Building talent evaluation system based on ideal solution

Abstract

This invention relates to the field of performance analysis technology for construction personnel, specifically to a construction personnel evaluation system based on an ideal solution. In this invention, a graph attention network is used to map personnel numbers to personnel nodes, project numbers to construction project nodes, and the proportions of six responsibilities to performance-related edge attributes. Attention scores between personnel nodes and construction project nodes are calculated. Based on the TOPSIS model, seven scores and seven weights are converted, and positive and negative reference values ​​are locked, providing a unified benchmark for multiple evaluation objects. Mahalanobis distance is used to introduce the joint variation relationship between multiple indicators. An XGBoost regression tree model is used to read the scalar values ​​of job suitability, safety performance score, professional and technical score, collaborative contribution score, positive distance, negative distance, and proximity. These values ​​are combined with five job grade intervals and job capping conditions to generate a roster of appointment levels, transforming the evaluation results from score ranking into a basis for job grade appointment.

Description

Technical Field

[0001] This invention relates to the field of performance analysis technology for construction personnel, and in particular to a construction personnel evaluation system based on ideal solutions. Background Technology

[0002] The field of performance analysis technology for construction personnel aims to apply mathematical evaluation logic to quantitatively characterize the professional abilities of construction practitioners. This includes constructing a mapping model based on project delivery standards and personnel performance behavior, and using a multi-criteria decision-making algorithm to perform value judgments. By determining the objective weights of multiple evaluation dimensions, the evaluation score is output using a schedule-based calculation formula.

[0003] The purpose of the ideal solution-based construction talent evaluation system is to construct a quantitative evaluation scheme for personnel performance using multi-criteria decision-making logic. It involves big data management, and by establishing positive and negative ideal solutions as performance extreme value benchmarks, and quantitatively calculating the geometric distance of multiple evaluation objects to the positive and negative ideal solutions, it can accurately classify the job adaptability, task completion and professional and technical level of the participants. The scheme achieves logically consistent evaluation scores, providing a quantitative basis for the allocation of human resources in construction projects. After the accumulation of the above-mentioned cross-project, cross-stage and cross-position evaluation records, the existing big data management usually only saves personnel scores and project results, without establishing a common source index for the job responsibility chain list, job weight matrix, Mahalanobis proximity sequence, threshold verification results, job capping conditions and appointment level roster, resulting in insufficient basis for score batch traceability, weight version audit and job level boundary review.

[0004] Existing performance analysis of construction personnel is mostly based on the mapping of project delivery standards and personnel performance behavior. It focuses on determining the objective weights of multiple evaluation dimensions, setting positive and negative ideal solutions, and calculating the geometric distance from the evaluation object to the two ideal solutions. In actual operation, it is easy to compress personnel performance behavior into multiple score fields. There is a lack of chain-like consistency between job sequences, project types, job positions, and project stages. This increases the risk of mixing records of the same person across projects, stages, and positions. Moreover, existing technologies usually rely on fixed weights, static objective weights, or single weight calculations. Safety officers, cost estimators, and technical managers have different boundaries in safety, cost, and technical projects, but the weight structure is difficult to adjust with changes in job responsibility share. This can easily lead to evaluation biases such as high task completion scores masking insufficient safety performance and high technical scores masking insufficient coordination. The calculation of ideal solution distance focuses on geometric distance. If quality performance is highly synchronized with technical review pass rate and safety performance is highly synchronized with rectification closure performance, duplicate contributions may amplify the closeness. If there is no job cap and job level boundary calibration after score ranking, it will affect the stability of human resource allocation judgment in construction projects. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing an architectural talent evaluation system based on ideal solutions.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: An architectural talent evaluation system based on ideal solutions includes:

[0007] Duty chain construction module: Based on the construction talent number, it performs consistency verification on the job sequence number, project business code, duty position code, and project stage code, registers the proportion of safety responsibility, quality responsibility, schedule responsibility, cost responsibility, technical responsibility, and collaborative responsibility, collects duty-related edges, and generates a job responsibility chain list;

[0008] Responsibility weight configuration module: Based on the job responsibility chain list, a graph attention network is used to calculate the responsibility share of project manager, safety officer, cost engineer, and technical person in charge, verify the bottom line items of the job, determine the evaluation weights of seven items: safety, quality, schedule, cost, technology, collaboration, and task completion, and generate a job performance weight matrix.

[0009] Ideal distance calculation module: Based on the performance weight matrix, the seven scores and weights are converted, the positive reference value and the negative reference value are locked, the personnel vector difference and square root distance are calculated, and the Mahalanobis proximity sequence is obtained.

[0010] The proximity constraint correction module: Based on the Mahalanobis proximity sequence, it verifies the thresholds of safety performance score, quality performance score, collaborative contribution score, professional and technical score, and task completion score, determines the restricted level, and obtains the job suitability scalar.

[0011] Job grade boundary determination module: Based on the job matching scalar, the XGBoost regression tree model is used to determine the five job grade intervals, verify the job capping conditions, take the lower-order result according to the category number, and generate a list of appointment grades.

[0012] As a further aspect of the present invention, the optimized parameter set includes the adjustment of coating speed, coating thickness, and paint type selection; the job responsibility chain list includes construction talent number, job sequence number, project business code, job performance code, project stage code, and six responsibility percentages; the job performance weight matrix is ​​specifically a matrix formed by seven evaluation dimensions and corresponding evaluation weights; the Mahalanobis proximity sequence includes personnel number, positive ideal solution distance, negative ideal solution distance, Mahalanobis proximity degree, and basic level; the job fit scalar specifically refers to a value used to characterize the degree of matching between construction talent and target job; and the appointment level roster includes personnel number, target job, comprehensive evaluation score, job level category, and job capping result.

[0013] As a further aspect of the present invention, the duty performance chain construction module includes:

[0014] The job performance field verification submodule: Based on the construction talent number, it retrieves the job sequence number, project business code, job performance code and project stage code item by item, checks the consistency of fields according to the same person, the same project and the same stage, marks the records of missing fields, duplicate numbers and conflicting job stages, and obtains the job performance verification list.

[0015] Responsibility Chain Generation Submodule: Based on the performance verification checklist, register the proportions of safety responsibility, quality responsibility, schedule responsibility, cost responsibility, technical responsibility and collaboration responsibility, verify the total value of the proportions of the six responsibilities, collect the performance association edges between personnel and projects, and generate a job responsibility chain.

[0016] As a further aspect of the present invention, the responsibility weight configuration module includes:

[0017] Share Conversion Submodule: Based on the aforementioned job responsibility chain list, using a graph attention network, it reads the corresponding personnel number, project number, job stage identifier, and six responsibility percentages for project manager, safety officer, cost estimator, and technical lead. It then categorizes the safety responsibility, quality responsibility, schedule responsibility, cost responsibility, technical responsibility, and collaborative responsibility percentages by job category, verifies whether the percentage of each individual item is within the range of 0 to 100%, verifies whether the total of the six percentages is 100%, removes null records, duplicate job records, and project stage mismatch records, and obtains a job share list.

[0018] Bottom-line verification submodule: Based on the job share list, match job bottom-line items according to the job categories of project manager, safety officer, cost engineer, and technical manager. Project manager matches collaboration bottom line and task completion bottom line, safety officer matches safety bottom line, cost engineer matches cost bottom line, and technical manager matches technical bottom line and quality bottom line. Bind job bottom-line items with responsibility share to obtain bottom-line constraint table;

[0019] The weighted table submodule: Based on the bottom line constraint table, it retrieves seven evaluation items: safety, quality, schedule, cost, technology, collaboration, and task completion. It determines the basic weights of the seven evaluation items according to the responsibility share, corrects the weights of evaluation items that are lower than the bottom line requirements of the position according to the bottom line ratio, and normalizes the seven weights to a total value of 1 to generate a performance weight matrix.

[0020] As a further embodiment of the present invention, the graph attention network first maps personnel numbers to talent nodes and project numbers to construction project nodes. It then maps job stage identifiers, safety responsibility percentages, quality responsibility percentages, schedule responsibility percentages, cost responsibility percentages, technical responsibility percentages, and collaborative responsibility percentages to performance-related edge attributes. Job category codes are written to talent nodes, project type codes are written to construction project nodes, and six responsibility percentage vectors are written to the performance-related edges. An adjacency index from talent nodes to construction project nodes is established. A node linear transformation matrix is ​​used to generate job representation vectors and project representation vectors respectively, and an edge attribute transformation matrix is ​​used to generate responsibility representation vectors. Finally, the job representation vector, project representation vector, and responsibility representation vector are concatenated into an edge input. The input vector is used to calculate the original edge attention score through the attention parameter vector. For the same talent node, multiple construction project nodes are connected to form a subarray and normalized index processing is performed to obtain the attention scores of multiple job performance related edges. The attention scores are used to weight and converge the construction project node representation vector and the responsibility representation vector to form the job responsibility aggregation vector. The job responsibility aggregation vector is split into safety responsibility, quality responsibility, schedule responsibility, cost responsibility, technical responsibility and collaboration responsibility according to the job categories of project manager, safety officer, cost engineer and technical person in charge. Combined with the boundary verification of the single item proportion from 0 to 100%, the verification of the total value of the six items as 100%, the removal of empty value records, the removal of duplicate job records and the removal of project stage mismatch records, the job share list is obtained.

[0021] As a further aspect of the present invention, the ideal distance calculation module includes:

[0022] The score conversion submodule: Based on the performance weight matrix, the TOPSIS model is used to read the scores and corresponding weights of seven items: safety, quality, schedule, cost, technology, collaboration, and task completion. The scores and weights of the same person in the same job category are matched item by item. The seven weighted scores and unweighted scores are retained. Missing scores, out-of-bounds scores, and weight mismatch items are marked to obtain the personnel conversion vector.

[0023] Reference value submodule: Based on the personnel conversion vector, divide the evaluation items into positive evaluation items and negative evaluation items according to seven evaluation items: safety, quality, schedule, cost, technology, collaboration, and task completion. For positive evaluation items, select the high value as the positive reference value and select the low value as the negative reference value. For negative evaluation items, select the low value as the positive reference value and select the high value as the negative reference value to obtain the reference value group.

[0024] Distance Array Submodule: Based on the reference value group, retrieve the seven weighted scores in the personnel converted vector, calculate the difference between the personnel vector and the positive reference value and the difference between the personnel vector and the negative reference value, perform square accumulation, square root distance measurement and positive and negative distance ratio conversion on multiple differences, arrange the closeness values ​​according to the personnel number, and obtain the Mahalanobis closeness sequence.

[0025] As a further aspect of the present invention, the TOPSIS model first reads the personnel number, job category, safety score, quality score, schedule score, cost score, technical score, collaboration score, task completion score, and seven evaluation weights. It then establishes scoring rows by personnel number and job category, verifies that the seven scores range from 0 to 100, and verifies that the sum of the seven weights is 1. Each of the seven scores is multiplied by its weight to form a weighted score. Finally, it summarizes the original scores, weights, weighted scores, and verification identifiers to obtain a personnel conversion vector. Based on this personnel conversion vector, positive evaluation items are then defined. For positive evaluation items, high values ​​are selected as positive reference values ​​and low values ​​as negative reference values. For negative evaluation items, low values ​​are selected as positive reference values ​​and high values ​​as negative reference values ​​to obtain a reference value group. Based on the reference value group, the seven weighted scores in the personnel converted vector are retrieved. The difference between the personnel vector and the positive reference value and the difference between the personnel vector and the negative reference value are calculated respectively. The sum of squares and the square root of the difference are calculated for multiple differences. The proximity value is converted by dividing the negative distance by the sum of the positive distance and the negative distance. The personnel numbers are arranged from high to low according to the proximity value to obtain the Mahalanobis proximity sequence.

[0026] As a further aspect of the present invention, the proximity constraint correction module includes:

[0027] Performance threshold verification submodule: Based on the Mahalanobis proximity sequence, read the personnel number, proximity value and basic sorting number, compare the thresholds of safety performance score, quality performance score, collaborative contribution score, professional and technical score and task completion score, list the items below the threshold, threshold difference and trigger level, and obtain the performance threshold list.

[0028] Adaptation Scalar Generation Submodule: Based on the performance threshold list, it matches the restricted level, registers the adaptation deduction value and the retained item value, determines the deduction range according to the number of low-scoring items, determines the adaptation value according to the trigger level, and obtains the job adaptation scalar.

[0029] As a further aspect of the present invention, the job level boundary determination module includes:

[0030] Job grade interval attribution submodule: Based on the job matching scalar, using the XGBoost regression tree model, read the matching values ​​and the personnel's target job, match five types of job grade intervals, mark the boundaries of four intervals of 60 points, 70 points, 80 points and 90 points, record the matched interval and category number, and obtain the job grade interval list;

[0031] Appointment level generation submodule: Based on the job level range list, superimposed job capping conditions, verifying the corresponding capping categories for safety performance points, professional and technical points, and collaborative contribution points, comparing the matched job level sequence number and the capping category sequence number, taking the lower-order category, and generating an appointment level roster.

[0032] As a further aspect of the present invention, the XGBoost regression tree model first reads the personnel number, target position, fit value, safety performance score, professional and technical score, collaborative contribution score, forward distance, backward distance, and proximity value. The historical job level evaluation score is set as the monitoring target. The fit value, safety performance score, professional and technical score, collaborative contribution score, forward distance, backward distance, and proximity value are used to form an input vector. Fifty to two hundred regression trees are set up. The squared error loss function is used to calculate the difference between the predicted score and the monitoring target. In the t-th round of tree construction, the first-order gradient and second-order gradient are calculated based on the predicted score of the previous round. The second-order Taylor expansion is used to approximate the objective function using the second-order Taylor expansion. Candidate split points are set for the fit value, safety performance score, professional and technical score, collaborative contribution score, and closeness value. The split gain of the nodes corresponding to multiple candidate split points is calculated. The split point with the first gain ranking and the number of leaf node samples meets the lower limit condition is selected to generate tree nodes. Leaf node weights are written for each leaf node. Fifty to two hundred regression trees are accumulated to output the leaf node weights to form the comprehensive evaluation score. The comprehensive evaluation score is matched with the boundaries of four intervals: 60, 70, 80, and 90 points. The matched intervals and category numbers are recorded to obtain a list of job level intervals.

[0033] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0034] In this invention, the construction talent number is used as an index to perform consistency verification on the job sequence number, project business code, job performance code, and project stage code, and to register the proportion of safety responsibility, quality responsibility, schedule responsibility, cost responsibility, technical responsibility, and collaborative responsibility, so that the personnel's performance facts are expanded from a single score record to a data chain jointly defined by job, project, stage, and responsibility.

[0035] In this invention, a graph attention network is used to map personnel numbers to talent nodes, project numbers to construction project nodes, and the proportion of six responsibilities to job performance-related edge attributes. The attention score between talent nodes and construction project nodes is calculated so that the weights of the seven evaluation criteria—safety, quality, schedule, cost, technology, collaboration, and task completion—change with job category, project type, and share of responsibility.

[0036] In this invention, seven scores and seven weights are converted based on the TOPSIS model, and positive and negative reference values ​​are locked to provide a unified reference benchmark for multiple evaluation objects. Mahalanobis distance is used to introduce the joint variation relationship between multiple indicators, reducing the impact of repeated measurement of strongly correlated indicators such as safety performance, quality performance, technical review, and rectification closure in distance calculation;

[0037] In this invention, an XGBoost regression tree model is used to read the job fit scalar, safety performance score, professional and technical score, collaborative contribution score, forward distance, backward distance, and proximity score. The comprehensive evaluation score is generated by using Taylor second-order expansion, node split gain, and leaf node weight accumulation. The appointment level list is generated by combining five job level ranges and job capping conditions, so that the evaluation results are transformed from score ranking into job level appointment basis. Attached Figure Description

[0038] Figure 1 This is a system flowchart of the present invention. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0040] Example 1

[0041] Please see Figure 1 This invention provides a technical solution: a construction talent evaluation system based on ideal solutions, comprising:

[0042] Duty chain construction module: Based on the construction talent number, it performs consistency verification on the job sequence number, project business code, duty position code, and project stage code, registers the proportion of safety responsibility, quality responsibility, schedule responsibility, cost responsibility, technical responsibility, and collaborative responsibility, collects duty-related edges, and generates a job responsibility chain list;

[0043] Responsibility weight configuration module: Based on the job responsibility chain list, using graph attention network, calculate the responsibility share of project manager, safety officer, cost engineer, and technical person in charge, verify the bottom line items of the job, determine the evaluation weights of seven items: safety, quality, schedule, cost, technology, collaboration, and task completion, and generate a performance weight matrix;

[0044] Ideal distance calculation module: Based on the performance weight matrix, the seven scores and weights are converted, the positive reference value and the negative reference value are locked, the personnel vector difference and square root distance are calculated, and the Mahalanobis proximity sequence is obtained.

[0045] Proximity Constraint Correction Module: Based on the Mahalanobis proximity sequence, it verifies the thresholds of safety performance score, quality performance score, collaborative contribution score, professional and technical score, and task completion score, determines the restricted level, and obtains the job suitability scalar.

[0046] Job grade boundary determination module: Based on job matching scalar, using XGBoost regression tree model, it determines the affiliation of five job grade intervals, verifies the job capping conditions, takes the lowest result according to the category number, and generates a list of appointment grades.

[0047] The optimized parameter set includes adjustments to the coating speed, coating thickness, and paint type selection. The job responsibility chain list includes the construction talent number, job sequence number, project business code, job performance code, project stage code, and the proportion of six responsibilities. The job performance weight matrix is ​​specifically a matrix formed by seven evaluation dimensions and their corresponding evaluation weights. The Mahalanobis proximity sequence includes the personnel number, positive ideal solution distance, negative ideal solution distance, Mahalanobis proximity degree, and basic level. The job fit scalar specifically refers to the numerical value used to characterize the degree of matching between construction talents and target positions. The appointment level roster includes the personnel number, target position, comprehensive evaluation score, job level category, and job cap result.

[0048] The duty performance chain construction module includes:

[0049] The performance field verification submodule: Based on the construction talent ID, it retrieves the job sequence number, project business code, performance position code, and project stage code item by item. It verifies the consistency of fields for the same person, the same project, and the same stage. It checks whether the job sequence number corresponds to the registered position, whether the project business code corresponds to the project type, whether the performance position code corresponds to the job responsibilities, and whether the project stage code corresponds to the stage in which the performance occurred. Records of missing job sequence numbers, project business codes, performance position codes, and project stage codes are listed as field missing records. Records of duplicate registrations of the same construction talent ID in the same project stage are listed as number duplicate records. Records of multiple performance position codes corresponding to the same person in the same project stage are listed as position stage conflict records. The verification pass records, field missing records, number duplicate records, and position stage conflict records are compiled by construction talent ID to obtain a performance verification list. The performance verification list includes the construction talent ID, job sequence number, project business code, performance position code, project stage code, verification status, field missing records, number duplicate records, and position stage conflict records.

[0050] The responsibility chain generation submodule: Based on the performance verification checklist, it extracts the construction talent number, project business code, job position code, and project stage code from the verification pass records. It registers the percentages of safety responsibility, quality responsibility, schedule responsibility, cost responsibility, technical responsibility, and collaborative responsibility. It verifies whether the percentage of each individual responsibility falls within the range of 0 to 100, and whether the total percentage of the six responsibilities is 100. Records with a total value not equal to 100 are written as percentage anomalies. It binds the responsibility percentage records of the same person in the same project stage to the job position code. It establishes a performance association edge between the construction talent number, project business code, and project stage code, and writes this association edge into the percentages of safety responsibility, quality responsibility, schedule responsibility, cost responsibility, technical responsibility, and collaborative responsibility. This generates a job responsibility chain list, which includes the construction talent number, project business code, job position code, project stage code, safety responsibility percentage, quality responsibility percentage, schedule responsibility percentage, cost responsibility percentage, technical responsibility percentage, collaborative responsibility percentage, and the performance association edge.

[0051] The responsibility weight configuration module includes:

[0052] The share conversion submodule, based on a job responsibility chain list, uses a graph attention network and the GATConv operator to read the personnel ID, project ID, job stage identifier, and six responsibility percentages for project managers, safety officers, cost estimators, and technical leaders. It writes the personnel ID into the talent node index, the project ID into the project node index, the job stage identifier into the stage code, and the percentages of safety responsibility, quality responsibility, schedule responsibility, cost responsibility, technical responsibility, and collaborative responsibility into the edge attribute vector. The input channel is set to the job category code dimension plus the project type code dimension, and the output channel is set to 16. Set the number of attention heads to 4, the edge attribute dimension to 6, the negative slope to 0.2, the deactivation ratio to 0.1, disable self-loop appending, create an edge index from talent node to project node, perform edge attention score calculation and normalize the scores of adjacent projects of the same talent node, and merge the proportions of safety responsibility, quality responsibility, schedule responsibility, cost responsibility, technical responsibility and collaboration responsibility according to job category, verify whether the proportion of each item is in the range of 0 to 100%, verify whether the total value of the six items is 100%, remove null records, duplicate job records and project stage mismatch records, and obtain a job share list;

[0053] Bottom-line verification submodule: Based on the job quota list, a matching key is established by personnel number and job category. The matching key for project manager is set to the collaboration bottom line and task completion bottom line, the matching key for safety officer is set to the safety bottom line, the matching key for cost engineer is set to the cost bottom line, and the matching key for technical person in charge is set to the technical bottom line and quality bottom line. The safety bottom line ratio is set to 0.30, the cost bottom line ratio is set to 0.25, the technical bottom line ratio is set to 0.25, the quality bottom line ratio is set to 0.20, the collaboration bottom line ratio is set to 0.20, and the task completion bottom line ratio is set to 0.20. The six responsibility quotas are read one by one, matched with the job bottom line items, and the bottom line items, bottom line ratios, restriction levels, and responsibility quota reference identifiers are bound to obtain the bottom line constraint table.

[0054] The weighted table submodule, based on the bottom-line constraint table, retrieves seven evaluation items: safety, quality, schedule, cost, technology, collaboration, and task completion. It writes six responsibility shares, bottom-line percentages, job category codes, and restriction levels into the gated input items, sets the number of gated outputs to 7, sets the lower limit of the seven basic weights to 0.05, sets the upper limit of individual weights to 0.45, sets the lower limit of the safety weight for safety officers to 0.30, the lower limit of the cost weight for cost estimators to 0.25, the lower limit of the technical weight for technical leaders to 0.25, and the lower limit of the collaboration weight for project managers to 0.20. It generates basic weights for the seven evaluation items based on responsibility shares, corrects the weights of evaluation items below the job bottom-line requirements based on the bottom-line percentage, truncates values ​​exceeding the upper limit of individual weights, and normalizes the seven weights to a total value of 1, generating a performance weight matrix.

[0055] The graph attention network first maps personnel IDs to talent nodes and project IDs to construction project nodes. It then maps job stage identifiers, safety responsibility percentages, quality responsibility percentages, schedule responsibility percentages, cost responsibility percentages, technical responsibility percentages, and collaboration responsibility percentages to performance-related edge attributes. Job category codes are written for talent nodes, project type codes for construction project nodes, and six responsibility percentage vectors are written for the performance-related edges. An adjacency index from talent nodes to construction project nodes is established. Linear transformation matrices are used to generate job and project representation vectors, respectively, and edge attribute transformation matrices are used to generate responsibility representation vectors. The job, project, and responsibility representation vectors are concatenated into an edge input vector, which is then processed by the attention network. The force parameter vector is used to calculate the original edge attention score. For the same talent node, multiple construction project nodes are connected to form a subarray and normalized index processing is performed to obtain the attention scores of multiple job performance related edges. The attention scores are used to weight and converge the construction project node representation vector and the responsibility representation vector to form the job responsibility aggregation vector. The job responsibility aggregation vector is split into safety responsibility, quality responsibility, schedule responsibility, cost responsibility, technical responsibility and collaboration responsibility according to the job categories of project manager, safety officer, cost engineer and technical person in charge. Combined with the boundary verification of the single item proportion from 0 to 100%, the verification of the total value of the six items as 100%, the removal of empty value records, the removal of duplicate job records and the removal of project stage mismatch records, the job share list is obtained.

[0056] The ideal distance calculation module includes:

[0057] The score conversion submodule, based on the job performance weight matrix and the TOPSIS model, reads the scores and corresponding weights for seven categories: safety, quality, schedule, cost, technology, collaboration, and task completion. It sets the range of the seven scores to 0-100, the total weight value to 1, the weight total error limit to 0.0001, and the score decimal places to 4. Evaluation rows are created by personnel number and job category, and evaluation columns are created by safety, quality, schedule, cost, technology, collaboration, and task completion. The module pairs the seven scores with the seven weights for the same personnel and job category, scales the seven scores to the 0-1 range, multiplies the scaled scores by their corresponding weights, retains the weighted and unweighted scores, marks empty values ​​as missing scores, marks scores below 0 and above 100 as out-of-bounds scores, and marks weight totals not equal to 1 as weight mismatches. Finally, it obtains the personnel conversion vector.

[0058] The reference value submodule: Based on the personnel conversion vector, a reference column is established according to seven evaluation items: safety, quality, schedule, cost, technology, collaboration, and task completion. Safety, quality, schedule, technology, collaboration, and task completion are set as positive evaluation items, and cost is set as a negative evaluation item. The weighted score corresponding to each evaluation item is read. For positive evaluation items, the highest value in the column is selected as the positive reference value, and the lowest value in the column is selected as the negative reference value. For negative evaluation items, the lowest value in the column is selected as the positive reference value, and the highest value in the column is selected as the negative reference value. The decimal places of the reference value are set to 4 digits. The reference value source identifier is set as a combination field of personnel number, job category, and evaluation item name. The evaluation item name, evaluation item type, positive reference value, negative reference value, and reference value source identifier are registered to obtain the reference value group.

[0059] The Distance Array Submodule: Based on the reference value group, it retrieves the weighted scores of seven items—safety, quality, schedule, cost, technology, collaboration, and task completion—from the personnel conversion vector. A personnel evaluation matrix is ​​established based on these seven weighted scores, with a matrix dimension of seven rows and seven columns and a matrix stability coefficient of 0.01. The covariance matrix between the seven evaluation items is calculated, and the matrix stability coefficient is written to the main diagonal of the covariance matrix. The inverse covariance matrix is ​​then obtained. The personnel vector is subtracted item by item from both the forward and reverse reference values ​​to generate forward and reverse difference vectors. The forward and reverse distances are calculated based on the inverse covariance matrix, with the distances retained to four decimal places. The proximity value is converted according to the ratio of the reverse distance to the sum of the forward and reverse distances, with the proximity value also retained to four decimal places. The proximity values ​​are arranged by personnel number to obtain the Mahalanobis proximity sequence.

[0060] The TOPSIS model first reads the personnel ID, job category, safety score, quality score, schedule score, cost score, technical score, collaboration score, task completion score, and seven evaluation weights. It then creates scoring rows by personnel ID and job category, verifies that the seven scores range from 0 to 100, and verifies that the sum of the seven weights is 1. Each of the seven scores is multiplied by its weight to form a weighted score. Finally, it summarizes the original scores, weights, weighted scores, and verification identifiers to obtain a personnel conversion vector. Based on this vector, it categorizes evaluation items into positive and negative evaluation items. For positive evaluation items, high values ​​are selected as positive reference values ​​and low values ​​are selected as negative reference values. For negative evaluation items, low values ​​are selected as positive reference values ​​and high values ​​are selected as negative reference values ​​to obtain a reference value group. Based on the reference value group, the seven weighted scores in the personnel converted vector are retrieved. The difference between the personnel vector and the positive reference value and the difference between the personnel vector and the negative reference value are calculated respectively. The sum of squares and the square root of the difference are calculated for multiple differences. The proximity value is converted by dividing the negative distance by the sum of the positive distance and the negative distance. The personnel numbers are arranged from high to low according to the proximity value to obtain the Mahalanobis proximity sequence.

[0061] The proximity constraint correction module includes:

[0062] The performance threshold verification submodule, based on Mahalanobis proximity sequences and a random forest classifier, reads personnel ID, proximity score, basic ranking number, safety performance score, quality performance score, collaborative contribution score, professional and technical score, and task completion score. It sets the number of decision trees to 150, the maximum depth of a single tree to 6 layers, the minimum number of leaf node samples to 10, the Gini coefficient decrease criterion for node splitting, and sets the safety performance score threshold and the quality performance score threshold to 70 points. Set the threshold for collaborative contribution score to 70 points, the threshold for professional and technical score to 75 points, and the threshold for task completion score to 70 points. Write the proximity value and the five performance scores into the tree node input items. Merge the output categories of multiple decision trees according to the number of low-scoring items, the type of low-scoring items, and the basic sorting number. Compare the safety performance score, quality performance score, collaborative contribution score, professional and technical score, and task completion score thresholds item by item. List the items below the threshold, the threshold difference, and the trigger level to obtain the performance threshold list.

[0063] The adaptation scalar generation submodule, based on the performance threshold list and majority voting rules, reads the personnel number, items below the threshold, threshold difference, trigger level, proximity value, and basic sorting number. It sets a deduction of 5 points for a single low score, 10 points for two low scores, 15 points for three low scores, and 20 points for four or more low scores. It sets the trigger level for safety performance to Level 1 restricted, quality performance to Level 2 restricted, collaborative contribution to Level 2 restricted, professional and technical performance to Level 1 restricted, and task completion to Level 3 restricted. It reads the restriction level according to the trigger level, determines the deduction range according to the number of low-scoring items, registers the adaptation deduction value according to the threshold difference, registers the retained item value according to the non-triggered items, converts the proximity value to the proximity benchmark value, deducts the adaptation deduction value from the proximity benchmark value, and registers the adaptation value to obtain the job adaptation scalar.

[0064] The job grade boundary determination module includes:

[0065] The job grade range attribution submodule, based on a job fit scalar, employs an XGBoost regression tree model. It reads the personnel number, fit value, target job, safety performance score, professional and technical score, collaborative contribution score, forward distance, backward distance, and proximity score. The module sets the number of regression trees to 120, the maximum depth of a single tree to 6 levels, the learning rate to 0.05, the minimum number of leaf node samples to 10, the subsample extraction ratio to 0.8, the column extraction ratio to 0.8, the objective function to squared error regression, and the evaluation score boundary to 0 to 100. The input vector order is: fit value, safety performance score, professional and technical score, collaborative contribution score, forward distance, backward distance, and proximity score, arranged according to personnel number. A sample index is established, and job labels are created according to the personnel's target positions. The input vector is fed into 120 regression trees for round-by-round calculation. In each round, the prediction residual of the previous round is read, and the first-order gradient and second-order gradient are calculated. The splitting field and splitting threshold are selected based on the node splitting gain. The weights of the tree nodes of multiple regression trees are accumulated to obtain the comprehensive evaluation score. Five job grade intervals are matched according to the boundaries of four intervals: 60, 70, 80 and 90 points. Those scoring below 60 points are marked as job re-evaluation personnel, those scoring 60 to 69 points are marked as restricted appointment personnel, those scoring 70 to 79 points are marked as job competent personnel, those scoring 80 to 89 points are marked as professional backbone personnel, and those scoring 90 points and above are marked as candidates for core project leaders. The hit interval and category number are recorded to obtain a list of job grade intervals.

[0066] The appointment level generation submodule, based on the job level range list, reads the personnel number, target position, comprehensive performance evaluation score, matched job level, category number, safety performance score, professional and technical score, and collaborative contribution score. It sets the candidate category number for project core leaders to 5, the professional backbone category number to 4, the job competent personnel category number to 3, the restricted appointment personnel category number to 2, and the job re-evaluation personnel category number to 1. It also sets a cap of 70 points for safety performance score, 75 points for professional and technical score, and 70 points for collaborative contribution score, and then verifies these parameters. The safety performance score corresponds to the capped category. If the safety performance score is below 70, it is written into the capped category number 3 for competent personnel. The professional and technical score is checked against the capped category. If the professional and technical score is below 75, it is written into the capped category number 4 for professional backbone. The collaborative contribution score is checked against the capped category. If the collaborative contribution score is below 70, it is written into the capped category number 4 for professional backbone. The hit job grade number and the capped category number are compared. The category number with the lower value is taken as the appointment grade number. The personnel number, personnel target position, comprehensive evaluation score, hit job grade, capped category and appointment grade are registered to generate an appointment grade roster.

[0067] The XGBoost regression tree model first reads the personnel number, target position, fit score, safety performance score, professional and technical score, collaborative contribution score, forward distance, backward distance, and proximity score. Historical job level evaluation scores are set as the monitoring target. The fit score, safety performance score, professional and technical score, collaborative contribution score, forward distance, backward distance, and proximity score are used to form the input vector. Fifty to two hundred regression trees are set up. The squared error loss function is used to calculate the difference between the predicted score and the monitoring target. In the t-th round of tree construction, the first and second gradients are calculated based on the predicted scores from the previous round. Using Tai... The second-order expansion of the objective function is used to approximate the value. Candidate split points are set for the fit value, safety performance score, professional and technical score, collaborative contribution score, and closeness value. The split gain of the nodes corresponding to multiple candidate split points is calculated. The split point with the first gain ranking and the number of leaf node samples meets the lower limit condition is selected to generate tree nodes. Leaf node weights are written for each leaf node. Fifty to two hundred regression trees are accumulated to output the leaf node weights to form the comprehensive evaluation score. The comprehensive evaluation score is matched with the boundaries of four intervals: 60, 70, 80, and 90 points. The matched intervals and category numbers are recorded to obtain the list of job level intervals.

[0068] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A talent evaluation system for architecture based on ideal solutions, characterized in that: The system includes: Duty chain construction module: Based on the construction talent number, it performs consistency verification on the job sequence number, project business code, duty position code, and project stage code, registers the proportion of safety responsibility, quality responsibility, schedule responsibility, cost responsibility, technical responsibility, and collaborative responsibility, collects duty-related edges, and generates a job responsibility chain list; Responsibility weight configuration module: Based on the job responsibility chain list, a graph attention network is used to calculate the responsibility share of project manager, safety officer, cost engineer, and technical person in charge, verify the bottom line items of the job, determine the evaluation weights of seven items: safety, quality, schedule, cost, technology, collaboration, and task completion, and generate a job performance weight matrix. Ideal distance calculation module: Based on the performance weight matrix, the seven scores and weights are converted, the positive reference value and the negative reference value are locked, the personnel vector difference and square root distance are calculated, and the Mahalanobis proximity sequence is obtained. The proximity constraint correction module: Based on the Mahalanobis proximity sequence, it verifies the thresholds of safety performance score, quality performance score, collaborative contribution score, professional and technical score, and task completion score, determines the restricted level, and obtains the job suitability scalar. Job grade boundary determination module: Based on the job matching scalar, the XGBoost regression tree model is used to determine the five job grade intervals, verify the job capping conditions, take the lower-order result according to the category number, and generate a list of appointment grades.

2. The architectural talent evaluation system based on ideal solution method according to claim 1, characterized in that, The optimized parameter set includes adjustments to the coating speed, coating thickness, and paint type selection. The job responsibility chain list includes the construction talent number, job sequence number, project business code, job performance code, project stage code, and the proportion of six responsibilities. The job performance weight matrix is ​​specifically a matrix formed by seven evaluation dimensions and their corresponding evaluation weights. The Mahalanobis proximity sequence includes the personnel number, positive ideal solution distance, negative ideal solution distance, Mahalanobis proximity degree, and basic level. The job fit scalar specifically refers to a value used to characterize the degree of matching between construction talent and target job. The appointment level roster includes the personnel number, target job, comprehensive evaluation score, job level category, and job cap result.

3. The architectural talent evaluation system based on ideal solution method according to claim 1, characterized in that, The duty performance chain construction module includes: The job performance field verification submodule: Based on the construction talent number, it retrieves the job sequence number, project business code, job performance code and project stage code item by item, checks the consistency of fields according to the same person, the same project and the same stage, marks the records of missing fields, duplicate numbers and conflicting job stages, and obtains the job performance verification list. Responsibility Chain Generation Submodule: Based on the performance verification checklist, register the proportions of safety responsibility, quality responsibility, schedule responsibility, cost responsibility, technical responsibility and collaboration responsibility, verify the total value of the proportions of the six responsibilities, collect the performance association edges between personnel and projects, and generate a job responsibility chain.

4. The architectural talent evaluation system based on ideal solution method according to claim 1, characterized in that, The responsibility weight configuration module includes: Share Conversion Submodule: Based on the aforementioned job responsibility chain list, using a graph attention network, it reads the corresponding personnel number, project number, job stage identifier, and six responsibility percentages for project manager, safety officer, cost estimator, and technical lead. It then categorizes the safety responsibility, quality responsibility, schedule responsibility, cost responsibility, technical responsibility, and collaborative responsibility percentages by job category, verifies whether the percentage of each individual item is within the range of 0 to 100%, verifies whether the total of the six percentages is 100%, removes null records, duplicate job records, and project stage mismatch records, and obtains a job share list. Bottom-line verification submodule: Based on the job share list, match job bottom-line items according to the job categories of project manager, safety officer, cost engineer, and technical manager. Project manager matches collaboration bottom line and task completion bottom line, safety officer matches safety bottom line, cost engineer matches cost bottom line, and technical manager matches technical bottom line and quality bottom line. Bind job bottom-line items with responsibility share to obtain bottom-line constraint table; The weighted table submodule: Based on the bottom line constraint table, it retrieves seven evaluation items: safety, quality, schedule, cost, technology, collaboration, and task completion. It determines the basic weights of the seven evaluation items according to the responsibility share, corrects the weights of evaluation items that are lower than the bottom line requirements of the position according to the bottom line ratio, and normalizes the seven weights to a total value of 1 to generate a performance weight matrix.

5. The architectural talent evaluation system based on ideal solution method according to claim 4, characterized in that, The graph attention network first maps personnel IDs to talent nodes and project IDs to construction project nodes. It then maps job stage identifiers, safety responsibility percentages, quality responsibility percentages, schedule responsibility percentages, cost responsibility percentages, technical responsibility percentages, and collaboration responsibility percentages to performance-related edge attributes. Job category codes are written to talent nodes, project type codes are written to construction project nodes, and six responsibility percentage vectors are written to the performance-related edges. An adjacency index from talent nodes to construction project nodes is established. Linear transformation matrices are used to generate job representation vectors and project representation vectors, respectively. Edge attribute transformation matrices are used to generate responsibility representation vectors. The job representation vector, project representation vector, and responsibility representation vector are concatenated into an edge input vector. After attention... The attention parameter vector is used to calculate the original edge attention score. For the same talent node, multiple construction project nodes are connected to form a subarray and normalized index processing is performed to obtain the attention scores of multiple job performance related edges. The attention scores are used to weight and converge the construction project node representation vector and the responsibility representation vector to form the job responsibility aggregation vector. The job responsibility aggregation vector is split into safety responsibility, quality responsibility, schedule responsibility, cost responsibility, technical responsibility and collaboration responsibility according to the job categories of project manager, safety officer, cost engineer and technical person in charge. Combined with the boundary verification of the single item proportion from 0 to 100%, the verification of the total value of the six items as 100%, the removal of empty value records, the removal of duplicate job records and the removal of project stage mismatch records, the job share list is obtained.

6. The architectural talent evaluation system based on ideal solution method according to claim 1, characterized in that, The ideal distance calculation module includes: The score conversion submodule: Based on the performance weight matrix, the TOPSIS model is used to read the scores and corresponding weights of seven items: safety, quality, schedule, cost, technology, collaboration, and task completion. The scores and weights of the same person in the same job category are matched item by item. The seven weighted scores and unweighted scores are retained. Missing scores, out-of-bounds scores, and weight mismatch items are marked to obtain the personnel conversion vector. Reference value submodule: Based on the personnel conversion vector, divide the evaluation items into positive evaluation items and negative evaluation items according to seven evaluation items: safety, quality, schedule, cost, technology, collaboration, and task completion. For positive evaluation items, select the high value as the positive reference value and select the low value as the negative reference value. For negative evaluation items, select the low value as the positive reference value and select the high value as the negative reference value to obtain the reference value group. Distance Array Submodule: Based on the reference value group, retrieve the seven weighted scores in the personnel converted vector, calculate the difference between the personnel vector and the positive reference value and the difference between the personnel vector and the negative reference value, perform square accumulation, square root distance measurement and positive and negative distance ratio conversion on multiple differences, arrange the closeness values ​​according to the personnel number, and obtain the Mahalanobis closeness sequence.

7. The architectural talent evaluation system based on ideal solution method according to claim 6, characterized in that, The TOPSIS model first reads the personnel number, job category, safety score, quality score, schedule score, cost score, technical score, collaboration score, task completion score, and seven evaluation weights. It then establishes score rows by personnel number and job category, verifies that the seven scores range from 0 to 100, and verifies that the sum of the seven weights is 1. Each of the seven scores is multiplied by its weight to form a weighted score. Finally, it summarizes the original scores, weights, weighted scores, and verification identifiers to obtain a personnel conversion vector. Based on this vector, it categorizes evaluation items into positive and negative evaluation items. For positive evaluation items, high values ​​are selected as positive reference values ​​and low values ​​are selected as negative reference values. For negative evaluation items, low values ​​are selected as positive reference values ​​and high values ​​are selected as negative reference values ​​to obtain a reference value group. Based on the reference value group, the seven weighted scores in the personnel converted vector are retrieved. The difference between the personnel vector and the positive reference value and the difference between the personnel vector and the negative reference value are calculated respectively. The sum of squares and the square root of the difference are calculated for multiple differences. The proximity value is converted by dividing the negative distance by the sum of the positive distance and the negative distance. The personnel numbers are arranged from high to low according to the proximity value to obtain the Mahalanobis proximity sequence.

8. The architectural talent evaluation system based on ideal solution method according to claim 1, characterized in that, The proximity constraint correction module includes: Performance threshold verification submodule: Based on the Mahalanobis proximity sequence, read the personnel number, proximity value and basic sorting number, compare the thresholds of safety performance score, quality performance score, collaborative contribution score, professional and technical score and task completion score, list the items below the threshold, threshold difference and trigger level, and obtain the performance threshold list. Adaptation Scalar Generation Submodule: Based on the performance threshold list, it matches the restricted level, registers the adaptation deduction value and the retained item value, determines the deduction range according to the number of low-scoring items, determines the adaptation value according to the trigger level, and obtains the job adaptation scalar.

9. The architectural talent evaluation system based on ideal solution method according to claim 1, characterized in that, The job level boundary determination module includes: Job grade interval attribution submodule: Based on the job matching scalar, using the XGBoost regression tree model, read the matching values ​​and the personnel's target job, match five types of job grade intervals, mark the boundaries of four intervals of 60 points, 70 points, 80 points and 90 points, record the matched interval and category number, and obtain the job grade interval list; Appointment level generation submodule: Based on the job level range list, superimposed job capping conditions, verifying the corresponding capping categories for safety performance points, professional and technical points, and collaborative contribution points, comparing the matched job level sequence number and the capping category sequence number, taking the lower-order category, and generating an appointment level roster.

10. The architectural talent evaluation system based on ideal solution method according to claim 9, characterized in that, The XGBoost regression tree model first reads the personnel number, target position, fit value, safety performance score, professional and technical score, collaborative contribution score, forward distance, backward distance, and proximity value. It sets the historical job level assessment score as the monitoring target. The fit value, safety performance score, professional and technical score, collaborative contribution score, forward distance, backward distance, and proximity value are used to form the input vector. Fifty to two hundred regression trees are set up. The squared error loss function is used to calculate the difference between the predicted score and the monitoring target. In the t-th round of tree construction, the first and second gradients are calculated based on the predicted scores from the previous round. The Taylor second-order expansion is used to express an approximate value of the objective function. Candidate split points are set for the fit value, safety performance score, professional and technical score, collaborative contribution score, and closeness value. The split gain of the nodes corresponding to multiple candidate split points is calculated. The split point with the first gain ranking and the number of leaf node samples meets the lower limit condition is selected to generate tree nodes. Leaf node weights are written for each leaf node. Fifty to two hundred regression trees are accumulated to output the leaf node weights to form the comprehensive evaluation score. The comprehensive evaluation score is matched with the boundaries of four intervals: 60, 70, 80, and 90 points. The matched intervals and category numbers are recorded to obtain a list of job level intervals.

Images