Supply chain network optimization method and system

Abstract

The invention discloses a supply chain network optimization method and system. The method comprises the following steps: establishing a first envelope analysis model; obtaining a sample data set, and training the first envelope analysis model by adopting the sample data set to obtain a second envelope analysis model; inputting the operation data of the to-be-evaluated supplier node into the second envelope analysis model to obtain a first efficiency value of the supplier in each time period and a change condition of the first efficiency value; according to the change condition of the first efficiency value, a key driving factor causing supplier efficiency reduction is obtained; and inputting the second operation data and the key driving factor after a preset time into a second envelope analysis model to obtain a second efficiency value of the supplier, and optimizing parameters of the second envelope analysis model to manage supply chain network nodes. By adopting the embodiment of the invention, the data of each node in the supply chain network can be automatically collected and processed, and automatic and effective supply chain network management is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of supply chain management, and in particular to a supply chain network optimization method and system. Background Art

[0002] During the procurement process of the power grid, suppliers need to be evaluated and monitored for efficiency in order to control costs while ensuring quality output, which is of great significance to the management of the power material supply chain.

[0003] In the existing technology, supply chain networks generally require manual determination of evaluation indicators for nodes in the supply chain network, and then analysis and management of each node in the supply chain network. However, manual management methods require analysts to have high professionalism. As the complexity of the supply chain network increases, and the architecture and market changes of the supply chain network accelerate, traditional supply chain management methods require a lot of manpower and time, have low analysis efficiency, and are difficult to avoid errors and omissions, resulting in low accuracy of the generated supply chain management solution. Summary of the invention

[0004] The technical problem to be solved by the present invention is how to automatically manage the supply chain network based on the data of the supply chain network nodes, and to provide a supply chain network management method and system that can automatically collect and process the data of the supply chain network nodes, identify the key driving factors that lead to efficiency changes, and realize automatic and effective supply chain network management.

[0005] An embodiment of the present invention provides a method for optimizing a supply chain network, comprising: Establishing a first envelope analysis model; the input of the first envelope analysis model includes resource input data, technology input data and environmental factor data of each node in the supply chain network; Acquire a sample data set, and use the sample data set to train the first envelope analysis model to obtain a second envelope analysis model; the input-output ratio of the second envelope analysis model is not less than a preset target ratio threshold; Inputting the operation data of the supplier node to be evaluated into the second envelope analysis model to obtain the first efficiency value of the supplier node in each time period and the change of the first efficiency value; According to the change of the first efficiency value, a key driving factor causing the supplier node efficiency to decrease is obtained from the operation data; Collecting second operating data of the supplier node after a preset time, inputting the second operating data and the key driving factors into the second envelope analysis model, and obtaining a second efficiency value of the supplier node; According to the difference between the first efficiency value and the second efficiency value, the parameters of the second envelope analysis model are optimized to obtain an optimized envelope analysis model to manage supply chain network nodes.

[0006] As an improvement of the above solution, the establishing of the first envelope analysis model includes: Normalizing the input resource input data, technology input data and environmental factor data of each node in the supply chain network to obtain normalized data; Calculate the correlation value of the normalized data by using the Pearson algorithm, and obtain the first output efficiency value of the efficiency evaluation stage according to the correlation value; Clustering the first output efficiency values ​​to obtain an evaluation grade; A first function expression of the evaluation level is established by using a linear regression method, and parameters of the first function expression are weighted to obtain a second output efficiency value of the efficiency decomposition stage; Calculate an actual deviation value according to the second output efficiency value and an actual production efficiency value of the supplier node; According to the actual deviation value, a standard normal distribution curve is constructed, and a standard deviation is calculated according to the standard normal distribution curve to obtain a third output efficiency value; Using the least square method to perform quadratic fitting on the third output efficiency value to obtain a fitting curve; A difference sequence node is obtained from the fitting curve, an environmental adjustment value is determined according to the coordinates of the difference sequence node at the inflection point position, and the third output efficiency value is corrected using the environmental adjustment value to establish the output of the first envelope analysis model.

[0007] As an improvement of the above solution, the acquiring of the sample data set and the use of the sample data set to train the first envelope analysis model to obtain the second envelope analysis model include: Get an initial sample dataset from the supplier node; The initial sample data set is expanded using a sample enhancement algorithm to generate a number of expanded data having a similarity with the initial sample data set greater than a preset similarity threshold; Obtaining a sample data set according to the initial sample data set and the expanded data; Using the sample data set to train the first envelope analysis model to obtain first weight parameters of each input indicator of the first envelope analysis model; The first weight parameter is used to weight the input index of the first envelope analysis model to obtain a second envelope analysis model.

[0008] As an improvement of the above solution, the sample enhancement algorithm is used to expand the initial sample data set to generate a number of expanded data with a similarity with the initial sample data set greater than a preset similarity threshold, including: Normalizing the initial sample data set on a monthly basis to obtain monthly normalized data, and calculating the Pearson correlation coefficient based on the monthly normalized data; The Pearson correlation coefficient is used to determine the sampling weight, and the monthly normalized data is calculated according to the sampling weight to obtain the first-stage data; The moving average method is used to obtain the numerical central trend of the data of the first stage, and the probability distribution fitting is performed on the numerical central trend to obtain the data of the second stage; A monthly time series curve is constructed based on the data of the second stage, a prediction function is established for the production achievement rate, the qualified product rate and the on-time delivery rate in the monthly time series curve using the least square method, and the data of the third stage is obtained based on the prediction function; Constructing a timing evaluation curve according to the third stage data, calculating the mean variance of the timing evaluation curve, segmenting the timing evaluation curve, and obtaining the slope of the segmented curve; The initial sample data set is expanded according to the segmented curve slope and a preset similarity threshold to generate expanded data.

[0009] As an improvement of the above solution, the first envelope analysis model is trained using the sample data set to obtain the first weight parameter of each input indicator of the first envelope analysis model, including: Performing stratified sampling processing on the sample data set according to the output values ​​of the supplier nodes corresponding to the sample data set to obtain a training sample data set and a test sample data set; Obtaining training sample input data of the first envelope analysis model from the training sample data set, calculating the correlation between the training sample input data, and obtaining a covariance matrix; Obtaining initial weight parameters according to the eigenvalues ​​of the covariance matrix; Constructing a linear programming function according to the initial weight parameters to obtain a fitting curve; Optimizing the initial weight parameters according to the fitting curve and the actual values ​​of the training sample data to obtain optimized weight parameters; Inputting the test sample data set into a first envelope analysis model weighted by the optimized weight parameter to obtain a verification output value; Performing normal distribution fitting on the verification output value and the test sample data set to obtain an error parameter; According to the optimized weight parameter and the error parameter, a first weight parameter of each input index of the first envelope analysis model is obtained.

[0010] As an improvement of the above solution, the operation data of the supplier node to be evaluated is input into the second envelope analysis model to obtain the first efficiency value of the supplier node in each time period and the change of the first efficiency value, including: Collecting the operation data of the supplier node to be evaluated at different time points; the operation data includes input data and output data; In the second envelope analysis model, a correlation matrix of the input data is calculated, and cluster analysis is performed on the correlation matrix to obtain matching parameters of the input data; Performing exponential smoothing processing on the output data to obtain an output data sequence; Calculate the difference between each point of the output data sequence to obtain a monthly node sequence; According to the matching parameters and the output data of each sequence point on the monthly node sequence, a first efficiency value of each time period is obtained; Performing least square fitting on the first efficiency value and segmenting according to the fitting result to obtain a slope sequence; According to the fluctuation range of the slope sequence, a key interval is obtained, and a first change point of the proportioning parameter and a second change point of the output data are extracted from the key interval; According to the first change point and the second change point, the change of the first efficiency value of the supplier node in each time period is obtained.

[0011] As an improvement of the above solution, the key driving factors causing the decline in supplier node efficiency are obtained from the operation data according to the change of the first efficiency value, including: According to the change of the first efficiency value, the operation data in the decreasing interval is obtained by screening from the operation data; the absolute value of the difference between the efficiency value of the first time period corresponding to the operation data in the decreasing interval and the efficiency values ​​of the remaining second time periods is greater than a preset efficiency value difference threshold; Calculating the correlation coefficient of the input data of the declining interval operation data, and grouping the input data of the declining interval operation data according to the correlation coefficient to obtain a plurality of declining interval input data groups; Performing principal component analysis on the output data of the declining interval operation data to obtain a feature vector; Calculating a month-on-month change rate according to the characteristic vector, and obtaining a significant change index from the output data of the operating data in the declining interval according to the month-on-month change rate and a preset month-on-month change rate threshold; Performing correlation analysis on the decline interval input data group and the significant change index to obtain a correlation strength value; According to the association strength value, key driving factors are obtained; the key driving factors include key input data and key output data that lead to a decrease in supplier node efficiency.

[0012] As an improvement of the above solution, the correlation analysis is performed on the decline interval input data group and the significant change index to obtain the correlation strength value, including: Perform difference calculations on the input data group in the declining interval and the significant change index respectively with the preset benchmark data to obtain an input fluctuation amplitude sequence and an output fluctuation amplitude sequence; Calculating the correlation between the input fluctuation amplitude sequence and the output fluctuation amplitude sequence to obtain a correlation curve; According to a preset correlation threshold, the correlation curve is divided into a plurality of intervals; According to the plurality of intervals, interval features are obtained; each of the intervals corresponds to an interval feature; According to the interval characteristics, the falling interval input data group and the significant change index are grouped into a strong correlation group and a non-strong correlation group; Establishing a decision tree for the strongly associated group, splitting the nodes of the decision tree according to a preset information gain criterion, and extracting numerical distribution rules from the nodes; According to the numerical distribution law, the correlation coefficient is calculated to obtain the association strength value.

[0013] As an improvement of the above solution, the method of optimizing the parameters of the second envelope analysis model according to the difference between the first efficiency value and the second efficiency value to obtain an optimized envelope analysis model to manage supply chain network nodes includes: Calculating the difference between the first efficiency value and the second efficiency value to obtain a fitting curve of efficiency change within the preset time; According to the fluctuation of the fitting curve, abnormal time points are obtained by screening the operation data at all times; Performing similarity calculation on the parameters of the second envelope analysis model corresponding to the abnormal time point to obtain parameter similarity; According to the parameter similarity and a preset parameter similarity threshold, adjusting the parameter within an iteration interval to obtain an adjusted parameter; The adjustment parameters are used to optimize the parameters of the second envelope analysis model to obtain an optimized envelope analysis model to manage supply chain network nodes.

[0014] The embodiment of the present invention also provides an evaluation system for improving supplier efficiency, including: A model building module, used to build a first envelope analysis model; the input of the first envelope analysis model includes resource input data, technology input data and environmental factor data of each node in the supply chain network; A model training module, used to obtain a sample data set, and use the sample data set to train the first envelope analysis model to obtain a second envelope analysis model; the input-output ratio of the second envelope analysis model is not less than a preset target ratio threshold; A first efficiency value calculation module, used for inputting the operation data of the supplier node to be evaluated into the second envelope analysis model to obtain the first efficiency value of the supplier node in each time period and the change of the first efficiency value; A key driving factor acquisition module, configured to obtain, from the operation data, key driving factors that cause a decrease in supplier node efficiency according to changes in the first efficiency value; A second efficiency value calculation module, used for collecting second operation data of the supplier node after a preset time, inputting the second operation data and the key driving factors into the second envelope analysis model, and obtaining a second efficiency value of the supplier; The supply chain network management module is used to optimize the parameters of the second envelope analysis model according to the difference between the first efficiency value and the second efficiency value to obtain an optimized envelope analysis model to manage the supply chain network nodes.

[0015] Compared with the prior art, the present invention discloses an evaluation method and system for improving supplier efficiency, which establishes a first envelope analysis model; the input of the first envelope analysis model includes resource input data, technology input data and environmental factor data of each node in the supply chain network; obtains a sample data set, and uses the sample data set to train the first envelope analysis model to obtain a second envelope analysis model; the input-output ratio of the second envelope analysis model is not less than a preset target ratio threshold; the operation data of the supplier node to be evaluated is input into the second envelope analysis model to obtain the first efficiency value of the supplier node in each time period, and the change of the first efficiency value; according to the change of the first efficiency value, the key driving factors that cause the supplier node efficiency to decrease are obtained from the operation data; the second operation data of the supplier node after a preset time is collected, and the second operation data and the key driving factors are input into the second envelope analysis model to obtain the second efficiency value of the supplier node; according to the difference between the first efficiency value and the second efficiency value, the parameters of the second envelope analysis model are optimized to obtain the optimized envelope analysis model to manage the supply chain network nodes. By adopting the embodiment of the present invention, the data of each node in the supply chain network is automatically collected and processed, and the parameters of the envelope analysis model are improved in multiple levels and dimensions, so that the key driving factors that cause changes in node efficiency can be accurately identified, and automatic and effective supply chain network management can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of the steps of a supply chain network optimization method provided by an embodiment of the present invention; Figure 2 It is a structural diagram of a supply chain network optimization system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] In the description of the specification and claims, it is to be understood that the terms first, second, etc. in the specification and claims are only used for the purpose of describing the same technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated, nor necessarily describing the order or time sequence. The terms are interchangeable where appropriate. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features.

[0019] The embodiment of the present invention provides a method for optimizing a supply chain network. Figure 1 In this embodiment, the supply chain network optimization method is specifically performed through steps S1 to S6: S1. Establish a first envelope analysis model; the input of the first envelope analysis model includes resource input data, technology input data and environmental factor data of each node in the supply chain network.

[0020] S2. Obtain a sample data set, and use the sample data set to train the first envelope analysis model to obtain a second envelope analysis model; the input-output ratio of the second envelope analysis model is not less than a preset target ratio threshold.

[0021] S3. Inputting the operation data of the supplier node to be evaluated into the second envelope analysis model to obtain the first efficiency value of the supplier node in each time period and the change of the first efficiency value.

[0022] S4. According to the change of the first efficiency value, obtain the key driving factors that cause the decline in supplier node efficiency from the operation data.

[0023] S5. Collect second operating data of the supplier node after a preset time, input the second operating data and the key driving factors into the second envelope analysis model, and obtain a second efficiency value of the supplier node.

[0024] S6. Optimize the parameters of the second envelope analysis model according to the difference between the first efficiency value and the second efficiency value to obtain an optimized envelope analysis model to manage supply chain network nodes.

[0025] In the above scheme, by automatically collecting and processing data from each node in the supply chain network, the parameters of the envelope analysis model are improved in multiple levels and dimensions, which can accurately identify the key driving factors that cause changes in node efficiency and realize automatic and effective supply chain network management.

[0026] As a preferred implementation, step S1, establishing a first envelope analysis model, includes: Normalizing the input resource input data, technology input data and environmental factor data of each node in the supply chain network to obtain normalized data; Calculate the correlation value of the normalized data by using the Pearson algorithm, and obtain the first output efficiency value of the efficiency evaluation stage according to the correlation value; Clustering the first output efficiency values ​​to obtain an evaluation grade; A first function expression of the evaluation level is established by using a linear regression method, and parameters of the first function expression are weighted to obtain a second output efficiency value of the efficiency decomposition stage; Calculate an actual deviation value according to the second output efficiency value and an actual production efficiency value of the supplier node; According to the actual deviation value, a standard normal distribution curve is constructed, and a standard deviation is calculated according to the standard normal distribution curve to obtain a third output efficiency value; Using the least square method to perform quadratic fitting on the third efficiency value to obtain a fitting curve; A difference sequence node is obtained from the fitting curve, an environmental adjustment value is determined according to the coordinates of the difference sequence node at the inflection point position, and the third output efficiency value is corrected using the environmental adjustment value to establish the output of the first envelope analysis model.

[0027] It should be noted that, in an embodiment of the present invention, the first envelope analysis model is a three-stage DEA model; the three-stage DEA (Data Envelopment Analysis) model includes an efficiency evaluation stage, an efficiency decomposition stage and an environmental factor correction stage. The input of the three-stage DEA model is the supplier's operating data, including resource input data, technology input data and environmental factor data. The above steps show how to process the input data in three stages to obtain the output of the first envelope analysis model.

[0028] Exemplarily, the resource input data includes equipment investment, raw material purchase amount and labor cost. Taking the annual operating data of a supplier in 2022 as an example, the equipment investment amount is 10 million yuan, the raw material purchase amount is 5 million yuan, and the labor cost is 3 million yuan. After normalization, they are 0.85, 0.42, and 0.25 respectively. The technology input data includes R&D investment, patent licensing fees and technological transformation fees. The R&D investment is 8 million yuan, the patent licensing fee is 2 million yuan, and the technological transformation fee is 1.5 million yuan. After normalization, they are 0.75, 0.19, and 0.14. The market condition values ​​include market share, product price index and the number of competitors. The market share is 0.35, the price index is 1.2, and there are 15 competitors. After normalization, they are 0.35, 0.65, and 0.48.

[0029] The Pearson algorithm calculated that the correlation between resource input and technology input is 0.82, the correlation between resource input and market conditions is 0.65, and the correlation between technology input and market conditions is 0.58. The corresponding weight parameters are set to 0.4, 0.35, and 0.25, respectively. The weighted summation results in the first output efficiency value of 0.72.

[0030] The first output efficiency values ​​are clustered and divided into five levels: excellent, good, medium, qualified and to be improved. The corresponding numerical ranges are above 0.8, 0.6 to 0.8, 0.4 to 0.6, 0.2 to 0.4, and below 0.2. The centroid values ​​of each level are 0.85, 0.7, 0.5, 0.3, and 0.15. The evaluation value of 0.72 is closest to the centroid of the good level. The evaluation level function is established through linear regression, and the second output efficiency value of 0.68 is obtained after weighting the function parameters.

[0031] The production efficiency value is calculated by the output-to-input ratio to be 0.75, and the difference between it and the second output efficiency value of 0.68 is 0.07. The mean of all differences is 0.05 and the variance is 0.02. The standard normal distribution curve is constructed, and the third output efficiency value is 0.03, which is within the normal fluctuation range of -1 to 1.

[0032] The third output efficiency value is fitted with a quadratic term, and the sequence of adjacent point difference values ​​is 0.02, 0.03, 0.04, 0.06, and 0.08. After sorting, the first 20% position corresponds to a difference of 0.03, and the inflection point coordinates are recorded as (15, 0.68). The slope before the inflection point is 0.04 and the slope after the inflection point is -0.02. According to the change in slope, the environmental adjustment value is 0.05. The environmental adjustment value reflects the impact of the intensified market competition, the fluctuation range of raw material prices, and the intensity of policy regulation on the production and operation of the enterprise. The price of raw materials rose from 100 yuan per ton to 120 yuan per ton, the number of competitors increased from 12 to 15, and the implementation standard of environmental protection policies increased by 10%. The superposition of these factors has led to an increase in the operating costs of the enterprise. The output efficiency value is corrected after adjustment by environmental factors to make the evaluation results more objective and accurate.

[0033] As a preferred implementation, step S2, using the sample data set to train the first envelope analysis model to obtain a second envelope analysis model, includes: Get an initial sample dataset from the supplier node; The initial sample data set is expanded using a sample enhancement algorithm to generate a number of expanded data having a similarity with the initial sample data set greater than a preset similarity threshold; Obtaining a sample data set according to the initial sample data set and the expanded data; Using the sample data set to train the first envelope analysis model to obtain first weight parameters of each input indicator of the first envelope analysis model; The first weight parameter is used to weight the input index of the first envelope analysis model to obtain a second envelope analysis model.

[0034] In an embodiment of the present invention, by expanding the initial sample data set and making the similarity between the expanded data and the initial sample data set greater than a preset similarity threshold, the data for model training can be increased, and the changing trend of the data of the supplier node in different time periods can be further accurately evaluated.

[0035] Further, preferably, the using of a sample enhancement algorithm to expand the initial sample data set to generate a number of expanded data having a similarity with the initial sample data set greater than a preset similarity threshold includes: Normalizing the initial sample data set on a monthly basis to obtain monthly normalized data, and calculating the Pearson correlation coefficient based on the monthly normalized data; The Pearson correlation coefficient is used to determine the sampling weight, and the monthly normalized data is calculated according to the sampling weight to obtain the first-stage data; The moving average method is used to obtain the numerical central trend of the data of the first stage, and the probability distribution fitting is performed on the numerical central trend to obtain the data of the second stage; A monthly time series curve is constructed based on the data of the second stage, a prediction function is established for the production achievement rate, the qualified product rate and the on-time delivery rate in the monthly time series curve using the least square method, and the data of the third stage is obtained based on the prediction function; Constructing a timing evaluation curve according to the third stage data, calculating the mean variance of the timing evaluation curve, segmenting the timing evaluation curve, and obtaining the slope of the segmented curve; The initial sample data set is expanded according to the segmented curve slope and a preset similarity threshold to generate expanded data.

[0036] It should be noted that in the embodiment of the present invention, data processing is performed on a monthly basis. In actual applications, other time units may also be used, such as quarterly, annual, etc. This is related to the chain owner's timeliness requirements for the efficiency value. The specific time unit settings do not affect the beneficial effects of the present invention.

[0037] Exemplarily, the production data, quality data and delivery data in the supplier's monthly raw data are normalized, the normalized value is subtracted from the mean and divided by the standard deviation, the upper and lower boundary values ​​of the normalized value are calculated, the values ​​exceeding the upper and lower boundaries are marked as outliers and deleted, the correlation between the remaining values ​​is calculated by the Pearson correlation coefficient, the sampling weight of the values ​​with a correlation greater than 0.7 is set to 0.6, the sampling weight of the values ​​with a correlation between 0.3 and 0.7 is set to 0.3, the sampling weight of the values ​​with a correlation less than 0.3 is set to 0.1, and the values ​​are sampled according to the sampling weights to form the first stage data.

[0038] The data of the first phase are divided into monthly time sequence, and the moving average is used to obtain the central trend of the values ​​in each monthly interval. The moving window length is set to 5, and the probability distribution is fitted to the moving average results. The similarity with the supplier data of the first phase is obtained from the fitting curve. The similarity is calculated by the Pearson coefficient, and the data points with similarity values ​​greater than 0.8 are added to the second phase data.

[0039] The monthly time series change curve is obtained from the second phase data, and the vacant positions in the monthly curve are filled by least square fitting. The monthly efficiency index is calculated for the filled monthly curve. The monthly efficiency index includes three items: production achievement rate, qualified product rate and on-time delivery rate. The monthly prediction function is constructed for the three indicators by least square fitting, and the positive and negative signs of the prediction function parameters are determined to obtain the third phase supplier data. The time series evaluation curve is constructed for the third phase supplier data, and the mean variance is used to calculate the fluctuation range of the evaluation curve. When the value of the calculated fluctuation upper limit minus the lower limit exceeds 0.5, the curve is segmented, and the segmented curve is fitted by cubic spline interpolation. The slope of the fitting curve is calculated by the difference in the values ​​of adjacent points. The slope is greater than 0.3 and is determined to be an upward trend. The slope is less than -0.3 and is determined to be a downward trend. The slope is between -0.3 and 0.3 and is determined to be a stable trend. Then, the initial sample data set is expanded in combination with the stable trend to generate expanded data. The similarity between the expanded sample generated by the above scheme and the initial sample data set is greater than the preset similarity threshold.

[0040] The monthly raw data of suppliers include production line output records, quality inspection records and logistics distribution records. Taking a certain auto parts supplier as an example, the monthly output data is 8,500 pieces, the quality qualified data is 8,075 pieces, and the on-time delivery data is 7,990 pieces. After normalization, the values ​​are 0.85, 0.81, and 0.79 respectively. The mean is 0.82 and the standard deviation is 0.03. The upper and lower boundaries are set to mean ± 2 times the standard deviation, that is, 0.88 and 0.76, and the outliers that deviate from the boundaries are deleted. The correlation coefficient shows that the correlation between output and quality is 0.85, the correlation between output and delivery is 0.72, and the correlation between quality and delivery is 0.68. The sampling weights are set to 0.6, 0.6, and 0.3 accordingly. In the monthly time period from January to December, the supplier data is calculated by a 5-point moving average. The output data in January are 0.85, 0.83, 0.86, 0.82, and 0.84, with an average of 0.84. The quality data are 0.81, 0.82, 0.80, 0.83, and 0.81, with an average of 0.81. The delivery data are 0.79, 0.78, 0.80, 0.77, and 0.81, with an average of 0.79. The data distribution curve is obtained by probability distribution fitting, and the similarity with the original data is calculated to be 0.85, which is higher than the 0.8 threshold. There is a breakpoint in the supplier data in the second quarter, and the data in April is missing. The least squares fitting is used to supplement the data in April. The production data is 0.83, the quality data is 0.80, and the delivery data is 0.78. The monthly efficiency indicators show that the production achievement rate has increased from 83% to 86%, the qualified product rate has decreased from 95% to 93%, and the on-time delivery rate has remained at 94%. The least squares fitting shows that the parameters of the production prediction function are positive, the quality prediction function parameters are negative, and the delivery prediction function parameters are close to 0. The time series evaluation curve fluctuated significantly in the third quarter, with an average of 0.82 and a variance of 0.06 from July to September. The upper limit of fluctuation 0.88 minus the lower limit 0.76 is 0.12, which is less than the 0.5 segmentation threshold. The slope of the cubic spline interpolation fitting curve was 0.35 in August, showing an upward trend, and the slope was -0.25 in September, showing a stable trend.

[0041] It should be noted that the above-mentioned first-phase data, second-phase data and third-phase data can not only be used for data expansion, but also can evaluate the overall performance level of suppliers in the three dimensions of production increase, quality control and delivery timeliness. The fluctuation of data at key nodes can reflect the supplier's ability to adjust to emergencies.

[0042] Preferably, the adopting the sample data set to train the first envelope analysis model to obtain the first weight parameter of each input indicator of the first envelope analysis model includes: Performing stratified sampling processing on the sample data set according to the output values ​​of the supplier nodes corresponding to the sample data set to obtain a training sample data set and a test sample data set; Obtaining training sample input data of the first envelope analysis model from the training sample data set, calculating the correlation between the training sample input data, and obtaining a covariance matrix; Obtaining initial weight parameters according to the eigenvalues ​​of the covariance matrix; Constructing a linear programming function according to the initial weight parameters to obtain a fitting curve; Optimizing the initial weight parameters according to the fitting curve and the actual values ​​of the training sample data to obtain optimized weight parameters; Inputting the test sample data set into a first envelope analysis model weighted by the optimized weight parameter to obtain a verification output value; Performing normal distribution fitting on the verification output value and the test sample data set to obtain an error parameter; According to the optimized weight parameter and the error parameter, a first weight parameter of each input index of the first envelope analysis model is obtained.

[0043] In some preferred embodiments, stratified sampling is used to process the supplier data set, and the data is grouped according to the output value range of 0 to 1000 as the first layer, 1000 to 5000 as the second layer, and 5000 or more as the third layer. The data with a ratio of 7:3 is extracted from each layer and divided into a training set and a test set. The output index, quality index and delivery index in the training set are normalized, and the processing result is divided by the standard deviation after subtracting the mean. The Pearson coefficient is used to calculate the correlation value between the three indicators, and the covariance matrix is ​​constructed based on the correlation value. The initial weight parameter is obtained from the eigenvalue of the covariance matrix. A linear programming function is constructed for the initial weight parameter, and the upper limit of the parameter is set to 1.5 times the initial weight parameter, and the lower limit is set to 0.5 times the initial weight parameter. The least squares method is used to fit the values ​​in the parameter interval. When the difference between the fitting curve and the actual value is less than 0.1, the parameter value is recorded, and the average of all parameter values ​​is obtained to obtain the intermediate weight parameter. The change rate is calculated from the intermediate weight parameter and the initial weight parameter. When the change rate is less than 0.05, the optimized weight parameter is obtained. Substitute the optimized weight parameters into the test set data for verification calculation, use the moving average to smooth the calculated result value, calculate the residual between the processed value and the actual value of the test set, fit the residual value to the normal distribution, use the median of the fitting curve as the error mean, and the standard deviation of the curve as the error fluctuation range, and obtain the verification parameters from the error fluctuation range. Perform secondary optimization on the verification parameters, set the optimization goal to minimize the error mean, and the constraint condition is that the error fluctuation range is less than 0.2. Use cubic spline interpolation to make the verification parameters continuous, and set the interpolation node at the inflection point of the error mean. Extract the parameter value corresponding to the minimum error from the interpolation curve as the final optimization parameter, and update the parameter according to the final optimization parameter.

[0044] For example, in the supplier data set, the monthly output of a certain automotive parts supplier is distributed in three levels, with a monthly output of 800 small batch products, 3,500 medium batch products, and 7,200 large batch products. After being divided in a 7:3 ratio, the training set contains 560 small batch data, 2,450 medium batch data, and 5,040 large batch data, and the test set contains 240 small batch data, 1,050 medium batch data, and 2,160 large batch data. The product qualification rates are 98.5%, 97.8%, and 96.2%, respectively, and the on-time delivery rates are 99.2%, 98.5%, and 97.1%, respectively. After normalization, the mean of the output index is 0.82 and the standard deviation is 0.05, the mean of the quality index is 0.97 and the standard deviation is 0.02, and the mean of the delivery index is 0.98 and the standard deviation is 0.01. The Pearson coefficient shows that the correlation between output and quality is 0.85, the correlation between output and delivery is 0.78, and the correlation between quality and delivery is 0.92. The optimization starting parameters are obtained from the covariance matrix eigenvalues ​​0.75, 0.82, and 0.88. The linear programming sets the upper limit of the parameter to 1.2 and the lower limit to 0.4. The least squares fitting obtains the parameter sequence of 0.85, 0.82, 0.79, 0.76, and 0.73, which differ from the actual values ​​of 0.82, 0.80, 0.78, 0.75, and 0.72 by 0.03, 0.02, 0.01, 0.01, and 0.01, respectively. The parameter mean is 0.79, and the rate of change from the starting parameter 0.82 is 0.037, which is less than the 0.05 threshold, and the optimization result parameter is determined to be 0.79. In the test set verification calculation, a 5-point moving average was used to smooth the results, and the residual sequence median of 0.025 was obtained as the error mean, and the standard deviation of 0.015 was obtained as the fluctuation range. During the secondary optimization of the verification parameters, spline interpolation nodes were set at the three inflection points of the error mean of 0.025, 0.022, and 0.018, and the minimum error of 0.018 corresponded to a parameter value of 0.77 as the final optimization parameter. Suppliers show regular differences in different batches of products. The greater the output, the lower the quality and delivery indicators, indicating that capacity expansion brings pressure on quality control. Stratified sampling is used to ensure the consistency of the distribution ratio of each batch level data in the training set and the test set. The parameter optimization process reflects the trend of convergence to a lower error. In the test set verification stage, the error mean and fluctuation range were controlled in a small range, verifying the effectiveness of parameter optimization. The final optimized parameters were applied to actual production. With a monthly output of 7,500 pieces, the supplier adjusted the production rhythm and quality control plan through parameter guidance, which increased the product qualification rate to 97.5% and the on-time delivery rate to 98.2%.

[0045] In the above scheme, optimizing the weight parameters provides a quantifiable management basis for further expanding the production capacity of supplier nodes, which can accurately measure and effectively improve the efficiency of supplier nodes in practical applications.

[0046] Furthermore, preferably, obtaining the initial weight parameters according to the eigenvalues ​​of the covariance matrix includes: Selecting diagonal element values ​​from the covariance matrix as reference values; Performing least square fitting on the reference value to obtain a first initial weight value; Iterating the first initial weight value by using a moving average value to obtain a decaying weight value; Setting a disturbance interval according to the attenuation weight value, and acquiring a random sampling point sequence from the disturbance interval; According to the random sampling point sequence, a quantile interval is obtained; A decision tree is used to establish a division rule for the quantile interval to obtain upper and lower limits of weight adjustment; According to the upper and lower limits of the weight adjustment, a local optimal value is determined from the attenuated weight value to obtain an initial weight parameter.

[0047] It should be noted that the diagonal element values ​​of the covariance matrix reflect the contribution of each indicator to the overall performance.

[0048] It should also be noted that the units of various indicators are different, and the normalization methods are also different. In the embodiment of the present invention, quantitative standards are constructed for the input and output indicators of each stage. The input quantity of the output indicator adopts the output value measurement unit, and the output quantity adopts the physical quantity measurement unit. The input quantity of the quality indicator adopts the quantity of raw materials, and the output quantity adopts the quantity of qualified products. The input quantity of the equipment utilization index adopts the equipment operating time, and the output quantity adopts the effective operating time. The indicator values ​​are normalized through quantitative standards.

[0049] Exemplarily, the corresponding value of the output index in the diagonal element value of the covariance matrix is ​​0.85, the corresponding value of the quality index is 0.82, and the corresponding value of the equipment utilization index is 0.78. After least squares fitting, the initial adjustment parameter 0.75 is obtained. The calculation window of the moving average takes 5 values, and the optimization starting weight value 0.75 is iterated, multiplied by the attenuation factor of 0.85 to obtain 0.64, and then standardized to obtain 0.62. The standardized sequence 0.75, 0.62, 0.53, 0.45, and 0.38 are obtained by continuous calculation. The differences between adjacent values ​​are 0.13, 0.09, 0.08, and 0.07. The difference gradually decreases and the last two differences are less than 0.1, indicating that the weight value tends to be stable. Based on the weight value of 0.38, a disturbance interval of ±0.2 is set, ranging from 0.18 to 0.58. 100 random points are generated in this interval, and the input-output ratio of each point is calculated. The 0.25 quantile in the ratio series is 1.25, and the 0.75 quantile is 1.75. The decision tree is used to divide the interval into 1.35 and 1.65 points as the upper and lower limits of the weight adjustment. The local optimal value is calculated from the upper and lower limits of the weight, and a 5-point moving average is used to generate a continuous curve. The difference between adjacent points on the curve gradually decreases from 0.08 to 0.02. When the difference drops to 0.01, the corresponding weight value is 0.42. At this time, the output value of the output indicator in the first stage production operation indicator is 1 million yuan, the output output is 1.25 million pieces, and the input-output ratio is 1.25. The operating income input in the second stage financial benefit indicator is 1.25 million yuan, the output profit is 1.5 million yuan, and the input-output ratio is 1.2. The raw material utilization rate input in the third stage resource utilization indicator is 100 tons of raw materials, and the output of finished products is 75 tons, and the input-output ratio is 0.75.

[0050] As a preferred implementation, step S3, inputting the operation data of the supplier node to be evaluated into the second envelope analysis model, obtaining the first efficiency value of the supplier node in each time period, and the change of the first efficiency value, includes: Collecting the operation data of the supplier node to be evaluated at different time points; the operation data includes input data and output data; In the second envelope analysis model, a correlation matrix of the input data is calculated, and cluster analysis is performed on the correlation matrix to obtain matching parameters of the input data; Performing exponential smoothing processing on the output data to obtain an output data sequence; Calculate the difference between each point of the output data sequence to obtain a monthly node sequence; According to the matching parameters and the output data of each sequence point on the monthly node sequence, a first efficiency value of each time period is obtained; Performing least square fitting on the first efficiency value and segmenting according to the fitting result to obtain a slope sequence; According to the fluctuation range of the slope sequence, a key interval is obtained, and a first change point of the proportioning parameter and a second change point of the output data are extracted from the key interval; According to the first change point and the second change point, the change of the first efficiency value of the supplier node in each time period is obtained.

[0051] It should be noted that through efficiency value tracking, it is found that suppliers have experienced two stages of efficiency improvement and stabilization after equipment transformation. The ratio change point reflects the adjustment of the resource input structure of the supplier node, and the composition change point shows that the output structure has changed accordingly. This change pattern provides a reference for the subsequent resource allocation of the supplier node, making the management of the supply chain more accurate and effective.

[0052] In some embodiments, the equipment investment amount, raw material cost and labor cost in the monthly input factors of the supplier are normalized, the normalized value is subtracted from the mean and divided by the standard deviation, the Pearson coefficient is used to calculate the correlation between the three factors, a 3x3 matrix is ​​constructed for the correlation value, the matrix is ​​grouped by a clustering algorithm, the central value of each group is extracted from the grouping result, and the input factor ratio is quantitatively assigned according to the size of the central value to obtain the initial ratio parameter. The product output, sales revenue and asset return values ​​are obtained from the monthly output factors of the supplier, the three factors are multiplied to obtain the comprehensive output value, the comprehensive output value is processed by exponential smoothing, the smoothing parameter is set to 0.3, the difference before and after the smoothed numerical sequence is calculated, and the absolute value of the difference value greater than 0.2 is recorded as the inflection point position, and the monthly time node sequence number is obtained from the inflection point position. The efficiency value is obtained by calculating the quotient of the matching parameter and the comprehensive output value on the monthly time node sequence, and the least squares method is used to fit the efficiency value. The fitting function contains quadratic terms and linear terms. The slope value is extracted from the fitting curve. A positive slope is recorded as an upward trend mark 1, and a negative slope is recorded as a downward trend mark -1. The trend marks are accumulated to form an efficiency change curve. The efficiency change curve is fitted piecewise linearly, and the fitting interval length is set to 3 months. The fluctuation range is obtained from the slope sequence of the fitting line. The interval with a difference between the upper and lower limits of the fluctuation range greater than 0.5 is recorded as a key interval. The data of the months at both ends of the key interval are extracted. The input factor matching difference rate of the corresponding month exceeds 0.3 and is recorded as a matching change point. The output factor composition difference rate exceeds 0.3 and is recorded as a composition change point.

[0053] For example, among the monthly input factors of suppliers, equipment investment is 5 million yuan for production line transformation, 3 million yuan for automation equipment, and 2 million yuan for testing equipment, which are 0.85, 0.65, and 0.42 respectively after normalization. Raw material costs include 2.8 million yuan for metal materials, 1.5 million yuan for plastic materials, and 0.700,000 yuan for packaging materials, with normalized values ​​of 0.78, 0.55, and 0.32. Labor costs include 1.8 million yuan for production personnel, 1.2 million yuan for technicians, and 1 million yuan for management personnel, with normalized values ​​of 0.72, 0.58, and 0.45. The Pearson coefficient shows that the correlation between equipment and raw materials is 0.82, the correlation between equipment and labor is 0.75, and the correlation between raw materials and labor is 0.68. Cluster analysis divides input factors into three groups. The first group with a central value of 0.8 includes equipment investment, the second group with a central value of 0.6 includes raw material costs, and the third group with a central value of 0.4 includes labor costs. Among the output factors, the output of products reached 85,000 pieces, the sales revenue reached 12 million yuan, and the return on assets reached 15%. After exponential smoothing, the product of the three factors showed a difference of 0.25 in April, 0.22 in July, and 0.28 in October, which was determined to be the turning point month. At the April time point, the input ratio parameter was 0.6, the comprehensive output value was 0.75, and the efficiency value was calculated to be 1.25. The ratio parameter in July was 0.55, the output value was 0.82, and the efficiency value was 1.49. The ratio parameter in October was 0.52, the output value was 0.78, and the efficiency value was 1.50. The least squares fitting shows that the slope of 0.08 from April to July is recorded as an upward trend, and the slope of -0.01 from July to October is recorded as a stable trend. The cumulative value of the trend mark rises from 1 to 2 and then decreases to 1. The slope of the efficiency curve from January to March is 0.05, from April to June is 0.08, from July to September is 0.02, and from October to December is -0.01. The fluctuation range is from 0.02 to 0.08, and the interval from April to June corresponding to the difference of 0.06 is determined as the key interval. In April, the proportion of equipment in the input factors increased from 0.45 to 0.52, the proportion of raw materials decreased from 0.35 to 0.30, and the proportion of labor decreased from 0.20 to 0.18. The ratio difference rate was 0.35, which was recorded as the ratio change point. The proportion of output in the output factors increased from 0.40 to 0.45, the proportion of income increased from 0.35 to 0.38, and the proportion of yield decreased from 0.25 to 0.17. The composition difference rate was 0.32, which was recorded as the composition change point.

[0054] As a preferred implementation, step S4, according to the change of the first efficiency value, obtaining the key driving factors that cause the supplier node efficiency to decrease from the operation data, includes: According to the change of the first efficiency value, the operation data in the decreasing interval is obtained by screening from the operation data; the absolute value of the difference between the efficiency value of the first time period corresponding to the operation data in the decreasing interval and the efficiency values ​​of the remaining second time periods is greater than a preset efficiency value difference threshold; Calculating the correlation coefficient of the input data of the declining interval operation data, and grouping the input data of the declining interval operation data according to the correlation coefficient to obtain a plurality of declining interval input data groups; Performing principal component analysis on the output data of the declining interval operation data to obtain a feature vector; Calculating a month-on-month change rate according to the characteristic vector, and obtaining a significant change index from the output data of the operating data in the declining interval according to the month-on-month change rate and a preset month-on-month change rate threshold; Performing correlation analysis on the decline interval input data group and the significant change index to obtain a correlation strength value; According to the association strength value, key driving factors are obtained; the key driving factors include key input data and key output data that lead to a decrease in supplier node efficiency.

[0055] In some embodiments, the monthly order delivery time deviation value, product quality score value and response speed delay value of the supplier are normalized, the normalized value is subtracted from the mean and divided by the standard deviation, the Pearson coefficient is used to calculate the correlation between the three indicators, the weight of the indicator with a correlation greater than 0.8 is set to 0.5, the weight of the indicator with a correlation between 0.5 and 0.8 is set to 0.3, and the weight of the indicator with a correlation less than 0.5 is set to 0.2, and the performance evaluation matrix is ​​constructed by weighting and summing the normalized values ​​according to the indicator weights. The evaluation value sequence is extracted from the performance evaluation matrix, combined with the production capacity indicators composed of the number of production lines, the number of equipment, and the number of personnel of the supplier, the process indicators composed of R&D investment, the number of patents, and process parameters, and the operational indicators composed of quality control, on-site management, and inventory management, the indicators are divided into three groups using cluster analysis, the distance between groups and the distance within groups are calculated, the cluster center is obtained from the distance ratio, and the cluster center is sorted to obtain the comprehensive attribute value of the supplier. The supplier comprehensive attribute values ​​are arranged in monthly time sequence, and exponential smoothing is used to process the numerical sequence. The smoothing coefficient is set to 0.3. The difference between the smoothed value and the original value is calculated. If the difference exceeds 0.2, it is recorded as a fluctuation point. The slope of the value before and after the fluctuation point is extracted. A positive slope is recorded as an upward trend, a negative slope is recorded as a downward trend, and a slope with an absolute value less than 0.1 is recorded as a stable trend. The trend mark is converted into a numerical sequence. A hierarchical structure is constructed for the trend numerical sequence and the supplier comprehensive attribute value. The first layer is the trend category, the second layer is the performance indicator group, and the third layer is the supplier attribute group. The entropy method is used to calculate the weight of each layer of indicators. The weight value is substituted into the discriminant function. The indicator combination with the largest discriminant function result is marked as the key driving factor. The driving factors are accumulated monthly to obtain the driving strength value.

[0056] For example, the monthly order data of auto parts suppliers show that the average deviation of delivery time is 3 days, the quality score is 85 points, and the response speed is delayed by 4 hours, which are 0.82, 0.75, and 0.68 respectively after normalization. The Pearson coefficient calculation shows that the correlation between delivery and quality is 0.85, the correlation between delivery and response is 0.72, and the correlation between quality and response is 0.65, with corresponding weights of 0.5, 0.3, and 0.3. The performance evaluation value after weighted summation is 0.78. Among the supplier's capacity indicators, there are 4 production lines, 35 CNC equipment, and 120 operators. The process indicators include R&D investment of 6 million yuan, 8 authorized patents, and 15 key process parameters. The operation indicators cover 12 quality inspection points, 85 on-site management specifications, and an inventory turnover rate of 4.5 times. Cluster analysis divides the indicators into capacity group, process group, and operation group. The distances between groups are 0.45, 0.38, and 0.32, respectively, and the distances within the groups are 0.15, 0.12, and 0.10. The distance ratios correspond to cluster centers of 0.85, 0.75, and 0.68. According to the time series from January to December, the comprehensive attribute value of suppliers increased from 0.75 to 0.82, with fluctuations in the middle. After exponential smoothing, the difference in April was 0.25, the difference in July was 0.22, and the difference in October was 0.28, corresponding to the position of the fluctuation point. The slope changed from 0.05 to 0.12 before and after April, which was recorded as an upward trend, the slope changed from 0.12 to -0.08 before and after July, which was recorded as a downward trend, and the slope changed from -0.08 to 0.03 before and after October, which was recorded as a stable trend. In the hierarchical structure, the trend category layer includes three categories: rising, falling, and stable; the performance indicator group includes three groups: delivery, quality, and response; and the supplier attribute group includes three groups: capacity, process, and operation. The entropy method calculates the trend layer weight of 0.4, the indicator layer weight of 0.35, and the attribute layer weight of 0.25. The discriminant function shows that the combination of rising trend, quality indicator, and process attribute has the highest score, which is marked as a key driver. The driving strength value continued to rise from 0.45 in January to 0.78 in June, and then fluctuated and dropped to 0.62 in December. Suppliers drive quality improvement through process improvement in annual operations, and quality improvement promotes delivery accuracy, ultimately achieving overall efficiency improvement. This improvement chain is reflected in the fluctuations and trend changes of various indicators, especially the turning points at three key time points in April, July, and October.

[0057] In the above scheme, through quantitative tracking of the improvement process, it is found that process innovation is the core driving force for overall improvement, which provides direction for the subsequent development of suppliers. Annual data shows that suppliers invested resources in process improvement in the first half of the year, which brought significant improvements in quality and efficiency. In the second half of the year, they entered the overall optimization stage, and various indicators tended to stabilize. This change pattern provides an important reference for supplier performance evaluation and prediction.

[0058] Further, preferably, the performing of correlation analysis on the decline interval input data group and the significant change index to obtain a correlation strength value includes: Difference calculations are performed on the input data group in the declining interval and the significant change index respectively with the preset benchmark data to obtain an input fluctuation amplitude sequence and an output fluctuation amplitude sequence; Calculating the correlation between the input fluctuation amplitude sequence and the output fluctuation amplitude sequence to obtain a correlation curve; According to a preset correlation threshold, the correlation curve is divided into a number of intervals; According to the plurality of intervals, interval features are obtained; each of the intervals corresponds to an interval feature; According to the interval characteristics, the falling interval input data group and the significant change index are grouped into a strong correlation group and a non-strong correlation group; Establishing a decision tree for the strongly associated group, splitting the nodes of the decision tree according to a preset information gain criterion, and extracting numerical distribution rules from the nodes; According to the numerical distribution law, the correlation coefficient is calculated to obtain the association strength value.

[0059] In an embodiment of the present invention, by analyzing the correlation between each key driving factor and the efficiency decline, the correlation is sorted, and the key driving factors within the target correlation range are extracted for key analysis. Based on causal inference, it is determined whether the key driving factors within the target correlation range are the main reasons for the efficiency decline. By combining the key driving factors judged as the main reasons for the efficiency decline, the supplier efficiency can be effectively evaluated, and the supply chain management can be further automatically and accurately performed.

[0060] In some embodiments, the standard deviation of the supplier's monthly efficiency value is calculated, and the calculated result is subtracted from the mean and divided by the standard deviation to obtain the standardized efficiency value. The standardized efficiency value is smoothed by a 5-point moving average. If the smoothed value is less than -0.5, it is recorded as a low efficiency interval. The least squares method is used to fit the low efficiency interval curve, and the month number of the trough position is obtained from the fitting curve. The time period corresponding to the month number is marked as the interval to be analyzed. The input data of the corresponding period is extracted according to the interval to be analyzed, and the total equipment investment, total raw material purchases, and total labor costs are calculated from the input data. The three totals are normalized, and the normalized value is subtracted from the minimum value and divided by the range. The Pearson coefficient is used to calculate the correlation of the input data. Data items with correlation values ​​greater than 0.8 are merged into the same group, and data items with correlation values ​​between 0.5 and 0.8 are recorded as independent groups. The output data of the analyzed interval is reduced in dimension through principal component analysis, and the feature vectors of three dimensions, namely, product output, sales revenue and inventory turnover, are extracted. The ratio of the feature vector to the value of the previous month is calculated to obtain the month-on-month change rate. The fluctuation threshold of 0.2 is set for the month-on-month change rate. The feature vectors exceeding the threshold are marked as significant change indicators. The weight coefficient is set according to the significant change indicator, and the product of the weight coefficient and the change rate is used as the output indicator score. The correlation analysis is used to calculate the correlation strength between the input data combination and the output indicator score. The correlation strength values ​​are sorted from large to small, and the combinations in the top 20% are marked as key driving combinations. The weighted average of the input and output items in the key driving combination is calculated to obtain the impact strength value, and the ratio of the impact strength value to the efficiency decline is used as the evaluation result.

[0061] For example, the monthly efficiency values ​​of auto parts suppliers throughout the year showed a fluctuating trend, with a mean of 0.82 and a standard deviation of 0.15. After standardization, the efficiency value sequence from January to December was 0.85, 0.83, 0.78, 0.65, 0.72, 0.75, 0.68, 0.62, 0.71, 0.76, 0.81, and 0.79. The 5-point moving average calculation shows that the value of 0.62 in August is lower than the -0.5 threshold, and the least squares fitting locates July to September as the low efficiency interval. In the low efficiency interval, the input data shows that the equipment investment is 5.8 million yuan, the raw material purchase amount is 4.2 million yuan, and the labor cost is 2.8 million yuan. After normalization, they are 0.85, 0.72, and 0.65 respectively. The Pearson correlation coefficient calculation shows that the correlation between equipment investment and raw material purchase is 0.85, the correlation between equipment investment and labor cost is 0.75, and the correlation between raw material purchase and labor cost is 0.82. Equipment investment and raw material procurement are combined into the resource input group, and labor costs are separately treated as the labor group. The eigenvectors extracted by principal component analysis show that the monthly output of the product is 85,000 pieces, the sales revenue is 12 million yuan, and the inventory turnover rate is 4.2 times. Compared with the data in June, the output decreased by 12%, the revenue decreased by 15%, and the turnover rate decreased by 25% month-on-month, all exceeding the fluctuation threshold of 0.2. The output weight is set to 0.4, the revenue weight is 0.35, and the turnover weight is 0.25, and the comprehensive score of the output index is calculated to be 0.72. Correlation analysis shows that the correlation strength between the resource input group and the output score is 0.82, and the correlation strength between the human resource group and the output score is 0.65. The resource input group is marked as a key driver combination, among which the equipment investment impact strength is 0.45 and the raw material procurement impact strength is 0.35. The efficiency value in August decreased by 0.2 compared with the annual average, and the ratio of impact strength to decline was 0.8. According to the supplier's annual operating data, the production line was renovated from July to September, and the equipment was shut down, resulting in a decrease in production capacity. The backlog of raw material inventory caused a decrease in turnover, but the labor cost expenditure was relatively stable. After the production line was renovated, the efficiency of the equipment capacity ramp-up stage failed to recover quickly, which was reflected in the significant fluctuations of various indicators.

[0062] In the above scheme, through quantitative tracking of various indicators, it is found that the synergy between equipment investment and raw material management is the key factor affecting efficiency. From the quantitative evaluation results, the short-term efficiency decline caused by equipment transformation is within the expected range, and the matching degree between the impact intensity and the decline is high, indicating that the efficiency fluctuation is mainly due to the planned capacity adjustment rather than the decline in management level. This precise judgment based on data provides an objective basis for supplier performance evaluation.

[0063] As a preferred implementation, step S6, optimizing the parameters of the second envelope analysis model according to the difference between the first efficiency value and the second efficiency value to obtain an optimized envelope analysis model to manage supply chain network nodes, includes: Calculating the difference between the first efficiency value and the second efficiency value to obtain a fitting curve of efficiency change within the preset time; According to the fluctuation of the fitting curve, abnormal time points are obtained by screening the operation data at all times; Performing similarity calculation on the parameters of the second envelope analysis model corresponding to the abnormal time point to obtain parameter similarity; According to the parameter similarity and a preset parameter similarity threshold, adjusting the parameter within an iteration interval to obtain an adjusted parameter; The adjustment parameters are used to optimize the parameters of the second envelope analysis model to obtain an optimized envelope analysis model to manage supply chain network nodes.

[0064] Furthermore, preferably, the embodiment of the present invention also provides a preset management decision library, which includes a number of management decisions, each of which corresponds to management measures for different types of suppliers under different efficiency value changes. Specifically, the suppliers are classified and sorted by their efficiency values, the types of the suppliers are identified, and the management decisions guiding the suppliers are matched in the preset management decision library according to the types of the suppliers.

[0065] In some embodiments, the ratio of the monthly evaluation result of the supplier to the value of the previous month is calculated to obtain the month-on-month change rate, and the least squares method is used to fit the change rate sequence. The upper and lower floating range of the fitting curve is set to ±0.2. The point beyond the range is recorded as an abnormal point, and the month mark is obtained from the abnormal point. The parameter values ​​of each stage of the three-stage DEA are extracted through the month mark, and the parameter values ​​are similarly calculated to obtain the parameter improvement list. The correlation analysis is used to calculate the correlation between each parameter and the evaluation result in the parameter improvement list. The correlation value is sorted from large to small, and the parameters in the top 30% position are taken as the optimization object. The iteration interval is set for the optimization object. The upper limit of the iteration interval is 1.2 times the original parameter value, and the lower limit is 0.8 times the original parameter value. Exponential smoothing is used to adjust the parameter value within the iteration interval. The supplier evaluation results are calculated according to the adjusted parameters, and the evaluation results are grouped by cluster analysis. The median of each group is extracted as the evaluation benchmark value, and the execution completion index value is calculated from the evaluation benchmark value. The execution completion index includes the plan achievement rate, the quality compliance rate, and the response satisfaction rate. The weighted sum is used to obtain the comprehensive execution score. A warning threshold of 0.8 is set for the comprehensive execution score, and double exponential smoothing is used to predict the trend of the score sequence. The predicted curve below the threshold is recorded as a risk interval. Feature vectors are extracted from the risk interval, and the frequency of the feature vectors is counted. The features with the top 50% frequency values ​​are taken as key factors, and the entries are matched and updated in the management decision library based on the key factors.

[0066] For example, the evaluation results of auto parts suppliers for 6 consecutive months show that the monthly evaluation values ​​are 0.82, 0.85, 0.79, 0.68, 0.72, and 0.75, respectively, and the calculated month-on-month change rates are 3.7%, -7.1%, -13.9%, 5.9%, and 4.2%. The least squares fitting curve shows that the change rate of -13.9% in April exceeds the floating range of ±0.2. The three-stage DEA parameters for April are extracted, with the input-output parameter of the first stage being 0.75, the efficiency decomposition parameter of the second stage being 0.68, and the environmental adjustment parameter of the third stage being 0.82. Correlation analysis shows that the correlation between the three-stage parameters and the evaluation results is 0.85, 0.92, and 0.78, respectively. The efficiency decomposition parameter of the second stage is used as the optimization object, and the iteration interval of the original parameter value of 0.68 is set to 0.544 to 0.816. After exponential smoothing, the parameter values ​​show an upward trend at 0.72, 0.75, and 0.78, indicating that the parameter optimization direction is correct. The adjusted evaluation results were divided into three groups through cluster analysis, with medians of 0.85, 0.75, and 0.65, respectively. The plan achievement rate increased from 85% to 92%, the quality compliance rate increased from 88% to 94%, and the response satisfaction rate increased from 90% to 95%. The weights of the three indicators were set to 0.4, 0.35, and 0.25, and the weighted calculation obtained a comprehensive execution score of 0.93. The score sequence was predicted by double exponential smoothing and showed that the score may drop to 0.78 next month, which is lower than the warning threshold of 0.8. The feature vector extraction showed that the frequencies of the three factors of rising raw material prices, declining equipment utilization, and increased quality fluctuations were 0.85, 0.78, and 0.72, respectively. From the perspective of specific applications, suppliers experienced a sharp decline in efficiency in April. Through parameter optimization, it was found that there were problems in the efficiency decomposition link. The adjusted efficiency decomposition more accurately reflects the actual operating conditions of suppliers. The improvement in execution scores verifies the correctness of parameter optimization.

[0067] In the above scheme, dynamic parameter optimization is reflected in three aspects: first, accurate positioning of optimization opportunities through anomaly identification, second, determination of optimization objects through correlation analysis, and third, smooth transition through exponential smoothing. This multi-dimensional parameter adjustment mechanism ensures that the model always maintains a high evaluation accuracy. At the same time, the execution effect evaluation and decision update form a closed loop to continuously improve the supplier management level. The monthly supplier tracking data shows that through parameter optimization and decision updates, 90% of problem suppliers have achieved efficiency improvements within 3 months, demonstrating the feasibility of this method in practical applications.

[0068] An evaluation method for improving supplier efficiency provided by an embodiment of the present invention improves the parameters of the envelope analysis model in a multi-level and multi-dimensional manner, thereby automatically collecting and processing data from supply chain network nodes, identifying key driving factors that lead to efficiency changes, and realizing automatic and effective supply chain network management.

[0069] The embodiment of the present invention provides an evaluation system for improving supplier efficiency. Figure 2 The evaluation system for improving supplier efficiency includes a model building module 11, a model training module 12, a first efficiency value calculation module 13, a key driving factor acquisition module 14, a second efficiency value calculation module 15 and a supply chain network management module 16, wherein: The model building module 11 is used to build a first envelope analysis model; the input of the first envelope analysis model includes resource input data, technology input data and environmental factor data of each node in the supply chain network; A model training module 12 is used to obtain a sample data set, and use the sample data set to train the first envelope analysis model to obtain a second envelope analysis model; the input-output ratio of the second envelope analysis model is not less than a preset target ratio threshold; A first efficiency value calculation module 13, used to input the operation data of the supplier node to be evaluated into the second envelope analysis model, to obtain the first efficiency value of the supplier node in each time period, and the change of the first efficiency value; A key driving factor acquisition module 14 is used to obtain key driving factors that cause supplier node efficiency to decrease from the operation data according to the change of the first efficiency value; A second efficiency value calculation module 15, used for collecting second operation data of the supplier node after a preset time, inputting the second operation data and the key driving factors into the second envelope analysis model, and obtaining a second efficiency value of the supplier node; The supply chain network management module 16 is used to optimize the parameters of the second envelope analysis model according to the difference between the first efficiency value and the second efficiency value to obtain an optimized envelope analysis model to manage the supply chain network nodes.

[0070] As a preferred implementation, the model building module 11 includes: A normalization unit is used to normalize the input resource input data, technology input data and environmental factor data of each node in the supply chain network to obtain normalized data; A first output efficiency value calculation unit, used to calculate the correlation value of the normalized data by using a Pearson algorithm, and obtain a first output efficiency value in an efficiency evaluation stage according to the correlation value; An evaluation grade classification unit, used for clustering the first output efficiency values ​​to obtain an evaluation grade; A second output efficiency value calculation unit, used for establishing a first function expression of the evaluation level by a linear regression method, and weighting the parameters of the first function expression to obtain a second output efficiency value of the efficiency decomposition stage; an actual deviation value calculation unit, configured to calculate an actual deviation value according to the second output efficiency value and an actual production efficiency value of the supplier node; a third output efficiency value calculation unit, configured to construct a standard normal distribution curve according to the actual deviation value, and perform standard deviation calculation according to the standard normal distribution curve to obtain a third output efficiency value; A quadratic term fitting unit, used for performing a quadratic term fitting on the third output efficiency value by using a least square method to obtain a fitting curve; The model output establishment unit is used to obtain the difference sequence nodes from the fitting curve, determine the environmental adjustment value according to the coordinates of the difference sequence nodes at the inflection point position, and use the environmental adjustment value to correct the third output efficiency value to establish the output of the first envelope analysis model.

[0071] As a preferred implementation, the model training module 12 includes: An initial sample data set acquisition unit, used to acquire an initial sample data set from a supplier node; A data expansion unit, used to expand the initial sample data set by using a sample enhancement algorithm to generate a number of expanded data having a similarity with the initial sample data set greater than a preset similarity threshold; A sample data set generating unit, configured to obtain a sample data set according to the initial sample data set and the expanded data; A first weight parameter calculation unit, configured to train the first envelope analysis model using the sample data set to obtain first weight parameters of each input index of the first envelope analysis model; The second envelope analysis model generating unit is used to use the first weight parameter to weight the input index of the first envelope analysis model to obtain a second envelope analysis model.

[0072] Further, preferably, the data expansion unit is specifically used for: Normalizing the initial sample data set on a monthly basis to obtain monthly normalized data, and calculating the Pearson correlation coefficient based on the monthly normalized data; The Pearson correlation coefficient is used to determine the sampling weight, and the monthly normalized data is calculated according to the sampling weight to obtain the first-stage data; The moving average method is used to obtain the numerical central trend of the data of the first stage, and the probability distribution fitting is performed on the numerical central trend to obtain the data of the second stage; A monthly time series curve is constructed based on the data of the second stage, a prediction function is established for the production achievement rate, the qualified product rate and the on-time delivery rate in the monthly time series curve using the least square method, and the data of the third stage is obtained based on the prediction function; Constructing a timing evaluation curve according to the third stage data, calculating the mean variance of the timing evaluation curve, segmenting the timing evaluation curve, and obtaining the slope of the segmented curve; The initial sample data set is expanded according to the segmented curve slope and a preset similarity threshold to generate expanded data.

[0073] Preferably, the first weight parameter calculation unit is specifically used to: Performing stratified sampling processing on the sample data set according to the output values ​​of the supplier nodes corresponding to the sample data set to obtain a training sample data set and a test sample data set; Obtaining training sample input data of the first envelope analysis model from the training sample data set, calculating the correlation between the training sample input data, and obtaining a covariance matrix; Obtaining initial weight parameters according to the eigenvalues ​​of the covariance matrix; Constructing a linear programming function according to the initial weight parameters to obtain a fitting curve; Optimizing the initial weight parameters according to the fitting curve and the actual values ​​of the training sample data to obtain optimized weight parameters; Inputting the test sample data set into a first envelope analysis model weighted by the optimized weight parameter to obtain a verification output value; Performing normal distribution fitting on the verification output value and the test sample data set to obtain an error parameter; According to the optimized weight parameter and the error parameter, a first weight parameter of each input index of the first envelope analysis model is obtained.

[0074] As a preferred implementation, the first efficiency value calculation module 13 includes: An operation data collection unit, used to collect operation data of the supplier node to be evaluated at different time points; the operation data includes input data and output data; a matching parameter calculation unit, configured to calculate a correlation matrix of the input data in the second envelope analysis model, perform cluster analysis on the correlation matrix, and obtain matching parameters of the input data; An output data sequence generating unit, used for performing exponential smoothing processing on the output data to obtain an output data sequence; A difference calculation unit, used to calculate the difference between each point of the output data sequence to obtain a monthly node sequence; A first efficiency value calculation unit, used for obtaining a first efficiency value of each time period according to the matching parameters and the output data of each sequence point on the monthly node sequence; A slope sequence division unit, used for performing least square fitting on the first efficiency value and dividing it into sections according to the fitting result to obtain a slope sequence; A change point extraction unit, used to obtain a key interval according to the fluctuation range of the slope sequence, and extract a first change point of the ratio parameter and a second change point of the output data from the key interval; The first efficiency value change monitoring unit is used to obtain the change of the first efficiency value of the supplier node in each time period according to the first change point and the second change point.

[0075] As a preferred implementation, the key driving factor acquisition module 14 includes: a decreasing interval operation data screening unit, configured to screen the decreasing interval operation data from the operation data according to the change of the first efficiency value; the absolute value of the difference between the efficiency value of the first time period corresponding to the decreasing interval operation data and the efficiency values ​​of the remaining second time periods is greater than a preset efficiency value difference threshold; A falling interval input data group division unit is used to calculate the correlation coefficient of the input data of the falling interval operation data, and group the input data of the falling interval operation data according to the correlation coefficient to obtain a plurality of falling interval input data groups; A principal component analysis unit, used for performing principal component analysis on the output data of the declining interval operation data to obtain a feature vector; A significant change index calculation unit, used to calculate the month-on-month change rate according to the characteristic vector, and obtain a significant change index from the output data of the operating data in the decline interval according to the month-on-month change rate and a preset month-on-month change rate threshold; A correlation analysis unit, used for performing correlation analysis on the decline interval input data group and the significant change index to obtain a correlation strength value; The key driving factor acquisition unit is used to obtain the key driving factors according to the association strength value; the key driving factors include key input data and key output data that cause the supplier node efficiency to decrease.

[0076] Further, preferably, the correlation analysis unit is specifically used for: Perform difference calculations on the input data group in the declining interval and the significant change index respectively with the preset benchmark data to obtain an input fluctuation amplitude sequence and an output fluctuation amplitude sequence; Calculating the correlation between the input fluctuation amplitude sequence and the output fluctuation amplitude sequence to obtain a correlation curve; According to a preset correlation threshold, the correlation curve is divided into a number of intervals; According to the plurality of intervals, interval features are obtained; each of the intervals corresponds to an interval feature; According to the interval characteristics, the falling interval input data group and the significant change index are grouped into a strong correlation group and a non-strong correlation group; Establishing a decision tree for the strongly associated group, splitting the nodes of the decision tree according to a preset information gain criterion, and extracting numerical distribution rules from the nodes; According to the numerical distribution law, the correlation coefficient is calculated to obtain the association strength value.

[0077] As a preferred implementation, the supply chain network management module 16 includes: an efficiency change fitting curve generating unit, configured to calculate a difference between the first efficiency value and the second efficiency value, and obtain the efficiency change fitting curve within the preset time; An abnormal time point screening unit, used for screening abnormal time points from the operation data at all times according to the fluctuation of the fitting curve; A similarity calculation unit, used for performing similarity calculation on the parameters of the second envelope analysis model corresponding to the abnormal time point to obtain parameter similarity; An adjustment parameter generating unit, configured to adjust the parameter within an iteration interval according to the parameter similarity and a preset parameter similarity threshold to obtain an adjustment parameter; A node management unit is used to optimize the parameters of the second envelope analysis model using the adjustment parameters to obtain an optimized envelope analysis model to manage supply chain network nodes.

[0078] An evaluation system for improving supplier efficiency provided by an embodiment of the present invention improves the parameters of the envelope analysis model in a multi-level and multi-dimensional manner, thereby automatically collecting and processing data from supply chain network nodes, identifying key driving factors that lead to efficiency changes, and realizing automatic and effective supply chain network management.

[0079] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0080] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A supply chain network optimization method, characterized in that: include: Establishing a first envelope analysis model; the input of the first envelope analysis model includes resource input data, technology input data and environmental factor data of each node in the supply chain network; Acquire a sample data set, and use the sample data set to train the first envelope analysis model to obtain a second envelope analysis model; the input-output ratio of the second envelope analysis model is not less than a preset target ratio threshold; Inputting the operation data of the supplier node to be evaluated into the second envelope analysis model to obtain the first efficiency value of the supplier node in each time period and the change of the first efficiency value; According to the change of the first efficiency value, a key driving factor causing the supplier node efficiency to decrease is obtained from the operation data; Collecting second operating data of the supplier node after a preset time, inputting the second operating data and the key driving factors into the second envelope analysis model, and obtaining a second efficiency value of the supplier node; According to the difference between the first efficiency value and the second efficiency value, the parameters of the second envelope analysis model are optimized to obtain an optimized envelope analysis model to manage supply chain network nodes.

2. A supply chain network optimization method as claimed in claim 1, characterized in that: The establishing of the first envelope analysis model comprises: Normalizing the input resource input data, technology input data and environmental factor data of each node in the supply chain network to obtain normalized data; Calculate the correlation value of the normalized data by using the Pearson algorithm, and obtain the first output efficiency value of the efficiency evaluation stage according to the correlation value; Clustering the first output efficiency values ​​to obtain an evaluation grade; A first function expression of the evaluation level is established by using a linear regression method, and parameters of the first function expression are weighted to obtain a second output efficiency value of the efficiency decomposition stage; Calculate an actual deviation value according to the second output efficiency value and an actual production efficiency value of the supplier node; According to the actual deviation value, a standard normal distribution curve is constructed, and a standard deviation is calculated according to the standard normal distribution curve to obtain a third output efficiency value; Using the least square method to perform quadratic fitting on the third output efficiency value to obtain a fitting curve; A difference sequence node is obtained from the fitting curve, an environmental adjustment value is determined according to the coordinates of the difference sequence node at the inflection point position, and the third output efficiency value is corrected using the environmental adjustment value to establish the output of the first envelope analysis model.

3. A supply chain network optimization method as claimed in claim 1, characterized in that: The acquiring of the sample data set and using the sample data set to train the first envelope analysis model to obtain the second envelope analysis model includes: Get an initial sample dataset from the supplier node; The initial sample data set is expanded using a sample enhancement algorithm to generate a number of expanded data having a similarity with the initial sample data set greater than a preset similarity threshold; Obtaining a sample data set according to the initial sample data set and the expanded data; Using the sample data set to train the first envelope analysis model to obtain first weight parameters of each input indicator of the first envelope analysis model; The first weight parameter is used to weight the input index of the first envelope analysis model to obtain a second envelope analysis model.

4. A supply chain network optimization method as claimed in claim 3, characterized in that: The method of using a sample enhancement algorithm to expand the initial sample data set to generate a number of expanded data having a similarity with the initial sample data set greater than a preset similarity threshold includes: Normalizing the initial sample data set in monthly units to obtain monthly normalized data, and calculating the Pearson correlation coefficient based on the monthly normalized data; The Pearson correlation coefficient is used to determine the sampling weight, and the monthly normalized data is calculated according to the sampling weight to obtain the first-stage data; The moving average method is used to obtain the numerical central trend of the data of the first stage, and the probability distribution fitting is performed on the numerical central trend to obtain the data of the second stage; A monthly time series curve is constructed based on the data of the second phase, a prediction function is established for the production achievement rate, qualified product rate and on-time delivery rate in the monthly time series curve using the least square method, and the data of the third phase is obtained based on the prediction function; Constructing a timing evaluation curve according to the third stage data, calculating the mean variance of the timing evaluation curve, segmenting the timing evaluation curve, and obtaining the slope of the segmented curve; The initial sample data set is expanded according to the segmented curve slope and a preset similarity threshold to generate expanded data.

5. A supply chain network optimization method as claimed in claim 3, characterized in that: The step of training the first envelope analysis model using the sample data set to obtain first weight parameters of each input indicator of the first envelope analysis model includes: Performing stratified sampling processing on the sample data set according to the output values ​​of the supplier nodes corresponding to the sample data set to obtain a training sample data set and a test sample data set; Obtaining training sample input data of the first envelope analysis model from the training sample data set, calculating the correlation between the training sample input data, and obtaining a covariance matrix; Obtaining initial weight parameters according to the eigenvalues ​​of the covariance matrix; Constructing a linear programming function according to the initial weight parameters to obtain a fitting curve; Optimizing the initial weight parameters according to the fitting curve and the actual values ​​of the training sample data to obtain optimized weight parameters; Inputting the test sample data set into a first envelope analysis model weighted by the optimized weight parameter to obtain a verification output value; Performing normal distribution fitting on the verification output value and the test sample data set to obtain an error parameter; According to the optimized weight parameter and the error parameter, a first weight parameter of each input index of the first envelope analysis model is obtained.

6. A supply chain network optimization method as claimed in claim 1, characterized in that: The step of inputting the operation data of the supplier node to be evaluated into the second envelope analysis model to obtain the first efficiency value of the supplier node in each time period and the change of the first efficiency value includes: Collecting the operation data of the supplier node to be evaluated at different time points; the operation data includes input data and output data; In the second envelope analysis model, a correlation matrix of the input data is calculated, and cluster analysis is performed on the correlation matrix to obtain matching parameters of the input data; Performing exponential smoothing processing on the output data to obtain an output data sequence; Calculate the difference between each point of the output data sequence to obtain a monthly node sequence; According to the matching parameters and the output data of each sequence point on the monthly node sequence, a first efficiency value of each time period is obtained; Performing least square fitting on the first efficiency value and segmenting according to the fitting result to obtain a slope sequence; According to the fluctuation range of the slope sequence, a key interval is obtained, and a first change point of the proportioning parameter and a second change point of the output data are extracted from the key interval; According to the first change point and the second change point, the change of the first efficiency value of the supplier node in each time period is obtained.

7. A supply chain network optimization method as claimed in claim 1, characterized in that: The step of obtaining, from the operation data according to the change of the first efficiency value, a key driving factor causing the supplier node efficiency to decrease, includes: According to the change of the first efficiency value, the operation data in the decreasing interval is obtained by screening from the operation data; the absolute value of the difference between the efficiency value of the first time period corresponding to the operation data in the decreasing interval and the efficiency values ​​of the remaining second time periods is greater than a preset efficiency value difference threshold; Calculating the correlation coefficient of the input data of the declining interval operation data, and grouping the input data of the declining interval operation data according to the correlation coefficient to obtain a plurality of declining interval input data groups; Performing principal component analysis on the output data of the declining interval operation data to obtain a feature vector; Calculating a month-on-month change rate according to the characteristic vector, and obtaining a significant change index from the output data of the operating data in the declining interval according to the month-on-month change rate and a preset month-on-month change rate threshold; Performing correlation analysis on the decline interval input data group and the significant change index to obtain a correlation strength value; According to the association strength value, key driving factors are obtained; the key driving factors include key input data and key output data that lead to a decrease in supplier node efficiency.

8. A supply chain network optimization method as claimed in claim 7, characterized in that: The performing correlation analysis on the decline interval input data group and the significant change index to obtain a correlation strength value includes: Difference calculations are performed on the input data group in the declining interval and the significant change index respectively with the preset benchmark data to obtain an input fluctuation amplitude sequence and an output fluctuation amplitude sequence; Calculating the correlation between the input fluctuation amplitude sequence and the output fluctuation amplitude sequence to obtain a correlation curve; According to a preset correlation threshold, the correlation curve is divided into a number of intervals; According to the plurality of intervals, interval features are obtained; each of the intervals corresponds to an interval feature; According to the interval characteristics, the falling interval input data group and the significant change index are grouped into a strong correlation group and a non-strong correlation group; Establishing a decision tree for the strongly associated group, splitting the nodes of the decision tree according to a preset information gain criterion, and extracting numerical distribution rules from the nodes; According to the numerical distribution law, the correlation coefficient is calculated to obtain the association strength value.

9. A supply chain network optimization method as claimed in claim 1, characterized in that: The step of optimizing the parameters of the second envelope analysis model according to the difference between the first efficiency value and the second efficiency value to obtain an optimized envelope analysis model to manage supply chain network nodes includes: Calculating the difference between the first efficiency value and the second efficiency value to obtain a fitting curve of efficiency change within the preset time; According to the fluctuation of the fitting curve, abnormal time points are obtained by screening the operation data at all times; Performing similarity calculation on the parameters of the second envelope analysis model corresponding to the abnormal time point to obtain parameter similarity; According to the parameter similarity and a preset parameter similarity threshold, adjusting the parameter within an iteration interval to obtain an adjusted parameter; The adjustment parameters are used to optimize the parameters of the second envelope analysis model to obtain an optimized envelope analysis model to manage supply chain network nodes.

10. A supply chain network optimization system, characterized in that: include: A model building module, used to build a first envelope analysis model; the input of the first envelope analysis model includes resource input data, technology input data and environmental factor data of each node in the supply chain network; A model training module, used to obtain a sample data set, and use the sample data set to train the first envelope analysis model to obtain a second envelope analysis model; The input-output ratio of the second envelope analysis model is not less than a preset target ratio threshold; A first efficiency value calculation module, used for inputting the operation data of the supplier node to be evaluated into the second envelope analysis model to obtain the first efficiency value of the supplier node in each time period and the change of the first efficiency value; A key driving factor acquisition module, configured to obtain, from the operation data, key driving factors that cause a decrease in supplier node efficiency according to changes in the first efficiency value; A second efficiency value calculation module, used for collecting second operation data of the supplier node after a preset time, inputting the second operation data and the key driving factors into the second envelope analysis model, and obtaining a second efficiency value of the supplier; The supply chain network management module is used to optimize the parameters of the second envelope analysis model according to the difference between the first efficiency value and the second efficiency value to obtain an optimized envelope analysis model to manage the supply chain network nodes.

Images