A supplier intelligent recommendation method and system

Abstract

The application provides a supplier intelligent recommendation method and system, and relates to the technical field of supply chain management. The method comprises the following steps: in step 1, supplier evaluation index data is mapped to a multidimensional space, independent space coordinates corresponding to each supplier are generated, and a space coordinate set is formed; in step 2, based on the space coordinate set, a space distribution core coordinate is calculated, two orthogonal reference direction lines are established with the core coordinate as the origin, and an evaluation range area is delimited according to the opening angle formed by the reference direction lines. The application can objectively and comprehensively recommend suppliers in multiple dimensions, and can be dynamically updated to guarantee the recommendation quality.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of supply chain management, in particular to a supplier intelligent recommendation method and system. BACKGROUND

[0002] In the traditional bidding decision-making process, when the suppliers complete the quotation, the procurement personnel usually need to manually collect and compare the quotation information of each supplier, which covers material price, total price, payment method, delivery cycle and other aspects. When the number of suppliers participating in the quotation is large or the projects involved in the quotation are complex, the manual processing process often consumes a lot of time and effort, and the overall efficiency is relatively low.

[0003] In terms of decision basis, such processes mostly rely on the personal experience and subjective judgment of procurement personnel, and sometimes follow some preset rules, such as preferring to choose the supplier with the lowest total price or unit price. However, these rules are relatively simple, and it may be difficult to conduct comprehensive and accurate quantitative evaluation when facing scenarios that need to consider multiple dimensions such as price, quality, delivery period, payment conditions, historical cooperation performance, etc. Moreover, due to the differences in procurement needs, such as some urgent orders that pay more attention to delivery period, and when the capital turnover is tight, they may pay more attention to payment conditions — fixed rules have certain limitations in flexibly combining different dimension priority strategies to adapt to diversified needs. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a supplier intelligent recommendation method and system that can process multi-dimensional supplier data and realize comprehensive quantitative evaluation.

[0005] To solve the above technical problems, the technical solutions of the present application are as follows:

[0006] In a first aspect, a supplier intelligent recommendation method and system, the method comprising:

[0007] Step 1, map the supplier evaluation index data to a multi-dimensional space, generate an independent space coordinate corresponding to each supplier, form a space coordinate set;

[0008] Step 2, based on the space coordinate set, calculate the space distribution core coordinate, establish two orthogonal reference direction lines with the core coordinate as the origin, and delimit the evaluation range area according to the opening angle formed by the reference direction lines;

[0009] Step 3, equally space five core calibration points along the divergent direction within the evaluation range area; symmetrically arrange two boundary calibration points outside the evaluation range area; connect all seven calibration points in sequence according to the data acquisition time sequence to form a ring-shaped evaluation track; extract the curvature feature quantity of the ring-shaped evaluation track to generate an offset compensation factor;

[0010] Step 4, using an improved clustering algorithm to perform initial grouping on the spatial coordinate set, correcting the center position of each group according to the offset compensation factor, and outputting the optimized grouping result;

[0011] Step 5, constructing a three-level evaluation index weight framework according to the analytic hierarchy process, and based on the optimized grouping result, applying the weight framework to calculate the comprehensive score of each group, and determining the final supplier group;

[0012] Step 6, calculating the weighted score of each supplier in the final supplier group, pushing the supplier with the highest score to the procurement terminal; when the amount of new supplier data reaches a preset threshold, re-executing the whole process from multi-dimensional data space conversion to supplier individual weighted score.

[0013] The second aspect is a supplier intelligent recommendation system, comprising:

[0014] A conversion module is configured to map supplier evaluation index data to a multi-dimensional space, generate independent spatial coordinates corresponding to each supplier, and form a spatial coordinate set;

[0015] A range construction module is configured to calculate a spatial distribution core coordinate based on the spatial coordinate set, establish two orthogonal reference direction lines with the core coordinate as the origin, and define an evaluation range area according to the opening angle formed by the reference direction lines;

[0016] A trajectory construction module is configured to equally distribute five core calibration points along the divergent direction within the evaluation range area, symmetrically distribute two boundary calibration points outside the evaluation range area, connect all seven calibration points in sequence according to the data acquisition time sequence to form a ring-shaped evaluation trajectory, and extract the curvature feature quantity of the ring-shaped evaluation trajectory to generate an offset compensation factor;

[0017] A grouping optimization module is configured to use an improved clustering algorithm to perform initial grouping on the spatial coordinate set, correct the center position of each group according to the offset compensation factor, and output the optimized grouping result;

[0018] A comprehensive score module is configured to construct a three-level evaluation index weight framework according to the analytic hierarchy process, apply the weight framework to calculate the comprehensive score of each group based on the optimized grouping result, and determine the final supplier group;

[0019] A supplier pushing module is configured to calculate the weighted score of each supplier in the final supplier group, push the supplier with the highest score to the procurement terminal, and when the amount of new supplier data reaches a preset threshold, re-execute the whole process from multi-dimensional data space conversion to supplier individual weighted score.

[0020] The third aspect is a computing device, comprising:

[0021] One or more processors;

[0022] a storage device storing one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method.

[0023] In a fourth aspect, a computer-readable storage medium stores a program, which, when executed by a processor, implements the method.

[0024] The above scheme of the present application at least has the following beneficial effects:

[0025] By mapping the supplier evaluation index to a multi-dimensional space and processing it automatically, the decision-making time is shortened, and the subjective bias caused by the experience difference of the purchasing personnel is reduced; with the help of spatial coordinate analysis, improved clustering algorithm, etc., the distribution characteristics of the supplier data can be accurately captured; by calculating the core coordinates, delimiting the evaluation range area, generating the offset compensation factor, etc., the grouping result is optimized, so that the supplier characteristics in the group are more similar, and the difference between the groups is more significant. The supplier comprehensive capability is decomposed into specific criteria and indexes, combined with weight calculation and consistency test, so that the importance of each index is quantified more reasonably, and the comprehensive score can fully reflect the overall level of the supplier group, ensuring the quality of the final supplier group; the individual weighted score is accurately calculated in the final supplier group, and the highest score supplier is pushed, providing clear decision-making basis for the purchasing terminal and improving the efficiency and quality of the purchasing decision. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is a flowchart of a supplier intelligent recommendation method provided by an embodiment of the present application.

[0027] Figure 2 is a schematic diagram of a supplier intelligent recommendation system provided by an embodiment of the present application. DETAILED DESCRIPTION

[0028] Exemplary embodiments of the present disclosure will be described more fully hereinafter with reference to the accompanying drawings; however, they are not limited to the embodiments set forth herein but can be implemented in various forms. The embodiments are provided so that this disclosure will be thorough, and will fully convey the scope of the disclosure to those skilled in the art.

[0029] As Figure 1 shown, an embodiment of the present application proposes a supplier intelligent recommendation method, which comprises the following steps:

[0030] Step 1, mapping the supplier evaluation index data to a multi-dimensional space to generate independent spatial coordinates corresponding to each supplier, forming a spatial coordinate set;

[0031] Step 2, based on the set of spatial coordinates, calculate the spatial distribution core coordinates, establish two orthogonal reference direction lines with the core coordinates as the origin, and delimit the evaluation range area according to the opening angle formed by the reference direction lines;

[0032] Step 3, five core calibration points are equally spaced along the divergent direction within the evaluation range area; two boundary calibration points are symmetrically arranged outside the evaluation range area; all seven calibration points are connected in sequence according to the data acquisition time sequence to form a ring-shaped evaluation track; the curvature feature quantity of the ring-shaped evaluation track is extracted to generate an offset compensation factor;

[0033] Step 4, an improved clustering algorithm is used to initially group the set of spatial coordinates, the center positions of each group are corrected according to the offset compensation factor, and the optimized grouping result is output;

[0034] Step 5, a three-level evaluation index weight framework is constructed according to the analytic hierarchy process, and based on the optimized grouping result, the weight framework is used to calculate the comprehensive score of each group to determine the final supplier group;

[0035] Step 6, the weighted score of each supplier in the final supplier group is calculated, and the supplier with the highest score is pushed to the procurement terminal; when the data volume of the new supplier reaches the preset threshold, the whole process of converting from multi-dimensional data space to supplier individual weighted score is re-executed.

[0036] In the embodiment of the present application, by mapping the supplier evaluation index to the multi-dimensional space and performing automatic processing, the decision-making time is shortened, and the subjective bias caused by the experience difference of the procurement personnel is reduced; with the help of spatial coordinate analysis, improved clustering algorithm, etc., the distribution characteristics of the supplier data can be accurately captured; through the steps of calculating the core coordinates, delimiting the evaluation range area, and generating the offset compensation factor, the grouping result is optimized, so that the supplier characteristics in the group are more similar, and the difference between the groups is more significant. The supplier comprehensive capability is decomposed into specific criteria and indexes, combined with weight calculation and consistency check, so that the importance of each index is quantified more reasonably, and the comprehensive score can fully reflect the overall level of the supplier group, ensuring the quality of the final supplier group; the individual weighted score is accurately calculated in the final supplier group and the supplier with the highest score is pushed, providing clear decision-making basis for the procurement terminal and improving the procurement decision-making efficiency and quality.

[0037] In a preferred embodiment of the present application, the above step 1, mapping the supplier evaluation index data to the multi-dimensional space to generate independent spatial coordinates corresponding to each supplier to form a set of spatial coordinates, can include:

[0038] Step 100, collecting the supplier's bid index data, the bid index data including material line price, total price, payment method, delivery cycle and comprehensive score;

[0039] Step 101, standardize each bid indicator data, map the standardized material line price to the first dimension coordinate value, the total price to the second dimension coordinate value, the payment method to the third dimension coordinate value, the delivery cycle to the fourth dimension coordinate value, and the comprehensive score to the fifth dimension coordinate value;

[0040] Step 102, generate a unique five-dimensional space coordinate point for each supplier, and the coordinate points of all suppliers form a space coordinate set.

[0041] In the embodiment of the present application, the relevant indicator data of all suppliers participating in the bidding is collected from the enterprise procurement management system, the electronic bid submitted by the supplier, the historical transaction records of both parties, etc.

[0042] The specific content and collection method of each indicator are as follows:

[0043] Material line price: collect the unit price of each material separately quoted by the supplier for the purchase list, and distinguish different specifications and models of materials to ensure that each material item corresponds to a unique quote;

[0044] Total price: collect the total price of all materials quoted by the supplier for the purchase list, and if there are discounts, taxes and other additional costs, they need to be included in the total price to ensure the final settlement total price;

[0045] Payment method: collect the payment conditions required by the supplier, including prepayment ratio (such as no prepayment, 30% prepayment, etc.), account period (such as cash on delivery, 30-day account period, 60-day account period, etc.), payment form (such as wire transfer, acceptance bill, etc.), and the text description needs to be converted into quantifiable information (such as account period in days, prepayment percentage);

[0046] Delivery cycle: collect the total length of time promised by the supplier from order confirmation to delivery of materials to the designated location, in units of "days", and if there are batch deliveries, the delivery time and corresponding material quantity of each batch need to be specified;

[0047] Comprehensive score: generate a comprehensive evaluation score based on the supplier's historical cooperation records (such as product quality pass rate, compliance rate, after-sales service satisfaction, etc.), which is a quantitative score of 0-100 points, and if it is a first-time cooperation supplier, the industry general evaluation standard or third-party rating agency data can be referred to.

[0048] Step 101, quantitative indicators (material line price, total price, delivery cycle): adopt min-max standardization method to eliminate the dimensional differences of different indicators. The specific process is: first, count the minimum and maximum values of all suppliers on the indicator, then convert the indicator value to a standardized value between 0 and 1 through the calculation of "(current supplier indicator value - minimum value of the indicator) / (maximum value of the indicator - minimum value of the indicator)"; if the indicator is "the smaller the better" (such as price, delivery cycle), use "1 - standardized result" for reverse processing to ensure that the larger the value, the better the indicator performance.

[0049] Semi-quantitative indicators (payment method): first, convert qualitative descriptions into quantitative scores (e.g. no prepayment and 60-day account period score 1 point, 30% prepayment and 30-day account period score 0.6 points, full prepayment and no account period score 0 points), then use the same min-max standardization method as quantitative indicators to convert to a standardized value between 0 and 1;

[0050] Comprehensive score: if the original score is 0-100 points, directly convert to a standardized value between 0 and 1 through "comprehensive score / 100"; if it is other score range, also use the min-max standardization method to adjust to the 0-1 interval;

[0051] Dimension mapping: the standardized values are mapped to the five dimensions of the five-dimensional space according to fixed rules - the standardized material line price corresponds to the first dimension coordinate value, the total price corresponds to the second dimension coordinate value, the payment method corresponds to the third dimension coordinate value, the delivery cycle corresponds to the fourth dimension coordinate value, and the comprehensive score corresponds to the fifth dimension coordinate value.

[0052] Step 102, the five standardized indicator values of each supplier are combined in the format "(first dimension coordinate value, second dimension coordinate value, third dimension coordinate value, fourth dimension coordinate value, fifth dimension coordinate value)" to form a unique five-dimensional space coordinate point, which corresponds to the supplier. Collect the five-dimensional coordinate points of all participating bidders to form a spatial coordinate set containing the location information of all suppliers. Each element in the set is a five-dimensional coordinate point and there is no repetition (the same supplier corresponds to only one coordinate point).

[0053] The standardization eliminates the dimensional differences of different indicators (such as price in "yuan" and cycle in "days"), making the dimension coordinate values comparable; abstract bid data is converted into visual five-dimensional space coordinate points, making the differences between suppliers into spatial distances, which facilitates the use of spatial analysis methods (such as clustering, trajectory analysis) to mine data rules and improve the objectivity and interpretability of the recommendation logic; the formation of the spatial coordinate set facilitates the centralized storage, batch processing and dynamic updating of all supplier data.

[0054] In a preferred embodiment of the present application, the step 2, based on the set of spatial coordinates, calculates the spatial distribution core coordinates, establishes two orthogonal reference direction lines with the core coordinates as the origin, and delimits the evaluation range area according to the opening angle formed by the reference direction lines, which can include:

[0055] Step 200, based on the set of spatial coordinates, calculates the arithmetic mean position of all independent coordinate points in the five-dimensional space, and defines the corresponding arithmetic mean position coordinates as the spatial distribution core coordinates;

[0056] Step 201, determine the first principal component direction and the second principal component direction of the coordinate set by principal component analysis, construct the first reference direction line along the first principal component direction and the second reference direction line along the second principal component direction with the core coordinates as the origin, and the two direction lines remain orthogonal;

[0057] Step 202, calculate the included angle between the two reference direction lines to delimit the sector evaluation range area.

[0058] In an embodiment of the present application, each supplier's coordinate point in the set of spatial coordinates is of five-dimensional structure, and the numerical value of each dimension is the result of the standardization processing in step 101 (range 0-1). The specific dimension correspondence is:

[0059] First dimension: material line price standardized value (for example, the first dimension value of supplier A is 0.7, supplier B is 0.5, and supplier C is 0.6, etc.);

[0060] Second dimension: total price standardized value (for example, supplier A is 0.8, supplier B is 0.6, and supplier C is 0.7, etc.);

[0061] Third dimension: payment method standardized value (for example, supplier A is 0.9, supplier B is 0.4, and supplier C is 0.6, etc.);

[0062] Fourth dimension: delivery cycle standardized value (for example, supplier A is 0.5, supplier B is 0.8, and supplier C is 0.7, etc.);

[0063] Fifth dimension: comprehensive score standardized value (for example, supplier A is 0.8, supplier B is 0.7, and supplier C is 0.9, etc.).

[0064] First, all the specific numerical values of the suppliers in the five dimensions are sorted out to form a "supplier-dimension-value" table (for example, 30 suppliers have 30 rows, and each row corresponds to the specific numerical value of 5 dimensions).

[0065] First dimension average value calculation: Extract all the first dimension values of the suppliers from the "Supplier-Dimension-Value" table (e.g. 30 suppliers correspond to 30 values: 0.7, 0.5, 0.6,...), add them one by one to get the total sum (e.g. total sum is 18.6), and then divide the total sum by the total number of suppliers (30) to get the arithmetic mean of the first dimension.

[0066] Second dimension average value calculation: Similarly, extract all the second dimension values of the suppliers (e.g. 0.8, 0.6, 0.7,...), sum them up and divide by the total number of suppliers (e.g. total sum is 21.3) to get the average value of the second dimension.

[0067] Third / fourth / fifth dimension average value calculation: According to the same logic, extract all the values of the third dimension (payment method), fourth dimension (delivery cycle), and fifth dimension (overall score) of the suppliers respectively, sum them up one by one and divide by the total number of suppliers to get the arithmetic mean of each dimension (e.g. third dimension average value 0.65, fourth dimension 0.68, fifth dimension 0.75).

[0068] Arrange the arithmetic mean of the above five dimensions in the order of "first dimension → second dimension → third dimension → fourth dimension → fifth dimension" to form a five-dimensional coordinate point. For example, if the average values of the five dimensions are 0.62, 0.71, 0.65, 0.68, and 0.75 respectively, the spatial distribution core coordinate is (0.62, 0.71, 0.65, 0.68, 0.75). This coordinate point is the "central position" of all suppliers in the five-dimensional space, reflecting the concentration trend of the overall data.

[0069] Step 201, based on the spatial coordinate set, arrange all the five-dimensional coordinate values of the suppliers into a "data matrix", the matrix structure is: number of rows = number of suppliers (e.g. 30 rows, each row represents a supplier); number of columns = 5 columns (each column corresponds to a dimension, respectively, the standardized values of the first to fifth dimensions);

[0070] The value of each cell in the matrix = the standardized value of the corresponding supplier in the corresponding dimension (consistent with the "Supplier-Dimension-Value" table in step 200). For example, the first 3 rows of the matrix of 30 suppliers are: (0.7, 0.8, 0.9, 0.5, 0.8); (0.5, 0.6, 0.4, 0.8, 0.7); (0.6, 0.7, 0.6, 0.7, 0.9); Ensure that all the values in the matrix are standardized results of 0-1, which are completely consistent with the processing logic of step 101, to avoid affecting subsequent analysis due to inconsistent data formats.

[0071] Calculate the principal component direction (first and second principal components): First step: calculate the covariance matrix:

[0072] The covariance matrix is used to reflect the correlation between the five dimensions (e.g. the degree of association between "material line price" and "total price", the degree of association between "delivery cycle" and "overall score", etc.). When calculating, the covariance of each pair of dimensions (a total of 10 pairs: first and second, first and third,..., fourth and fifth) needs to be calculated respectively: for example, calculate the covariance of "first dimension (material line price) and second dimension (total price)": first extract the first dimension values and the second dimension values of all suppliers, calculate the average deviation product of the two (i.e. the difference between the first dimension value of each supplier and the average value of the first dimension, multiplied by the difference between the second dimension value of the supplier and the average value of the second dimension, and then all the products are added and divided by "the total number of suppliers - 1"); according to this logic, the covariances of all 10 pairs of dimensions are calculated in turn, and finally a 5x5 covariance matrix is formed (the diagonal line is the variance of each dimension itself, and the non-diagonal line is the covariance between dimensions). For example, the value in position (1,2) of the matrix is 0.3 (indicating a positive correlation between the first and second dimensions), the value in position (3,4) is -0.2 (indicating a negative correlation between the third and fourth dimensions), etc.

[0073] Second step: extract eigenvalues and eigenvectors:

[0074] The eigenvalue reflects the "degree of dispersion of data in a certain direction" (the larger the value, the more obvious the difference between suppliers in that direction); the eigenvector reflects the "specific direction of that direction" (consisting of 5 numbers, corresponding to the weights of the five dimensions, such as (0.4, 0.3, 0.2, 0.1, 0.0) indicating that this direction is mainly affected by the first and second dimensions). Extract all eigenvalues (a total of 5, as the matrix is 5x5) and corresponding eigenvectors (each eigenvalue corresponds to a 5-dimensional eigenvector) from the covariance matrix: sort the eigenvalues from large to small (for example, after sorting, they are: 1.2, 0.8, 0.5, 0.3, 0.2); the eigenvector corresponding to the largest eigenvalue (1.2) is the "first principal component direction" (for example, the vector is (0.5, 0.4, 0.1, 0.0, 0.0), indicating that this direction is mainly driven by the first and second dimensions, i.e. the difference between suppliers is mainly reflected in "material line price" and "total price"); the eigenvector corresponding to the second largest eigenvalue (0.8) needs to be verified for orthogonality with the first principal component direction (i.e. the "vector dot product" of the two eigenvectors is 0: multiply the corresponding dimension values of the two vectors and sum them up, the result is 0). If orthogonal, it is determined as the "second principal component direction" (for example, the vector is (0.0, 0.0, 0.3, 0.4, 0.3), indicating that this direction is mainly driven by the third, fourth and fifth dimensions, i.e. the difference between suppliers is significantly reflected in "payment method", "delivery cycle" and "overall score").

[0075] Third step: build an orthogonal reference direction line:

[0076] The "spatial distribution core coordinates" (such as (0.62, 0.71, 0.65, 0.68, 0.75)) calculated in step 200 are taken as the origin, and a first reference direction line is formed by extending along the first principal component direction (for example, the line extends infinitely in the direction of the "first principal component vector" from the origin, that is, all points on the line satisfy "the numerical values of each dimension change in proportion to the weight of the first principal component vector"); similarly, a second reference direction line is formed by extending along the second principal component direction. Since the eigenvectors of the first and second principal component directions are orthogonal, the two reference direction lines are perpendicular in space (similar to the x-axis and y-axis in the plane coordinate system). The meanings of the two direction lines need to be explained in combination with the business scenario: for example, if the first principal component direction is mainly driven by "material line price" and "total price", the first reference direction line can be understood as the "price sensitive direction"; if the second principal component direction is driven by "delivery cycle" and "comprehensive score", the second reference direction line can be understood as the "compliance ability direction", ensuring that the direction line corresponds to the core indicators actually concerned by the procurement.

[0077] In step 202, the angle calculation is based on the eigenvectors of the first and second principal component directions, reflecting the distribution span of the data in the two core directions:

[0078] Since the two direction lines are orthogonal (the dot product of the eigenvectors is 0), the theoretical maximum angle is 90°, but in actual data, the angle may be smaller (such as 60° or 75°) due to different degrees of concentration of the distribution; if the eigenvalues of the two principal component directions are small (such as 1.2 and 0.8), it means that the dispersion of the data in the two directions is close, and the angle is large (such as 80°); if the eigenvalues are large (such as 1.2 and 0.3), it means that the data is mainly dispersed along the first principal component direction, and the angle is small (such as 30°); the finally determined angle needs to be verified by "data coverage": to ensure that at least 70% of the coordinate points of the suppliers are included in the angle range (for example, 22 out of 30 suppliers fall within the region corresponding to the angle), to avoid missing most data due to too small angle or including too many abnormal points due to too large angle.

[0079] The "spatial distribution core coordinates" calculated in step 200 are taken as the vertex (origin), the two reference direction lines are taken as the left and right boundaries, and the calculated angle (such as 60°) is taken as the opening angle, to define a sector region in the five-dimensional space:

[0080] All points in the region need to meet the "projection in the first principal component direction does not exceed the boundary" and "projection in the second principal component direction does not exceed the boundary"; for example, if the included angle is 60°, the sector region covers all the space range deviated by 60° from the first principal component direction to the second principal component direction, and all the supplier coordinate points falling in the region belong to the "core group" in the data set (excluding abnormal suppliers deviating too far from the two principal component directions, such as suppliers with much higher prices and much longer delivery cycles).

[0081] The spatial distribution core coordinates as the average position of all suppliers objectively reflect the concentration trend of the overall data, ensuring the stability of the evaluation benchmark; the two orthogonal reference direction lines extracted by the principal component analysis method accurately capture the difference dimension in the data (such as the main change direction of price and delivery time), so that the evaluation range is more in line with the actual data law, avoiding the interference of irrelevant dimensions. The sector evaluation range area based on the opening angle focuses on the core area in the data set, covering the distribution characteristics of most suppliers and excluding extreme abnormal points through boundary limitation, improving the evaluation efficiency.

[0082] In a preferred embodiment of the present application, the above step 3, five core calibration points are equally spaced along the divergent direction within the evaluation range area; two boundary calibration points are symmetrically arranged outside the evaluation range area; all seven calibration points are connected in series according to the data acquisition time sequence to form a ring-shaped evaluation track; the curvature feature of the ring-shaped evaluation track is extracted to generate an offset compensation factor, which can include:

[0083] Step 300, five core calibration points are arranged along the angle bisector direction within the evaluation range area, including dividing the region arc into five segments; extending radially to the region boundary at each division point; defining the boundary intersection coordinates as the core calibration point; arranging two boundary calibration points outside the evaluation range area, extending the region radius by 1.2 times along the first reference direction line to generate the first boundary calibration point; extending the region radius by 1.2 times along the second reference direction line to generate the second boundary calibration point;

[0084] Step 301, connecting seven calibration points in series according to the data acquisition time sequence to form a closed loop path, and extracting the curvature features of the closed loop path, including maximum curvature and average curvature;

[0085] Step 302, according to the numerical combination of the maximum curvature and the average curvature, querying the preset compensation rule table, and outputting the offset compensation factor.

[0086] In the embodiment of the present application, seven calibration points (five core calibration points + two boundary calibration points) are arranged:

[0087] The evaluation range area is a sector area defined in step 202, and its core parameters include: the vertex is the spatial distribution core coordinate (origin), the two boundaries are the reference direction lines of the first and second principal component directions, the opening angle is the included angle between the two reference direction lines (such as 60°), and the area radius is the distance from the core coordinate to the arc edge of the sector (i.e. the average distance from the core coordinate to the farthest supplier coordinate point in the sector, for example, 5 units of length).

[0088] Five core calibration points are arranged in the evaluation range area: first step: determine the arc line of the sector area:

[0089] The arc line of the sector area refers to the arc-shaped boundary formed between the two reference direction lines with the core coordinate as the center and the area radius as the radius (i.e. the "arc edge" of the sector). For example, for a sector with an opening angle of 60° and a radius of 5 units, the arc line is a circular arc segment with a central angle of 60° and a radius of 5.

[0090] Second step: divide the arc line into five segments:

[0091] Along the arc line from the endpoint of the first reference direction line to the endpoint of the second reference direction line, divide it into five equal parts by length, resulting in four division points, plus the two endpoints of the arc line, for a total of five division points (i.e. "division points"). For example, a 60° arc line divided into five segments corresponds to a central angle of 12° for each segment, and the five division points are located at distances of 0°, 12°, 24°, 36°, and 48° from the first reference direction line (the endpoint is 60°, which is the endpoint of the second reference direction line).

[0092] Third step: extend radially to the area boundary to determine the core calibration point:

[0093] For each division point, draw a straight line (i.e. "radial", consistent with the radius direction) from the core coordinate (origin) to the division point. This straight line intersects the arc edge of the sector at the division point, and this intersection point is the core calibration point. For example, the first division point is located at 0° (the intersection point of the first reference direction line and the arc line), and its coordinate is the position 5 units away from the core coordinate in the 0° direction. The second division point is located at 12°, and its coordinate is the position 5 units away from the core coordinate in the 12° direction. Similarly, a total of five core calibration points are obtained.

[0094] Two boundary calibration points are arranged outside the evaluation range area:

[0095] Determine the area radius: the area radius is the distance from the core coordinate to the arc edge of the sector (such as 5 units of length).

[0096] Lay the first boundary calibration point: along the first reference direction line (from the core coordinate, extend to the outside of the sector, that is, the opposite direction of the core calibration point), the extension distance is 1.2 times the radius of the region (5x1.2=6 units of length), and the extension endpoint is the first boundary calibration point.

[0097] Lay the second boundary calibration point: along the second reference direction line (also extend to the outside of the sector), the extension distance is 1.2 times the radius of the region (6 units of length), and the extension endpoint is the second boundary calibration point. The two boundary calibration points are symmetric about the angle bisector of the sector, and both are outside the evaluation range area (sector).

[0098] Step 301, connect the seven calibration points in chronological order to form a closed loop path:

[0099] Determine the collection time of the calibration point: each calibration point corresponds to the original data (such as supplier quotation data) with collection time (for example, the supplier data corresponding to the five core calibration points are collected on March 1, March 3, March 5, March 7, and March 9; the supplier data corresponding to the two boundary calibration points are collected on February 28 and March 10).

[0100] Sort by time: sort the seven calibration points in chronological order (for example: the second boundary calibration point on February 28→the first core calibration point on March 1→the second core calibration point on March 3→…→the fifth core calibration point on March 9→the first boundary calibration point on March 10).

[0101] Connect to form a closed loop path: connect adjacent calibration points in order according to the sorted order, and finally connect the last calibration point with the first calibration point to form a closed loop track (for example, starting from the second boundary calibration point, connecting the five core calibration points in turn, then connecting the first boundary calibration point, and finally returning to the second boundary calibration point to form a loop).

[0102] Extract the curvature feature quantity (maximum curvature and average curvature) of the closed loop path:

[0103] The meaning of curvature: curvature reflects the degree of bending of the path, the more acute the bending, the greater the curvature; the curvature of a straight line is 0.

[0104] Calculate the curvature of each segment of the path: the loop track is connected by 7 straight lines (because there are 7 calibration points to form 7 edges), but the connection between adjacent two straight lines (i.e. the calibration point position) will form a corner, the larger the corner, the greater the curvature of the point. For example, the first core calibration point connects the second boundary calibration point and the second core calibration point, if the angle between the two connecting lines is 30°, the curvature of the point is smaller; if the angle is 150°, the curvature is larger.

[0105] Determine the maximum curvature: traverse the 7 corner points of the circular trajectory (i.e. 7 calibration points), record the curvature value of each point, and the maximum value is the "maximum curvature" (reflecting the most curved position in the trajectory).

[0106] Calculate the average curvature: add the curvature values of the 7 corner points and divide by 7 to get the "average curvature" (reflecting the overall bending degree of the trajectory).

[0107] Step 302, the rule table is a table based on historical data or business experience, containing the corresponding relationship between "maximum curvature range", "average curvature range" and "offset compensation factor". For example:

[0108] If the maximum curvature is less than 0.2 and the average curvature is less than 0.1, the offset compensation factor is 0.9;

[0109] If the maximum curvature is between 0.2 and 0.5 and the average curvature is between 0.1 and 0.3, the offset compensation factor is 1.2;

[0110] If the maximum curvature is greater than 0.5 and the average curvature is greater than 0.3, the offset compensation factor is 1.5;

[0111] (Note: the range and factor values in the rule table need to be set according to the actual business scenario, which is used to quantify the influence of trajectory bending degree on subsequent clustering).

[0112] Query the rule table to output the offset compensation factor:

[0113] Match the maximum curvature and average curvature calculated in step 301 with the ranges in the rule table: for example, if the maximum curvature is 0.3 and the average curvature is 0.2, corresponding to the range of "maximum curvature 0.2-0.5, average curvature 0.1-0.3" in the rule table, then output the corresponding offset compensation factor (such as 1.2). This factor will be used to correct the position of the clustering center in the future, offsetting the influence of data distribution bias.

[0114] The core calibration points are evenly distributed along the fan-shaped area, covering the core data area of the evaluation range; the boundary calibration points are symmetrically distributed outside the area, taking into account the edge data, ensuring that the calibration points can fully reflect the spatial distribution characteristics of the supplier data; the calibration points are connected in series according to the data collection time, so that the trajectory reflects both spatial distribution and time dimension, and can capture the trend of data change over time (such as price fluctuation pattern). The maximum curvature and average curvature quantify the bending degree of the trajectory, indirectly reflecting the concentration or dispersion state of data distribution; through the pre-set rule table, the curvature characteristics are converted into the offset compensation factor, which can correct the position of the clustering center, reducing the grouping deviation caused by data distribution bias.

[0115] In a preferred embodiment of the present application, the step 4 above, using the improved clustering algorithm to perform initial grouping on the spatial coordinate set, correcting the position of each grouping center according to the offset compensation factor, and outputting the optimized grouping result, can include:

[0116] Step 400, input the spatial coordinate set into the clustering algorithm, divide the supplier coordinate points into K initial groupings through iterative calculation, and output the clustering center coordinates of each initial grouping, wherein the number of groupings K is determined according to the inflection point of the within-group sum of squared errors corresponding to different clustering numbers;

[0117] Step 401, calculate the direction vector between each initial clustering center coordinate and the spatial distribution core coordinate respectively, and apply the offset compensation factor to the length of the direction vector to obtain the adjusted length;

[0118] Step 402, taking the spatial distribution core coordinate as the starting point, pointing in the original direction vector, and determining the new clustering center coordinate according to the adjusted length;

[0119] Step 403, for each supplier coordinate point in the spatial coordinate set, calculate the Euclidean distance from the supplier coordinate point to all new clustering center coordinates respectively, and divide the supplier coordinate point into the grouping corresponding to the new clustering center with the smallest Euclidean distance value, to form the optimized grouping result.

[0120] In an embodiment of the present application, the spatial coordinate set (five-dimensional coordinate points of all suppliers) formed in step 102 is input into the improved clustering algorithm, and the initial grouping is completed through multiple rounds of iteration:

[0121] First round of iteration: randomly initialize the clustering center:

[0122] Randomly select several coordinate points from the spatial coordinate set as the initial clustering center (the number is tentatively set as the candidate K value, such as 3 are selected first), and each center represents an initial core of a potential grouping.

[0123] Assign the coordinate point to the nearest clustering center:

[0124] Calculate the spatial distance from each supplier coordinate point to all initial clustering centers (reflecting the similarity between the supplier and the center), and assign the coordinate point to the grouping to which the nearest clustering center belongs, for example, the distance from the coordinate point of supplier A to center 1 is 0.3, and the distance to center 2 is 0.5, then A is divided into the grouping of center 1.

[0125] Update the clustering center:

[0126] For each grouping, calculate the five-dimensional arithmetic mean of all supplier coordinate points in the group (similar to the core coordinate calculation method in step 200), and take the average value as the new clustering center to replace the initial center.

[0127] Repeat iteration until stable:

[0128] Recalculate the distance of all coordinate points to the new cluster center and reassign the group, then update the cluster center again, repeat this process until the position of the cluster center changes less than a preset threshold (such as each dimension coordinate value changes less than 0.01) in two consecutive iterations, at this time the grouping result tends to be stable, stop iteration, get the initial grouping corresponding to the candidate K value.

[0129] Calculate the sum of squares of errors within groups (SSE) for different K values:

[0130] The sum of squares of errors within groups is an index to measure the compactness of grouping: for each candidate K value (such as from 2 to 10), after completing the initial grouping according to the above iteration process, calculate the sum of squares of spatial distances of all supplier coordinate points within each group to the cluster center of the group, then add up the sum of squares of all groups to get the SSE corresponding to the K value, and so on.

[0131] Determine K by "inflection point method":

[0132] Draw a curve with K value as the horizontal axis and SSE as the vertical axis, and observe the trend of the curve: as K increases, SSE will gradually decrease (the finer the grouping, the more concentrated the data), but when K exceeds a certain value, the decline of SSE will slow down significantly (i.e. "inflection point" appears). For example, the curve at K=4, SSE decreases from 15.0 at K=3 to 12.5 (decrease of 2.5), and at K=5, it decreases to 11.8 (decrease of 0.7), so K=4 is the inflection point, and the initial grouping result (4 groups and corresponding cluster center coordinates) at this time is the final initial grouping.

[0133] Step 401, calculate the direction vector of the initial cluster center and the spatial distribution core coordinate:

[0134] The spatial distribution core coordinate is the five-dimensional average coordinate determined in step 200 (such as (0.62, 0.71, 0.65, 0.68, 0.75)), and each initial cluster center is also a five-dimensional coordinate (such as when K=4, the four initial centers are C1(0.5, 0.6, 0.7, 0.6, 0.8), C2(0.7, 0.8, 0.5, 0.7, 0.6), etc.). The direction vector is a "direction indicator" from the core coordinate to the initial cluster center, which is calculated as follows: for each dimension, subtract the corresponding dimension value of the core coordinate from the coordinate value of the initial cluster center to get the component of that dimension. For example, the direction vector of C1 and the core coordinate has the following dimension components:

[0135] First dimension: 0.5-0.62=-0.12; second dimension: 0.6-0.71=-0.11; third dimension: 0.7-0.65=0.05; fourth dimension: 0.6-0.68=-0.08; fifth dimension: 0.8-0.75=0.05;

[0136] Thus, the direction vector of C1 is (-0.12, -0.11, 0.05, -0.08, 0.05), which embodies both direction (the positive and negative of each component indicates the direction of deviation from the core coordinate) and distance (the "modulus" of the vector). The modulus of the direction vector is the "length" of the vector (i.e. the spatial straight-line distance from the core coordinate to the initial cluster center), which is calculated by synthesizing the differences of the components of each dimension (for example, the modulus of C1 above reflects the overall deviation of C1 from the core coordinate).

[0137] Adjust the modulus with the offset compensation factor:

[0138] The offset compensation factor is a quantized value generated in step 302 (such as 1.2), which serves to adjust the length of the direction vector:

[0139] If the offset compensation factor > 1 (such as 1.2), multiply the original modulus by 1.2 to obtain a longer adjusted modulus (i.e. the initial cluster center needs to be further away from the core coordinate);

[0140] If the offset compensation factor < 1 (such as 0.9), multiply the original modulus by 0.9 to obtain a shorter adjusted modulus (i.e. the initial cluster center needs to be closer to the core coordinate);

[0141] Step 402, the position of the new cluster center is determined by the "spatially distributed core coordinate", the "original direction vector points to", and the "adjusted modulus":

[0142] Keep the direction unchanged: the new cluster center is still on the straight line pointed to by the original direction vector (i.e. the deviation direction of each dimension is consistent with the initial cluster center, such as C1 deviates from the core coordinate in the first dimension in the negative direction, and the new center is still negative);

[0143] Distance adjustment: from the core coordinate, move a distance of "adjusted modulus" along the direction of the original direction vector to obtain the new coordinate value.

[0144] For example, the core coordinate is (0.62, 0.71, 0.65, 0.68, 0.75), the original direction vector of C1 points to the direction of "first dimension decreasing, second dimension decreasing, third dimension increasing", and the adjusted modulus is 0.24 (original modulus 0.2 x 1.2), then the coordinate values of each dimension of the new cluster center C1' are the core coordinate values plus the values of "scaling the components of the direction vector to the adjusted modulus" (i.e. ensuring that the distance between the new center and the core coordinate is 0.24, and the direction is unchanged).

[0145] Step 403, calculate the Euclidean distance from the supplier coordinate point to the new cluster center:

[0146] The Euclidean distance is the straight-line distance between two points in space. For each supplier's five-dimensional coordinate point, the distance to all new cluster centers (such as C1', C2', C3', C4' when K=4) is calculated: the distance calculation takes into account the differences in the five dimensions (for example, the coordinates of Supplier A are (0.6, 0.7, 0.6, 0.7, 0.8), the distance to C1' needs to compare the difference in the first to fifth dimension, and finally get a comprehensive distance value). For each supplier coordinate point, it is assigned to the group corresponding to the "closest new cluster center":

[0147] For example, the distance from Supplier A to C1' is 0.15, to C2' is 0.2, to C3' is 0.3, and to C4' is 0.25, so it is assigned to the group corresponding to C1'; after all supplier coordinate points are assigned, K new groups are formed, which is the optimization grouping result.

[0148] The K value is determined by iterative calculation and "inflection point method", which avoids too coarse grouping (information loss) or too fine grouping (redundancy), and ensures that the initial grouping can reflect the natural clustering characteristics of the supplier data; the offset compensation factor combines the curvature characteristics of data distribution, adjusts the length of the cluster center position, and effectively offsets the deviation in the initial grouping (such as the center deviation caused by local data concentration). Based on the new cluster center, the suppliers are re-assigned, making the suppliers in the group more similar and the difference between the groups more significant, improving the accuracy of supplier recommendation.

[0149] In a preferred embodiment of the present application, step 5 above, the three-level evaluation index weight framework is constructed according to the analytic hierarchy process, and based on the optimization grouping result, the weight framework is used to calculate the comprehensive score of each group to determine the final supplier group, which can include:

[0150] Step 500, based on the five-dimensional coordinate set of each supplier group in the optimization grouping result, a three-level evaluation system composed of target layer, criterion layer and index layer is constructed; the target layer corresponds to the comprehensive ability of the supplier, the criterion layer includes product quality, transaction cost and fulfillment ability, and the index layer is decomposed into five evaluation indexes;

[0151] Step 501, construct the pairwise comparison judgment matrix of the criterion layer to the target layer and the index layer to the criterion layer by the analytic hierarchy process, calculate the weight coefficients of each level element and perform consistency check, and form a three-level evaluation index weight framework;

[0152] Step 502, for each optimization group, the coordinate components of all suppliers in the group in the five-dimensional space are extracted respectively, and the five index values are weighted and aggregated according to the weight framework, and the comprehensive score of each group is calculated;

[0153] Step 503, according to the comprehensive score, all supplier groups are sorted in descending order, and the group with the highest ranking and a score higher than the preset qualified threshold is selected as the final supplier group; if the first group does not reach the threshold, then the next order is matched in turn until the qualified group is selected.

[0154] In the embodiment of the application, based on the optimization grouping result (each supplier group formed in step 403, each group containing multiple five-dimensional coordinate points of suppliers), a three-level evaluation system of "target layer-criterion layer-index layer" is built, and the specific corresponding relationship and content are as follows:

[0155] Target layer: the only element is "supplier comprehensive ability", that is, the core target that needs to be finally evaluated, which is the comprehensive level of measuring the overall performance of the supplier.

[0156] Criterion layer: contains three dimensions, which are specific decompositions of the target layer, respectively:

[0157] Product quality: reflects the quality level and reliability of the materials provided by the supplier;

[0158] Transaction cost: reflects the cost control ability in the cooperation process with the supplier;

[0159] Performance: reflects the ability of the supplier to complete the delivery, payment and other cooperation processes according to the agreement.

[0160] Index layer: the criterion layer is further refined into five evaluation indexes (one-to-one corresponding to the five-dimensional coordinates), and the specific corresponding relationship is:

[0161] Product quality→fifth-dimensional index (comprehensive score, because the comprehensive score contains historical quality, cooperation satisfaction and other quality-related information);

[0162] Transaction cost→first-dimensional index (material line price), second-dimensional index (total price), third-dimensional index (payment method, such as prepayment ratio, account period and other factors directly affecting the cost of funds);

[0163] Performance→fourth-dimensional index (delivery cycle, directly reflecting the ability to deliver on time).

[0164] Thus, a three-level structure of "target layer (1), criterion layer (3), index layer (5)" is formed, and the affiliation of each level is clear.

[0165] Step 501, the weight framework (including judgment matrix, weight calculation and consistency check) is constructed by the analytic hierarchy process

[0166] Construct two two comparison judgment matrix:

[0167] The analytic hierarchy process determines the relative importance of each element by "two two comparison", using 1-9 scale (1 means two elements are equally important, 3 means the former is slightly more important than the latter, 5 means the former is obviously more important than the latter, 7 means the former is strongly more important than the latter, 9 means the former is extremely more important than the latter, 2, 4, 6, 8 are intermediate values, and the inverse indicates the reverse comparison), respectively construct two types of judgment matrix:

[0168] The judgment matrix of the criterion layer to the target layer: compare the importance of the three elements of the criterion layer (product quality, transaction cost, and performance ability) to the target layer (supplier comprehensive ability). For example, if "transaction cost" is obviously more important than "product quality" (scale 5), "transaction cost" is slightly more important than "performance ability" (scale 3), and "product quality" is equally important as "performance ability" (scale 1), construct a 3x3 judgment matrix (rows and columns are product quality, transaction cost, and performance ability), and the matrix elements are the corresponding scale values.

[0169] The judgment matrix of the index layer to the criterion layer: construct the judgment matrix of the index layer for each criterion layer element:

[0170] The index layer corresponding to product quality only has the fifth dimension index (comprehensive score), which does not need to be compared, and the weight is directly determined as 1;

[0171] The index layer corresponding to transaction cost includes the first dimension (material line price), the second dimension (total price), and the third dimension (payment method), which need to be compared for the importance of the three to "transaction cost". For example, "material line price" is slightly more important than "total price" (scale 3), "material line price" is obviously more important than "payment method" (scale 5), and "total price" is slightly more important than "payment method" (scale 3), construct a 3x3 judgment matrix;

[0172] The index layer corresponding to performance ability only has the fourth dimension index (delivery cycle), which does not need to be compared, and the weight is 1.

[0173] Calculate the weight coefficients of each level element:

[0174] For each judgment matrix, the weight is calculated by the "sum method" (add up the elements in each row of the matrix to get the row sum, then divide each row sum by the total of all row sums to get the weight of the element): For example, the row sums of the judgment matrix of the index layer of transaction cost are: the first dimension (material line price) row sum = 3 + 1 + 5 = 9 (hypothetical), the second dimension (total price) row sum = 1 / 3 + 1 + 3 = 4.33, the third dimension (payment method) row sum = 1 / 5 + 1 / 3 + 1 = 1.53; total row sum = 9 + 4.33 + 1.53 = 14.86; then the first dimension weight = 9 ÷ 14.86, the second dimension ≈ 4.33 ÷ 14.86, the third dimension ≈ 1.53 ÷ 14.86.

[0175] According to this logic, the weights of the criterion layer to the target layer are calculated in turn (such as product quality weight 0.2, transaction cost 0.5, and performance capability 0.3), and the weights of the index layer to the criterion layer are calculated, finally forming a three-level weight chain of "index layer → criterion layer → target layer" (such as the fifth dimension index weight 1 to product quality, and product quality weight 0.2 to the target layer, then the final weight of the fifth dimension to the target layer = 1 × 0.2 = 0.2).

[0176] Consistency test:

[0177] Test whether the judgment matrix is reasonable (avoid the contradiction that "A is more important than B, B is more important than C, and C is more important than A"):

[0178] Calculate the consistency index (reflecting the degree of deviation of the matrix from consistency), and then compare it with the average random consistency index of the same order (preset standard value, such as 0.58 for a 3-order matrix) to get the consistency ratio (consistency index ÷ standard value).

[0179] If the consistency ratio < 0.1, the matrix is reasonable and the weight is effective; if ≥ 0.1, the scale value of the judgment matrix needs to be adjusted, and the weight is calculated repeatedly until the test is passed. Finally, a three-level evaluation index weight framework that passes the consistency test is formed (clearly showing the final weight of the five indexes to the target layer).

[0180] Step 502, based on the optimization grouping result and the weight framework, calculate the comprehensive score of each group in steps:

[0181] For each optimization grouping (such as the 4 groupings formed in step 403), extract the five-dimensional coordinate components (i.e. the standardized values of the five indexes) of all suppliers in the group. For example, Group 1 contains 5 suppliers, whose first dimension (material line price) values are 0.6, 0.7, 0.5, 0.6, and 0.7 respectively; the second dimension (total price) values are 0.5, 0.6, 0.4, 0.5, and 0.6 respectively, and so on.

[0182] Calculate the average value of the indexes in the group:

[0183] For each group, the average value of the five indicators of all suppliers in the group is calculated (eliminating individual differences, reflecting the overall level of the group), so as to obtain the average value of the five indicators of each group.

[0184] The comprehensive score is calculated by weighted aggregation:

[0185] According to the maximum weight of the target layer in the five indicators in the three-level weight framework (such as the first dimension 0.2, the second dimension 0.2, the third dimension 0.1, the fourth dimension 0.3, and the fifth dimension 0.2), the average value of the five indicators of each group is multiplied by the corresponding weight and summed up: for example, the comprehensive score of group 1 = (the average value of the first dimension) + (the average value of the second dimension) + (the average value of the third dimension) + (the average value of the fourth dimension) + (the average value of the fifth dimension). The comprehensive scores of all groups are calculated in this way (such as group 2 score 0.71, group 3 score 0.58, and group 4 score 0.67).

[0186] Step 503: Sort all optimized groups according to the comprehensive score from high to low (such as group 2: 0.71→ group 4: 0.67→ group 1: 0.643→ group 3: 0.58).

[0187] A preset qualified threshold (such as 0.6, set according to the enterprise procurement standard) is checked from the top of the ranking:

[0188] Group 2 score 0.71>0.6, meets the condition, and is directly selected as the final supplier group; if group 2 score 0.59<0.6, skip group 2, check group 4 (0.67>0.6), select group 4; if all groups are lower than the threshold, re-evaluate or adjust the threshold (in actual application, it is usually ensured that at least one group is qualified).

[0189] The three-level evaluation system decomposes the abstract "comprehensive ability" into quantifiable criteria and indicators, combines the weight distribution of the analytic hierarchy process, makes the evaluation logic clear and the hierarchy distinct, avoids the one-sidedness of single indicator decision-making; pairwise comparison and consistency check ensure the rationality of the weight, reduce subjective speculation, make the importance of the five indicators more objective, and meet the actual priority of enterprise procurement (such as higher weight of transaction cost).

[0190] Based on the weighted calculation of the average value of the indicators in the group, the overall level of the group is reflected, and the multi-dimensional information is integrated through the weight, so that the score can accurately measure the comprehensive advantage of the group.

[0191] In a preferred embodiment of the present application, step 6 above, the weighted score of each supplier in the final supplier group is calculated, and the supplier with the highest score is pushed to the procurement terminal; when the amount of new supplier data reaches a preset threshold, the whole process of converting from multi-dimensional data space to supplier individual weighted score is re-executed, which can include:

[0192] Step 600, in the final supplier group, the five indicators of each supplier individual in the group are calculated by a three-level evaluation index weight framework;

[0193] Step 601, according to the five indicators of each supplier individual, all suppliers in the group are sorted, and the supplier information with the highest score is pushed to the enterprise procurement terminal in real time;

[0194] Step 602, continuously monitor the new supplier data volume of the inter-enterprise electronic transaction platform, when the new data volume reaches the preset threshold, combine the new data with the historical data into an updated spatial coordinate set, retrigger the whole process of spatial distribution core coordinate calculation, offset compensation factor generation, cluster center correction, grouping comprehensive score calculation and individual score pushing; if the new data does not reach the threshold, the current optimized grouping result is used to continuously receive supplier data updates.

[0195] In the embodiment of the application, the final supplier group is the group with the highest comprehensive score determined in step 503 (such as containing 8 suppliers), and the five-dimensional coordinate data (i.e. the standardized values of the five indicators) of each supplier in the group needs to be extracted:

[0196] The first dimension: material line price standardized value (such as supplier A is 0.65, supplier B is 0.72, etc.);

[0197] The second dimension: total price standardized value (such as supplier A is 0.58, supplier B is 0.63, etc.);

[0198] The third dimension: payment method standardized value (such as supplier A is 0.80, supplier B is 0.75, etc.);

[0199] The fourth dimension: delivery cycle standardized value (such as supplier A is 0.62, supplier B is 0.59, etc.);

[0200] The fifth dimension: comprehensive score standardized value (such as supplier A is 0.85, supplier B is 0.90, etc.).

[0201] The three-level evaluation index weight framework formed in step 501 (clearly indicating the maximum weight of the five indicators to the target layer, such as the first dimension 0.2, the second dimension 0.2, the third dimension 0.1, the fourth dimension 0.3, and the fifth dimension 0.2) is used to calculate the weighted score of each supplier's five indicators. According to the same logic, the individual weighted total score of all suppliers in the final supplier group is calculated (such as supplier B gets 0.725, supplier C gets 0.650, etc.).

[0202] Step 601, sort the individual weighted total scores of all suppliers in the final supplier group from high to low: for example, the sorting result is: supplier B (0.725) → supplier A (0.682) → supplier D (0.670) → … → supplier H (0.610); select the supplier at the top of the ranking (such as supplier B), organize its detailed information (including original bid data: material line price, total price, payment method, delivery cycle, specific value of comprehensive score, not standardized value), and push it to the enterprise procurement terminal (such as procurement management system, client interface of procurement personnel) in real time. The push information needs to include the basis for the score (such as “because of the best performance in price, delivery cycle and comprehensive score, the weighted score is the highest”), so as to facilitate the procurement personnel to make reference and decision.

[0203] Step 602, connect the inter-enterprise electronic transaction platform in real time through the system interface, and monitor the newly added supplier bid data on the platform: the new data includes the bid information (five indicators such as material line price and total price) of the newly registered supplier, and the bid update data (such as adjusting price, modifying delivery cycle, etc.) of the original supplier; the system automatically counts the number of new data (in “pieces”, each piece corresponds to a complete bid data of a supplier), and compares it with the preset threshold (such as 50, which can be adjusted according to the business volume of the enterprise).

[0204] When the amount of new data reaches the threshold: for example, the new supplier data reaches 50, the system automatically merges these new data with the historical data (original spatial coordinate set) and re-executes the following whole process:

[0205] Standardize the new data to generate an updated spatial coordinate set (steps 100-102);

[0206] Recalculate the spatial distribution core coordinates (step 200), establish the reference direction line and delimit the evaluation range area (steps 201-202); lay out the calibration points, generate the ring-shaped trajectory and offset compensation factor (steps 300-302);

[0207] Optimize the grouping by using the improved clustering algorithm (steps 400-403); construct the weight framework and calculate the grouping comprehensive score to determine the new final supplier group (steps 500-503); calculate the individual score of the suppliers in the new group and push the supplier with the highest score (steps 600-601). When the amount of new data does not reach the threshold: for example, the new data is only 30, the system does not trigger the whole process, continues to use the current optimized grouping result, the final supplier group and the individual score, and at the same time continuously receives and temporarily stores the new data, until the cumulative amount reaches the threshold before updating.

[0208] The individual weighted score is calculated through the three-level weight framework, the importance difference of the five indexes is comprehensively considered, the individual advantage of the supplier can be accurately reflected by the score, and the decision is avoided by a single index; the highest one is pushed according to the score ranking, so that the procurement terminal can obtain the supplier with the best comprehensive performance in the final supplier group, and the procurement decision efficiency is improved. By monitoring the amount of new data and triggering the whole process update, the recommended result can be optimized in real time with market changes (new suppliers join, price adjustment), so as to avoid the recommended deviation caused by outdated data and ensure the accuracy of long-term decision.

[0209] As shown in Figure 2 The embodiment of the application also provides a supplier intelligent recommendation system, comprising:

[0210] A conversion module is configured to map the supplier evaluation index data to a multi-dimensional space, generate independent space coordinates corresponding to each supplier, and form a space coordinate set.

[0211] A range construction module is configured to calculate a space distribution core coordinate based on the space coordinate set, establish two orthogonal reference direction lines with the core coordinate as the origin, and delimit an evaluation range area according to the opening angle formed by the reference direction lines.

[0212] A trajectory construction module is configured to equally arrange five core calibration points along the divergent direction within the evaluation range area, symmetrically arrange two boundary calibration points outside the evaluation range area, form a ring-shaped evaluation trajectory by connecting all seven calibration points in sequence according to the data acquisition time sequence, and extract the curvature feature quantity of the ring-shaped evaluation trajectory to generate an offset compensation factor.

[0213] A grouping optimization module is configured to perform initial grouping on the space coordinate set by using an improved clustering algorithm, correct the positions of the grouping centers according to the offset compensation factor, and output an optimized grouping result.

[0214] A comprehensive score module is configured to construct a three-level evaluation index weight framework according to the analytic hierarchy process, calculate the comprehensive scores of each group based on the optimized grouping result and the weight framework, and determine a final supplier group.

[0215] A supplier pushing module is configured to calculate individual weighted scores of each supplier in the final supplier group, push the supplier with the highest score to the procurement terminal, and re-execute the whole process from multi-dimensional data space conversion to supplier individual weighted score calculation when the amount of new supplier data reaches a preset threshold.

[0216] It should be noted that the system corresponds to the above method, all the implementation manners in the above method embodiment are applicable to this embodiment, and the same technical effects can be achieved.

[0217] The embodiment of the present application also provides a computing device, comprising a processor, a memory storing a computer program, the computer program being executed by the processor to perform the method as described above. All implementation manners in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0218] The embodiment of the present application also provides a computer readable storage medium storing instructions, which, when executed on a computer, cause the computer to perform the method as described above. All implementation manners in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0219] The above is the preferred embodiment of the present application, it should be pointed out that, for those skilled in the art, without departing from the principles of the present application, can make a number of improvements and refinements, these improvements and refinements should also be considered as the protection scope of the present application.

Claims

1. A method for intelligent recommendation of suppliers, characterized in that, The method comprises: Step 1, mapping the supplier evaluation index data to a multi-dimensional space to generate independent space coordinates corresponding to each supplier, forming a space coordinate set; Step 2, based on the space coordinate set, calculating the space distribution core coordinates, establishing two orthogonal reference direction lines with the core coordinates as the origin, and delimiting the evaluation range area according to the opening angle formed by the reference direction lines; Step 3, arranging five core calibration points equidistantly along the divergent direction within the evaluation range area; symmetrically arranging two boundary calibration points outside the evaluation range area; connecting all seven calibration points in series according to the data acquisition time sequence to form a ring-shaped evaluation track; extracting the curvature feature quantity of the ring-shaped evaluation track to generate an offset compensation factor; Step 4, using an improved clustering algorithm to initially group the space coordinate set, correcting the position of each grouping center according to the offset compensation factor, and outputting the optimized grouping result; Step 5, constructing a three-level evaluation index weight framework according to the analytic hierarchy process, and based on the optimized grouping result, applying the weight framework to calculate the comprehensive score of each group to determine the final supplier group; Step 6, calculating the weighted score of each supplier in the final supplier group, and pushing the supplier with the highest score to the procurement terminal; when the amount of new supplier data reaches a preset threshold, re-executing the whole process from multi-dimensional data space conversion to supplier individual weighted score. 2.The vendor intelligent recommendation method of claim 1, wherein, Mapping the supplier evaluation index data to a multi-dimensional space to generate independent space coordinates corresponding to each supplier, forming a space coordinate set, comprising: Collecting the bid index data of the supplier, which includes material line price, total price, payment method, delivery cycle and comprehensive score; Standardizing each bid index data, mapping the standardized material line price to the first-dimensional coordinate value, the total price to the second-dimensional coordinate value, the payment method to the third-dimensional coordinate value, the delivery cycle to the fourth-dimensional coordinate value, and the comprehensive score to the fifth-dimensional coordinate value; Each supplier generates a unique five-dimensional space coordinate point, and all supplier coordinate points constitute a space coordinate set. 3.The vendor intelligent recommendation method of claim 2, wherein, Based on the space coordinate set, calculating the space distribution core coordinates, establishing two orthogonal reference direction lines with the core coordinates as the origin, and delimiting the evaluation range area according to the opening angle formed by the reference direction lines, comprising: Based on the space coordinate set, calculating the arithmetic average position of all independent coordinate points in the five-dimensional space, and defining the corresponding arithmetic average position coordinates as the space distribution core coordinates; Determining the first principal component direction and the second principal component direction of the coordinate set by principal component analysis, constructing the first reference direction line along the first principal component direction and the second reference direction line along the second principal component direction with the core coordinates as the origin, and the two direction lines remain orthogonal; Calculating the included angle between the two reference direction lines to delimit the sector evaluation range area. 4.The method according to claim 3, wherein, Arranging five core calibration points equidistantly along the divergent direction within the evaluation range area; symmetrically arranging two boundary calibration points outside the evaluation range area; Connecting all seven calibration points in series according to the data acquisition time sequence to form a ring-shaped evaluation track; Extracting the curvature feature quantity of the ring-shaped evaluation track to generate an offset compensation factor, comprising: Five core calibration points are arranged along the angular bisector direction in the evaluation range area, including dividing the area arc into five segments; extending radially to the area boundary at each division point; defining the boundary intersection coordinates as the core calibration points; arranging two boundary calibration points outside the evaluation range area, extending 1.2 times the radius of the area along the first reference direction line to generate the first boundary calibration point; extending 1.2 times the radius of the area along the second reference direction line to generate the second boundary calibration point; Seven calibration points are connected in time sequence to form a closed loop path, and the curvature characteristics of the closed loop path are extracted, including maximum curvature and average curvature; According to the numerical combination of the maximum curvature and the average curvature, the preset compensation rule table is queried, and the offset compensation factor is output. 5.The method of claim 4, wherein, The spatial coordinate set is initially grouped by using an improved clustering algorithm, and the center positions of each group are corrected according to the offset compensation factor, and the optimized grouping result is output, including: The spatial coordinate set is input into the clustering algorithm, and the supplier coordinate points are divided into K initial groups through iterative calculation, and the clustering center coordinates of each initial group are output, wherein the number of groups K is determined according to the inflection point of the within-group error sum of squares corresponding to different clustering numbers; The direction vectors between each initial clustering center coordinate and the spatial distribution core coordinate are calculated respectively, and the offset compensation factor is applied to the length of the direction vector to obtain the adjusted length; The new clustering center coordinates are determined according to the adjusted length along the original direction vector from the spatial distribution core coordinate as the starting point; For each supplier coordinate point in the spatial coordinate set, the Euclidean distances between the supplier coordinate point and all new clustering center coordinates are calculated respectively, and the supplier coordinate point is divided into the group corresponding to the new clustering center with the smallest Euclidean distance, forming the optimized grouping result. 6.The method of claim 5, wherein, A three-level evaluation index weight framework is constructed according to the analytic hierarchy process, and based on the optimized grouping result, the weight framework is applied to calculate the comprehensive scores of each group to determine the final supplier group, including: Based on the five-dimensional coordinate set of each supplier group in the optimized grouping result, a three-level evaluation system composed of a target layer, a criterion layer and an index layer is constructed; wherein the target layer corresponds to the supplier comprehensive capability, the criterion layer includes product quality, transaction cost and performance capability, and the index layer is decomposed into five evaluation indexes; The two-by-two comparison judgment matrix of the criterion layer to the target layer and the index layer to the criterion layer is constructed by the analytic hierarchy process, the weight coefficients of each level element are calculated and consistency test is performed, and a three-level evaluation index weight framework is formed; For each optimized group, the coordinate components of all suppliers in the five-dimensional space are extracted, and the five index values are weighted and aggregated according to the weight framework to calculate the comprehensive scores of each group; According to the descending order of the comprehensive scores of all supplier groups, the group with the highest score and the score exceeding the preset qualified threshold is selected as the final supplier group; if the first group does not meet the threshold, the qualified group is selected in turn from the next order. 7.The vendor intelligent recommendation method of claim 6, wherein, The weighted scores of each supplier in the final supplier group are calculated, and the supplier with the highest score is pushed to the procurement terminal. When the amount of newly added supplier data reaches a preset threshold, the whole process of converting from a multi-dimensional data space to individual weighted scores of suppliers is re-executed, including: In the final supplier group, the weighted scores of five indicators of each individual supplier in the group are calculated through a three-level evaluation indicator weight framework; According to the weighted scores of five indicators of each individual supplier, all suppliers in the group are sorted, and the supplier information with the highest score is pushed to the enterprise procurement terminal in real time; The amount of newly added supplier data on the inter-enterprise electronic transaction platform is continuously monitored, and when the amount of newly added data reaches a preset threshold, the newly added data and historical data are combined into an updated spatial coordinate set, triggering the whole process of spatial distribution core coordinate calculation, offset compensation factor generation, cluster center correction, group comprehensive score calculation, and individual score pushing to be re-executed; if the amount of newly added data does not reach the threshold, the current optimized grouping result is used to continuously receive supplier data updates.

8. A vendor intelligent recommendation system, the system implements the method as claimed in any one of claims 1 to 7, characterized in that, It includes: A conversion module for mapping supplier evaluation indicator data to a multi-dimensional space, generating independent spatial coordinates corresponding to each supplier, and forming a spatial coordinate set; A range construction module for calculating spatial distribution core coordinates based on the spatial coordinate set, establishing two orthogonal reference direction lines with the core coordinates as the origin, and defining the evaluation range area according to the opening angle formed by the reference direction lines; A trajectory construction module for equally spacing five core calibration points in the evaluation range area along the divergent direction, symmetrically arranging two boundary calibration points outside the evaluation range area, and forming a ring-shaped evaluation trajectory by connecting all seven calibration points in sequence according to the data acquisition time sequence, and extracting the curvature feature quantity of the ring-shaped evaluation trajectory to generate an offset compensation factor; A grouping optimization module for performing initial grouping on the spatial coordinate set using an improved clustering algorithm, correcting the positions of the centers of each group according to the offset compensation factor, and outputting an optimized grouping result; A comprehensive score module for constructing a three-level evaluation indicator weight framework based on the analytic hierarchy process, calculating the comprehensive scores of each group based on the optimized grouping result using the weight framework, and determining the final supplier group; A supplier pushing module for calculating the weighted scores of individual suppliers in the final supplier group, and pushing the supplier with the highest score to the procurement terminal; When the amount of newly added supplier data reaches a preset threshold, the whole process of converting from a multi-dimensional data space to individual weighted scores of suppliers is re-executed.

9. A computing device, comprising: One or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, so that the one or more processors implement the method of any one of claims 1-7. The computer readable storage medium stores a program which, when executed by a processor, implements the method of any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, ​

Images