Material supply risk prediction method and system
By using the XGBoost model to extract and model the historical data of power grid material suppliers, predicting supplier performance risks, solving the problem that traditional methods are difficult to cope with complex market environments, and improving the accuracy and efficiency of supply chain risk management.
Patent Information
- Application Number
- CN202411752643.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-02
- Publication Date
- 2025-03-25
AI Technical Summary
Traditional supply chain risk management methods rely on experience and intuition, and are difficult to cope with complex market environments and cannot effectively predict supply chain risks.
The XGBoost model is used to extract and model the historical data of power grid material suppliers, build a contract performance probability prediction model, and predict supplier performance risks through machine learning technology.
It improves the accuracy of forecasting supplier performance risks, helps enterprises identify high-risk suppliers in advance, take preventive measures to reduce losses, and improves the economy and flexibility of the supply chain.
Smart Images
Figure CN119692768A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of material management and control, and in particular to a material supply risk prediction method and system. Background Art
[0002] With the acceleration of globalization, the supply chain of enterprises has become more and more complex. This complexity not only increases the difficulty of management, but also brings more uncertainties, such as supplier default, shortage of raw materials and other problems. Effective supply chain management and risk control have become the key to maintaining normal operations of enterprises.
[0003] Traditional supply chain risk management methods often rely on experience and intuition, which are difficult to cope with the increasingly complex market environment. Therefore, it is urgent to develop a method that can predict supply chain risks. Summary of the invention
[0004] The purpose of the present invention is to provide a material supply risk prediction method and system to solve the problems raised in the above background technology.
[0005] In the first aspect, the present application provides a material supply risk prediction method, which adopts the following technical solution:
[0006] Step S1, obtaining historical data of several power grid material suppliers;
[0007] Step S2, extracting several characteristic indicators from the historical data, and characteristic data corresponding to each characteristic indicator, and constructing a characteristic data set according to the characteristic indicators and the corresponding characteristic data;
[0008] Step S3, randomly extract one feature data from the feature data corresponding to each feature index in the feature data set to form a data group, and construct a data set L containing several data groups:
[0009] L=(S1、S2…S n )
[0010] Among them, S n is the nth data group;
[0011] Step S4, using the XGBoost model to train the data group in the feature data set to obtain a contract performance probability prediction model for power grid material suppliers;
[0012] Step S5, obtaining characteristic data of each characteristic indicator of the target power grid material supplier and recording it as target data;
[0013] Step S6, input the target data into the contract performance probability prediction model to obtain the contract performance probability of the target power grid material supplier.
[0014] Preferably, before step S2, the historical data is also cleaned and normalized.
[0015] Preferably, the step of extracting several characteristic indicators from the historical data is specifically as follows:
[0016] Step S21, dividing the historical data into two categories according to the performance status, wherein the first category of data is the performance data T1, and the second category of data is the non-performance data T2;
[0017] Step S22, calculate the mean of each data in the performance data T1 and variance α1, and the mean of each data in the default data T2 and variance α2;
[0018] Step S23, based on the mean Variance α1, number of data in fulfillment data T1 m1, mean Variance α2 and m2 of the data in the default data T2, calculate the statistic t and degrees of freedom df;
[0019] Step S24, calculating the P value according to the statistic t and the degree of freedom df;
[0020] Step S25: If the P value is less than the preset value of 0.05, it is considered that there is a significant relationship between the data and the probability of contract performance, and the index corresponding to the data is marked as a characteristic indicator.
[0021] Preferably, in step S23, the statistical calculation formula is as follows:
[0022]
[0023] The statistic t is calculated according to the statistic calculation formula.
[0024] Preferably, in step S23, the calculation formula of the degree of freedom is as follows:
[0025] df=m1+m2 -2
[0026] The degrees of freedom df is calculated according to the calculation formula of degrees of freedom.
[0027] Preferably, the step of constructing a feature data set based on the feature indicators and the corresponding feature data is specifically as follows:
[0028] Step S26, forming a characteristic indicator set according to all characteristic data corresponding to the characteristic indicators:
[0029] A1=(A 11 , A 12 …A 1n )
[0030] A2=(A 21 , A 22 …A 2n )
[0031] …
[0032] A n =(A n1 , A n2 …A nn )
[0033] Among them, A1, A2, ... A n are all characteristic index sets, A 11 is the first characteristic data of the first characteristic indicator, A 1n is the nth characteristic data of the first characteristic indicator; A 21 is the first characteristic data of the second characteristic indicator, A 2n is the nth characteristic data of the second characteristic index; A n1 is the nth characteristic data of the second characteristic index, A nn is the nth characteristic data of the nth characteristic indicator;
[0034] Step S27, forming a feature data set according to the feature indicator set:
[0035] F=(A1、A2…A n )
[0036] Among them, F is the feature data set.
[0037] Preferably, the step of using the XGBoost model to train the data group in the feature data set to obtain the contract performance probability prediction model of the power grid material supplier is specifically:
[0038] Step S41, inputting the data group in the feature data set into the XGBoost model for training to obtain a contract performance probability prediction model for power grid material suppliers;
[0039] Step S42: the contract performance probability prediction model is:
[0040]
[0041] in, is the predicted value of the contract performance probability, k is the activation function, K is the total number of activation functions, fk is the kth classification decision tree, S n is the nth data group, Γ is the mapping space of the classification tree.
[0042] In the second aspect, the present application provides a material supply risk prediction system, which adopts the following technical solution:
[0043] A material supply risk prediction system, comprising:
[0044] The first acquisition module acquires historical data of several power grid material suppliers;
[0045] An extraction module extracts several characteristic indicators from the historical data, as well as characteristic data corresponding to each characteristic indicator, and constructs a characteristic data set based on the characteristic indicators and the corresponding characteristic data;
[0046] The construction module randomly extracts one feature data from the feature data corresponding to each feature index in the feature data set to form a data group, and constructs a data set L containing several data groups:
[0047] L=(S1、S2…S n )
[0048] Among them, S n is the nth data group;
[0049] The training module uses the XGBoost model to train the data group in the feature data set to obtain the contract performance probability prediction model of the power grid material supplier;
[0050] The second acquisition module acquires characteristic data of each characteristic indicator of the target power grid material supplier and records it as target data;
[0051] The execution module inputs the target data into the contract fulfillment probability prediction model to obtain the contract fulfillment probability of the target power grid material supplier.
[0052] Preferably, the execution module includes a memory and a processor, the memory is used to store the feature data, and the processor is used to process the feature data.
[0053] In summary, the present application includes at least one of the following beneficial technical effects:
[0054] 1. By using advanced XGBoost machine learning technology, the supplier's performance risk can be predicted more accurately. Compared with traditional statistical methods or manual judgment, this method can better capture complex data patterns and relationships. Identifying high-risk suppliers in advance helps companies take preventive measures, thereby reducing losses caused by supplier defaults and effectively improving the economic efficiency of the entire supply chain.
[0055] 2. Through effective prediction of supplier performance risks, it can help enterprises to plan supply chains more scientifically, improve the flexibility and responsiveness of the entire supply chain, and reduce additional costs caused by supplier default. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] Figure 1It is a schematic diagram of specific steps of an embodiment of a material supply risk prediction method of the present invention.
[0057] Figure 2 It is a module connection diagram of an embodiment of a material supply risk prediction system of the present invention. DETAILED DESCRIPTION
[0058] Below is a combination of the embodiments and Figure 1-Figure 2 The present invention is further described in detail, but the embodiments of the present invention are not limited thereto.
[0059] The present invention discloses a material supply risk prediction method, which specifically comprises the following steps:
[0060] Step S1, obtaining historical data of several power grid material suppliers.
[0061] During the implementation process, several power grid material suppliers can be selected as data collection objects, and a certain amount of historical data can be collected from each power grid material supplier. Among them, the historical data contains several indicators, and the indicator data corresponding to each indicator. For example, the indicators can include the material unique code, whether the price is linked, the sub-bid number, the subcontract number, the raw material category, the raw material price, the historical default rate, the contract execution cycle, the historical plan submission cycle average, the plan submission cycle, the total price of materials, the variance of the unit price of materials, and the historical default amount average. The indicator data is the material unique code data, whether the price is linked data, the sub-bid number data, the subcontract number data, the raw material category data, the raw material price data, the historical default rate data, the contract execution cycle data, the historical plan submission cycle average data, the plan submission cycle data, the material total price data, the material unit price variance data, and the historical default amount average data.
[0062] Step S2, extracting a number of characteristic indicators from the historical data, and characteristic data corresponding to each characteristic indicator, and constructing a characteristic data set according to the characteristic indicators and the corresponding characteristic data.
[0063] Step S3, randomly extract one feature data from the feature data corresponding to each feature index in the feature data set to form a data group, and construct a data set L containing several data groups:
[0064] L=(S1、S2…S n )
[0065] Among them, S n is the nth data group.
[0066] Step S4, using the XGBoost model to train the data group in the feature data set to obtain a contract performance probability prediction model for power grid material suppliers.
[0067] Step S5, obtaining characteristic data of each characteristic indicator of the target power grid material supplier and recording it as target data.
[0068] During the implementation process, if it is necessary to predict the contract performance probability of a target power grid material supplier, the characteristic indicator data of the target power grid material supplier should be collected first, such as historical default rate data, contract execution cycle data, historical plan submission cycle mean data, plan submission cycle data, material total price data, material unit price variance data and historical default amount mean data.
[0069] Step S6, input the target data into the contract performance probability prediction model to obtain the contract performance probability of the target power grid material supplier.
[0070] During the implementation process, the characteristic indicator data of the target power grid material supplier, such as historical default rate data, contract execution cycle data, historical plan submission cycle mean data, plan submission cycle data, total material price data, material unit price variance data and historical default amount mean data, are input into the contract performance probability prediction model to obtain the contract performance probability of the target power grid material supplier. After obtaining the contract performance probability of the target power grid material supplier, the default risk of the target power grid material supplier can be judged according to the contract performance probability of the target power grid material supplier. If the default risk of the target power grid material supplier is high, the cooperation with the target power grid material supplier can be stopped, or the cooperation with the target power grid material supplier can be terminated. If the default risk of the target power grid material supplier is low, the cooperation with the target power grid material supplier can be considered, or the cooperation with the target power grid material supplier can be continued.
[0071] In actual application, during the implementation process, any historical default rate data is extracted from several historical default rate data of the first characteristic indicator (historical default rate), any contract execution cycle data is extracted from several contract execution cycle data of the second characteristic indicator (contract execution cycle)... any historical default amount average data is extracted from several historical default amount average data of the nth characteristic indicator (historical default amount average), and any historical default rate data, any contract execution cycle data... any historical default amount average data are classified into one data group. S1 is the first data group, S1 = (S 11 , S 12 …S 1n ), where S 11 It can be defined as the historical default rate data in the data set S1, S 12 It can be defined as the contract execution cycle data in data group S1...S 1n It can be defined as the historical default amount average data in data group S1. S2 is the second data group, S2 = (S 21, S 22 …S 2n ), where S 21 It can be defined as the historical default rate data in the data set S2, S 22 It can be defined as the contract execution cycle data in data group S2...S 2n It can be defined as the historical default amount average data in data group S2...S n is the nth data group, S n =(S n1 , S n2 …S nn ), where S n1 It can be defined as a data set S n The historical default rate data in S n2 It can be defined as a data set S n Contract execution cycle data in ... nn It can be defined as a data set S n The historical average default amount data.
[0072] Before step S2, the historical data is also cleaned and normalized.
[0073] In actual application, for example, if there are identical records in the historical data, one record is retained and other duplicate records are deleted. If a record has missing values, it is filled with the average value of other records. After the data cleaning is completed, the cleaned historical data is normalized so that the indicator data is at a similar order of magnitude.
[0074] The steps for extracting several characteristic indicators from historical data are as follows:
[0075] Step S21, divide the historical data into two categories according to the performance status, wherein the first category of data is the performance data T1, and the second category of data is the non-performance data T2.
[0076] During the implementation process, if the power grid material supplier delivers the goods that meet the quality requirements on time in a certain record and does not violate any terms in the contract, then this record can be defined as fulfillment, and the data in this record is the fulfillment data. If the power grid material supplier fails to deliver the goods that meet the quality requirements on time in a certain record, or violates any terms in the contract, then this record can be defined as non-fulfillment, and the data in this record is the non-fulfillment data.
[0077] Step S22, calculate the mean of each data in the performance data T1 and variance α1, and the mean of each data in the default data T2 and variance α2.
[0078] During the implementation process, extract several indicator data (historical default rate data) of a certain indicator (such as historical default rate) in the performance data T1, and calculate the average of the several indicator data (historical default rate data) Extract the indicator data (historical default rate data) of a certain indicator (such as historical default rate) in the default data T2 in several records, and calculate the mean of the several indicator data (historical default rate data) And variance α2. Among them, the number of indicator data of a certain indicator in the fulfillment data T1 may also be 0, 1, 2 or several, and the number in the non-fulfillment data T2 may also be 0, 1, 2 or several.
[0079] Step S23, based on the mean Variance α1, number of data in fulfillment data T1 m1, mean Variance α2 and m2 of the data in the default data T2, calculate the statistic t and degrees of freedom df.
[0080] Step S24, calculating the P value according to the statistic t and the degrees of freedom df.
[0081] In implementation, the P value in the t-distribution table is found based on the statistic t and the degrees of freedom df.
[0082] Step S25: If the P value is less than the preset value of 0.05, it is considered that there is a significant relationship between the data and the probability of contract performance, and the index corresponding to the data is marked as a characteristic indicator.
[0083] In actual application, during the implementation process, the P value indicates the probability of observing the current sample difference or a more extreme difference when the null hypothesis (i.e., the means of the two groups of data are the same) is established. If the P value is less than the preset significance level (such as 0.05), the null hypothesis is rejected, and it is considered that there is a significant difference between the two groups of data in this characteristic indicator. If the P value of an indicator is less than 0.05, it is considered that there is a significant difference between the performance data and the non-performance data of the indicator, that is, there is a significant relationship between the indicator data and the probability of contract performance, and the indicator is marked as a characteristic indicator, and the indicator data corresponding to the indicator is marked as characteristic indicator data.
[0084] Traverse all indicators in the historical data to select characteristic indicators that have a significant relationship with the probability of contract performance. For example, historical default rate, contract execution cycle, historical plan submission cycle mean, plan submission cycle, total material price, material unit price variance and historical default amount mean. Among them, historical default rate = supplier's historical default times / historical supply times, contract execution cycle = contract specified delivery date - contract signing date, historical plan submission cycle mean = total plan submission cycle of material code / number of orders, plan submission cycle = planned delivery date - planned submission date, total material price = total material price in the contract, material unit price variance = total material unit price / number of orders, historical default amount mean = total contract amount / number of contracts.
[0085] In step S23, the statistical calculation formula is as follows:
[0086]
[0087] The statistic t is calculated according to the statistic calculation formula.
[0088] In step S23, the calculation formula of the degree of freedom is as follows:
[0089] df=m1+m2 -2
[0090] The degrees of freedom df is calculated according to the calculation formula of degrees of freedom.
[0091] The steps for constructing a feature data set based on feature indicators and corresponding feature data are as follows:
[0092] Step S26, forming a characteristic indicator set according to all characteristic data corresponding to the characteristic indicators:
[0093] A1=(A 11 , A 12 …A 1n )
[0094] A2=(A 21 , A 22 …A 2n )
[0095] …
[0096] A n =(A n1 , A n2 …A nn )
[0097] Among them, A1, A2, ... A n are all characteristic index sets, A 11 is the first characteristic data of the first characteristic indicator, A 1nis the nth characteristic data of the first characteristic indicator. 21 is the first characteristic data of the second characteristic indicator, A 2n is the nth characteristic data of the second characteristic index. n1 is the nth characteristic data of the second characteristic index, A nn is the nth feature data of the nth feature indicator.
[0098] Step S27, forming a feature data set according to the feature indicator set:
[0099] F=(A1、A2…A n )
[0100] Among them, F is the feature data set.
[0101] The steps of using the XGBoost model to train the data group in the feature data set to obtain the contract performance probability prediction model of the power grid material supplier are as follows:
[0102] Step S41, inputting the data group in the feature data set into the XGBoost model for training to obtain a contract performance probability prediction model for power grid material suppliers.
[0103] Step S42: the contract performance probability prediction model is:
[0104] in, is the predicted value of the contract performance probability, k is the activation function, K is the total number of activation functions, fk is the kth classification decision tree, S n is the nth data group, Γ is the mapping space of the classification tree.
[0105] A material supply risk prediction system, by applying a material supply risk prediction method as described above, comprises:
[0106] The first acquisition module acquires the historical data of several power grid material suppliers. It can be a database server, a web crawler or an API calling tool, which is responsible for collecting and integrating the historical data of power grid material suppliers from the enterprise resource planning (ERP) system, supply chain management system (SCM) or other related business systems.
[0107] The extraction module extracts several characteristic indicators from the historical data, as well as the characteristic data corresponding to each characteristic indicator, and constructs a characteristic data set based on the characteristic indicators and the corresponding characteristic data. It can be a data processing server, such as a Hadoop cluster or a Spark cluster, which can process large-scale data sets and extract specific characteristic indicators from them. It can also be a data analysis software, such as a Python or R language environment, which uses libraries such as Pandas and NumPy for data cleaning and feature extraction.
[0108] The construction module randomly extracts one feature data from the feature data corresponding to each feature index in the feature data set to form a data group, and constructs a data set L containing several data groups:
[0109] L=(S1、S2…S n )
[0110] Among them, S n For the nth data set
[0111] This can be a data analysis software (Python / R), using Pandas or similar tools to create a DataFrame, and then selecting feature data to form data groups.
[0112] The training module uses the XGBoost model to train the data group in the feature data set to obtain the electric
[0113]
[0114] The contract performance probability prediction model of online material suppliers can be a deep learning framework, such as TensorFlow or PyTorch, which supports the XGBoost algorithm and can train the model efficiently on GPU or CPU.
[0115] The second acquisition module acquires the characteristic data of each characteristic indicator of the target power grid material supplier and records it as the target data. It can be a Web crawler or an API call tool, which acquires the latest characteristic indicator data of the target power grid material supplier through database query, API call or Web crawler. It can also be a dedicated program interface (API) for collecting information from multiple data sources in real time or periodically.
[0116] The execution module inputs the target data into the contract performance probability prediction model to obtain the contract performance probability of the target power grid material supplier. The execution module includes a memory and a processor, the memory is used to store the feature data, and the processor is used to process the feature data. It can be an application server, such as Tomcat or Nginx, running a Web application or microservice, receiving front-end requests and sending the target supplier's data to the prediction model.
[0117] In actual application, by using advanced machine learning techniques (such as XGBoost), the performance risk of suppliers can be predicted more accurately. Compared with traditional statistical methods or manual judgment, this method can better capture complex data patterns and relationships. Identifying high-risk suppliers in advance can help companies take preventive measures, such as finding alternative suppliers, adjusting procurement strategies, etc., thereby reducing losses caused by supplier default. Through effective prediction of supplier performance risks, it can help companies plan supply chains more scientifically and improve the flexibility and responsiveness of the entire supply chain. Reduce additional cost expenditures caused by supplier default, such as production stagnation costs caused by delayed delivery and the cost of finding emergency substitutes.
[0118] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.
Claims
1. A material supply risk prediction method, characterized in that: The following steps are involved: Step S1, obtaining historical data of several power grid material suppliers; Step S2, extracting several characteristic indicators from the historical data, and characteristic data corresponding to each characteristic indicator, and constructing a characteristic data set according to the characteristic indicators and the corresponding characteristic data; Step S3, randomly extract one feature data from the feature data corresponding to each feature index in the feature data set to form a data group, and construct a data set L containing several data groups: L=(S1、S2…S n ) Among them, S n is the nth data group; Step S4, using the XGBoost model to train the data group in the feature data set to obtain a contract performance probability prediction model for power grid material suppliers; Step S5, obtaining characteristic data of each characteristic indicator of the target power grid material supplier and recording it as target data; Step S6, input the target data into the contract performance probability prediction model to obtain the contract performance probability of the target power grid material supplier.
2. A material supply risk prediction method according to claim 1, characterized in that: Before step S2, the historical data is also cleaned and normalized.
3. A material supply risk prediction method according to claim 2, characterized in that: The step of extracting several characteristic indicators from the historical data is specifically as follows: Step S21, dividing the historical data into two categories according to the performance status, wherein the first category of data is the performance data T1, and the second category of data is the non-performance data T2; Step S22, calculate the mean of each data in the performance data T1 and variance α1, and the mean of each data in the default data T2 and variance α2; Step S23, based on the mean Variance α1, number of data in fulfillment data T1 m1, mean Variance α2 and m2 of the data in the default data T2, calculate the statistic t and degrees of freedom df; Step S24, calculating the P value according to the statistic t and the degree of freedom df; Step S25: If the P value is less than the preset value of 0.05, it is considered that there is a significant relationship between the data and the probability of contract performance, and the index corresponding to the data is marked as a characteristic indicator.
4. A material supply risk prediction method according to claim 3, characterized in that: In step S23, the statistical calculation formula is as follows: The statistic t is calculated according to the statistic calculation formula.
5. A material supply risk prediction method according to claim 4, characterized in that: In step S23, the calculation formula of the degree of freedom is as follows: df=m1+m2 -2 The degrees of freedom df is calculated according to the calculation formula of degrees of freedom.
6. A material supply risk prediction method according to claim 3, characterized in that: The step of constructing a feature data set according to the feature index and the corresponding feature data is specifically as follows: Step S26, forming a characteristic indicator set according to all characteristic data corresponding to the characteristic indicators: A1=(A 11 、A 12 …A 1n ) A2=(A 21 、A 22 …A 2n ) …… A n =(A n1 、A n2 …A nn ) Among them, A1, A2, ... A n are all characteristic index sets, A 11 is the first characteristic data of the first characteristic indicator, A 1n is the nth characteristic data of the first characteristic indicator; A 21 is the first characteristic data of the second characteristic indicator, A 2n is the nth characteristic data of the second characteristic index; A n1 is the nth characteristic data of the second characteristic index, A nn is the nth characteristic data of the nth characteristic indicator; Step S27, forming a feature data set according to the feature indicator set: <h2 style=";text-align:left;direction:ltr">F=(A1、A2…A<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> ) Among them, F is the feature data set.
7. A material supply risk prediction method according to claim 1, characterized in that: The step of using the XGBoost model to train the data group in the feature data set to obtain the contract performance probability prediction model of the power grid material supplier is specifically as follows: Step S41, inputting the data group in the feature data set into the XGBoost model for training to obtain a contract performance probability prediction model for power grid material suppliers; Step S42: the contract performance probability prediction model is: in, is the predicted value of the contract performance probability, k is the activation function, K is the total number of activation functions, fk is the kth classification decision tree, S n is the nth data group, Γ is the mapping space of the classification tree.
8. A material supply risk prediction system, characterized in that: By applying a material supply risk prediction method as described in any one of claims 1 to 7, comprising: The first acquisition module acquires historical data of several power grid material suppliers; An extraction module extracts several characteristic indicators from the historical data, as well as characteristic data corresponding to each characteristic indicator, and constructs a characteristic data set based on the characteristic indicators and the corresponding characteristic data; The construction module randomly extracts one feature data from the feature data corresponding to each feature index in the feature data set to form a data group, and constructs a data set L containing several data groups: L=(S1、S2…S n ) Among them, S n is the nth data group; The training module uses the XGBoost model to train the data group in the feature data set to obtain the contract performance probability prediction model of the power grid material supplier; The second acquisition module acquires characteristic data of each characteristic indicator of the target power grid material supplier and records it as target data; The execution module inputs the target data into the contract fulfillment probability prediction model to obtain the contract fulfillment probability of the target power grid material supplier.
9. A material supply risk prediction system according to claim 8, characterized in that: The execution module includes a memory and a processor, the memory is used to store feature data, and the processor is used to process feature data.