Large model-based laboratory station dynamic configuration method and system
Patent Information
- Application Number
- CN202611084538.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-21
- Publication Date
- 2026-10-23
AI Technical Summary
第一，工位占用真实性难以识别
1.本发明采用“云端语义检测+本地规则融合”的架构，利用大模型的强大语义理解能力，通过结构化Prompt将开放式视觉理解任务转化为封闭式标注任务，能够精准识别工位占用状态，有效解决了传统打卡方式中代打卡、打卡后离开等虚假占用问题；针对实验室摄像头安装高度和角度限制导致的遮挡问题，采用头部检测模式，提高了检测的可靠性和一致性；同时本地承担轻量级任务，降低了硬件门槛，普通边缘计算单元即可满足需求，无需配置GPU。
Smart Images
Figure CN122965724A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the technical field of laboratory resource management, specifically relating to a method and system for dynamic configuration of laboratory workstations based on a large model. Background Technology
[0002] University research laboratories are laboratories that rely on discipline development and primarily aim at scientific research activities. They undertake the teaching and research work of master's and doctoral students, as well as the implementation of research projects at all levels. They serve as a fundamental platform for talent cultivation, scientific research, and innovative practice. A laboratory workstation refers to a fixed work position within the laboratory provided for researchers or students, equipped with the necessary equipment and resources.
[0003] In recent years, there has been a significant imbalance between the supply and demand for laboratory workstations. From the demand side, the AI for Science paradigm has become mainstream, leading to an increasing number of research teams needing to conduct on-site work in laboratories; universities are undertaking more and more vertical and horizontal projects; and more graduate students are participating in research work comprehensively and deeply, resulting in a continuous increase in demand for workstations. From the supply side, building new laboratories requires substantial financial investment and has a long construction period; the number of workstations is limited by the original design, making it difficult to achieve a significant increase in quantity through renovations.
[0004] Therefore, optimizing laboratory workstation configuration and improving workstation utilization have become important means to resolve the supply-demand imbalance. However, traditional workstation management methods face two core challenges: First, the authenticity of workstation occupancy is difficult to verify. Traditional clocking-in methods have significant drawbacks: users can create a false impression of occupancy by having someone else clock in on their behalf; users can leave their workstations after clocking in, making the workstations nominally used but actually idle. Existing visual detection methods mostly use traditional deep learning models, which require pre-training on fixed categories and are sensitive to complex occlusions, abnormal postures, and changes in lighting. They are difficult to adapt to the complex and interfering environment of laboratory workstations, which includes tables, chairs, monitors, partitions, bags, and other objects.
[0005] Secondly, the rationality of workstation idleness is difficult to determine. Once workstations are assigned to fixed users, these users cannot remain at their workstations continuously; they may leave due to objective reasons such as attending classes, taking exams, or participating in group meetings. If utilization is calculated solely based on simple occupancy without considering reasonable idleness factors, conflicts with users can easily arise, affecting the stable operation of the laboratory. Current technology lacks quantitative methods for measuring the rationality of workstation idleness, making it difficult for managers to scientifically distinguish between "reasonable idleness" and "ineffective idleness."
[0006] Therefore, there is an urgent need for a new method that can accurately identify the actual occupancy of workstations, reasonably quantify the factors of idle workstations, and realize dynamic allocation of workstations accordingly. Summary of the Invention
[0007] This invention aims to solve the technical problem of how to identify the authenticity of workstation occupancy and quantitatively determine the rationality of idleness, and to make scientific dynamic allocation accordingly, and provides a method for dynamic allocation of laboratory workstations based on a large model.
[0008] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for dynamic configuration of laboratory workstations based on a large model, comprising: Collect workstation scene images and upload them to the cloud big model for personnel head detection. Locally, match the coordinates of the detected head bounding boxes with the predefined workstation area to determine the occupancy status of each workstation and calculate the actual occupancy time of the workstation. A hierarchical evaluation model for the rationality of idle workstations is constructed, comprising a target layer, a criterion layer, and an indicator layer. The criterion layer includes four dimensions: rationality of research tasks, differences in personnel status, time characteristics, and adjustability. The indicator layer includes several evaluation indicators belonging to each criterion dimension. The combined weight of each evaluation indicator relative to the target layer is determined using the analytic hierarchy process, and the idle workstation weight coefficient for workstation users is determined based on the combined weight. The weighted utilization rate of the workstation is calculated based on the actual occupancy time, the expected arrival time, and the idle weight coefficient. Workstations are classified according to the weighted utilization rate and preset threshold system, and the corresponding workstation configuration strategy is executed according to the classification results.
[0009] In one possible implementation, the workstation scene images are collected at minute intervals, compressed and encoded, and then uploaded to a large cloud model through an authentication interface. The inference calculation for personnel head detection is completely offloaded to the cloud graphics processing unit, and the local terminal only performs lightweight edge tasks such as image decoding, coordinate mapping, and status recording.
[0010] In one possible implementation, the process of acquiring workstation scene images and uploading them to a cloud-based large model for personnel head detection includes: The collected workstation scene images are preprocessed and then encoded and uploaded to the cloud-based large model API. The large model performs head detection tasks according to structured detection constraint instructions and outputs structured data containing normalized coordinates of the head bounding box. The detection constraint instructions include at least limiting the detection target to the human head, excluding non-human interference objects, ensuring the bounding box is close to the edge of the detection target, and limiting the output format to a structured array.
[0011] In one possible implementation, the step of collecting workstation scene images and uploading them to a large cloud model for head detection also includes a multi-layered fault tolerance mechanism: the first layer is to automatically retry and record the exception when the cloud request fails; the second layer is to perform format verification on the data returned by the cloud and perform fault tolerance processing when the format is abnormal; the third layer is to isolate the exception when a single frame image fails to be determined, without interrupting the overall detection process; and the fourth layer is to fully record the request parameters, return results, and exception information for each detection.
[0012] In one possible implementation, the evaluation indicators, each belonging to a different criterion dimension, include at least the following: Research task rationality dimension: research phase, courses and group meetings, reporting of temporary affairs, and unexcused idleness; Doctoral students, master's students, and corporate partners are distinguished by their individual identities. Duration, frequency of occurrence, and periodic regularity under the time-related characteristics; The complexity of equipment relocation and the cost of negotiation under the dimension of adjustability.
[0013] In one possible implementation, the step of determining the combined weights of each evaluation index relative to the target layer using the analytic hierarchy process (AHP), and determining the workstation idle weight coefficient for workstation users based on the combined weights, includes: Using the target layer as a reference, pairwise comparisons and scores are performed on each dimension of the criterion layer based on expert scoring, and a target-criterion layer judgment matrix is constructed. Using each dimension of the criteria layer as a reference, the evaluation indicators belonging to each dimension in the indicator layer are compared and scored in pairs based on the expert scoring method to construct the criteria layer-indicator layer judgment matrix. A consistency check is performed on the target layer-criteria layer judgment matrix and each criteria layer-indicator layer judgment matrix; Solve the judgment matrix that passes the test, extract the relative weights within the level, and obtain the combined weights of each evaluation index relative to the target level through the overall hierarchical sorting. The idle weight coefficient of a workstation is determined by comprehensively considering the evaluation indicators actually triggered by the workstation user and their corresponding combined weights.
[0014] In one possible implementation, the workstation weighted utilization rate includes individual weighted utilization rate and team average weighted utilization rate; The calculation of the weighted utilization rate of workstations based on the actual occupancy time, expected arrival time, and idle weight coefficient includes: The idle time is determined based on the actual occupancy time and the expected arrival time of the workstation; Multiply the idle time by the idle weight coefficient to obtain the converted effective idle time. The effective usage time is obtained by adding the actual occupied time to the effective idle time. Divide the effective usage time by the expected arrival time to obtain the individual weighted usage rate; The team average weighted utilization rate is obtained by taking the arithmetic mean of the individual weighted utilization rates of each workstation within the same team.
[0015] In one possible implementation, the step of classifying workstations according to the weighted utilization rate and a preset threshold system, and executing the corresponding workstation configuration strategy according to the classification result, includes: When an individual's weighted utilization rate is continuously within the first threshold range during the first preset period, it is determined to be a high utilization rate level for the individual, and the status quo is maintained. When an individual's weighted utilization rate is continuously within the second threshold range during the first preset period, it is determined to be the individual's utilization rate level and the status quo is maintained. When an individual's weighted utilization rate is within the third threshold range during the second preset period, it is determined to be a low utilization level for that individual, and the corresponding workstation user is reminded to improve efficiency. When an individual's weighted utilization rate is within the fourth threshold range during the second preset period, it is judged as an extremely low utilization rate level, and the team leader is reminded to adjust the user of the workstation. When the average weighted utilization rate of a team is within the first threshold range for a continuous period of the first preset period, the team is judged to be at a high utilization level, and workstations will be allocated to the team first when laboratory resources are expanded. When the team's average weighted utilization rate is continuously within the second threshold range for the first preset period, it is determined to be the utilization level of the team and the status quo is maintained. When the team's average weighted utilization rate is within the third threshold range during the second preset period, it is judged as a low utilization level of the team, and the team leader is reminded to pay attention to the use of workstations. When the team's average weighted utilization rate is within the fourth threshold range during the second preset period, it is determined to be an extremely low utilization rate level for the team, and a batch recall and redistribution within the laboratory will be initiated for all workstations of the team.
[0016] Secondly, the present invention also provides a laboratory workstation dynamic configuration system based on a large model, comprising: The image acquisition module is used to acquire images of the workstation scene and upload them to the cloud-based large model; The cloud-based detection module is used to call a large cloud model to detect human heads and return the coordinates of the human head bounding box. The local fusion module is used to match the detected head bounding box coordinates with the predefined workstation area, determine the occupancy status of each workstation, and calculate the actual occupancy time of the workstation. The idle quantity quantification module is used to construct a hierarchical evaluation model for the rationality of idle workstations, including a target layer, a criterion layer, and an indicator layer. The criterion layer includes four dimensions: rationality of scientific research tasks, differences in personnel status, time characteristics, and adjustability. The indicator layer includes several evaluation indicators belonging to each criterion dimension. The combined weight of each evaluation indicator relative to the target layer is determined by the analytic hierarchy process, and the idle weight coefficient of the workstation user is determined based on the combined weight. The utilization rate calculation module is used to calculate the weighted utilization rate of the workstation based on the actual occupation time of the workstation, the expected arrival time, and the idle weight coefficient. The configuration decision module is used to classify workstations according to the weighted utilization rate and preset threshold system, and execute the corresponding workstation configuration strategy according to the classification result.
[0017] One possible implementation also includes a fault tolerance module, which is used to perform automatic retries when cloud requests fail, perform format validation and fault tolerance when returned data is abnormal, perform process isolation when single frame processing fails, and record full process logs.
[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention adopts an architecture of "cloud-based semantic detection + local rule fusion". It leverages the powerful semantic understanding capabilities of large models and transforms open-ended visual understanding tasks into closed-ended annotation tasks through structured Prompt. This enables accurate identification of workstation occupancy status and effectively solves the problems of false occupancy such as proxy clocking and leaving after clocking in in traditional clocking-in methods. To address the occupancy problem caused by the height and angle limitations of laboratory cameras, a head detection mode is adopted, which improves the reliability and consistency of detection. At the same time, it undertakes lightweight tasks locally, reducing the hardware threshold. Ordinary edge computing units can meet the requirements without the need for GPU configuration.
[0019] 2. This invention constructs a quantitative model for workstation idleness based on the analytic hierarchy process (AHP). It establishes 12 evaluation indicators from four dimensions and scientifically calculates the idleness weight coefficient through expert scoring and consistency testing. This enables a quantitative measurement of the rationality of workstation idleness, solves the drawback of simple occupancy calculation that does not consider reasonable idleness factors, and makes workstation management both efficient and fair.
[0020] 3. This invention proposes a weighted utilization rate index that integrates actual occupancy and reasonable idleness. When α=0, it degenerates into the traditional utilization rate formula. When α=1, all idle spaces are considered to be in effective use. The dynamic value of α allows the utilization rate to reflect both efficiency and fairness, and can more realistically reflect the actual utilization of workstations, providing a scientific basis for dynamic allocation.
[0021] 4. This invention designs a dynamic configuration strategy based on a weighted utilization rate threshold system, realizing hierarchical management of individual workstations and team workstations, forming a complete closed loop of "identification → quantification → calculation → configuration", providing an operable technical solution for the refined management of laboratory workstations, and helping to improve the level of laboratory management and optimize resource allocation. Attached Figure Description
[0022] Figure 1 This is a flowchart of a laboratory workstation dynamic configuration method based on a large model, according to an embodiment of the present invention. Figure 2 This is an embodiment of the present invention. Figure 1 Flowchart of step S2; Figure 3 This is an embodiment of the present invention. Figure 1 Flowchart of step S3; Figure 4 This is a structural diagram of a laboratory workstation dynamic configuration system based on a large model, according to an embodiment of the present invention. Detailed Implementation
[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0024] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.
[0025] The present invention will be further described below with reference to specific embodiments, but these are not intended to limit the scope of the invention.
[0026] Example 1 Firstly, such as Figure 1 As shown, this invention provides a method for dynamic configuration of laboratory workstations based on a large model, including: S1. Collect workstation scene images and upload them to the cloud big model for personnel head detection. Locally, match the detected head bounding box coordinates with the predefined workstation area to determine the occupancy status of each workstation and obtain the actual occupancy time of the workstation. S2. Construct a hierarchical evaluation model for the rationality of idle workstations, including a target layer, a criterion layer, and an indicator layer. The criterion layer includes four dimensions: rationality of research tasks, differences in personnel status, time characteristics, and adjustability. The indicator layer includes several evaluation indicators belonging to each criterion dimension. The combined weight of each evaluation indicator relative to the target layer is determined by the analytic hierarchy process, and the idle workstation weight coefficient of the workstation user is determined based on the combined weight. S3. Calculate the weighted utilization rate of the workstation based on the actual occupancy time, the expected arrival time, and the idle weight coefficient; S4. Classify workstations according to their weighted utilization rate and preset threshold system, and execute the corresponding workstation configuration strategy according to the classification results.
[0027] Regarding the identification of workstation occupancy authenticity, this embodiment collects workstation scene images and uploads them to a cloud-based large model for head detection. The detection frame coordinates are matched with predefined workstation areas to determine occupancy status and calculate actual occupancy time. Large-model visual detection replaces traditional clocking in to eliminate false occupancy. Head detection avoids complex occupancy issues in the laboratory, and precise matching with predefined workstation areas ensures a one-to-one correspondence between detection results and physical space. Regarding the determination of workstation idleness rationality, this embodiment constructs a hierarchical evaluation model comprising a target layer, a criterion layer, and an indicator layer. The criterion layer covers four dimensions: the rationality of research tasks, differences in personnel identity, time characteristics, and adjustability. The indicator layer sets several evaluation indicators belonging to each criterion dimension. The combined weight of each indicator relative to the target layer is determined using the analytic hierarchy process, and the idleness weight coefficient of the workstation user is determined based on the combined weight. This multi-dimensional idleness assessment... Factors are quantified into evaluable indicators, transforming the qualitative experience of domain experts into quantitative weights that have undergone consistency testing, thus achieving personalized and refined management. Building upon this, this embodiment further calculates the weighted utilization rate of workstations based on actual occupancy time, expected arrival time, and idle weight coefficients. By adjusting the traditional utilization rate calculation method by incorporating idle time into the weighted formula of effective usage time after converting it into a coefficient, the utilization rate indicator simultaneously reflects both actual occupancy and reasonable idleness, achieving a balance between efficiency and fairness. Finally, this embodiment classifies workstations based on the weighted utilization rate and a preset threshold system, and implements differentiated configuration strategies, including resource allocation, maintaining the status quo, communication reminders, and reclaiming and redistributing, according to the classification results. This achieves a technological upgrade from experience-based management to data-driven management, effectively solving the technical problem of how to accurately identify workstation occupancy and quantitatively determine the rationality of idleness, and then scientifically and dynamically allocate resources accordingly.
[0028] Regarding step S1, in one possible implementation, workstation scene images are collected at minute intervals, compressed and encoded, and then uploaded to the cloud-based large model through an authentication interface. The inference calculation for personnel head detection is completely offloaded to the cloud-based graphics processing unit, and the local terminal only performs lightweight edge tasks such as image decoding, coordinate mapping, and status recording.
[0029] Preferably, camera equipment is deployed in the laboratory to extract keyframes from the video stream every 5 minutes. Users use an interactive polygon annotation tool to delineate the boundaries of each workstation on the first frame of the video, predefining the workstation area and saving it as a JSON configuration file. It should be noted that polygon annotation precisely fits irregular workstation layouts, effectively reducing missed and false positives. After dynamic scaling and JPEG compression, the keyframes are Base64 encoded and sent to the cloud-based large model API via an HTTP POST request. A key authentication and model ID routing mechanism is used to access the cloud-based large model.
[0030] Furthermore, in one possible implementation, images of the workstation scene are collected and uploaded to a large cloud model for head detection of personnel, including: The collected workstation scene images are preprocessed and encoded before being uploaded to the cloud-based large model API. The large model executes the head detection task according to the structured detection constraint instructions and outputs structured data containing the normalized coordinates of the head bounding box. The detection constraint instructions include at least limiting the detection target to the human head, excluding non-human interference objects, ensuring that the bounding box is close to the edge of the detection target, and limiting the output format to a structured array.
[0031] Preferably, the large model performs the detection task based on a structured Prompt. The Prompt includes the following core constraints: "only detect heads, not bodies, chairs, and computers," "bounding boxes are close to the head edge," and "only return JSON arrays." This Prompt transforms the open-ended visual understanding task into a closed-ended annotation task, reducing post-processing complexity. The local program maps the normalized coordinates of the head detection boxes returned by the large model to actual pixel coordinates. After size filtering and boundary clipping, it calculates the center point of the detection box, determines whether it falls within the pre-annotated workstation polygon area, and overlays the detection box and occupancy status indicator (e.g., green box indicates occupancy, red box indicates vacancy) on the original frame for easy manual verification. A structured record of "workstation number - occupancy status - timestamp" is generated and written to Excel or a database for subsequent utilization calculations.
[0032] Furthermore, in one possible implementation, the step of collecting workstation scene images and uploading them to a large cloud model for head detection also includes a multi-layered fault tolerance mechanism: the first layer is to automatically retry and record exceptions when cloud requests fail; the second layer is to perform format verification on the data returned from the cloud and perform fault tolerance processing when the format is abnormal; the third layer is to isolate exceptions when a single frame image fails to be determined, without interrupting the overall detection process; and the fourth layer is to fully record the request parameters, return results, and exception information for each detection.
[0033] Regarding step S2, it should be noted that the hierarchical evaluation model for the rationality of idle chemical plants constructed in this embodiment is the AHP model, namely the Analytic Hierarchy Process (AHP). AHP is a systematic analysis method that decomposes complex decision-making problems into three levels: the goal level, the criterion level, and the indicator level. It constructs a judgment matrix by comparing pairs of factors, calculates eigenvectors, and performs consistency checks, transforming qualitative judgments into quantitative weights. Its principle lies in utilizing the human ability to intuitively judge the relative importance of two factors. Through the quantitative comparison results of 1-9 levels of scaling, the weights are extracted using the matrix eigenvalue method and logical self-consistency is checked. Finally, the combined weights of each factor to the overall goal are obtained through hierarchical overall ranking.
[0034] In this embodiment, the objective layer of the model is to "determine the idle time weight coefficient". α The criteria layer comprises four dimensions: "Research Task Rationality (B1)," "Personnel Identity Differences (B2)," "Time Characteristics (B3)," and "Adjustability (B4)." In one possible implementation, the evaluation indicators belonging to each criterion dimension include at least the following: the research stage under the research task rationality dimension (C...). 11 ), Courses and Group Meetings (C) 12 ), temporary affairs reporting (C 13 Unjustified idleness (C) 14 ); Doctoral students under the dimension of personnel identity differences (C 21 Master's students (C) 22 ), Corporate Partners (C) 23 Duration under the time characteristic dimension (C) 31 ), frequency of occurrence (C) 32 ), time period regularity (C) 33 ); Equipment relocation complexity under the adjustability dimension (C 41 Negotiation costs (C) 42 ).
[0035] Furthermore, such as Figure 2 In one possible implementation, the combined weights of each evaluation index relative to the target layer are determined using the analytic hierarchy process (AHP), and the workstation idleness weight coefficient for workstation users is determined based on these combined weights, including: S21. Using the target layer as a reference, pairwise comparisons and scores are performed on each dimension of the criterion layer based on the expert scoring method to construct a target layer-criterion layer judgment matrix. S22. Using each dimension of the criteria layer as a reference, pairwise comparison and scoring are performed on each evaluation indicator belonging to that dimension in the indicator layer based on the expert scoring method, and a criteria layer-indicator layer judgment matrix is constructed. S23. Perform a consistency check on the target-criteria-level judgment matrix and the judgment matrices of each criteria-indicator layer; S24. Solve for the judgment matrix that passes the test, extract the relative weights within the level, and obtain the combined weights of each evaluation index relative to the target level through the overall hierarchical sorting; S25. Based on the evaluation indicators actually triggered by the workstation user and their corresponding combined weights, comprehensively determine the workstation idle weight coefficient for the workstation user.
[0036] Specifically, four experts in the field of laboratory management were invited to score independently, and a 9-level scoring system was used to construct judgment matrices for each level. Tables 1-5 are the judgment matrices for AB, B1-C, B2-C, B3-C, and B4-C, respectively.
[0037] The judgment matrix of Table 1A-B
[0038] The judgment matrix of Table 2B1-C
[0039] The judgment matrix of Table 3B2-C
[0040] The judgment matrix in Table 4B3-C
[0041] The judgment matrix in Table 5B4-C
[0042] Calculate the maximum eigenvalue λ of each judgment matrix. max Consistency Indicators and consistency ratio , RI It is a random consistency index, obtained through random simulation.
[0043] Taking a partial matrix from this embodiment as an example, the λ of the AB matrix... max =4.051, CI =0.017, CR =0.019<0.1; λ of the B1-C matrix max =4.080, CI =0.027, CR =0.030<0.1. After calculation, each matrix... CR The values are all less than 0.1, indicating satisfactory consistency.
[0044] After hierarchical ranking, the absolute weights of each indicator to the target layer are obtained, as detailed in the table below: Table 6. Absolute weights of each indicator to the target layer
[0045] Based on the above weights, the idle weight coefficient α can be calculated for each workstation user. For example, a doctoral student (C 21 Weight 0.159) Due to the research stage (C) 11 (Weight 0.267) and Courses and Groups (C) 12 A weight of 0.167) leads to idleness, and its α value should be significantly higher than that of unexplained idleness (C). 14 The case with a weight of 0.034.
[0046] Regarding step S3, in one possible implementation, the workstation weighted utilization rate includes the individual weighted utilization rate and the team average weighted utilization rate; like Figure 3 As shown, step S3 calculates the weighted utilization rate of the workstation based on the actual occupancy time, expected arrival time, and idle weight coefficient, including: S31. Determine the idle time based on the actual time the workstation is occupied and the expected arrival time; S32. Multiply the idle time by the idle weight coefficient to obtain the converted effective idle time; S33. Add the actual occupied time to the effective idle time to obtain the effective usage time; S34. Divide the effective usage time by the required arrival time to obtain the individual weighted usage rate; S35. Take the arithmetic mean of the individual weighted utilization rates of each workstation within the same team to obtain the team average weighted utilization rate.
[0047] For example, suppose a certain workstation is supposed to be available for 160 hours in a certain month, but is actually occupied for 80 hours and idle for 80 hours. Traditional calculation methods treat all idle time as invalid, resulting in a utilization rate of only 50%. This "one-size-fits-all" approach fails to differentiate between the reasons for idle time. Reasonable idle time due to objective reasons such as research projects or group meetings is treated the same as unjustified idle time, easily leading to conflicts between managers and users and failing to accurately reflect the actual utilization value of workstations. This embodiment introduces an idle time weighting coefficient α to differentiate idle time. Actual occupied time is fully included in the effective usage time. Idle time is determined based on the workstation user's identity, task, idle time period, and workstation adjustability using the Analytic Hierarchy Process (AHP). The reasonableness of the idle time is then calculated proportionally to α and included in the effective usage time. For example, for a certain workstation, the AHP model determines its idle time weighting coefficient α = 0.5. Therefore, out of 80 hours of idle time, 40 hours are considered effective usage, resulting in an effective usage time of 80 + 40 = 120 hours, with a weighted utilization rate of 75%. This embodiment quantifies the rationality of idle time by introducing the analytic hierarchy process and uses the idle weight coefficient α to adjust the idle time, so that the weighted utilization rate can reflect both actual work efficiency and reasonable idle value, providing a scientific basis for the refined management of laboratory workstations.
[0048] Regarding step S4, in one possible implementation, workstations are classified according to a weighted utilization rate and a preset threshold system, and the corresponding workstation configuration strategy is executed according to the classification results, including: When an individual's weighted utilization rate is continuously within the first threshold range during the first preset period, it is determined to be a high utilization rate level for the individual, and the status quo is maintained. When an individual's weighted utilization rate is continuously within the second threshold range during the first preset period, it is determined to be the individual's utilization rate level and the status quo is maintained. When an individual's weighted utilization rate is within the third threshold range during the second preset period, it is determined to be a low utilization level for that individual, and the corresponding workstation user is reminded to improve efficiency. When an individual's weighted utilization rate is within the fourth threshold range during the second preset period, it is judged as an extremely low utilization rate level, and the team leader is reminded to adjust the user of the workstation. When the average weighted utilization rate of a team is within the first threshold range for a continuous period of the first preset period, the team is judged to be at a high utilization level, and workstations will be allocated to the team first when laboratory resources are expanded. When the team's average weighted utilization rate is continuously within the second threshold range for the first preset period, it is determined to be the utilization level of the team and the status quo is maintained. When the team's average weighted utilization rate is within the third threshold range during the second preset period, it is judged as a low utilization level of the team, and the team leader is reminded to pay attention to the use of workstations. When the team's average weighted utilization rate is within the fourth threshold range during the second preset period, it is determined to be an extremely low utilization rate level for the team, and a batch recall and redistribution within the laboratory will be initiated for all workstations of the team.
[0049] The implementation process of steps S3-S4 above will be described below with reference to a specific embodiment: It is understood that this embodiment provides a dynamic configuration strategy based on weighted utilization.
[0050] (1) Calculation of Individual Weighted Utility Rate Assuming a team has 10 workstations, what is the expected arrival time for a certain month? =160 hours. Based on the large-scale model identification, the actual occupancy time of a certain workstation is... =80 hours, idle time =80 hours. According to AHP calculation, due to the rationality of the research task and the difference in personnel status, the idle weight coefficient α of this workstation user is 0.5 (indicating that half of the idle time is reasonable).
[0051] Then the weighted utilization rate of this workstation for:
[0052] According to the threshold system, this workstation is in the B-level good range (60%-80%), and the status quo should be maintained.
[0053] (2) Calculation of team average weighted utilization rate Assuming the weighted utilization rates of the 10 workstations are 75%, 82%, 45%, 90%, 60%, 30%, 88%, 70%, 55%, and 65%, respectively, then the team's average weighted utilization rate is... for:
[0054] The team as a whole is in the B-level good range (60%-80%), and should maintain the status quo.
[0055] ① If the weighted utilization rate of a certain workstation is 35% (<40%), it is in a Class D hazard, triggering the following response: Individual workstation: Remind the team leader to adjust the user of this workstation; Team: Initiate the workstation recovery procedure and redistribute workstations within the laboratory area.
[0056] ② If the weighted utilization rate of a certain workstation is 50% (40%-60%), it will be under Level C alert, triggering the following response: Individual workstations: Communicate with and remind users to improve workstation utilization efficiency; Team: Communicate and remind team leaders to pay attention to the usage of workstations.
[0057] ③ If the weighted utilization rate of a certain workstation is ≥85% for two consecutive months, it is rated as excellent (A level) and the team will be given priority for additional workstations when laboratory resources are expanded.
[0058] like Figure 4 Secondly, the present invention also provides a laboratory workstation dynamic configuration system 40 based on a large model, comprising: Image acquisition module 410 is used to acquire images of the workstation scene and upload them to the cloud-based large model; The cloud-based detection module 420 is used to call a large cloud model to detect human heads and return the coordinates of the human head bounding box. The local fusion module 430 is used to match the detected head bounding box coordinates with the predefined workstation area, determine the occupancy status of each workstation, and calculate the actual occupancy time of the workstation. The idle quantity quantification module 440 is used to construct a hierarchical evaluation model for the rationality of idle workstations, including a target layer, a criterion layer, and an indicator layer. The criterion layer includes four dimensions: rationality of scientific research tasks, differences in personnel status, time characteristics, and adjustability. The indicator layer includes several evaluation indicators belonging to each criterion dimension. The combined weight of each evaluation indicator relative to the target layer is determined by the analytic hierarchy process, and the idle weight coefficient of the workstation user is determined based on the combined weight. The utilization rate calculation module 450 is used to calculate the weighted utilization rate of the workstation based on the actual occupancy time, the expected arrival time, and the idle weight coefficient. The configuration decision module 460 is used to classify workstations according to the weighted utilization rate and preset threshold system, and execute the corresponding workstation configuration strategy according to the classification result.
[0059] In one possible implementation, a fault tolerance module 470 is also included, which is used to perform automatic retries when cloud requests fail, perform format validation and fault tolerance when returned data is abnormal, perform process isolation when single frame processing fails, and record full process logs.
[0060] Furthermore, in this embodiment, the local fusion module 430 specifically includes: a coordinate mapping unit for mapping normalized coordinates to actual pixel coordinates; a size filtering unit for filtering abnormal detection boxes that do not conform to the size of a human head; a boundary cropping unit for cropping detection boxes that exceed the image boundary to the image range; a center point calculation unit for calculating the coordinates of the center point of the detection box; and a polygon region determination unit for determining whether the center point falls within the pre-labeled polygon region of the workstation. System 40 adopts a layered architecture of "light local, heavy cloud". All large model inference is completed on the cloud GPU server, and the local terminal only undertakes lightweight tasks such as video reading, image preprocessing, interface display, and result saving.
[0061] The above are merely preferred embodiments of the present invention and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should recognize that any equivalent substitutions and obvious changes made based on the content of this specification should be included within the protection scope of the present invention.
Claims
1. A method for dynamic configuration of laboratory workstations based on a large model, characterized in that, include: Collect workstation scene images and upload them to the cloud big model for personnel head detection. Locally, match the coordinates of the detected head bounding boxes with the predefined workstation area to determine the occupancy status of each workstation and calculate the actual occupancy time of the workstation. A hierarchical evaluation model for the rationality of idle workstations is constructed, comprising a target layer, a criterion layer, and an indicator layer. The criterion layer includes four dimensions: rationality of research tasks, differences in personnel status, time characteristics, and adjustability. The indicator layer includes several evaluation indicators belonging to each criterion dimension. The combined weight of each evaluation indicator relative to the target layer is determined using the analytic hierarchy process, and the idle workstation weight coefficient for workstation users is determined based on the combined weight. The weighted utilization rate of the workstation is calculated based on the actual occupancy time, the expected arrival time, and the idle weight coefficient. Workstations are classified according to the weighted utilization rate and preset threshold system, and the corresponding workstation configuration strategy is executed according to the classification results.
2. The method for dynamic configuration of laboratory workstations based on a large model according to claim 1, characterized in that, The workstation scene images are collected at minute intervals, compressed and encoded, and then uploaded to the cloud-based large model through an authentication interface. The inference calculation for personnel head detection is completely offloaded to the cloud-based graphics processing unit, and the local terminal only performs lightweight edge tasks such as image decoding, coordinate mapping, and status recording.
3. The method for dynamic configuration of laboratory workstations based on a large model according to claim 1, characterized in that, The process of collecting workstation scene images and uploading them to a cloud-based large model for personnel head detection includes: The collected workstation scene images are preprocessed and then encoded and uploaded to the cloud-based large model API. The large model performs head detection tasks according to structured detection constraint instructions and outputs structured data containing normalized coordinates of the head bounding box. The detection constraint instructions include at least limiting the detection target to the human head, excluding non-human interference objects, ensuring the bounding box is close to the edge of the detection target, and limiting the output format to a structured array.
4. The method for dynamic configuration of laboratory workstations based on a large model according to claim 1, characterized in that, The step of collecting workstation scene images and uploading them to the cloud-based large model for personnel head detection also includes a multi-layered fault tolerance mechanism: the first layer is to automatically retry and record the exception when the cloud request fails; The second layer performs format verification on the data returned from the cloud and performs fault tolerance processing when the format is abnormal; the third layer isolates the exception when the single frame image judgment fails, without interrupting the overall detection process. The fourth layer records the request parameters, return results, and exception information for each detection.
5. The method for dynamic configuration of laboratory workstations based on a large model according to claim 1, characterized in that, The evaluation indicators, each belonging to one of the criteria dimensions, include at least the following: Research task rationality dimension: research phase, courses and group meetings, reporting of temporary affairs, and unexcused idleness; Doctoral students, master's students, and corporate partners are distinguished by their individual identities. Duration, frequency of occurrence, and periodic regularity under the time-related characteristics; The complexity of equipment relocation and the cost of negotiation under the dimension of adjustability.
6. The method for dynamic configuration of laboratory workstations based on a large model according to claim 1, characterized in that, The process of determining the combined weights of each evaluation index relative to the target layer using the analytic hierarchy process (AHP), and determining the workstation idle weight coefficient for workstation users based on the combined weights, includes: Using the target layer as a reference, pairwise comparisons and scores are performed on each dimension of the criterion layer based on expert scoring, and a target-criterion layer judgment matrix is constructed. Using each dimension of the criteria layer as a reference, the evaluation indicators belonging to each dimension in the indicator layer are compared and scored in pairs based on the expert scoring method to construct the criteria layer-indicator layer judgment matrix. A consistency check is performed on the target layer-criteria layer judgment matrix and each criteria layer-indicator layer judgment matrix; Solve the judgment matrix that passes the test, extract the relative weights within the level, and obtain the combined weights of each evaluation index relative to the target level through the overall hierarchical sorting. The idle weight coefficient of a workstation is determined by comprehensively considering the evaluation indicators actually triggered by the workstation user and their corresponding combined weights.
7. The method for dynamic configuration of laboratory workstations based on a large model according to claim 1, characterized in that, The weighted utilization rate of the workstation includes the individual weighted utilization rate and the team average weighted utilization rate; The calculation of the weighted utilization rate of workstations based on the actual occupancy time, expected arrival time, and idle weight coefficient includes: The idle time is determined based on the actual occupancy time and the expected arrival time of the workstation; Multiply the idle time by the idle weight coefficient to obtain the converted effective idle time. The effective usage time is obtained by adding the actual occupied time to the effective idle time. Divide the effective usage time by the expected arrival time to obtain the individual weighted usage rate; The team average weighted utilization rate is obtained by taking the arithmetic mean of the individual weighted utilization rates of each workstation within the same team.
8. The method for dynamic configuration of laboratory workstations based on a large model according to claim 7, characterized in that, The step of classifying workstations according to the weighted utilization rate and a preset threshold system, and executing the corresponding workstation configuration strategy according to the classification results, includes: When an individual's weighted utilization rate is continuously within the first threshold range during the first preset period, it is determined to be a high utilization rate level for the individual, and the status quo is maintained. When an individual's weighted utilization rate is continuously within the second threshold range during the first preset period, it is determined to be the individual's utilization rate level and the status quo is maintained. When an individual's weighted utilization rate is within the third threshold range during the second preset period, it is determined to be a low utilization level for that individual, and the corresponding workstation user is reminded to improve efficiency. When an individual's weighted utilization rate is within the fourth threshold range during the second preset period, it is judged as an extremely low utilization rate level, and the team leader is reminded to adjust the user of the workstation. When the average weighted utilization rate of a team is within the first threshold range for a continuous period of the first preset period, the team is judged to be at a high utilization level, and workstations will be allocated to the team first when laboratory resources are expanded. When the team's average weighted utilization rate is continuously within the second threshold range for the first preset period, it is determined to be the utilization level of the team and the status quo is maintained. When the team's average weighted utilization rate is within the third threshold range during the second preset period, it is judged as a low utilization level of the team, and the team leader is reminded to pay attention to the use of workstations. When the team's average weighted utilization rate is within the fourth threshold range during the second preset period, it is determined to be an extremely low utilization rate level for the team, and a batch recall and redistribution within the laboratory will be initiated for all workstations of the team.
9. A dynamic configuration system for laboratory workstations based on a large model, characterized in that, include: The image acquisition module is used to acquire images of the workstation scene and upload them to the cloud-based large model; The cloud-based detection module is used to call a large cloud model to detect human heads and return the coordinates of the human head bounding box. The local fusion module is used to match the detected head bounding box coordinates with the predefined workstation area, determine the occupancy status of each workstation, and calculate the actual occupancy time of the workstation. The idle quantity quantification module is used to construct a hierarchical evaluation model for the rationality of idle workstations, including a target layer, a criterion layer, and an indicator layer. The criterion layer includes four dimensions: rationality of scientific research tasks, differences in personnel status, time characteristics, and adjustability. The indicator layer includes several evaluation indicators belonging to each criterion dimension. The combined weight of each evaluation indicator relative to the target layer is determined by the analytic hierarchy process, and the idle weight coefficient of the workstation user is determined based on the combined weight. The utilization rate calculation module is used to calculate the weighted utilization rate of the workstation based on the actual occupation time of the workstation, the expected arrival time, and the idle weight coefficient. The configuration decision module is used to classify workstations according to the weighted utilization rate and preset threshold system, and execute the corresponding workstation configuration strategy according to the classification result.
10. The laboratory workstation dynamic configuration system based on a large model according to claim 9, characterized in that, It also includes a fault tolerance module, which is used to perform automatic retries when cloud requests fail, perform format validation and fault tolerance when returned data is abnormal, perform process isolation when single frame processing fails, and record full process logs.