information to generate a clustering explanation based on the comparison results; and the clustering explanation of the material to be predicted is provided.

12. The material price forecasting method as described in claim 1, wherein the regression model includes an ensemble voting regressor model, the ensemble voting regressor model being a combination of multiple basic regressors, and the multiple basic regressors corresponding to different component weights.

13. A material price forecasting system, comprising: A storage device that stores multiple instructions; A processor, coupled to the storage device and accessing the instructions, executes: receiving a specification description text of a material to be predicted; obtaining specification data of the material to be predicted from the specification description text using a large language model based on a specified data format; selecting at least one material with similar specifications from a plurality of historical purchased materials based on the specification data of the material to be predicted; establishing a regression model based on multiple actual purchase prices of the at least one material with similar specifications, and determining a predicted purchase price of the material to be predicted based on the regression model; and outputting the predicted purchase price of the material to be predicted.

14. The material price forecasting system as claimed in claim 13, wherein the processor further performs: extracting a key specification parameter of the material to be forecasted from the specification description text using the large language model, and outputting the key specification parameter as specification data of the material to be forecasted conforming to the specified data format.

15. The material price forecasting system of claim 14, wherein the processor further performs: using a regular expression to perform a validity check on the key specification parameters in the specification data of the material to be forecasted, in order to revise the key specification parameters that fail the validity check.

16. The material price forecasting system as claimed in claim 14, wherein the processor further performs: if the key specification parameter in the specification data of the material to be forecast fails the validity check, the key specification parameter of the material to be forecast is extracted again from the specification description text using the large language model to update the key specification parameter.

17. The material price forecasting system of claim 16, wherein the processor further performs: determining that the key specification parameter is a default value when the number of times the key specification parameter in the specification data of the material to be forecast fails the validity check reaches a threshold value.

18. The material price forecasting system of claim 13, wherein the processor further performs: in response to the specification data of the material to be predicted being different from the historical specification data of each of the plurality of historically purchased materials, obtaining a plurality of material clusters and a plurality of sub-clusters of each of the plurality of material clusters determined by cluster analysis based on the historical purchase information of the plurality of historically purchased materials; and determining, based on the specification data of the material to be predicted, that the material to be predicted belongs to a target material cluster among the plurality of material clusters and to a target sub-cluster among the plurality of sub-clusters of the target material cluster, wherein the plurality of historically purchased materials in the target sub-cluster are the at least one material with similar specifications.

19. The material price forecasting system of claim 18, wherein the processor further performs: establishing a first regression model based on the multiple actual purchase prices of the at least one similar material, specification data of the at least one similar material, and a purchase variable; and determining the predicted purchase price of the material to be forecasted based on the first regression model.

20. The material price forecasting system of claim 18, wherein the processor further performs: in response to the historical specification data of one of the plurality of historically purchased materials being identical to the specification data of the material to be forecasted, selecting one of the plurality of historically purchased materials as the at least one material with similar specifications.

21. The material price forecasting system of claim 18, wherein the processor further performs: establishing a second regression model based on the multiple actual purchase prices of the at least one similar material and a purchase variable, wherein the specification data of the at least one similar material is the same as the specification data of the material to be forecasted; and determining the forecasted purchase price of the material to be forecasted based on the second regression model.

22. The material price forecasting system of claim 18, wherein the processor further performs: obtaining historical specification data of each of the plurality of historically purchased materials from historical purchase information of the plurality of historically purchased materials based on the specified data format using the large language model; performing a first cluster analysis based on the historical specification data of each of the plurality of historically purchased materials to group the plurality of historically purchased materials into the plurality of material clusters; and performing a second cluster analysis based on the historical specification data of the plurality of historically purchased materials in one of the plurality of material clusters to group the plurality of historically purchased materials in one of the plurality of material clusters into the plurality of sub-clusters.

23. The material price forecasting system of claim 18, wherein the processor further performs: analyzing specification parameter distribution information of the historical specification data of the historically purchased materials in the target sub-cluster; comparing the specification data of the material to be predicted with the specification parameter distribution information using the large language model to generate a clustering explanation based on the comparison result; and providing the clustering explanation of the material to be predicted.

24. The material price forecasting system as described in claim 13, wherein the regression model includes an ensemble voting regressor model, the ensemble voting regressor model being a combination of multiple basic regressors, and the multiple basic regressors each corresponding to different component weights.

Images