An industrial design project resource scheduling management method and system

Abstract

The application relates to the field of industrial design project management, and discloses an industrial design project resource scheduling management method and system, which integrates hardware performance data, designer behavior data and project context information, constructs a more accurate resource release prediction mechanism, introduces elastic task scheduling on the basis, effectively alleviates the efficiency loss caused by resource conflicts and waiting, and realizes intelligentization and dynamic optimization of resource scheduling.

Description

Technical Field

[0001] This application relates to the field of industrial design project management, and more specifically, to a method and system for resource scheduling and management of industrial design projects. Background Technology

[0002] In the field of modern industrial design, project teams often need to handle multiple design tasks simultaneously, covering the entire process from initial concept development, 3D model building, rendering, prototype production, to final review. To ensure efficient project progress and on-time delivery, many design agencies have introduced resource scheduling and management systems. These systems aim to optimize the allocation of various resources, including designers with specific skills, high-performance graphics workstations, 3D printers, CNC machining centers, and various professional design software licenses. However, in practice, due to designers' informal "soft reservation" of resources and the dynamic changes in the inherent complexity of tasks, discrepancies often arise between the actual occupancy of physical resources and the system records, leading to a series of efficiency losses and management challenges. Summary of the Invention

[0003] This application discloses a resource scheduling and management method and system for industrial design projects, aiming to solve a series of management problems in the field of industrial design, such as inconsistency between the actual physical resource occupancy information and the system records caused by the designer's informal "soft reservation" behavior of resources and the dynamic changes in the inherent complexity of tasks, which leads to efficiency loss, hidden conflicts, resource waste and data pollution.

[0004] The technical solution of this application is as follows:

[0005] Firstly, this application discloses a resource scheduling and management method for industrial design projects, specifically including:

[0006] By deploying a hardware monitoring sensor group on a graphics workstation dedicated to industrial design, the real-time operating parameters of the CPU, GPU, memory and disk I / O subsystems of the graphics workstation are continuously collected according to a preset sampling period, and the resource operation instability index is calculated based on the real-time operating parameters.

[0007] By deploying an operation behavior acquisition agent on the designer's terminal, the operation behavior event flow of the designer currently occupying the graphics workstation during the period when the workstation is occupied is captured in real time during the period when the designer is performing 3D modeling or rendering tasks. Based on the operation behavior event flow, non-productive high-frequency operation patterns are identified, and combined with the designer's historical operation baseline, an abnormal operation index of the designer is generated.

[0008] Using the resource operation instability index, designer operation anomaly index, remaining estimated execution time of the current task, and current project stage information as input features, the system outputs the release probability of the graphics workstation at each discrete time point in the future through a pre-trained resource release probability prediction model, and generates a resource release time probability distribution based on the release probability.

[0009] When a target designer waiting for a graphics workstation is unable to start their core task due to the continuous occupation of the graphics workstation, the probability distribution of resource release time is used to determine whether the target designer should enter the elastic waiting mode.

[0010] If the target designer enters the flexible waiting mode, the system will start from the target designer's skill tag set, filter out flexible tasks from the candidate task pool that match the skill tags, do not rely on a graphics workstation, and have an estimated waiting time shorter than that of a graphics workstation, and then recommend the selected flexible tasks to the target designer in descending order of their value contribution index to the downstream stages of the project.

[0011] The system receives the target designer's response to the elastic task, updates the allocation status of the elastic task and the target designer's task execution time data, and associates the updated task execution time data with the actual release time of the graphics workstation. This data is then fed back as a new training sample to the resource release probability prediction model for online incremental training, so that the subsequent resource release time probability distribution prediction results are continuously optimized as scheduling data accumulates.

[0012] Secondly, this application also discloses an industrial design project resource scheduling and management system, specifically including:

[0013] The hardware monitoring module is used to continuously collect real-time operating parameters of the CPU, GPU, memory and disk I / O subsystems of the graphics workstation at a preset sampling period through a hardware monitoring sensor group deployed in the graphics workstation for industrial design, and calculate the resource operation instability index based on the real-time operating parameters.

[0014] The behavior acquisition module is used to capture the operation behavior event flow of the designer currently occupying the graphics workstation during the period when the graphics workstation is occupied, through the operation behavior acquisition agent deployed on the designer terminal. Based on the operation behavior event flow, it identifies non-productive high-frequency operation patterns and generates a designer operation anomaly index by combining the designer's historical operation baseline.

[0015] The prediction module takes the resource operation instability index, the designer operation anomaly index, the remaining estimated execution time of the current task, and the current project stage information as input features, and outputs the release probability of the graphics workstation at each discrete time point in the future through a pre-trained resource release probability prediction model, and generates a resource release time probability distribution based on the release probability.

[0016] The status judgment module is used to determine whether the target designer should enter the elastic waiting mode when the target designer is unable to start its core task due to the continuous occupation of the graphics workstation.

[0017] The task filtering and recommendation module is used to filter out flexible tasks from the candidate task pool that match the skill tags, do not rely on a graphics workstation, and have an estimated waiting time shorter than that of a graphics workstation if the target designer enters the flexible waiting mode. The selected flexible tasks are then sorted from high to low according to their value contribution index to the downstream links of the project and recommended to the target designer.

[0018] The feedback update module is used to receive the target designer's response to the elastic task, update the allocation status of the elastic task and the target designer's task execution time data, and associate the updated task execution time data with the actual release time of the graphics workstation. This data is then fed back as a new training sample to the resource release probability prediction model for online incremental training, so that the subsequent resource release time probability distribution prediction results are continuously optimized as scheduling data accumulates.

[0019] The resource scheduling and management method and system for industrial design projects disclosed in this application effectively solves problems such as information blind spots, hidden conflicts, resource waste, and data pollution caused by resource "soft reservation" and dynamic changes in task complexity in existing technologies through multi-dimensional data collection and intelligent prediction. Specifically, this method collects the operating parameters of the graphics workstation in real time through a hardware monitoring sensor group and calculates the resource operation instability index, thereby quantifying the probability of resource release delay caused by hardware failure and overcoming the limitation of insufficient hardware status perception in traditional systems. At the same time, it captures the designer's operation behavior event stream through an operation behavior acquisition agent, identifies non-productive high-frequency operation patterns, and generates a designer operation anomaly index by combining historical operation baselines, effectively quantifying the impact of the designer's subjective factors on resource occupation time and making up for the shortcomings of traditional systems that only focus on task planning and ignore human factors.

[0020] Building upon this foundation, this method uses resource instability index, designer operation anomaly index, remaining estimated task execution time, and project stage information as inputs. Through a pre-trained resource release probability prediction model, it outputs the release probability of the graphics workstation at each discrete future time point and generates a resource release time probability distribution. This prediction mechanism can dynamically and accurately assess the uncertainty of resource release, providing a scientific basis for subsequent scheduling decisions and significantly outperforming traditional static resource calendars. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating a resource scheduling and management method for industrial design projects provided in an embodiment of the present invention;

[0022] Figure 2 This is a flowchart of a method for calculating a resource instability index based on real-time operating parameters, provided by an embodiment of the present invention.

[0023] Figure 3 This is a schematic diagram of the structure of an industrial design project resource scheduling and management system provided in an embodiment of the present invention. Detailed Implementation

[0024] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0025] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0026] In modern industrial design projects, resource scheduling and management face numerous challenges, including discrepancies between actual physical resource usage and system records, inaccurate time estimates due to the dynamic changes in the inherent complexity of tasks, and implicit conflicts arising from designers' informal "soft reservations." These problems not only lead to project delays and resource waste but also make it difficult for project management to accurately grasp project progress and risks, ultimately affecting design quality and delivery timelines.

[0027] In this regard, refer to Figure 1 , Figure 1 This is a flowchart illustrating a resource scheduling and management method for industrial design projects provided in an embodiment of the present invention, including:

[0028] S11, by deploying a hardware monitoring sensor group on a graphics workstation dedicated to industrial design, the real-time operating parameters of the CPU, GPU, memory and disk I / O subsystems of the graphics workstation are continuously collected according to a preset sampling period, and the resource operation instability index is calculated based on the real-time operating parameters.

[0029] S12, by deploying an operation behavior acquisition agent on the designer's terminal, during the time period when the graphics workstation is occupied, the operation behavior event stream of the designer currently occupying the graphics workstation during the execution of 3D modeling or rendering tasks is captured in real time, and non-productive high-frequency operation patterns are identified based on the operation behavior event stream, and a designer operation anomaly index is generated by combining the designer's historical operation baseline.

[0030] S13, using the resource operation instability index, the designer operation anomaly index, the remaining estimated execution time of the current task, and the current project stage information as input features, the graphics workstation is output as a release probability at each discrete time point in the future through a pre-trained resource release probability prediction model, and a resource release time probability distribution is generated based on the release probability.

[0031] S14, when the target designer waiting for the graphics workstation is unable to start its core task due to the continuous occupation of the graphics workstation, determine whether the target designer enters the elastic waiting mode based on the probability distribution of the resource release time.

[0032] S15, if the target designer enters the flexible waiting mode, then starting from the target designer's skill tag set, select flexible tasks from the candidate task pool that match the skill tags, do not depend on the graphics workstation, and have an expected waiting time less than the expected waiting time of the graphics workstation, and recommend the selected flexible tasks to the target designer after sorting them from high to low according to their value contribution index to the downstream links of the project.

[0033] S16, receive the target designer's response to the elastic task, update the allocation status of the elastic task and the target designer's task execution time data, and associate the updated task execution time data with the actual release time of the graphics workstation, feeding it back as a new training sample to the resource release probability prediction model for online incremental training, so that the subsequent resource release time probability distribution prediction results are continuously optimized as scheduling data accumulates.

[0034] This application integrates hardware performance data, designer behavior data, and project context information to construct a more accurate resource release prediction mechanism. Based on this, it introduces elastic task scheduling, which effectively alleviates the efficiency loss caused by resource conflicts and waiting, and realizes intelligent and dynamic optimization of resource scheduling.

[0035] The term "graphics workstation" as used in this application refers to a high-performance computer system specifically designed for handling complex graphics calculations and 3D design tasks. It is typically equipped with a powerful central processing unit (CPU), graphics processing unit (GPU), large-capacity memory, and a high-speed disk I / O subsystem. This hardware configuration enables it to run professional design software such as 3D modeling, rendering, and animation smoothly. In industrial design projects, graphics workstations are key tools for designers to perform core creative work and implement technical solutions.

[0036] The "hardware monitoring sensor suite" refers to various sensors integrated inside or outside the graphics workstation to collect real-time hardware operating status data. Examples include CPU temperature and load sensors, GPU utilization and memory usage sensors, memory read / write speed sensors, and disk I / O throughput and latency sensors. These sensors work together to provide the system with a comprehensive view of its hardware performance.

[0037] "Operation Behavior Acquisition Agent" refers to a lightweight software module deployed on the designer's terminal (usually the graphics workstation itself). Its main function is to monitor and record the designer's interactive behaviors in the operating system and design software in real time, such as keyboard input, mouse clicks, drags and drops, software menu selections, file saving, undo and redo operations, etc. These events constitute the designer's operation behavior event flow, which is an important data source for analyzing the designer's work patterns and efficiency.

[0038] The "Resource Instability Index" is a quantitative indicator used to assess the stability of a graphics workstation hardware system's operating status over a specific time period. A higher index indicates a greater probability of performance fluctuations, potential failures, or abnormal behavior in the hardware system, which may affect the normal execution of tasks and the timely release of resources.

[0039] The "Designer Operation Anomaly Index" is another quantitative indicator used to assess the degree to which a designer's operational behavior deviates from normal productive patterns while occupying a graphics workstation. The higher the index, the more likely the designer is to engage in unproductive, repetitive, or inefficient operations due to subjectively perceived performance issues, task obstruction, or distraction, which may prolong the actual time spent on the graphics workstation.

[0040] The "Resource Release Probability Prediction Model" is a prediction model based on machine learning or deep learning techniques. Its core function is to predict the likelihood of a graphics workstation being released at different points in the future, based on multi-source input features. By learning complex patterns in historical data, this model can output a probability distribution rather than a simple fixed time point, thus better reflecting the uncertainty of resource release.

[0041] "Elastic Waiting Mode" refers to an alternative work state provided by the system when a target designer is unable to immediately begin their main task due to the core resource (such as a graphics workstation) being occupied. In this mode, the designer no longer waits passively but can actively choose to perform some auxiliary tasks that do not depend on the core resource and are expected to take less time, so as to make full use of the waiting time and improve overall work efficiency.

[0042] "Flexible tasks" refer to auxiliary tasks that do not rely on specific high-performance resources (such as graphics workstations) and contribute value to the overall project schedule. These tasks typically have a short estimated duration and can be flexibly completed by designers while waiting for core resources, such as document preparation, design review preparation, concept sketching, and market research.

[0043] The "Value Contribution Index" is a quantitative metric used to assess the importance or impact of a task on the achievement of the overall project goals. This index comprehensively considers factors such as the task's output's influence on downstream project stages, the importance of those stages, and the task's own priority, aiming to help designers select the most valuable task from multiple flexible assignments.

[0044] This application utilizes a hardware monitoring sensor array deployed in a graphics workstation designed for industrial design to continuously collect real-time operating parameters of the workstation's CPU, GPU, memory, and disk I / O subsystems at preset sampling intervals. For example, the sensor array can collect data every 5 seconds on CPU utilization and temperature, GPU memory usage and rendering queue length, available memory capacity and swap space utilization, and disk I / O read / write speed and queue depth. These real-time operating parameters reflect the current load and health status of the graphics workstation. Based on these parameters, the system can calculate a resource instability index. For instance, it can determine whether there is a hardware instability trend by analyzing drastic fluctuations in CPU or GPU utilization, signs of memory leaks, or abnormal increases in disk I / O latency.

[0045] Simultaneously, through an operation behavior acquisition agent deployed on the designer's terminal, the system captures in real-time the flow of operational behavior events during the time the graphics workstation is occupied, specifically during 3D modeling or rendering tasks performed by the designer currently using the workstation. For example, the agent can record frequent "save," "undo," and "redo" operations in CAD software, or frequent switching between different design software. These operational behavior event flows reflect the designer's work habits and efficiency. Based on these event flows, the system can identify non-productive, high-frequency operation patterns, such as saving the same file multiple times in a short period or frequently adjusting minor parameters. Combined with the designer's historical operational baseline (e.g., the designer's average operation frequency and patterns in similar tasks), the system can generate a designer operation anomaly index. For example, if a designer's frequency of undoing and redoing operations in a task is significantly higher than their historical average, it may indicate a high operation anomaly index.

[0046] Subsequently, using resource instability index, designer operation anomaly index, remaining estimated execution time of the current task, and current project stage information as input features, a pre-trained resource release probability prediction model outputs the release probability of the graphics workstation at various discrete time points in the future. For example, the model can predict the probability of the graphics workstation being released at time points of 10 minutes, 30 minutes, 1 hour, and 2 hours. Based on these release probabilities, the system can generate a resource release time probability distribution, intuitively displaying the probability trend of when the graphics workstation might be released.

[0047] When a target designer waiting for a graphics workstation is unable to start their core tasks due to the workstation being continuously occupied, the system will determine whether the target designer should enter the flexible waiting mode based on the probability distribution of resource release time. For example, if the prediction model shows that the cumulative probability of the graphics workstation being released in the next hour is lower than a certain preset threshold (such as 30%), the system will consider the waiting time to be long and the uncertainty to be high, and will prompt the target designer to consider entering the flexible waiting mode.

[0048] If the target designer chooses to enter the flexible waiting mode, the system will start from the target designer's skill tag set and filter from the candidate task pool for flexible tasks that match the skill tags, do not rely on a graphics workstation, and have an estimated waiting time shorter than that of a graphics workstation. For example, if the target designer is skilled in "concept sketching" and "user interface design," and the graphics workstation is expected to take another 2 hours, the system will filter from the task pool for tasks that require "concept sketching" or "user interface design" skills, do not require a high-performance graphics workstation, and have an estimated waiting time of less than 2 hours. The selected flexible tasks will be recommended to the target designer after being sorted from high to low according to their value contribution index to downstream stages of the project. For example, a sketching task that provides a preliminary solution for subsequent structural design may have a higher value contribution index than a task that organizes historical documents.

[0049] Finally, the system receives the target designer's response to the elastic task (e.g., accepting or rejecting a recommended task) and updates the assignment status of the elastic task and the target designer's task execution time data. For example, if a designer accepts an elastic task, the task status will be updated to "in progress," and the designer's time spent on that task will begin to be recorded. The updated task execution time data is correlated with the actual release time of the graphics workstation and fed back as new training samples to the resource release probability prediction model for online incremental training. For example, if the graphics workstation is released one hour after the designer executes the elastic task, then this correlation data between the "one-hour elastic task execution time" and the "actual release time" will be used for model training, so that the subsequent resource release time probability distribution prediction results will be continuously optimized as scheduling data accumulates.

[0050] The resource scheduling and management method for industrial design projects proposed in this application effectively addresses the shortcomings of traditional scheduling systems in dealing with the uncertainties of industrial design projects by introducing multi-dimensional data collection and intelligent prediction mechanisms.

[0051] Traditional resource scheduling systems rely primarily on static resource calendars and preset task priority rules, making it difficult to accurately capture the actual occupancy of physical resources and the dynamic behavior of designers. When graphics workstations are informally "soft-reserved" or their occupancy time is extended due to task complexity, the system fails to detect this in a timely manner, resulting in designers waiting for resources being idle for extended periods, leading to a waste of high-value skills and software licenses. Furthermore, traditional systems lack the ability to identify designers' non-productive operations and cannot correlate these behaviors with resource occupancy time, causing a significant accumulation of "noise" in the estimated work hours data.

[0052] In summary, this application constructs a dynamic and adaptive resource scheduling and management method for industrial design projects through multi-source data fusion, intelligent prediction, and flexible task scheduling. This method significantly improves resource utilization efficiency, shortens project cycles, and provides more accurate decision-making basis for project management, thereby effectively solving many problems existing in the prior art.

[0053] In some embodiments described above, the resource instability index is calculated by collecting real-time operating parameters from the graphics workstation. However, in practical applications, these real-time operating parameters may be affected by various factors, such as sensor noise, transient performance fluctuations caused by background processes, and some non-critical instantaneous performance jitters. Without effective processing and verification, the resource instability index calculated directly based on the original real-time operating parameters may be inaccurate and prone to misjudgment, thereby affecting the reliability of subsequent resource release probability prediction.

[0054] In this regard, refer to Figure 2 , Figure 2 This is a flowchart of a method for calculating a resource instability index based on real-time operating parameters according to an embodiment of the present invention. S11 includes:

[0055] S111, perform noise suppression processing on the real-time operating parameters based on sliding median filtering to obtain a sequence of performance parameters after noise suppression;

[0056] S112, Match the noise-suppressed performance parameter sequence with the pre-established baseline operating mode library to identify and eliminate background performance fluctuations caused by system background processes;

[0057] S113, determine whether the duration of the remaining performance fluctuation after removing the background performance fluctuation is greater than the first duration threshold and whether the fluctuation amplitude is greater than the first amplitude threshold. If both are satisfied, cross-validate the remaining performance fluctuation with the system error log of the graphics workstation.

[0058] S114, when the cross-validation confirms that the remaining performance fluctuation is caused by intermittent hardware failure, the resource operation instability index is generated; wherein, the resource operation instability index is used to quantify the probability that the graphics workstation will fail in the future and cause a delay in resource release time.

[0059] Specifically, the real-time operating parameters refer to the real-time operating data of the CPU, GPU, memory, and disk I / O subsystems continuously collected by a hardware monitoring sensor array. To eliminate potential random noise and transient spikes in this raw data, noise suppression processing based on sliding median filtering is applied to the real-time operating parameters. Sliding median filtering is a nonlinear digital filtering technique that effectively suppresses impulse noise by replacing the value of each point in the data sequence with the median of its neighborhood data, while also preserving the edge information of the signal, thus obtaining a smoother and more realistic performance parameter sequence.

[0060] The noise-suppressed performance parameter sequence may still contain periodic or sporadic performance fluctuations caused by system background processes (such as operating system updates, virus scans, data backups, etc.). To distinguish these normal background fluctuations from genuine resource instability, the noise-suppressed performance parameter sequence is matched against a pre-established baseline operating pattern library. This library stores typical performance fluctuation patterns caused by various known system background processes during normal operation of the graphics workstation. Through pattern matching, these background performance fluctuations can be identified and eliminated, thereby focusing on abnormal fluctuations related to actual resource instability.

[0061] Furthermore, after removing the background performance fluctuations, the remaining performance fluctuations are analyzed. It is determined whether the duration of the remaining performance fluctuations exceeds a preset first duration threshold, and whether the fluctuation amplitude exceeds a preset first amplitude threshold. These two thresholds are used to filter out short-lived, low-impact instantaneous fluctuations, ensuring that only significant and persistent performance anomalies that may indicate potential problems are considered. If both conditions are met, the remaining performance fluctuations are cross-validated with the system error log of the graphics workstation. The system error log records key events such as hardware failures, driver errors, and system crashes. By correlating performance fluctuations with the error log, it is possible to more accurately determine whether the performance anomalies are indeed caused by intermittent hardware failures.

[0062] Therefore, the resource instability index is generated only when the cross-validation confirms that the remaining performance fluctuations are caused by intermittent hardware failures. The resource instability index is a quantitative indicator used to assess the probability that the graphics workstation will fail in the future, resulting in a delay in resource release time.

[0063] This application's solution effectively improves the calculation accuracy of the resource operation instability index through multi-stage refined processing. Using the aforementioned technical solution, this application significantly improves the calculation accuracy and reliability of the resource operation instability index. Compared to direct calculation based on raw real-time operating parameters, this solution effectively eliminates various interference factors through noise suppression, background fluctuation removal, threshold judgment, and cross-validation with system error logs. This allows the generated resource operation instability index to more accurately quantify the probability of future failures of the graphics workstation, leading to resource release time delays. This helps avoid resource scheduling decision errors caused by misjudgments, thus providing higher-quality and more reliable input features for subsequent resource release probability prediction models, thereby improving the overall efficiency and accuracy of the resource scheduling management method for the entire industrial design project.

[0064] In some preferred embodiments, suppose the GPU temperature of a graphics workstation experiences multiple small fluctuations within a short period, but the duration of these fluctuations is less than a first duration threshold (e.g., 5 seconds), and the fluctuation amplitude is less than a first amplitude threshold (e.g., 2 degrees Celsius). In this case, even if fluctuations exist, they will not be judged as potential resource instability. For example, if the system is performing a weekly automatic virus scan, causing a sustained increase in CPU utilization over a period of time, the system can identify this as normal background process behavior by matching it with a pre-established baseline runtime mode library, thus removing it and avoiding misjudging it as resource instability. Further, if the GPU temperature continues to rise and exceeds the first amplitude threshold, and the duration exceeds the first duration threshold, the system will further check the system error log. If the error log records information such as "GPU driver crash" or "memory error," the system will confirm that the performance fluctuation is caused by intermittent hardware failure and generate a high resource instability index accordingly. Conversely, if the performance fluctuation is significant but there are no related records in the error log, further analysis may be necessary, or it may not be directly attributed to hardware failure, thus avoiding unnecessary resource scheduling adjustments. This sophisticated judgment mechanism ensures that the generation of the resource instability index is based on sufficient and reliable evidence.

[0065] In some embodiments described above in this application, an operation behavior acquisition agent deployed on a designer's terminal is proposed to capture, in real time, the operation behavior event stream of the designer currently occupying the graphics workstation during the period when the workstation is occupied, while performing 3D modeling or rendering tasks. Based on the operation behavior event stream, non-productive high-frequency operation patterns are identified, and combined with the designer's historical operation baseline, a designer operation anomaly index is generated. Specifically, the step of generating the designer operation anomaly index includes:

[0066] The operation behavior acquisition agent captures the keyboard, mouse and software interface operation events of the designer on the graphics workstation to obtain the operation behavior event stream.

[0067] Based on predefined operation pattern recognition rules, non-productive high-frequency operation patterns are identified in the operation behavior event stream; the non-productive high-frequency operation patterns include at least one of repeated save operations, frequent undo and redo operations, unnecessary parameter fine-tuning operations, and abnormally frequent software switching operations.

[0068] Statistics were compiled on the types, frequency per unit time, and cumulative duration of each operation identified as a non-productive high-frequency operation.

[0069] Based on the type, the frequency of occurrence per unit time, and the cumulative duration, combined with the designer's historical operation baseline, the designer's operation anomaly index is calculated using a pre-trained anomaly scoring model; wherein, the designer's operation anomaly index is used to quantify the probability that the designer takes non-productive operations due to subjective perception of the graphics workstation's performance instability, thereby prolonging the actual occupancy time of the graphics workstation.

[0070] Specifically, the operation behavior acquisition agent is a software component deployed on the designer's terminal. Its main function is to monitor and record all interactive behaviors performed by the designer on the graphics workstation in real time. These interactive behaviors cover keyboard input, mouse clicks, drags, scrolling, and other operations, as well as menu selections, parameter adjustments, and tool usage events on various design software interfaces. All these captured events are aggregated in chronological order to form the operation behavior event stream.

[0071] Furthermore, to identify non-productive high-frequency operation patterns from the operational event stream, the system analyzes the data based on predefined operation pattern recognition rules. These rules are a series of pre-defined logical conditions or behavioral patterns designed to distinguish between effective productive operations by the designer and non-productive operations that may lead to inefficiency or extended time. The non-productive high-frequency operation patterns can include various specific manifestations. For example, repeated save operations may indicate the designer's concern about data loss or distrust of the graphics workstation's performance stability; frequent undo / redo operations may reflect the designer's indecisiveness in design decisions or a high rate of operational errors; unnecessary parameter fine-tuning operations refer to the designer excessively or repeatedly adjusting minor parameters without affecting the final design effect, which may be a sign of procrastination or distraction; abnormally frequent software switching operations may mean that the designer is frequently jumping between different applications, unable to focus on the core task, or attempting to circumvent the difficulty of the current task. When at least one of these patterns is identified, it is determined to be a non-productive high-frequency operation.

[0072] After identifying high-frequency, non-productive operations, the system will perform detailed statistics on these operations. These statistics include the specific type of each operation (e.g., save, undo, or software switch), its frequency of occurrence per unit of time (e.g., per minute), and the cumulative duration of these non-productive operations. This statistical data provides a quantitative basis for subsequent anomaly index calculations.

[0073] Ultimately, the designer's operational anomaly index is calculated based on the types, frequency per unit time, and cumulative duration obtained from the aforementioned statistics, combined with the designer's historical operational baseline. The designer's historical operational baseline is a personalized dataset recording the designer's typical operational patterns and efficiency levels under normal working conditions, helping to differentiate individual differences. These input features are comprehensively evaluated through a pre-trained anomaly scoring model to calculate the designer's operational anomaly index. The anomaly scoring model is a machine learning-trained model capable of judging the degree of anomaly in operational behavior based on input data. The designer's operational anomaly index is a quantitative indicator aimed at assessing the probability that a designer, due to a subjective perception of the graphics workstation's performance instability, engages in unproductive operations, thereby prolonging the actual occupancy time of the graphics workstation.

[0074] This application's solution, by introducing in-depth analysis of designer operational behavior, can capture potential factors prolonging resource occupancy from a subjective human perspective. Through the aforementioned technical solution, this application can more comprehensively assess the actual release time of the graphics workstation. The introduction of the designer operation anomaly index compensates for the shortcomings of relying solely on hardware monitoring, effectively capturing non-productive operations caused by designers' subjectively perceived performance instability, which often prolong the actual occupancy time of the graphics workstation. Therefore, the resource release probability prediction model can obtain more accurate input features, thereby outputting a more reliable resource release time probability distribution, providing a more accurate decision-making basis for flexible task scheduling, and ultimately improving the overall efficiency and accuracy of resource scheduling in industrial design projects.

[0075] In some embodiments of this application described above, in order to improve the accuracy and comprehensiveness of the prediction, the step of generating the resource release time probability distribution based on the release probability can be further refined into the following process.

[0076] The step of generating a resource release time probability distribution based on the release probability includes:

[0077] Obtain the remaining estimated execution time and task type of the current task, as well as the current project stage information and team collaboration density, to obtain project context information;

[0078] The resource operation instability index, the designer operation anomaly index, the remaining estimated execution time, the task type, the project stage information, and the team collaboration density are concatenated into a multi-source input feature vector;

[0079] The multi-source input feature vector is input into the resource release probability prediction model, and the release probability value of the graphics workstation at each preset discrete time point in the future is output.

[0080] The resource release time probability distribution is constructed based on the release probability values ​​at each preset discrete time point.

[0081] Specifically, before generating the probability distribution of resource release times, it is first necessary to obtain the remaining estimated execution time and task type of the current task, as well as the current project stage information and team collaboration density, thereby obtaining project context information. The task type can be a classification of the nature of the task, such as 3D modeling, rendering, post-processing, etc.; the project stage information can indicate the current lifecycle stage of the project, such as conceptual design, detailed design, prototype development, etc.; and the team collaboration density can quantify the degree of collaboration among team members in the current project. This information collectively constitutes a comprehensive description of the current task and project environment, aiming to provide richer contextual data for subsequent predictive models.

[0082] Furthermore, the resource operation instability index, the designer operation anomaly index, the remaining estimated execution time, the task type, the project stage information, and the team collaboration density are concatenated into a multi-source input feature vector. This feature vector integrates multi-dimensional information such as hardware performance, designer behavior, task attributes, and project environment, providing comprehensive input for the resource release probability prediction model.

[0083] Subsequently, the multi-source input feature vector is input into a pre-trained resource release probability prediction model, which is configured to output the release probability value of the graphics workstation at various preset discrete time points in the future. These discrete time points can be preset time intervals, such as every 15 minutes, 30 minutes, or 1 hour.

[0084] Finally, based on the release probability values ​​at each preset discrete time point, the resource release time probability distribution is constructed. This distribution intuitively shows the probability that the graphics workstation will be released at different future time points, providing a quantitative basis for subsequent elastic waiting mode judgment.

[0085] This application's solution, when generating the resource release time probability distribution, considers not only hardware operating status and designer operational behavior, but also incorporates project context information such as task type, project stage information, and team collaboration density. Because this additional contextual information is integrated into the multi-source input feature vector, the resource release probability prediction model can more comprehensively understand the current graphics workstation occupancy and its potential release dynamics. For example, different types of tasks (such as rendering and modeling) have significantly different occupancy patterns and durations on the graphics workstation; the project stage (such as initial concept design and later detail optimization) also affects the designer's work rhythm and resource needs; and team collaboration density may reflect the frequency of task handover and resource sharing. By taking these factors into account, the model can capture more refined and complex release patterns, thereby improving the accuracy and reliability of the prediction.

[0086] This application further proposes a method for training the above-mentioned resource release probability prediction model. By adopting a time-series prediction model that integrates a multi-head self-attention mechanism and a gated recurrent unit, and designing a specific training process, the performance and accuracy of the prediction model are improved.

[0087] The above method also includes training the resource release probability prediction model, which is a temporal prediction model that integrates a multi-head self-attention mechanism and a gated recurrent unit. Its training process includes:

[0088] Multi-source time-series data from historical project data are collected as training samples. The multi-source time-series data includes historical resource operation instability index sequence, historical designer operation anomaly index sequence, historical task remaining execution time sequence, and historical project stage sequence.

[0089] Using the multi-source time-series data as input and the actual release time of historical graphics workstations as supervision labels, the initial time-series prediction model is trained.

[0090] During training, the contribution weights of each data source in the multi-source time series data to the release probability prediction at each time step are adaptively adjusted through the multi-head self-attention mechanism to obtain the trained resource release probability prediction model.

[0091] Specifically, the resource release probability prediction model is designed as a time-series prediction model that integrates a multi-head self-attention mechanism and a gated recurrent unit (GRU). The gated recurrent unit is a variant of a recurrent neural network (RNN), particularly suitable for processing and predicting sequential data, effectively capturing long-term dependencies in time series. The multi-head self-attention mechanism allows the model to simultaneously focus on information at different locations when processing the input sequence, assigning different weights to these information, thereby better understanding complex patterns in the data and the interactions between different features.

[0092] When training this model, it is first necessary to collect multi-source time-series data from historical project data as training samples. This multi-source time-series data specifically includes historical resource instability index sequences, historical designer operation anomaly index sequences, historical task remaining execution time sequences, and historical project stage sequences. These sequences reflect the operating status of the graphics workstation, designer behavior patterns, task progress, and the macro-stage of the project, providing the model with comprehensive contextual information.

[0093] Subsequently, these collected multi-source time-series data were used as input features for the model, and the actual release times of historical graphics workstations were used as supervision labels. The supervision labels are the targets of model learning, i.e., the true values ​​the model needs to predict. The initial time-series prediction model was trained by pairing the input data with the corresponding true release times. The training process aims to enable the model to learn the mapping relationship from input features to release times.

[0094] Furthermore, a multi-head self-attention mechanism plays a crucial role during training. This mechanism is used to adaptively adjust the contribution weights of each data source in the multi-source time-series data to the release probability prediction at each time step. This means that the model can dynamically determine which historical data sources (e.g., resource instability index, designer operation anomaly index, etc.) are more important to the prediction result at which time point according to the needs of the current prediction task, and accordingly increase their weights while decreasing the weights of less important data sources. In this way, the model can more accurately capture the key factors affecting resource release, thereby obtaining a trained resource release probability prediction model.

[0095] This application proposes a time-series prediction model that integrates a multi-head self-attention mechanism with a gated recurrent unit (GRU) and its specific training process. This aims to address the limitations of traditional prediction models when processing complex, multi-source time-series data. The GRU, as an effective sequence processing network, can capture long-term dependencies in time-series data such as graphics workstation operating parameters and designer actions, thus providing a better understanding of the dynamic processes of resource occupancy and release. However, a single GRU model may struggle to effectively distinguish the impact of different data sources (such as hardware status, user behavior, and task progress) on the final release probability at different time points.

[0096] To address this, a multi-head self-attention mechanism is integrated into the model, enabling it to adaptively learn and adjust the contribution weights of each data source to the release probability prediction at different time steps during training. Specifically, when the graphics workstation experiences intermittent hardware failures, the multi-head self-attention mechanism may assign higher weights to the "resource instability index sequence"; while when designers frequently perform unproductive operations due to task setbacks, the mechanism may focus more on the "designer operation anomaly index sequence." This dynamic weight adjustment mechanism allows the model to focus more flexibly and accurately on the most relevant input information, overcoming the shortcomings of traditional models that treat all input features equally, and significantly improving the model's ability to understand complex situations and its prediction accuracy.

[0097] This application further proposes the above-mentioned steps for determining whether the target designer has entered the flexible waiting mode, in order to provide a more explicit and user-friendly decision-making mechanism. In the above method, the step of determining whether the target designer has entered the flexible waiting mode based on the resource release time probability distribution includes:

[0098] When the core task of the target designer cannot be started due to the continuous occupation of the graphics workstation, query the probability distribution of the resource release time;

[0099] Calculate the cumulative release probability of the graphics workstation within a preset time window; if the cumulative release probability is lower than a preset probability threshold, it is determined that the release of the graphics workstation is uncertain.

[0100] In response to the determination that the release is uncertain, a flexible waiting prompt is pushed to the target designer, and an option to enter the flexible waiting mode is provided;

[0101] If the target designer chooses to enter the flexible waiting mode, then the target designer's work status is marked as flexible waiting mode.

[0102] Specifically, when a target designer needs to use a graphics workstation to start their core tasks, but finds that the workstation is being continuously occupied by other designers and cannot be used immediately, the system will proactively query the previously generated resource release time probability distribution. This probability distribution reflects the likelihood that the graphics workstation will be released at different discrete points in the future.

[0103] To quantify this uncertainty, the system calculates the cumulative release probability of the graphics workstation within a preset time window. This preset time window can be flexibly configured based on factors such as actual project needs, task urgency, or the designer's patience; for example, it can be set to a period of 1 hour, 2 hours, or longer. The cumulative release probability refers to the sum of the release probabilities at all discrete points in time within this preset time window. If the calculated cumulative release probability is lower than a preset probability threshold, such as 30% or 50%, the system determines that the release of the graphics workstation has a high degree of uncertainty, meaning that the probability of the graphics workstation being released within an acceptable waiting time is low. This preset probability threshold can be adjusted according to the actual business scenario and the tolerance for waiting uncertainty.

[0104] In practical applications, once the system determines that the release of the graphics workstation is uncertain, it will immediately push a flexible waiting prompt to the waiting target designer. This prompt aims to inform the designer of the current resource status and provide a clear option, allowing the designer to decide whether to enter flexible waiting mode. For example, the prompt message may include, "The probability of the graphics workstation being released in the short term is low. Would you like to enter flexible waiting mode? The system will recommend flexible tasks that do not rely on the graphics workstation."

[0105] Furthermore, if the target designer selects to enter the flexible waiting mode as prompted, the system will mark the designer's current work status as flexible waiting mode. This marking will trigger a subsequent task filtering and recommendation mechanism to provide the designer with suitable flexible tasks, thereby effectively utilizing the waiting time.

[0106] This application's solution addresses the ambiguity in determining whether a designer should enter the flexible waiting mode in the basic solution by introducing clear judgment criteria and a user interaction mechanism. Through this technical solution, the application effectively resolves the issue of unclear triggering mechanisms for the flexible waiting mode in the basic solution. By quantifying the uncertainty of resource release and combining it with user autonomy, the system avoids blindly or untimely guiding designers into the flexible waiting mode, thereby improving the accuracy and flexibility of resource scheduling. After clearly understanding the resource status, designers can independently decide whether to use the waiting time to perform flexible tasks. This not only improves designer job satisfaction but also maximizes the use of designers' fragmented time, improving overall project efficiency and resource utilization. Furthermore, this clear judgment and interaction mechanism makes the entire scheduling process more transparent and controllable, contributing to optimized user experience.

[0107] In some preferred embodiments, a specific example is given below. Suppose designer A in an industrial design team needs to use a graphics workstation for a core 3D rendering task, but the workstation is currently being used by designer B. The system first queries the resource release time probability distribution of the graphics workstation. For example, this distribution might show a 20% probability of release within the next hour, a 30% probability within 1-2 hours, and a 40% probability within 2-3 hours. The system sets a preset time window of 2 hours and a preset probability threshold of 60%. The system calculates the cumulative release probability within the next 2 hours as 20% + 30% = 50%. Since 50% is lower than 60%, the system determines that the release of the graphics workstation is uncertain. In response to this determination, the system immediately pushes a flexible waiting prompt to designer A's terminal, such as: "The probability of your required graphics workstation being released within the next 2 hours is low (cumulative release probability 50%). Would you like to consider entering flexible waiting mode? The system will recommend flexible tasks that do not depend on the graphics workstation." If Designer A selects "Yes," agreeing to enter the flexible waiting mode, the system will mark Designer A's current work status as "Flexible Waiting Mode." Subsequently, based on Designer A's skill tags and current project stage, the system will select flexible tasks from the candidate task pool—such as "organizing project documents," "conducting market research," or "participating in online design reviews"—that do not rely on a graphics workstation and are expected to have short execution times. These tasks will be sorted according to their value contribution index and recommended to Designer A. Designer A can then choose one task to execute, effectively utilizing the time spent waiting for the graphics workstation, avoiding idle waiting, and improving work efficiency.

[0108] Specifically, the steps described above for selecting flexible tasks from the candidate task pool that match skill tags, do not rely on a graphics workstation, and have an estimated waiting time shorter than that of a graphics workstation can be performed as follows.

[0109] First, semantic parsing is performed on the description text of each task in the candidate task pool to identify the skill tags required for each task, the resource type tags it depends on, and the estimated completion time tag. Specifically, semantic parsing involves natural language processing of the task description text to extract key information. For example, deep learning-based text classification models or named entity recognition technology can be used to automatically identify the specific skills required to complete the task (such as "CAD modeling," "rendering optimization," "material mapping," etc.), the types of resources that the task may depend on (such as "CPU-intensive," "GPU-accelerated," "memory-sensitive," etc.), and the estimated completion time of the task from the task description. This extracted information is then structured into skill tags, resource type tags, and estimated completion time tags for subsequent automated filtering and matching. The target designer's skill tag set can be understood as a list of the designer's professional skills and abilities, maintained and updated through methods such as the designer's personal profile, historical project performance, skill certifications, or self-assessment. This set ensures that the recommended tasks match the designer's capabilities, thereby improving the efficiency and quality of task completion.

[0110] Further, tasks whose skill tags match the target designer's skill tag set are selected from the candidate task pool to obtain the first candidate subset. Then, tasks with resource type tags identical to the graphics workstation type information are removed from the first candidate subset to obtain the second candidate subset. The graphics workstation type information typically refers to its dedicated computing resource for 3D modeling or rendering tasks, characterized by high-performance CPUs, GPUs, and large memory capacity. This step aims to ensure that the selected elastic tasks do not depend on currently occupied graphics workstations, thus preventing the target designer from encountering resource bottlenecks again in elastic waiting mode. Next, tasks with estimated time tags shorter than the estimated waiting time of the graphics workstation are selected from the second candidate subset to obtain the third candidate subset. The estimated waiting time refers to the predicted time point when the graphics workstation might be released based on the resource release time probability distribution. Its purpose is to ensure that elastic tasks can be completed before the graphics workstation is released, thereby effectively utilizing the designer's waiting time.

[0111] Finally, the value contribution index of each task in the third candidate subset is calculated, and the tasks are sorted from highest to lowest value contribution index to obtain the recommended task list. The value contribution index quantifies the strategic value and impact of each task on the downstream stages of the entire industrial design project. The calculation of this index comprehensively considers the importance of task outputs to subsequent design phases, production processes, and even market feedback, aiming to prioritize flexible tasks that have a greater positive impact on the overall progress and success of the project.

[0112] This application's solution employs a multi-dimensional, phased, and refined screening process for tasks in the candidate task pool. First, key task attributes are obtained through semantic parsing. Then, skill matching, resource dependency exclusion, and time-limited filtering are performed sequentially to ensure that recommended flexible tasks align with the target designer's skill expertise, are independent of currently occupied graphics workstations, and can be completed within the estimated waiting time. Finally, tasks meeting the criteria are prioritized based on a value contribution index, ensuring that recommended tasks not only effectively fill the designer's waiting time but also maximize the overall progress and value of the project.

[0113] This application further proposes steps for calculating and ranking the value contribution index of each task in the aforementioned third candidate subset, including:

[0114] Semantic parsing is performed on the description text of each task in the third candidate subset to identify the downstream links of the project affected by the output of each task.

[0115] Based on a predefined project hierarchy model, obtain the importance weights of each downstream link of the project to the overall strategic goal of the project.

[0116] Based on the predefined downstream impact coefficient table, obtain the expected impact coefficient of each task's output on each downstream link of the project.

[0117] The value contribution index of each task is calculated by summing the products of the importance weight of each downstream link of the project and the expected impact coefficient of each task.

[0118] The tasks in the third candidate subset are sorted from highest to lowest according to their value contribution index.

[0119] Specifically, semantic parsing of the description text of each task in the aforementioned third candidate subset refers to using Natural Language Processing (NLP) technology to analyze textual information such as the task's title, content, and requirements specifications, and automatically identify which subsequent stages of the project the task's output will affect. For example, a task of "designing a user interface prototype" may affect downstream stages such as "front-end development" and "user experience testing."

[0120] Specifically, based on a predefined project hierarchy model, the importance weights of each downstream stage of the project to the overall strategic goals are obtained. This can be understood as, during the project initiation or planning phase, assessing the importance of each project stage, module, or specific stage based on the overall project objectives and priorities, and assigning corresponding weight values. For example, in a new product development project, downstream stages related to "core function implementation" may be assigned higher importance weights, while "non-core function optimization" may have lower weights. These weights reflect the degree to which different stages contribute to the project's ultimate success and strategic value.

[0121] In practical applications, obtaining the expected impact coefficient of each task's output on each downstream stage of the project, based on a predefined downstream impact coefficient table, refers to establishing a knowledge base or rule set to quantify the degree to which the output of a specific type of task promotes or influences different downstream stages. For example, a high-quality "3D model" output may have a high positive impact coefficient on the "rendering" stage, while its impact coefficient on the "marketing material preparation" stage may be relatively low. This coefficient can be defined and adjusted based on historical project data, expert experience, or project management standards.

[0122] Therefore, calculating the value contribution index of each task by summing the products of the importance weights of each downstream link of the project and the expected impact coefficients of each task means comprehensively considering the impact of the task's output on each downstream link (expected impact coefficient) and the importance of the downstream links it affects (importance weight). This summation calculation method can comprehensively and quantitatively assess the overall value contribution of a flexible task to the overall strategic goals of the project.

[0123] Finally, the tasks in the third candidate subset are sorted in descending order of their value contribution index, ensuring that when recommending flexible tasks to the target designer, those tasks that contribute the most to the overall value of the project are given priority, thereby optimizing the efficiency of the designer's flexible waiting time utilization.

[0124] This application's solution constructs a multi-dimensional, quantitative value contribution index calculation mechanism by introducing the identification of downstream project links affected by task outputs, the importance weights of the project hierarchy model, and a downstream impact coefficient table. This mechanism enables the system to go beyond simple task attribute matching and deeply analyze the potential impact of each flexible task on the overall strategic goals of the project. Through the above technical solution, this application ensures that when designers enter flexible waiting mode, the recommended flexible tasks not only meet their skill and time requirements but also contribute the highest value to the overall strategic goals of the project. This significantly improves the efficiency of designers' flexible waiting time utilization and avoids designers investing time in tasks with little impact on the project. Furthermore, by prioritizing the completion of tasks that have a positive impact on key downstream links of the project, this application helps accelerate the overall project progress, reduces the potential delay risk caused by resource bottlenecks, thereby optimizing the resource scheduling and management of industrial design projects and improving project success rate and team collaboration efficiency.

[0125] In some preferred embodiments, a specific example is given below. Assume an industrial design project comprises key phases such as "conceptual design," "detailed modeling," "rendering and visualization," "prototyping," and "client review." The project hierarchy model pre-defines the importance weights of these phases to the overall strategic goals of the project. For example, "client review" has the highest importance weight (0.9), followed by "detailed modeling" (0.7), "conceptual design" and "rendering and visualization" (0.6), and "prototyping" (0.5).

[0126] When a target designer enters a resilient waiting mode due to a occupied graphics workstation, the system identifies the following resilient tasks from the third candidate subset:

[0127] Task A: "Optimize the material texture of product A" - its output mainly affects the "rendering and visualization" stage.

[0128] Task B: "Prepare the preliminary design specification document for Product B" - its output mainly affects the "customer review" stage.

[0129] Task C: "Research new rendering software plugins" - its output mainly affects the "rendering and visualization" stage.

[0130] The system will make an assessment based on a predefined table of downstream impact coefficients:

[0131] The expected impact coefficient of Task A on the "Rendering and Visualization" stage is 0.8.

[0132] The expected impact coefficient of Task B on the "Customer Review" stage is 0.9.

[0133] The expected impact coefficient of Task C on the "Rendering and Visualization" stage is 0.7.

[0134] Based on the above data, calculate the value contribution index of each task:

[0135] The value contribution index of task A = 0.8 (influence coefficient) * 0.6 (importance weight of "rendering and visualization") = 0.48.

[0136] The value contribution index of Task B = 0.9 (impact coefficient) * 0.9 (importance weight of "client review") = 0.81.

[0137] The value contribution index of task C = 0.7 (influence coefficient) * 0.6 (importance weight of "rendering and visualization") = 0.42.

[0138] Ultimately, the system will sort tasks from highest to lowest based on their value contribution index, recommending task B (0.81) to the target designer, followed by task A (0.48), and finally task C (0.42). This approach ensures that designers prioritize tasks that contribute most to the overall strategic goals of the project during the flexible waiting period, thereby maximizing the value of the flexible waiting time.

[0139] This application further proposes the following steps for correlated task execution time data with the actual release time of the graphics workstation, and used as new training samples to feed into the resource release probability prediction model for online incremental training:

[0140] The task execution time data of the target designer during the execution of flexible tasks is associated with the actual release time of the graphics workstation and stored to form an associated data pair;

[0141] The associated data pairs are used as new training samples and input into the online update module of the resource release probability prediction model;

[0142] The online update module uses an incremental learning method based on gradient descent to locally update the model parameters of the resource release probability prediction model based on newly added training samples, so that the subsequent resource release time probability distribution prediction results are continuously optimized as actual scheduling data accumulates.

[0143] Specifically, task execution time data refers to the actual working time a target designer spends on a task after being recommended and accepting a flexible task, such as the task's start time, end time, and total duration. The actual release time of a graphics workstation refers to the specific point in time when the workstation is released from its current use by the designer and becomes available for other designers. Storing these two types of data together means recording them as a pair of data with a time-corresponding relationship. For example, this could be stored in a structured database containing fields such as task ID, designer ID, flexible task execution duration, graphics workstation ID, and the corresponding actual release time, thus forming a related data pair.

[0144] The associated data pairs can be understood as real feedback samples containing actual operation results and corresponding resource states. These samples, as new training samples, are input into the online update module of the resource release probability prediction model. The online update module is a component of the resource release probability prediction model, specifically responsible for processing real-time feedback data and iteratively updating the model.

[0145] In practical applications, the online update module uses a gradient descent-based incremental learning method to locally update the model parameters of the resource release probability prediction model based on newly added training samples. Gradient descent-based incremental learning is a commonly used machine learning optimization technique. It allows the model to gradually optimize its performance by calculating the gradient of the loss function with respect to the model parameters and adjusting the parameters in small steps in the opposite direction of the gradient when receiving new data. Local updates refer to adjusting only the parameters most relevant to the current new samples, rather than retraining the entire model. This helps improve update efficiency and avoids the "catastrophic forgetting" phenomenon. The goal is to ensure that subsequent resource release time probability distribution prediction results can be continuously optimized as actual scheduling data accumulates, ensuring that the model always reflects the latest resource usage patterns and designer behavior habits.

[0146] This application's solution establishes a correlation between task execution time data and the actual release time of the graphics workstation, providing high-quality, realistic feedback data for the resource release probability prediction model. Because these correlated data pairs accurately reflect the actual working conditions of designers during flexible waiting periods and the true availability of the graphics workstation, the model can obtain reliable supervision signals. By inputting these correlated data pairs as new training samples into the online update module and using a gradient descent-based incremental learning method to locally update the model parameters, the model can learn new data patterns and potential association rules in an efficient and progressive manner. This online incremental learning mechanism enables the resource release probability prediction model to continuously adapt to the ever-changing resource usage habits, task characteristics, and designer behavior patterns in the industrial design environment, thereby avoiding the problem of decreased prediction accuracy due to outdated data.

[0147] In some preferred embodiments, suppose the target designer, Xiao Wang, enters a flexible waiting mode because his graphics workstation is occupied, and is recommended and accepted a flexible task that does not depend on the graphics workstation, such as organizing project documents. During Xiao Wang's execution of this flexible task, the system records his task execution time data, for example, from 9:00 AM to 10:30 AM, totaling 1.5 hours. Simultaneously, the system detects that the graphics workstation Xiao Wang was waiting for was released by the previous designer at 10:15 AM. At this point, the system correlates Xiao Wang's 1.5-hour flexible task execution time data with the actual time the graphics workstation was released at 10:15 AM, forming a correlation data pair. This correlation data pair is then input into the online update module of the resource release probability prediction model. The online update module uses an incremental learning method based on gradient descent to fine-tune the parameters such as weights and biases within the model based on this new real feedback sample. For example, if the model previously tended to overestimate workstation usage time for a specific task type or designer behavior pattern, this new correlation data pair will prompt the model to adjust relevant parameters, making its predictions for similar scenarios in the future more accurate. In this way, as more designers complete flexible tasks and release their graphics workstations, the model will continuously learn from practical experience, and its ability to predict future resource release times will be continuously optimized and improved.

[0148] refer to Figure 3 , Figure 3 This is a schematic diagram of the structure of an industrial design project resource scheduling and management system provided in an embodiment of the present invention, including:

[0149] The hardware monitoring module is used to continuously collect real-time operating parameters of the CPU, GPU, memory and disk I / O subsystems of the graphics workstation at a preset sampling period through a hardware monitoring sensor group deployed on the graphics workstation for industrial design, and calculate the resource operation instability index based on the real-time operating parameters.

[0150] The behavior acquisition module is used to capture the operation behavior event stream of the designer currently occupying the graphics workstation during the period when the graphics workstation is occupied, through the operation behavior acquisition agent deployed on the designer terminal. Based on the operation behavior event stream, non-productive high-frequency operation patterns are identified, and combined with the designer's historical operation baseline, a designer operation anomaly index is generated.

[0151] The prediction module is used to take the resource operation instability index, the designer operation anomaly index, the remaining estimated execution time of the current task, and the current project stage information as input features, and output the release probability of the graphics workstation at each discrete time point in the future through a pre-trained resource release probability prediction model, and generate a resource release time probability distribution based on the release probability.

[0152] The status judgment module is used to determine whether the target designer has entered the elastic waiting mode when the target designer waiting for the graphics workstation is unable to start its core task due to the continuous occupation of the graphics workstation.

[0153] The task filtering and recommendation module is used to, if the target designer enters the flexible waiting mode, start from the target designer's skill tag set, filter out flexible tasks from the candidate task pool that match the skill tags, do not depend on the graphics workstation, and have an expected waiting time less than the expected waiting time of the graphics workstation, and recommend the filtered flexible tasks to the target designer after sorting them from high to low according to their value contribution index to the downstream links of the project.

[0154] The feedback update module is used to receive the target designer's response to the elastic task, update the allocation status of the elastic task and the target designer's task execution time data, and associate the updated task execution time data with the actual release time of the graphics workstation, feeding it back as a new training sample to the resource release probability prediction model for online incremental training, so that the subsequent resource release time probability distribution prediction results are continuously optimized as scheduling data accumulates.

[0155] This application presents an industrial design project resource scheduling and management system designed to address issues in traditional industrial design project resource scheduling, such as inconsistencies between actual physical resource occupancy and system records, inaccurate time estimates due to dynamically changing task complexity, and implicit conflicts arising from designers' informal "soft reservation" behavior. This system integrates hardware performance monitoring, designer behavior analysis, and intelligent prediction mechanisms to achieve dynamic and precise scheduling of key resources such as graphics workstations. Specifically, the system acquires multi-dimensional data in real time through hardware monitoring and behavior acquisition modules. A prediction module makes probabilistic predictions of resource release times, a status judgment module guides designers into a flexible waiting mode, a task selection and recommendation module provides valuable flexible tasks, and a feedback update module continuously optimizes the prediction model, thereby significantly improving resource utilization efficiency and project management level.

[0156] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A resource scheduling and management method for industrial design projects, characterized in that, include: By deploying a hardware monitoring sensor group on a graphics workstation dedicated to industrial design, the real-time operating parameters of the CPU, GPU, memory and disk I / O subsystems of the graphics workstation are continuously collected at a preset sampling period, and the resource operation instability index is calculated based on the real-time operating parameters. By deploying an operation behavior acquisition agent on the designer's terminal, the operation behavior event stream of the designer currently occupying the graphics workstation is captured in real time during the period when the graphics workstation is occupied, and non-productive high-frequency operation patterns are identified based on the operation behavior event stream. Combined with the designer's historical operation baseline, a designer operation anomaly index is generated. Using the resource operation instability index, the designer operation anomaly index, the remaining estimated execution time of the current task, and the current project stage information as input features, the resource release probability prediction model, pre-trained, outputs the release probability of the graphics workstation at each discrete time point in the future, and generates a resource release time probability distribution based on the release probability. When a target designer waiting for the graphics workstation is unable to start its core task due to the continuous occupation of the graphics workstation, it is determined whether the target designer should enter the elastic waiting mode based on the probability distribution of the resource release time. If the target designer enters the flexible waiting mode, then starting from the target designer's skill tag set, flexible tasks that match the skill tags, do not depend on the graphics workstation, and have an expected waiting time less than the expected waiting time of the graphics workstation are selected from the candidate task pool. The selected flexible tasks are then sorted from high to low according to their value contribution index to the downstream links of the project and recommended to the target designer. The system receives the target designer's response to the elastic task, updates the allocation status of the elastic task and the target designer's task execution time data, and associates the updated task execution time data with the actual release time of the graphics workstation. This data is then fed back as a new training sample to the resource release probability prediction model for online incremental training, so that the subsequent resource release time probability distribution prediction results are continuously optimized as scheduling data accumulates.

2. The method according to claim 1, characterized in that, The step of calculating the resource operation instability index based on the real-time operating parameters includes: The real-time operating parameters are subjected to noise suppression processing based on sliding median filtering to obtain a sequence of performance parameters after noise suppression; The noise-suppressed performance parameter sequence is matched with a pre-established baseline operating mode library to identify and eliminate background performance fluctuations caused by system background processes. Determine whether the duration of the remaining performance fluctuation after removing the background performance fluctuation is greater than a first duration threshold and whether the fluctuation amplitude is greater than a first amplitude threshold. If both conditions are met, then cross-validate the remaining performance fluctuation with the system error log of the graphics workstation. When the cross-validation confirms that the remaining performance fluctuations are caused by intermittent hardware failures, the resource operation instability index is generated; wherein, the resource operation instability index is used to quantify the probability that the graphics workstation will fail in the future and cause a delay in resource release time.

3. The method according to claim 1, characterized in that, The steps for generating the designer operation anomaly index include: The operation behavior acquisition agent captures the keyboard, mouse and software interface operation events of the designer on the graphics workstation to obtain the operation behavior event stream. Based on predefined operation pattern recognition rules, non-productive high-frequency operation patterns are identified in the operation behavior event stream; the non-productive high-frequency operation patterns include at least one of repeated save operations, frequent undo and redo operations, unnecessary parameter fine-tuning operations, and abnormally frequent software switching operations. Statistics were compiled on the types, frequency per unit time, and cumulative duration of each operation identified as a non-productive high-frequency operation. Based on the type, the frequency of occurrence per unit time, and the cumulative duration, combined with the designer's historical operation baseline, the designer's operation anomaly index is calculated using a pre-trained anomaly scoring model; wherein, the designer's operation anomaly index is used to quantify the probability that the designer takes non-productive operations due to subjective perception of the graphics workstation's performance instability, thereby prolonging the actual occupancy time of the graphics workstation.

4. The method according to claim 1, characterized in that, The step of generating a resource release time probability distribution based on the release probability includes: Obtain the remaining estimated execution time and task type of the current task, as well as the current project stage information and team collaboration density, to obtain project context information; The resource operation instability index, the designer operation anomaly index, the remaining estimated execution time, the task type, the project stage information, and the team collaboration density are concatenated into a multi-source input feature vector; The multi-source input feature vector is input into the resource release probability prediction model, and the release probability value of the graphics workstation at each preset discrete time point in the future is output. The resource release time probability distribution is constructed based on the release probability values ​​at each preset discrete time point.

5. The method according to claim 4, characterized in that, The method further includes: training the resource release probability prediction model, wherein the resource release probability prediction model is a temporal prediction model that integrates a multi-head self-attention mechanism and a gated recurrent unit, and its training process includes: Multi-source time-series data from historical project data are collected as training samples. The multi-source time-series data includes historical resource operation instability index sequence, historical designer operation anomaly index sequence, historical task remaining execution time sequence, and historical project stage sequence. Using the multi-source time-series data as input and the actual release time of historical graphics workstations as supervision labels, the initial time-series prediction model is trained. During training, the contribution weights of each data source in the multi-source time series data to the release probability prediction at each time step are adaptively adjusted through the multi-head self-attention mechanism to obtain the trained resource release probability prediction model.

6. The method according to claim 1, characterized in that, The step of determining whether the target designer has entered the elastic waiting mode based on the resource release time probability distribution includes: When the core task of the target designer cannot be started due to the continuous occupation of the graphics workstation, query the probability distribution of the resource release time; Calculate the cumulative release probability of the graphics workstation within a preset time window; if the cumulative release probability is lower than a preset probability threshold, it is determined that the release of the graphics workstation is uncertain. In response to the determination that the release is uncertain, a flexible waiting prompt is pushed to the target designer, and an option to enter the flexible waiting mode is provided; If the target designer chooses to enter the flexible waiting mode, then the target designer's work status is marked as flexible waiting mode.

7. The method according to claim 1, characterized in that, The step of selecting flexible tasks from the candidate task pool that match skill tags, do not depend on the graphics workstation, and have an expected processing time less than the expected waiting time of the graphics workstation includes: Semantic parsing is performed on the description text of each task in the candidate task pool to identify the skill tags required for each task, the resource type tags it depends on, and the estimated time consumption tags. The first candidate subset is obtained by filtering tasks from the candidate task pool whose skill tags match the skill tag set of the target designer; The second candidate subset is obtained by removing tasks whose resource type labels are the same as the type information of the graphics workstation from the first candidate subset; From the second candidate subset, tasks whose estimated time consumption labels are less than the estimated waiting time of the graphics workstation are selected to obtain a third candidate subset; Calculate the value contribution index of each task in the third candidate subset, and sort them from high to low according to the value contribution index to obtain a recommended task list.

8. The method according to claim 7, characterized in that, The step of calculating and ranking the value contribution index of each task in the third candidate subset includes: Semantic parsing is performed on the description text of each task in the third candidate subset to identify the downstream links of the project affected by the output of each task. Based on a predefined project hierarchy model, obtain the importance weights of each downstream link of the project to the overall strategic goal of the project. Based on the predefined downstream impact coefficient table, obtain the expected impact coefficient of each task's output on each downstream link of the project. The value contribution index of each task is calculated by summing the products of the importance weight of each downstream link of the project and the expected impact coefficient of each task. The tasks in the third candidate subset are sorted from highest to lowest according to their value contribution index.

9. The method according to claim 1, characterized in that, The step of associating the updated task execution time data with the actual release time of the graphics workstation and feeding it back as a new training sample to the resource release probability prediction model for online incremental training includes: The task execution time data of the target designer during the execution of the flexible task is associated with the actual release time of the graphics workstation and stored to form an associated data pair; The associated data pairs are used as new training samples and input into the online update module of the resource release probability prediction model; The online update module uses the newly added training samples and employs an incremental learning method based on gradient descent to locally update the model parameters of the resource release probability prediction model, so that the subsequent resource release time probability distribution prediction results are continuously optimized as actual scheduling data accumulates.

10. A resource scheduling and management system for industrial design projects, characterized in that, include: The hardware monitoring module is used to continuously collect real-time operating parameters of the CPU, GPU, memory and disk I / O subsystems of the graphics workstation at a preset sampling period through a hardware monitoring sensor group deployed on the graphics workstation for industrial design, and calculate the resource operation instability index based on the real-time operating parameters. The behavior acquisition module is used to capture the operation behavior event stream of the designer currently occupying the graphics workstation during the period when the graphics workstation is occupied, through the operation behavior acquisition agent deployed on the designer terminal. Based on the operation behavior event stream, non-productive high-frequency operation patterns are identified, and combined with the designer's historical operation baseline, a designer operation anomaly index is generated. The prediction module is used to take the resource operation instability index, the designer operation anomaly index, the remaining estimated execution time of the current task, and the current project stage information as input features, and output the release probability of the graphics workstation at each discrete time point in the future through a pre-trained resource release probability prediction model, and generate a resource release time probability distribution based on the release probability. The status judgment module is used to determine whether the target designer has entered the elastic waiting mode when the target designer waiting for the graphics workstation is unable to start its core task due to the continuous occupation of the graphics workstation. The task filtering and recommendation module is used to, if the target designer enters the flexible waiting mode, start from the target designer's skill tag set, filter out flexible tasks from the candidate task pool that match the skill tags, do not depend on the graphics workstation, and have an expected waiting time less than the expected waiting time of the graphics workstation, and recommend the filtered flexible tasks to the target designer after sorting them from high to low according to their value contribution index to the downstream links of the project. The feedback update module is used to receive the target designer's response to the elastic task, update the allocation status of the elastic task and the target designer's task execution time data, and associate the updated task execution time data with the actual release time of the graphics workstation, feeding it back as a new training sample to the resource release probability prediction model for online incremental training, so that the subsequent resource release time probability distribution prediction results are continuously optimized as scheduling data accumulates.

Images