Method and system for dynamic matching of metrology talents and metrology tasks based on reinforcement learning
By employing a reinforcement learning-based approach, utilizing a multimodal fusion encoder and knowledge graph, and combining GAT and K-means clustering, dynamic matching of metrology personnel with tasks was achieved. This solved the problems of low efficiency and strong subjectivity in traditional matching methods, and improved the quality and efficiency of metrology tasks.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANDONG MEASUREMENT SCI RES INST
- Filing Date
- 2026-06-15
- Publication Date
- 2026-07-14
AI Technical Summary
Traditional methods of matching talent with tasks are inefficient and highly subjective, making it difficult to meet the process-oriented and collaborative needs of metrological research tasks, resulting in low quality and efficiency in the execution of metrological tasks.
We employ a reinforcement learning-based approach, extracting comprehensive features of metrology talents through a multimodal fusion encoder, constructing a knowledge graph, combining GAT and K-means clustering, and using reinforcement learning algorithms to dynamically match tasks with talents, generating a talent combination recommendation list.
It improves the quality and efficiency of metrology task completion, adapts to the team collaboration needs of metrology tasks, ensures that core tasks are prioritized for high-quality personnel, avoids resource waste, and achieves fully automated calculation and dynamic optimization.
Smart Images
Figure CN122390410A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the interdisciplinary field of computer science and metrology research management, and in particular to a method and system for dynamically matching metrology talent with metrology tasks based on reinforcement learning. Background Technology
[0002] Traditional talent-task matching methods rely on manual experience for screening, which suffers from low efficiency, strong subjectivity, and the tendency to overlook key qualifications, making them unsuitable for the large-scale, high-precision metrology research needs. Currently, some industries adopt talent-task matching methods based on keyword matching and simple attribute comparison. Others have proposed constructing a research talent knowledge graph to determine talent matching priorities and then implementing a task-talent matching strategy. However, existing methods mostly match talent with the current stage of the task as it progresses, achieving only a simple match between a single talent and a single task stage. Metrology tasks are multi-role team collaboration projects. In addition to considering the contribution of individual talent's professional abilities to the overall task, the matching process must also focus on the combined matching ability of different talents to balance the overall collaborative efficiency and professional coverage of the talent team. Otherwise, mismatches and ineffective connections between key technical stages can easily occur, hindering the process of metrology tasks and failing to meet the process-oriented and collaborative team matching requirements of metrology tasks, ultimately affecting the quality of metrology R&D task completion. Summary of the Invention
[0003] To address the problems existing in the prior art, this invention provides a dynamic matching method and system for metrology talent and metrology tasks based on reinforcement learning. By extracting refined talent and task features and using reinforcement learning algorithms for team-based dynamic intelligent matching, this invention solves the problem that existing matching methods lack the ability to analyze talent combination matching, which affects the task execution process and completion quality. This method meets the team collaboration needs of metrology tasks and improves the completion quality and efficiency of metrology tasks.
[0004] Firstly, this invention provides a method for dynamically matching metrology talent with metrology tasks based on reinforcement learning.
[0005] A reinforcement learning-based method for dynamically matching metrology talent with metrology tasks includes: Multiple candidate metrology tasks were obtained, and multimodal basic competency data and historical metrology task data of each metrology talent were extracted from the multi-source metrology database. Based on multimodal basic capability data, the basic capability features of metrology talents are extracted through a multimodal fusion encoder, and a metrology talent knowledge graph containing talent nodes and technology-related edges is constructed. The comprehensive features of metrology talents are obtained by GAT aggregation. Based on multiple candidate metrology tasks and historical metrology task data, feature vectorization and clustering are performed, and integrated to obtain comprehensive features of candidate metrology tasks. A reinforcement learning matching model is constructed and trained with historical econometrics task data. The trained matching model is then used to dynamically match the comprehensive features of each candidate econometrics task with the comprehensive features of all econometrics talents according to task priority, thereby generating a recommended list of econometrics talent combinations.
[0006] A further technical solution is that the multimodal basic capability data includes structured data, text data, and numerical data; wherein, the structured data includes professional qualifications, certification status, qualification level, years of service, affiliated institution, and types of equipment proficient; the text data includes professional research direction, description of technical expertise, and published metrological technical papers / patents; the numerical data includes equipment operation proficiency scores, project result acceptance scores, and error analysis accuracy. The historical metrology task data includes the characteristics of the historical metrology tasks, the list of collaborating personnel who completed the historical metrology tasks, the basic ability data of each person in the list of collaborating personnel and their role in the task, the quality of task completion, the task completion cycle, and the metrology equipment used; among which, the characteristics of the metrology tasks include task number, task type, sub-specialty direction, technical indicator requirements, technical difficulty level, required metrology equipment, and completion time limit.
[0007] A further technical solution, the process of extracting the comprehensive characteristics of metrology personnel, includes: Using metrology professionals as nodes and their basic competency characteristics as node features, and establishing technical relationships between nodes based on historical metrology task data, forming technical relationship edges, and building a metrology professional knowledge graph; Based on the knowledge graph of metrology talents, the comprehensive characteristics of candidate metrology talents are obtained by GAT aggregation. The weight of the technology-related edge is a weighted sum of collaboration frequency, technology complementarity, and project fit. The collaboration frequency is the normalized value of the number of historical measurement tasks jointly participated in by the two individuals. The technology complementarity is calculated based on the cosine similarity of the basic ability feature vectors of the two individuals, and then calculated according to the 1-cosine similarity value. The project fit is a weighted value of the matching degree of the roles of the two individuals in the collaborative measurement task and the quality of task completion.
[0008] A further technical solution involves using feature vectorization and clustering based on multiple candidate metrology tasks and historical metrology task data to integrate and obtain comprehensive features for the candidate metrology tasks, including: The features of multiple candidate econometric tasks and historical econometric tasks are vectorized, and K-means clustering is used to cluster all econometric tasks. Historical econometric tasks that are in the same cluster as candidate econometric tasks are selected as similar historical econometric tasks. Among them, K-means clustering uses the sub-specialty direction, task type and technical difficulty level of the econometric task as the core clustering dimensions, and calculates the feature similarity between tasks through Euclidean distance. An attention mechanism is used to weight and fuse the initial features of the candidate measurement task with the features of similar historical measurement tasks to obtain the comprehensive features of the candidate measurement task.
[0009] A further technical solution employs reinforcement learning algorithms to dynamically match multiple candidate metrology tasks with metrology professionals. The process is as follows: Construct a reinforcement learning matching model, including: using the comprehensive feature vector of measurement tasks, the comprehensive feature vector of measurement talents, the priority of measurement tasks, and the state vector of measurement talents as the state space, using different combinations of measurement talents as actions, designing a reward function for task-talent matching, and building an agent and policy network. Using historical measurement task data as the training set, the agent interacts with the environment to sample state, action, and reward samples, and continuously iterates and optimizes the policy network parameters until the model converges. Based on the strategic needs of the metrology field, multiple candidate metrology tasks are prioritized. Using a trained reinforcement learning model, tasks and talents are matched in descending order of priority. The state corresponding to each candidate metrology task is constructed, the corresponding optimal action is matched, the optimal metrology talent combination is determined, and a recommended list of metrology talent combinations for multiple candidate metrology tasks is generated.
[0010] In a further technical solution, the reward function is a weighted sum of feature similarity reward, technology complementarity reward, priority matching reward, talent load penalty, and ability deviation penalty; The action space is a discrete set, formally defined as: ,in, This represents the total number of candidate combinations of metrology talents; For the first Each action corresponds to a set of measurement talents; the action constraint is that the same measurement talent cannot appear in multiple roles in the same task, and a matched measurement talent can only be selected when the task measurement priority is greater than or equal to the priority of its matched measurement tasks.
[0011] Secondly, this invention provides a dynamic matching system for metrology talent and metrology tasks based on reinforcement learning.
[0012] A reinforcement learning-based dynamic matching system for metrology talent and metrology tasks includes: The data acquisition module is used to acquire multiple candidate metrology tasks and extract multimodal basic ability data and historical metrology task data for each metrology talent from the multi-source metrology database. The metrology talent feature extraction module is used to extract the basic ability features of metrology talents based on multimodal basic ability data through a multimodal fusion encoder, and to construct a metrology talent knowledge graph containing talent nodes and technology-related edges. The comprehensive features of metrology talents are obtained by aggregation through GAT. The candidate metrology task feature extraction module is used to vectorize and cluster features based on multiple candidate metrology tasks and historical metrology task data, and integrate them to obtain comprehensive features of candidate metrology tasks. The econometrics task-talent matching module is used to build a reinforcement learning matching model and train it with historical econometrics task data. The trained matching model is used to dynamically match the comprehensive features of each candidate econometrics task with the comprehensive features of all econometrics talents in order of task priority, and generate a recommended list of econometrics talent combinations.
[0013] Thirdly, the present invention also provides an electronic device, comprising: a memory for storing executable instructions; and a processor for implementing the above-mentioned reinforcement learning-based dynamic matching method for metrology talent and metrology tasks when executing the executable instructions stored in the memory.
[0014] Fourthly, the present invention also provides a computer-readable storage medium storing executable instructions for causing a processor to execute the executable instructions to implement the above-described reinforcement learning-based dynamic matching method for metrology talent and metrology tasks.
[0015] Fifthly, the present invention also provides a computer program product comprising executable instructions stored in a computer-readable storage medium; wherein, when the processor of the electronic device reads the executable instructions from the computer-readable storage medium and executes the executable instructions, the aforementioned method for dynamic matching of metrology talent and metrology tasks based on reinforcement learning is implemented.
[0016] The above one or more technical solutions have the following beneficial effects: This invention provides a method and system for dynamic matching of metrology talent and metrology tasks based on reinforcement learning. It comprehensively extracts the integrated features of metrology talent by multimodal fusion encoding of their fundamental abilities and incorporating the technical correlation value between talents using graph attention networks. Simultaneously, it refines and optimizes the features of metrology tasks through K-means clustering and attention mechanisms. Based on the refined feature extraction, a reinforcement learning strategy is employed to make matching decisions between tasks and talent. During the matching process, talent combinations are used as the matching unit. Collaborative talent combinations are built based on the procedural characteristics of metrology tasks, and a reward function is used to enhance the technical complementarity of talent combinations, meeting the needs of multi-role team collaboration in metrology tasks. The solution aims to ensure the quality of metrology R&D tasks from a personnel allocation perspective. Furthermore, the entire matching process dynamically matches tasks from high to low priority, ensuring that core tasks are prioritized for high-quality talent, avoiding waste of core resources. The fully automated calculation process significantly improves the matching efficiency between large-scale tasks and a massive talent pool. Additionally, the matching process updates talent occupancy status in real time to prevent overloading of talent. Multi-dimensional model evaluation indicators and parameter optimization rules are set up, allowing for dynamic model optimization based on actual effects such as matching accuracy, resource utilization, and talent load balance. This ensures the solution can continuously adapt to the diversified tasks and dynamically changing talent capabilities in the metrology field, making it practical for long-term implementation.
[0017] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description
[0018] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.
[0019] Figure 1 This is a flowchart of the dynamic matching method for metrology talents and metrology tasks based on reinforcement learning in Embodiment 1 of the present invention. Detailed Implementation
[0020] It should be noted that the following detailed descriptions are exemplary and are intended only to describe specific embodiments and to provide further explanation of the invention, and are not intended to limit the scope of exemplary embodiments of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0021] Example 1 As the background technology points out, metrology is a fundamental support field for technological innovation and industrial development. It involves coordinating large-scale metrological research tasks and scheduling multi-disciplinary metrological personnel across provincial / national metrology research institutes, large enterprise metrology testing centers, and cross-regional metrology technology research platforms. Taking the centralized allocation of annual metrological research tasks at a provincial metrology institute as an example, this institute typically has several specialized metrology research institutes, including electromagnetics and quantum mechanics, with hundreds of registered metrology technical personnel. Each year, it undertakes dozens or even hundreds of research tasks, including metrological benchmark development, emergency metrological testing technology research, optimization of metrological instrument calibration methods, and improvement of metrological data models. The traditional task execution process is as follows: First, at the beginning of the year, the Research Management Office of the Academy will compile the tasks submitted by each research institute, marking the task priority, professional direction, technical difficulty, number of equipment and personnel required, etc. Among them, there may be cross-regional tasks (such as the task of formulating environmental monitoring metrology standards in multiple regions) and cross-professional tasks (such as the task of measuring and testing high-end equipment involving geometry, electromagnetics, mechanics and other disciplines). Secondly, the Research Management Office, in conjunction with the heads of various research institutes, screened candidates who met the professional requirements from the talent files based on their experience, and verified their professional qualifications, certifications, and other explicit information. Then, a team of personnel is manually assigned to each task. This manual assignment process has several problems: (1) For historically similar tasks, the historical team assignment method is usually followed. However, considering the potential increase or decrease in measurement personnel (such as resignations or new hires), the historical team assignment method becomes ineffective and needs to be reconsidered and reassigned; (2) For new tasks, the assignment method needs to be redesigned. Traditional methods usually only ensure that the number of personnel meets the standard and that the professionals are matched, without clear consideration of team collaboration. Measurement tasks are team collaboration projects, and when unexpected problems occur during execution, the team must collaborate in real time to resolve them. For example, if there are data anomalies during task execution, this... When equipment operators and error analysis personnel need to work together to locate the problem, or when there is a discrepancy between the standards, the person in charge, quality review personnel, and measurement R&D personnel need to work together to adjust the plan. Therefore, the traditional pairing method may result in low team collaboration and troubleshooting efficiency due to the lack of consideration for the collaborative ability between different personnel, which affects the quality of task completion. (3) Since the above allocation is for the selection and coordination of hundreds of metrology personnel and tasks, in order to improve efficiency, it is usually allocated by a team of multiple people from the scientific research management department. However, when multiple managers are allocating, they need to repeatedly coordinate problems such as talent conflicts and uneven talent load, which takes a long time and has low allocation efficiency.
[0022] After the allocation is completed, a task allocation notice will be issued. Metrology personnel will carry out the metrology tasks for the year according to the allocation. After the tasks are completed, the task acceptance results will be archived.
[0023] To address the aforementioned scenarios of metrology task allocation and execution, considering the problems of time-consuming, inefficient, subjective, and prone to overlooking key qualifications in manual allocation, and the fact that existing matching methods applicable to general fields still match single task stages with single talents, which cannot meet the needs of team matching for the process-oriented and collaborative nature of metrology tasks, thus affecting the completion quality and efficiency of metrology R&D tasks, this embodiment proposes a dynamic matching method for metrology talents and metrology tasks based on reinforcement learning. This method comprehensively extracts the comprehensive characteristics of metrology talents by multimodal fusion encoding of the basic capabilities of metrology talents and introducing the technical correlation value between talents using graph attention networks (GAT). Simultaneously, it refines and optimizes the metrology task characteristics through K-means clustering and attention mechanisms. Combined with task priority, reinforcement learning artificial intelligence technology is used to complete the dynamic intelligent matching of candidate metrology tasks and metrology talent combinations, improving matching accuracy and efficiency, optimizing the allocation of metrology research resources, and adapting to the team collaboration needs of the metrology field.
[0024] The dynamic matching method proposed in this embodiment, such as Figure 1 As shown, the specific steps include: Step S1: Obtain multiple candidate metrology tasks and extract multimodal basic capability data and historical metrology task data for each metrology talent from the multi-source metrology database.
[0025] In this embodiment, the server is equipped with a metrology research management platform. This platform serves as the carrier for central task allocation and management functions and interfaces with a pre-set metrology multi-source database on the server. This metrology multi-source database includes a metrology talent file database (including professional qualifications, years of service, certification status, project participation experience, etc.), a metrology research project database (including project type, technical parameters, completed results, etc.), a metrology standard literature database (including standard texts, technical specifications, etc.), a metrology institution cooperation database (including cross-institutional collaborative projects, joint research records, etc.), a metrology equipment operation history database (including equipment model, operation proficiency, maintenance experience, etc.), and other databases. Through this platform, the corresponding metrology tasks and multi-dimensional data of metrology talents stored therein can be directly retrieved.
[0026] Specifically, the process begins by collecting multiple candidate metrology research and development tasks (referred to as candidate metrology tasks, including annual metrology research tasks and cross-regional tasks) within the metrology field. Basic characteristics of each candidate metrology task are extracted, such as structured features like task type, sub-specialty direction, technical indicator requirements, required metrology equipment, completion deadline, resource investment scale, technical difficulty level, and strategic priority, as well as unstructured features like task description text, technical requirement specifications, and standard reference text, forming a candidate metrology task feature library. This approach leverages a multi-source metrology database to achieve unified retrieval of data across all dimensions, integrating data from talent profiles, research projects, standard documents, and equipment operations. This avoids data silos and lays a complete and professional data foundation for subsequent feature extraction, effectively adapting to the professional characteristics and requirements of the metrology industry.
[0027] In this embodiment, the metrology tasks include tasks such as developing metrology standards, researching and developing calibration methods for metrology instruments, tackling key technical challenges in metrology data traceability, and constructing metrology error correction models. As a further implementation, based on the characteristics of each extracted metrology task, multiple task matching priorities can be predefined according to factors such as the strategic priority of the metrology field (e.g., the priority of national metrology benchmark construction), the urgency of the task (e.g., the calibration of emergency testing equipment), the scale of resource investment, and the level of technical difficulty, forming a priority sequence to lay the foundation for subsequent data processing and matching.
[0028] Secondly, data on metrology professionals to be matched is collected, specifically multimodal basic competency data for each metrology professional within the database. This multimodal basic competency data encompasses structured data, text data, and numerical data. Structured data includes professional qualifications, certification status, qualification level, years of experience, affiliated institution, and types of equipment proficient. Text data includes professional research direction, descriptions of technical expertise, and published metrology technical papers / patents. Numerical data includes equipment operation proficiency scores, project outcome acceptance scores, and error analysis accuracy rates. By characterizing the basic competencies of metrology professionals from multiple perspectives, a foundation is laid for subsequent precise feature matching.
[0029] Preferably, the extracted multimodal data of metrology personnel may also include unstructured practical data, such as calibration videos of metrology personnel actually operating equipment. One frame of image is extracted per second from this video data, and the action area of the personnel is extracted using a detection network. Background-irrelevant frames are removed, and a lightweight image compression algorithm is used to reduce the amount of data. Furthermore, existing lightweight CNN (Convolutional Neural Networks) is used to extract action features. These action features are then used as practical sub-features and concatenated with the above-mentioned multiple data features to further enrich the dimensions of talent ability representation, making the basic ability characteristics of talents more in line with the industry characteristics of metrology that emphasize practical operation and equipment operation.
[0030] At the same time, historical metrology task data is extracted from the multi-source metrology database, which includes relevant data of previously completed historical metrology tasks, such as the characteristics of historical metrology tasks, the list of collaborating personnel who completed the historical metrology tasks, the basic ability data of each person in the list of collaborating personnel and their roles in the tasks (as well as role matching degree, etc.), task completion quality, task completion cycle, and metrology equipment used.
[0031] By extracting structured and unstructured data from candidate measurement task features as described above, and covering both multimodal basic ability data of talents and full historical task completion data, we can achieve a comprehensive representation of task and talent data. This includes both explicit and implicit qualifications of talents, full-dimensional features of tasks, and historical execution data. This avoids omissions and biases in manual data sorting and breaks through the limitations of traditional matching that only extracts a single explicit attribute.
[0032] As a further implementation method, the collected data is preprocessed, including: data cleaning of multimodal basic capability data to remove missing and abnormal data; natural language processing of text data such as word segmentation, deduplication, and stop word deletion; normalization of numerical data to map to the [0,1] interval; and standardization of task features to unify feature description dimensions, thereby constructing a multidimensional database of metrology talents and a candidate metrology task feature library to improve the accuracy and efficiency of subsequent feature extraction and matching calculations and avoid interference from data noise on the model.
[0033] Step S2: Based on multimodal basic capability data, extract the basic capability features of metrology talents through multimodal fusion encoder, and construct a metrology talent knowledge graph containing talent nodes and technology-related edges. The comprehensive features of metrology talents are obtained by GAT aggregation.
[0034] Specifically, considering that the accuracy of talent attribute feature extraction will affect the accuracy and rationality of subsequent matching, although existing technologies have proposed feature extraction methods based on knowledge graphs, they mostly build graph relationships around the basic attributes of talents, and are only based on the superficial associations between collaborative papers or projects among talents. Their graph associations are simple and lack the mining of implicit value. The feature extraction is incomplete, which easily makes the final matching results unreasonable and affects the quality of the metrological task.
[0035] To address the aforementioned issues and achieve a more accurate and comprehensive representation of the professional competence of metrology personnel, this embodiment first employs a multimodal fusion encoder to extract and fuse multimodal features from the candidate metrology personnel's basic competence data. This includes: (1) for structured data, converting discrete features into low-dimensional dense vector features through an embedding layer; (2) for text data, using a pre-trained language model such as BERT to extract semantic features and generate text feature vectors; and (3) for numerical data, constructing numerical feature vectors after normalization. Next, using an attention-based multimodal fusion strategy, the feature vectors extracted from the different modalities are weighted and fused to unify the feature dimensions, resulting in the basic competence feature vector for each metrology personnel. This achieves deep fusion of multi-dimensional data, including structured, textual, and numerical data, solving the problem of one-sided representation by a single modality and accurately extracting the basic competence features of the personnel.
[0036] Next, considering that metrology tasks require teamwork, in addition to the aforementioned surface-level professional skills of metrology talents, task-talent matching should also consider the actual technical skills and implicit abilities such as the ability to collaborate and complement others. To this end, this embodiment constructs a metrology talent knowledge graph. By establishing the technical correlation value between talents and combining graph neural network (GAT) technology, it comprehensively extracts the comprehensive characteristics of metrology talents.
[0037] In this embodiment, metrology professionals are used as knowledge graph nodes, and node attributes / features are the basic ability feature vectors of the corresponding metrology professionals. Based on historical metrology task data, technical association edges are established between each professional node. The weight of each edge is the weighted sum of the frequency of collaboration, technical complementarity, and project fit between metrology professionals, calculated using the following formula: ; In the above formula, , The weighting coefficients are determined based on the actual task requirements in the metrology field; the collaboration frequency is the normalized value of the number of historical metrology tasks in which the two individuals jointly participated; the technical complementarity is calculated based on the cosine similarity of the basic ability feature vectors of the two individuals, using the formula: complementarity = 1 - cosine similarity; the project fit is the weighted value of the matching degree of the roles of the two individuals in the collaborative task and the quality of task completion. Both the matching degree of roles and the quality of task completion are obtained based on the archived data of historically completed tasks, and are used to measure the comprehensive performance of metrology personnel in historical team collaboration.
[0038] Finally, the knowledge graph of metrology talents is input into the graph attention network (GAT). The multi-head attention mechanism of GAT is used to weight and aggregate the features of the neighboring nodes of each talent node, and to integrate the basic ability features of the talent with the correlation features of the neighboring talents to achieve comprehensive extraction of the comprehensive features of the talents. The aggregated feature vectors are then concatenated and normalized to obtain the comprehensive feature vector of each candidate metrology talent. This feature is more in line with the needs of team collaboration in metrology tasks.
[0039] The above methods can transform implicit collaborative relationships and technological complementarity among metrology professionals into quantifiable feature data, overcoming the shortcomings of traditional matching methods that ignore implicit technological connections among professionals and ensuring the rationality and accuracy of task-talent matching. Furthermore, it should be noted that the entire feature extraction process described above is customized for the metrology field. From data types to feature fusion logic, it is adapted to the characteristics of metrology's specialized specialization and strict qualification requirements, significantly improving the professional adaptability and representation accuracy of talent features.
[0040] Step S3: Based on multiple candidate metrology tasks and historical metrology task data, perform feature vectorization and clustering, and integrate to obtain the comprehensive features of candidate metrology tasks.
[0041] Specifically, considering the diversification and increasing complexity of metrological research tasks, as well as the growing number of cross-disciplinary metrological projects, the extracted features for metrological tasks in emerging professional directions may be new feature descriptions. Directly using these features for matching can easily lead to matching bias. Therefore, this embodiment, based on the initial extraction of features for subsequent metrological tasks, refines and optimizes these features by introducing relevant historical metrological task features, so that the task features can fully and comprehensively reflect the actual technical needs of metrological research and development, thus avoiding subsequent matching bias.
[0042] In this embodiment, the basic features of candidate measurement tasks and historical measurement tasks are first uniformly vectorized, including: structured features are converted into vectors through the embedding layer, and unstructured features are extracted into semantic vectors through a pre-trained language model. After fusion, the initial feature vectors of each candidate measurement task and historical measurement task are obtained, and a measurement task feature vector library is constructed to provide a unified computational basis for subsequent clustering and feature fusion, thereby improving the standardization of task feature processing.
[0043] Then, using the sub-specialty direction, task type, and technical difficulty level of the econometric task as the core clustering dimensions (the core clustering dimensions can be selected according to the specific situation), K-means clustering is performed on the initial feature vector of the candidate econometric task and the initial feature vector of all historical econometric tasks. That is, for each candidate econometric task, the number of clusters is set to K, and the feature similarity between tasks is calculated by Euclidean distance. Historical econometric tasks that are clustered in the same cluster (i.e., the same class) as the candidate econometric task are judged as similar historical econometric tasks, forming a set of similar historical tasks for each candidate econometric task.
[0044] Finally, the selected set of similar historical tasks is used to provide historical knowledge-based guidance for matching emerging professional directions or complex metrology tasks, avoiding matching bias issues. In this embodiment, an attention mechanism is employed to weightedly fuse the initial features of candidate metrology tasks with the features of similar historical metrology tasks. This integrates the implementation experience and technical requirements of historical tasks into the candidate task features, achieving refined optimization of task features. This makes the comprehensive features of candidate tasks more closely match the actual technical needs of metrology R&D, improving the accuracy of subsequent matching. The weighted fusion calculation formula based on the attention mechanism is as follows: ; In the above formula, This is the optimized candidate econometric task comprehensive feature vector. This represents the initial feature vector for the candidate task. For the first Attention weights for similar historical tasks For the first Feature vectors of similar historical tasks The number of similar historical tasks.
[0045] Preferably, the attention weight is the feature similarity between similar historical tasks and candidate measurement tasks. This attention weight enables differentiated feature fusion, avoids interference from invalid features of similar historical tasks, and ensures the core and relevance of the optimized task features.
[0046] The above approach involves a multimodal fusion encoder extracting the structured, textual, and numerical fundamental capabilities of talents, quantifying implicit technical connections between talents using the GAT knowledge graph, and fusing the two to achieve a complete representation of the comprehensive characteristics of talents. At the same time, K-means clustering and attention mechanisms are used to fuse similar historical task features, refining and optimizing task features. This ensures that both talent and task features align with the characteristics of the metrology field, which emphasizes specialization, practical skills, and collaboration, significantly improving the accuracy of feature representation and laying the foundation for accurate matching of talent and task features in the future.
[0047] Step S4: Construct a reinforcement learning matching model and train it with historical econometric task data. Use the trained matching model to dynamically match the comprehensive features of each candidate econometric task with the comprehensive features of all econometric talents according to task priority, and generate a recommended list of econometric talent combinations.
[0048] Specifically, considering that existing methods lack matching designs based on talent combinations, making it difficult to meet the team matching needs of streamlined and collaborative measurement tasks, this embodiment introduces a reinforcement learning strategy. It dynamically matches measurement tasks with measurement talent according to task priority, matching candidate measurement tasks with different combinations of measurement talent to meet the team collaboration needs of measurement tasks and improve the quality and efficiency of task completion. The specific process of step S4 above is as follows: Step S4.1: Construct a reinforcement learning matching model. The elements of a reinforcement learning matching model are defined as follows: (1) Intelligent agent: Matching decision-making subject, responsible for completing the matching calculation of the comprehensive characteristics of the measurement task and the comprehensive characteristics of the talent; (2) State Space S: The state space integrates the comprehensive characteristics of the task, the characteristics of the talent combination, the task priority, and the talent occupancy status to achieve a full-dimensional digital representation of the matching scenario. This state vector It can be characterized as: , This is the comprehensive feature vector of the candidate econometric task. This represents the comprehensive feature vector of the candidate portfolio of metrology talents. This is the task priority coefficient. This is the talent occupancy state vector, which is 1 when occupied and 0 when idle. If the current talent is matched with a task, the state vector changes from 0 to 1.
[0049] Among them, the comprehensive feature vector of the candidate econometrics talent pool The calculation formula is: ; In the above formula, The number of personnel required to match the task; For the first The role weights for each metrology professional can be predefined. For example, high weights can be assigned to roles such as task leaders and core R&D personnel, medium weights to roles such as equipment operators and data analysts, and low weights to support staff. ; For the first A comprehensive feature vector of a metrology professional.
[0050] Furthermore, the aforementioned state vector For high-dimensional vectors, dimension ,in As a comprehensive feature dimension of the task, As a dimension of talent combination characteristics, For task priority dimension, Match talent status dimensions.
[0051] (3) Action Space A: Action space A is a discrete set, formally defined as... , The total number of candidate combinations of metrology talents. For the first Each action corresponds to a set of talent combinations; the action constraint is that the same talent cannot appear in multiple roles in the same task. Matched talents can only be selected when the task priority is greater than or equal to the priority of their own matched tasks. This is to adapt to the industry characteristics of measurement tasks that require multi-role collaboration, while avoiding the problem of duplicate talent allocation and core talents being occupied by low-priority tasks.
[0052] (4) Reward function R This reward function employs a multi-dimensional weighted reward and penalty mechanism, integrating positive indicators such as feature similarity, technological complementarity, and priority matching with negative indicators such as talent load and capability bias. This ensures that the optimization objective of model training closely aligns with the matching requirements of the econometrics domain, guiding the model to learn more reasonable matching strategies. Specifically, the reward function is a weighted sum of feature similarity reward, technological complementarity reward, priority matching reward, talent load penalty, and capability bias penalty, calculated as follows: ; in, The core purpose of feature similarity rewards is to ensure a high degree of match between the professional capabilities of talent combinations and the technical requirements of the measurement task. Rewards are given by calculating the similarity between the comprehensive features of the talent combination and the comprehensive features of the task, guiding the model to prioritize matching talent combinations whose professional capabilities, technical directions, and core task requirements align. This fundamentally guarantees the accuracy of the matching and avoids a disconnect between talent capabilities and task requirements. It can be represented as: ; The above The core purpose of the technology complementarity reward is to strengthen the collaboration and complementarity within the talent pool. Given the multi-role and cross-disciplinary nature of metrology tasks, rewards are given by quantifying the degree of technology complementarity among talents. This guides the pairing of models with talent combinations that have complementary technical capabilities and suitable roles, avoids overlapping technical skills within the team, ensures smooth transitions between key technical aspects, and improves the overall collaborative efficiency of the team. It can be represented as: ; ; The above Priority-based reward matching aims to optimize the allocation of metrology research resources. It guides the model to match high-quality talent combinations according to task priority, ensuring that core high-priority tasks such as the development of national mandatory metrology standards and emergency metrology technology breakthroughs are allocated to personnel with stronger professional capabilities and better collaboration skills. This avoids core talent being tied up in low-priority tasks, guaranteeing the progress of core metrology R&D tasks. This can be represented as: ; The above The core purpose of talent load penalty is to avoid over-allocation of talent and ensure a balanced workload. It penalizes talent combinations with a large number of assigned tasks and saturated workloads, guiding the model to allocate talent resources rationally. This prevents individual talents from being over-assigned tasks, leading to a decline in work quality, and avoids the waste of resources caused by both idle and over-worked talent. It also ensures the efficiency of the execution of measurement tasks. This can be expressed as: ; The above The core purpose of capability deviation penalty is to control the range of deviation between the overall capability of the talent pool and the task requirements. It penalizes talent pools with excessively large gaps between their overall capability and the task's technical specifications, guiding the model to match talent pools with capabilities suitable for the task's difficulty and technical requirements. This avoids wasting resources by using overqualified individuals while also preventing underqualified individuals from failing to complete the task as required, thus ensuring the quality of metrological task completion. This can be expressed as: ; In the above formula, , These are the weighting coefficients; M This indicates the number of metrology professionals required to match a single metrology task, i.e., the number of people in the talent pool corresponding to that metrology task. i, j This refers to two distinct metrology talent nodes within a metrology talent pool that matches a single metrology task; that is, any two metrology talents within the metrology talent pool. It is the number of combinations; This is the normalized task priority coefficient. The normalized talent portfolio quality score can be obtained by weighted summation of the comprehensive ability characteristics of each talent in the talent portfolio based on role weights. This represents the average number of tasks already assigned to all metrology professionals in the current metrology talent pool. This represents the maximum number of tasks a single metrology professional can undertake. It is an L2 norm.
[0053] (5) Policy Network: A deep neural network (DNN) is used as the policy network to output the probability distribution of each action, i.e., the matching and adaptation probability of each talent combination to the current measurement task. In this embodiment, the hidden layer activation function is ReLU, with Dropout regularization, and the output layer activation function is Softmax to ensure the model's feature fitting ability and generalization ability, avoid overfitting, and enable the model to efficiently output the adaptation probability of each talent combination.
[0054] Step S4.2: Using historical metrology task data as the training set, which includes historical metrology task characteristics, metrology talent characteristics, matching results, task completion quality, etc., initialize the model parameters and train the reinforcement learning matching model. This allows the model to learn matching experience that aligns with the actual metrology industry, avoiding the problem of poor professional adaptability of general matching models and improving the model's matching decision-making ability in the metrology field. In this embodiment, the PPO algorithm is used to train the model, setting hyperparameters such as learning rate, optimizer, discount factor, advantage function coefficient, batch size, number of iterations, and shearing coefficient. Then, through the interaction between the agent and the environment, state, action, and reward samples are sampled, and the parameters of the policy network are continuously adjusted to maximize the model's cumulative reward value until the model's cumulative reward value converges, completing the model training.
[0055] Step S4.3: Based on factors such as strategic needs in the field of metrology, urgency of tasks, and level of technical difficulty, candidate metrology tasks are prioritized and matched to form a priority sequence. At the same time, the reinforcement learning matching model performs matching calculations on the comprehensive characteristics of each candidate metrology task and the comprehensive characteristics of all talents in descending order of priority. The model outputs the optimal metrology talent combination matching action based on the current state, and generates a recommended list of metrology talent combinations for each candidate metrology task. The list is sorted from high to low suitability.
[0056] As one implementation method, this embodiment takes 60 candidate metrology tasks as an example and divides them into 3 priorities, in the following sequence: microwave metrology standard development (15 tasks) > metrology instrument calibration method optimization (25 tasks) > metrology data model improvement (20 tasks). The reinforcement learning model matches tasks in descending order of priority and generates a state vector for each task. The model outputs the optimal action corresponding to the state vector, i.e., the optimal talent combination. For example, for the microwave metrology standard development task, a team composed of corresponding microwave standard development talents, microwave equipment calibration talents, and electromagnetic error analysis talents is recommended.
[0057] The above matching process is based on automated intelligent computing, which can significantly improve the matching efficiency between large-scale metrology tasks and massive talent pools. Compared with traditional manual screening and simple keyword matching, it saves a lot of manpower and time costs and adapts to the diversified and complex development trend of scientific research tasks in the field of metrology.
[0058] As a further implementation method, the above model is evaluated and its parameters are optimized. Optimization rules are set for core indicators such as matching accuracy, resource utilization, and talent load balance to achieve dynamic iterative upgrades of the model. This allows the model to continuously adapt to the ever-changing task and talent needs in the metrology field, ensuring the long-term stability of the matching effect. In this embodiment, the model evaluation indicators and target values are set as follows: Matching accuracy ≥ 95%, Matching accuracy = Task completion qualification rate after matching / Total number of tasks; Resource utilization ≥ 90%, Resource utilization = Number of core talents allocated to high-priority tasks / Total number of core talents allocated; Talent load balance ≥ 0.8, Talent load balance = 1 - (Maximum number of tasks - Minimum number of tasks) / Average number of tasks. If the model evaluation indicators do not reach the target values, optimization is performed according to the set rules, i.e., when the matching accuracy is below 90%, the value is increased. Reduce to 0.5. Up to 0.1; when resource utilization is below 80%, increase. Reduce to 0.3. When the talent load balance is below 0.7, adjust to 0.3; Once the value reaches 0.2, retrain the model.
[0059] Step S4.4: Based on the talent combination recommendation list, complete the matching recommendation of metrology tasks and metrology talents, and simultaneously update the metrology talent task allocation file and task progress management system to realize the systematic management of the matching relationship between metrology tasks and metrology talents. Clarify the recommended person in charge, collaborating team, required resources, time nodes and other information for each task, and provide a clear management basis for the subsequent execution of metrology tasks.
[0060] As a further implementation method, the above-mentioned solution proposed in this embodiment can also be adapted to other typical scenarios in the field of metrology, such as the task allocation scenario of a large enterprise metrology and testing center, such as the metrology and testing of production lines in automobile and aviation enterprises. This solution can adapt to the metrology needs of multiple processes and multiple equipment in the production line, match professional talents for each process, and improve the metrology and testing efficiency of the production line.
[0061] In this embodiment, taking the composite metrological testing of the three-dimensional morphology and high-temperature field distribution of high-pressure turbine blades for aero-engines in an aviation enterprise as an example, this metrological task involves professional aspects such as length, temperature, and electromagnetics, requiring the operation of equipment such as 3D scanning measuring instruments, high-temperature sensors, and electromagnetic acquisition modules. The technical specifications, technical difficulty level, completion time, and other requirements of this task are also clearly defined. For this task, structured and textual data of currently employed metrological personnel are automatically extracted from the multi-source metrology database of the aviation enterprise's metrology and testing center. This data includes qualifications, certifications, years of experience, region, preferred equipment, research direction, papers and patents, technical descriptions, equipment operation scores, acceptance scores, and error accuracy. Simultaneously, historical similar aero-engine metrological tasks are extracted to obtain a list of collaborating personnel, collaboration frequency, task completion quality, and team role matching. Through the above methods, the comprehensive characteristics of metrological personnel and metrological tasks are extracted. Using reinforcement learning for dynamic matching, all personnel combinations are traversed to determine the optimal metrological personnel combination.
[0062] Furthermore, it can also be applied to cross-regional metrology technology research scenarios, such as the joint development of metrology standards across provinces. This solution can integrate cross-regional talent data, construct a cross-regional talent knowledge graph, match the optimal talent combination for cross-regional collaboration, solve problems such as unfamiliarity and poor complementarity among cross-regional collaborative talents, and improve the quality and efficiency of metrology task completion.
[0063] Example 2 This embodiment provides a dynamic matching system for metrology talent and metrology tasks based on reinforcement learning, specifically including: The data acquisition module is used to acquire multiple candidate metrology tasks and extract multimodal basic ability data and historical metrology task data for each metrology talent from the multi-source metrology database. The metrology talent feature extraction module is used to extract the basic ability features of metrology talents based on multimodal basic ability data through a multimodal fusion encoder, and to construct a metrology talent knowledge graph containing talent nodes and technology-related edges. The comprehensive features of metrology talents are obtained by aggregation through GAT. The candidate metrology task feature extraction module is used to vectorize and cluster features based on multiple candidate metrology tasks and historical metrology task data, and integrate them to obtain comprehensive features of candidate metrology tasks. The econometrics task-talent matching module is used to build a reinforcement learning matching model and train it with historical econometrics task data. The trained matching model is used to dynamically match the comprehensive features of each candidate econometrics task with the comprehensive features of all econometrics talents in order of task priority, and generate a recommended list of econometrics talent combinations.
[0064] Example 3 This embodiment provides an electronic device, including: a memory for storing executable instructions; and a processor for executing the executable instructions stored in the memory to implement the method provided in this embodiment.
[0065] Example 4 This embodiment also provides a computer-readable storage medium storing executable instructions, which, when executed by a processor, will cause the processor to execute the method described above in this embodiment.
[0066] Example 5 This embodiment provides a computer program product including executable instructions, which are computer instructions; the executable instructions are stored in a computer-readable storage medium. When the processor of an electronic device reads the executable instructions from the computer-readable storage medium and executes the executable instructions, the electronic device performs the method described in this embodiment.
[0067] The steps and methods involved in Embodiments 2 to 5 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.
[0068] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.
[0069] The above description is only a preferred embodiment of the present invention. Although the specific implementation of the present invention has been described in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that, based on the technical solution of the present invention, various modifications or variations that can be made by those skilled in the art without creative effort are still within the scope of protection of the present invention.
Claims
1. A method for dynamically matching metrology talent with metrology tasks based on reinforcement learning, characterized in that, include: Multiple candidate metrology tasks were obtained, and multimodal basic competency data and historical metrology task data of each metrology talent were extracted from the multi-source metrology database. Based on multimodal basic capability data, the basic capability features of metrology talents are extracted through a multimodal fusion encoder, and a metrology talent knowledge graph containing talent nodes and technology-related edges is constructed. The comprehensive features of metrology talents are obtained by GAT aggregation. Based on multiple candidate metrology tasks and historical metrology task data, feature vectorization and clustering are performed, and integrated to obtain comprehensive features of candidate metrology tasks. A reinforcement learning matching model is constructed and trained with historical econometrics task data. The trained matching model is then used to dynamically match the comprehensive features of each candidate econometrics task with the comprehensive features of all econometrics talents according to task priority, thereby generating a recommended list of econometrics talent combinations.
2. The method for dynamic matching of metrology talent and metrology tasks based on reinforcement learning as described in claim 1, characterized in that, The multimodal basic capability data includes structured data, text data, and numerical data. Among them, the structured data includes professional qualifications, certification status, qualification level, years of work experience, affiliated institution, and types of equipment proficient; the text data includes professional research direction, description of technical expertise, and published metrological technical papers / patents; and the numerical data includes equipment operation proficiency scores, project result acceptance scores, and error analysis accuracy. The historical metrology task data includes the characteristics of the historical metrology tasks, the list of collaborating personnel who completed the historical metrology tasks, the basic ability data of each person in the list of collaborating personnel and their role in the task, the quality of task completion, the task completion cycle, and the metrology equipment used; among which, the characteristics of the metrology tasks include task number, task type, sub-specialty direction, technical indicator requirements, technical difficulty level, required metrology equipment, and completion time limit.
3. The method for dynamic matching of metrology talent and metrology tasks based on reinforcement learning as described in claim 1, characterized in that, The process of extracting the comprehensive characteristics of metrology talents includes: Using metrology professionals as nodes and their basic competency characteristics as node features, and establishing technical relationships between nodes based on historical metrology task data, forming technical relationship edges, and building a metrology professional knowledge graph; Based on the knowledge graph of metrology talents, the comprehensive characteristics of candidate metrology talents are obtained by GAT aggregation. The weight of the technology-related edge is a weighted sum of collaboration frequency, technology complementarity, and project fit. The collaboration frequency is the normalized value of the number of historical measurement tasks jointly participated in by the two individuals. The technology complementarity is calculated based on the cosine similarity of the basic ability feature vectors of the two individuals, and then calculated according to the 1-cosine similarity value. The project fit is a weighted value of the matching degree of the roles of the two individuals in the collaborative measurement task and the quality of task completion.
4. The method for dynamic matching of metrology talent and metrology tasks based on reinforcement learning as described in claim 1, characterized in that, Based on multiple candidate metrology tasks and historical metrology task data, feature vectorization and clustering are performed, and the integrated features of the candidate metrology tasks are obtained, including: The features of multiple candidate econometric tasks and historical econometric tasks are vectorized, and K-means clustering is used to cluster all econometric tasks. Historical econometric tasks that are in the same cluster as candidate econometric tasks are selected as similar historical econometric tasks. Among them, K-means clustering uses the sub-specialty direction, task type and technical difficulty level of the econometric task as the core clustering dimensions, and calculates the feature similarity between tasks through Euclidean distance. An attention mechanism is used to weight and fuse the initial features of the candidate measurement task with the features of similar historical measurement tasks to obtain the comprehensive features of the candidate measurement task.
5. The method for dynamic matching of metrology talent and metrology tasks based on reinforcement learning as described in claim 1, characterized in that, A reinforcement learning algorithm is used to dynamically match multiple candidate metrology tasks with metrology professionals. The process is as follows: Construct a reinforcement learning matching model, including: using the comprehensive feature vector of measurement tasks, the comprehensive feature vector of measurement talents, the priority of measurement tasks, and the state vector of measurement talents as the state space, using different combinations of measurement talents as actions, designing a reward function for task-talent matching, and building an agent and policy network. Using historical measurement task data as the training set, the agent interacts with the environment to sample state, action, and reward samples, and continuously iterates and optimizes the policy network parameters until the model converges. Based on the strategic needs of the metrology field, multiple candidate metrology tasks are prioritized. Using a trained reinforcement learning model, tasks and talents are matched in descending order of priority. The state corresponding to each candidate metrology task is constructed, the corresponding optimal action is matched, the optimal metrology talent combination is determined, and a recommended list of metrology talent combinations for multiple candidate metrology tasks is generated.
6. The method for dynamic matching of metrology talent and metrology tasks based on reinforcement learning as described in claim 5, characterized in that, The reward function is a weighted sum of feature similarity reward, technology complementarity reward, priority matching reward, talent load penalty, and ability deviation penalty; The action space is a discrete set, formally defined as: ,in, This represents the total number of candidate combinations of metrology talents. For the first Each action corresponds to a set of measurement talents; the action constraint is that the same measurement talent cannot appear in multiple roles in the same task, and a matched measurement talent can only be selected when the task measurement priority is greater than or equal to the priority of its matched measurement tasks.
7. A dynamic matching system for metrology talent and metrology tasks based on reinforcement learning, characterized in that, include: The data acquisition module is used to acquire multiple candidate metrology tasks and extract multimodal basic ability data and historical metrology task data for each metrology talent from the multi-source metrology database. The metrology talent feature extraction module is used to extract the basic ability features of metrology talents based on multimodal basic ability data through a multimodal fusion encoder, and to construct a metrology talent knowledge graph containing talent nodes and technology-related edges. The comprehensive features of metrology talents are obtained by aggregation through GAT. The candidate metrology task feature extraction module is used to vectorize and cluster features based on multiple candidate metrology tasks and historical metrology task data, and integrate them to obtain comprehensive features of candidate metrology tasks. The econometrics task-talent matching module is used to build a reinforcement learning matching model and train it with historical econometrics task data. The trained matching model is used to dynamically match the comprehensive features of each candidate econometrics task with the comprehensive features of all econometrics talents in order of task priority, and generate a recommended list of econometrics talent combinations.
8. An electronic device, characterized in that, include: Memory, used to store executable instructions; The processor, when executing executable instructions stored in the memory, implements the reinforcement learning-based dynamic matching method for metrology talent and metrology tasks as described in any one of claims 1-6.
9. A computer-readable storage medium, characterized in that, The device stores executable instructions that, when executed by a processor, implement the reinforcement learning-based dynamic matching method for metrology talent and metrology tasks as described in any one of claims 1-6.
10. A computer program product, characterized in that, The computer program product includes executable instructions stored in a computer-readable storage medium; When the processor of the electronic device reads the executable instructions from the computer-readable storage medium and executes the executable instructions, it implements the reinforcement learning-based dynamic matching method for metrology talent and metrology tasks as described in any one of claims 1-6.