A collaborative control system for intelligent testing equipment with vehicle-mounted domain control

Abstract

This invention discloses a collaborative control system for intelligent test equipment with vehicle-mounted domain control, belonging to the field of intelligent test equipment collaborative control technology. It includes a control resource dynamic spectrum determination module, an occupancy feature analysis module, a scheduling determination tensor construction module, a multi-dimensional scheduling suppression execution module, and a self-evolving closed-loop control module. The occupancy feature analysis module, under the condition that the equipment is in a state where control resources are not released, inputs the dynamic occupancy spectrum of control resources into a convolutional attention residual network, extracts depth variables, and generates a set of real-time control resource occupancy parameters to express the true occupancy status of the equipment's internal control resources. This invention solves the problem that the state of unreleased control resources during test task switching cannot be perceived and managed, realizing dynamic scheduling suppression and adaptive feedback updates based on the internal resource occupancy status, effectively improving the stability of the test process and the validity of the data.

Description

Technical Field

[0001] This invention relates to the field of collaborative control technology for intelligent testing equipment, and more specifically to a collaborative control system for intelligent testing equipment with vehicle-mounted domain control. Background Technology

[0002] Intelligent testing equipment collaborative control for vehicle domain controllers refers to the use of intelligent testing systems to replace traditional manual testing methods for multiple functional verifications required before delivery of vehicle-mounted domain controllers (such as ADAS systems, infotainment modules, and communication gateways). This achieves automatic execution of test commands, automatic acquisition and comparison of test data, and automatic judgment and recording of test results, forming a closed-loop control system. Currently, the technical system for implementing this collaborative control mainly consists of five core components: ① Test host computer software, responsible for executing automated test scripts, issuing commands, and making test judgments; ② HMI (Human-Machine Interface) module, used for manual assistance in performing test operations such as touch and display confirmation; ③ Interface module, connecting and controlling the device under test, sending test signals such as CAN, SPI, and Ethernet, and receiving feedback; ④ Device under test, i.e., the specific vehicle domain controller; ⑤ Peripherals, such as high-definition digital cameras and speakers, used for automatic image or sound acquisition and verification. The entire testing process begins with scanning the product barcode. The system automatically loads and executes preset test scripts, completing various test tasks sequentially according to a set order. Automated tests (such as communication protocol testing) are completed fully automatically by the host computer program, while semi-automatic tests (such as screen display and touch operation) are completed by the operator through an HMI interface. Test results are automatically recorded into the system. The system incorporates error-proofing mechanisms; for example, barcode verification failure, test interruption, or result saving failure will prevent the next product from entering the test, ensuring test integrity and consistency. Finally, the information of products that pass the test is bound to their base plate barcode, and the test results are uploaded to a designated location on the server, achieving electronic management and traceability of the test results.

[0003] The existing technology has the following shortcomings:

[0004] In the collaborative control process of performing multiple functional tests on vehicle-mounted domain control equipment, the test system typically switches between multiple test tasks sequentially according to a preset procedure to achieve comprehensive coverage of all functional items of the device under test. During the transition from one test task to the next, some devices, although reporting an "executable status," may not actually have fully released the control resources used for the previous task. For example, control signal channels may not be disconnected, data buffers may not be cleared, and input / output paths may still be occupied. Because the aforementioned occupancy status of control resources is an implicit state within the device, the existing control system cannot directly perceive it and still relies on the "task completed" signal reported from the device's surface layer to advance the process, thereby triggering the scheduling and execution of the next test task. Existing vehicle-mounted domain control intelligent test equipment collaborative control technology cannot dynamically prevent the scheduling and execution of the next test task based on the actual occupancy information of the equipment in the state where control resources are not released during the test task switching process. As a result, multiple test tasks run intermittently before the control resources have been switched over, causing overlapping test instructions, signal conflicts, and disordered execution logic. This not only invalidates the current test results but may also cause subsequent test data anomalies. In severe cases, it may lead to the test system misjudging equipment failures or even causing equipment damage, thus undermining the collaborative control stability and data validity of the entire test process.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a collaborative control system for intelligent testing equipment with vehicle-mounted domain control, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a collaborative control system for intelligent test equipment with vehicle-mounted domain control, comprising a control resource dynamic spectrum determination module, an occupancy feature analysis module, a scheduling determination tensor construction module, a multi-dimensional scheduling suppression execution module, and a self-evolving closed-loop control module;

[0008] The control resource dynamic spectrum determination module collects the control path on / off status, cache refresh level, signal acknowledgment delay, interface level stability and thread blocking time during test task switching, constructs a control resource dynamic occupancy spectrum, and determines whether the device is in a state where control resources are not released based on the feature cross-referencing in the control resource dynamic occupancy spectrum.

[0009] The occupancy feature parsing module, under the condition that the device is in a state where control resources are not released, inputs the dynamic occupancy spectrum of control resources into the convolutional attention residual network, extracts depth variables, and generates a set of real-time control resource occupancy parameters to express the true occupancy status of control resources inside the device.

[0010] The scheduling decision tensor construction module constructs a task scheduling decision tensor based on the actual parameter set of control resource occupancy, combined with task type, historical execution time and scheduling success rate, and calculates the blocking intention coefficient.

[0011] The multi-dimensional scheduling inhibition execution module constructs a scheduling inhibition structure based on the calculated blocking intention coefficient. It divides the blocking intention coefficient into prohibited scheduling segments, lazy scheduling segments, and allowed scheduling segments, and executes three types of scheduling inhibition behaviors: concurrent instruction cancellation, delay lock setting, and task advancement, respectively.

[0012] After completing the scheduling suppression behavior, the self-evolving closed-loop control module constructs a state convergence matrix based on the scheduling feedback signal and task results, generates a self-evolving tensor mapping by combining the actual control resource occupancy parameter set, and updates the actual control resource occupancy parameter set and the task scheduling decision tensor.

[0013] Preferably, the implementation of the control resource dynamic spectrum determination module is as follows:

[0014] During the test task switching process, for the control path on / off status, on / off data is collected by monitoring the continuous conduction level changes of the control command signal; for the cache refresh rate, the refresh ratio is collected by detecting the cache write rate and the data clear rate; for the signal acknowledgment delay, the delay sequence is collected by recording the time difference between signal transmission and acknowledgment; for the interface level stability, the level waveform is formed by sampling the interface voltage fluctuation amplitude; and for the thread blocking time, the blocking interval is calculated by recording the blocking start and end times of the task thread.

[0015] The control path on / off data, cache refresh ratio, signal acknowledgment delay sequence, interface level waveform, and thread blocking interval are synchronized and aligned according to the time sequence to generate a multi-dimensional time sequence matrix. A dynamic occupancy spectrum of control resources is constructed through frequency domain decomposition and amplitude normalization, where the horizontal axis represents the task switching duration and the vertical axis represents the coupling strength of the control dimension.

[0016] The system extracts the intersection points of feature lines from the dynamic occupancy spectrum of control resources, calculates the density gradient and offset angle of the intersection region, and determines that the device is in a state where control resources are not released when the density gradient changes continuously beyond a preset threshold and the displacement direction of the offset angle is inconsistent with the historical release trend. The determination result is then output for subsequent occupancy feature analysis processing.

[0017] Preferably, the implementation of the occupancy feature parsing module is as follows:

[0018] Under the condition that the device is in a state where control resources are not released, the dynamic occupancy spectrum of control resources is processed by temporal block and feature normalization and then input into the input end of the convolutional attention residual network. The spectrum data is passed sequentially in a fixed time window. Multiple sets of convolution kernels are used to perform cross-convolution operations on the time axis and the control dimension axis to extract the spatial feature correlation between the distribution of control resources and temporal changes in the spectrum.

[0019] The multi-channel feature maps obtained by cross-convolution are input into the attention weight allocation layer. The weight distribution is dynamically adjusted based on the coupling degree between the channel response intensity and the control dimension. Then, the input signal and weighted features are superimposed through the residual structure to generate a high-dimensional feature matrix with multi-layer feature fusion capability.

[0020] Deep variables representing the relationship between the coupling degree of control signals, resource retention strength and thread response latency are extracted from the high-dimensional feature matrix. Based on the extracted deep variables, a set of control resource occupancy parameters is constructed. The set of control resource occupancy parameters is used to express the actual occupancy status of control resources inside the device and is used for subsequent task scheduling judgment.

[0021] Preferably, deep variables characterizing the relationship between control signal coupling degree, resource retention strength, and thread response latency are extracted from the high-dimensional feature matrix, and a set of control resource occupancy parameters is constructed based on the extracted deep variables, specifically:

[0022] In the high-dimensional feature matrix, channels are divided based on the response amplitude and time change rate of the feature channels, into control signal coupling channel region, resource holding signal channel region and thread response delay channel region, respectively. Feature point masks are marked in each channel region to extract the target response distribution features within the region.

[0023] In the control signal coupling channel region, the channel response density is extracted, the signal attenuation ratio is calculated in the resource retention signal channel region, and the variation range of the blocking duration is statistically analyzed in the thread response delay channel region. In these ways, depth variables corresponding to the relationship between the control signal coupling degree, resource retention strength and thread response delay are generated.

[0024] The deep variables relating control signal coupling degree, resource retention strength and thread response latency are numerically normalized to generate a set of deep variables at the same data scale, and a control resource occupancy real-time parameter set is constructed in a matrix structure.

[0025] Preferably, in the scheduling decision tensor construction module, the control resource occupancy real-time parameter group is expanded according to the control dimension and time series, a task association vector is established in combination with the task type, and time domain alignment is completed based on historical execution time to construct a data matrix that reflects the impact of control state on task scheduling efficiency; the data matrix is ​​fused with the scheduling success rate information obtained during task scheduling, response weights are set according to task type, and tensor is constructed based on parameter coupling relationship to generate a task scheduling decision tensor with task type, control state parameters, and execution feedback information as dimensions; according to the weight change trend between the control state parameter dimension and the task type dimension in the task scheduling decision tensor, the gradient change region is extracted, and the blocking intention coefficient used for scheduling control is calculated in combination with the control disturbance amplitude and the task sensitivity change rate.

[0026] Preferably, based on the weight change trend between the control state parameter dimension and the task type dimension in the task scheduling determination tensor, the gradient change region is extracted, and combined with the control disturbance amplitude and the task sensitivity change rate, the blocking intention coefficient used for scheduling control is calculated, specifically:

[0027] In the task scheduling decision tensor, the weight distribution intervals of the control state parameter dimension and the task type dimension are selected. The weight change curve is expanded over time to calculate the gradient change rate within the continuous window. The sensitive interval of weight change is formed by filtering through the gradient threshold, which is used to identify the key nodes of sudden changes in resource state during task scheduling.

[0028] Extract control disturbance amplitude data within the weight change sensitive interval, calculate the change difference of each control dimension in adjacent time periods, and generate a disturbance sensitivity matrix by combining the sensitivity change rate of task type. The disturbance sensitivity matrix is ​​used to characterize the relationship between the fluctuation intensity of control resource changes and time offset during task execution.

[0029] The perturbation sensitivity matrix is ​​input into the blocking calculation model, and the blocking intention coefficient is generated based on the coupling slope between the control perturbation amplitude and the rate of change of task sensitivity. The blocking intention coefficient is used to characterize whether the dynamic scheduling suppression condition needs to be triggered during the current task scheduling process.

[0030] Preferably, the implementation of the multidimensional scheduling suppression execution module is as follows:

[0031] Based on the calculated blocking intention coefficient, a scheduling inhibition structure is constructed. By analyzing the amplitude fluctuation of the blocking intention coefficient in the process of controlling changes in resource occupation, a mapping model is established. The blocking intention coefficient is divided into multiple threshold intervals according to risk level in the numerical space, forming a scheduling inhibition structure that includes prohibited scheduling segments, lazy scheduling segments, and allowed scheduling segments.

[0032] The blocking intention coefficient is mapped to the corresponding segment in the scheduling suppression structure. If the blocking intention coefficient is located in the prohibited scheduling segment, the concurrent instruction cancellation behavior is triggered. The instruction priority filtering and mutual exclusion conflict detection method is used to clear all pending concurrent control instructions in the scheduling instruction queue to prevent task conflict execution when the resource state has not been switched.

[0033] If the blocking intention coefficient is in the lazy scheduling segment, a delay lock setting behavior is executed, and a time offset is inserted into the task scheduling logic to postpone the scheduling action; if the blocking intention coefficient is in the allowed scheduling segment, a task advancement behavior is executed, and the scheduling instructions for subsequent tasks are executed immediately after the scheduling conditions are confirmed to be met through feedback.

[0034] Preferably, the self-evolving closed-loop control module is implemented as follows:

[0035] After completing the scheduling suppression behavior, the system collects the control path on / off state recovery time, cache refresh completion flag, signal acknowledgment delay fall-off value, and interface level stabilization time based on the scheduling feedback signal. Combined with the task completion flag, output data consistency flag, and resource release delay value recorded in the task results, a state convergence matrix is ​​constructed to describe the dynamic regression characteristics of control resources after the task is completed.

[0036] The state convergence matrix and the actual parameter set of control resource occupancy are bidirectionally coupled in the control and time dimensions. By constructing a multi-head attention mapping structure, the response trend between the history of control resources and feedback is extracted, and a self-evolving tensor mapping containing the change trend of control resources and the evolution state is generated.

[0037] Based on the change magnitude and time offset gradient of the control dimension evolution path in the self-evolving tensor mapping, the corresponding parameters in the control resource occupancy real-time parameter set are updated. The updated control resource occupancy real-time parameter set is then used as input to jointly reconstruct the task scheduling decision tensor with the original scheduling strategy, so as to realize dynamic iterative control with scheduling state awareness.

[0038] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0039] 1. This invention constructs a complete dynamic scheduling suppression and feedback adaptive control mechanism based on control resource status awareness, achieving accurate identification and scheduling control of the unreleased control resources of on-board domain control equipment during test task switching. By constructing a dynamic control resource occupancy spectrum and extracting depth variables using a convolutional attention residual network, a set of real-time control resource occupancy parameters is generated, thereby obtaining real-time occupancy characteristics of resources such as control paths, caches, and signal channels within the equipment. Furthermore, during task scheduling, task type, execution history, and feedback information are introduced to construct a task scheduling decision tensor. Tensor analysis is then used to extract blocking intention coefficients, effectively identifying resource state change nodes and proactively intervening in scheduling behavior to prevent task conflicts, instruction overlaps, and test errors when resources are not released.

[0040] 2. This invention introduces a multi-dimensional scheduling inhibition structure and a self-evolving closed-loop control module. It divides scheduling state segments based on blocking intention coefficients and executes instruction cancellation, delay lock setting, and task advancement behaviors to achieve refined dynamic control of scheduling behavior. After the scheduling behavior is executed, a state convergence matrix is ​​constructed by combining scheduling feedback signals and task results. Through a multi-head attention mapping structure and bidirectional coupling with historical resource parameters, a self-evolving tensor mapping is generated to continuously update resource state perception and scheduling judgment conditions, forming a closed-loop learning capability. This mechanism significantly improves the accuracy of the testing system's perception of equipment status and the dynamic adaptability of control strategies, enhances the stability and effectiveness of multi-task testing, and effectively avoids system anomalies and equipment risks caused by resource conflicts and misscheduling. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0042] Figure 1 This is a schematic diagram of a collaborative control system for intelligent testing equipment with vehicle-mounted domain control, as described in this invention. Detailed Implementation

[0043] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0044] This invention provides, for example Figure 1The intelligent test equipment collaborative control system shown includes a control resource dynamic spectrum determination module, an occupancy feature analysis module, a scheduling determination tensor construction module, a multi-dimensional scheduling suppression execution module, and a self-evolving closed-loop control module.

[0045] The control resource dynamic spectrum determination module collects the control path on / off status, cache refresh level, signal acknowledgment delay, interface level stability and thread blocking time during test task switching, constructs a control resource dynamic occupancy spectrum, and determines whether the device is in a state where control resources are not released based on the feature cross-referencing in the control resource dynamic occupancy spectrum.

[0046] In this embodiment, the implementation of the control resource dynamic spectrum determination module is as follows:

[0047] During the test task switching process, for the control path on / off status, on / off data is collected by monitoring the continuous conduction level changes of the control command signal; for the cache refresh rate, the refresh ratio is collected by detecting the cache write rate and the data clear rate; for the signal acknowledgment delay, the delay sequence is collected by recording the time difference between signal transmission and acknowledgment; for the interface level stability, the level waveform is formed by sampling the interface voltage fluctuation amplitude; and for the thread blocking time, the blocking interval is calculated by recording the blocking start and end times of the task thread.

[0048] During the transition from one test task to the next, to accurately determine the occupancy of device control resources, five types of control states need to be acquired and calculated in parallel. The control path connectivity status can be determined by real-time acquisition of the duration of control command signal levels. For example, on a CAN or Ethernet bus, the duration of the high and low levels of each control command is acquired. If a control signal remains high during the release phase, it indicates that the path has not yet been disconnected. The cache refresh rate can be determined by the ratio of the number of write operations to the number of clear operations per unit time. For example, if the write rate is 200 times per second while the clear rate is only 80 times per second, it indicates that the cache has not been fully released. Signal acknowledgment delay can be formed by recording the difference sequence between the timestamp of the command sent and the timestamp of the received acknowledgment signal. For example, if the expected acknowledgment delay is less than 10ms, but multiple consecutive delays exceeding 30ms occur, it can be considered that the bus is still occupied by the previous task. Interface level stability is analyzed by sampling the fluctuation values ​​of the interface voltage signal and forming complete waveform data. Continuous voltage fluctuations may indicate that the connected device is still in an active connection state. Thread blocking time is recorded by tracking the start and end times of each test thread from suspension to resumption, and accumulating the occupied intervals of each thread. For example, if a thread of a certain test logic does not return to zero between two test intervals, it means that it is still waiting for resource release. Through these specific data collection and calculation methods, the real state of the device during test switching can be constructed.

[0049] Monitoring the connectivity of control paths reflects whether underlying hardware signals are physically disconnected. Essentially, it determines whether control commands have completed their release operations, forming the basis for judging whether the conditions for the next test task to enter are met. Monitoring cache refresh reflects the clearing status of short-term data retention within the device. If residual data is not cleared, new tasks may read residual information from the previous test, leading to incorrect judgments. Signal acknowledgment delay is a direct measure of the health status of the device's feedback link; persistent delay anomalies usually indicate that resources are in a state of occupancy. Interface level stability monitoring reflects whether external connections and internal interfaces are completely released. For example, in USB, Ethernet, or LVDS connections, slight level jitter indicates that the signal link is connected rather than disconnected. Thread blocking time reflects the instruction blocking status during the execution of the control program. If a thread is continuously locked, it indicates that the program flow has not completely ended. Combining these characteristics not only comprehensively restores the internal control resource status of the device but also provides a multi-dimensional, time-continuous state input structure, providing a stable foundation for subsequent spectrum construction and task scheduling judgments. These five characteristics are complementary and synergistic, significantly enhancing the accuracy of judging the device resource release status.

[0050] The control path on / off data, cache refresh ratio, signal acknowledgment delay sequence, interface level waveform, and thread blocking interval are synchronized and aligned according to the time sequence to generate a multi-dimensional time sequence matrix. A dynamic occupancy spectrum of control resources is constructed through frequency domain decomposition and amplitude normalization, where the horizontal axis represents the task switching duration and the vertical axis represents the coupling strength of the control dimension.

[0051] Control path on / off data, cache refresh ratios, signal acknowledgment delay sequences, interface level waveforms, and thread blocking intervals are all time-sensitive data. Therefore, it is essential to first unify the time reference of these data sources. All data can be mapped to the same time axis using timestamp resampling and interpolation alignment, constructing a multi-dimensional time-series matrix with a unified sampling frequency. Each column of this matrix corresponds to a control dimension, and each row corresponds to a sampling time. To reveal the dynamic characteristics of each control dimension during task switching, a Fast Fourier Transform can be performed on each column of data to extract frequency distribution features, followed by amplitude normalization to eliminate the influence of data units and amplitude. Finally, a two-dimensional dynamic control resource occupancy spectrum is generated through spectrum mapping, where the horizontal axis represents the unified time axis during task switching, and the vertical axis represents the coupling strength of the normalized frequency domain features of each control dimension. For example, if the control path on / off frequency suddenly increases while a significant peak appears in the thread blocking time distribution, a highlighted overlapping area will form in the spectrum, indicating a risk of abnormal control resource release and concurrency. By constructing this spectrum, dispersed control state signals can be visually integrated, facilitating subsequent feature discrimination and state analysis.

[0052] The core purpose of integrating control path on / off data, cache refresh ratios, signal acknowledgment delay sequences, interface level waveforms, and thread blocking intervals into a multi-dimensional time series matrix is ​​to establish a unified state representation carrier, enabling the comparison and joint analysis of the changing trends of various control resources across the time dimension. Time synchronization alignment ensures causal matching between different signals, avoiding logical inference distortion. Frequency domain decomposition is used to mine the temporal fluctuation patterns of control resource states, which is particularly effective in detecting abrupt changes, residual events, or abnormal behaviors during resource release. Amplitude normalization is used to eliminate differences in the original physical magnitudes of different control dimensions. For example, although cache refresh and interface levels have different units, they can be interactively analyzed on a unified scale through normalization. The dynamic occupancy spectrum of control resources, as the final presentation structure, retains the time clues of state evolution and reveals the cooperative strength between control resources, serving as an indispensable intermediate layer technology structure for subsequent feature extraction and scheduling decisions. This spectrum essentially constructs a bridge between low-dimensional time series and high-dimensional control resource states, providing a highly compressed and information-rich data structure for the input of intelligent discrimination models.

[0053] The system extracts the intersection points of feature lines from the dynamic occupancy spectrum of control resources, calculates the density gradient and offset angle of the intersection region, and determines that the device is in a state where control resources are not released when the density gradient changes continuously beyond a preset threshold and the displacement direction of the offset angle is inconsistent with the historical release trend. The determination result is then output for subsequent occupancy feature analysis processing.

[0054] After generating the dynamic occupancy spectrum of control resources, it is necessary to identify the intersection points of characteristic lines to determine whether there is abnormal occupancy of control resources during task switching. Specifically, local gradient analysis and density clustering algorithms can be used to detect the changing trend of energy distribution in the spectrum. First, the intersection positions of the control dimension intensity curves are scanned on the continuous time axis. When different control dimensions show high intensity superposition within the same time window, it is defined as a candidate intersection region. Then, the density gradient of each intersection region is calculated, i.e., the continuous incremental trend of control intensity change per unit time, to assess the stability of the resource state. If the density gradient shows a continuous upward trend in a short period, it indicates that the resource has not been released in time. Next, geometric fitting is performed on the intersection points, and the direction of the offset angle displacement is obtained by comparing the extension direction of the intersection lines. If this offset direction is opposite to the typical trend direction formed during historical task release phases—for example, the offset direction in past release phases was always a downward slope, while the current offset is upward—it can be determined that the device control resources are still occupied. This analysis method can capture subtle abnormal release behaviors at the data dimension. For example, when the signal acknowledgment delay and thread blocking time curves intersect within the same time period and form a positive offset angle, the system can determine that a release exception has occurred and output a occupancy status signal for subsequent occupancy feature analysis.

[0055] Density gradient measures the intensity of changes in the control resource occupancy status over time. Its continuous change reflects the concentration or dispersion of energy between control dimensions during task switching. A continuously increasing density gradient indicates that the resource release process is hindered, while a decreasing density gradient indicates successful release. The preset threshold is an empirical boundary derived from multiple batches of normal test data during system initialization, used to distinguish between normal fluctuations and abnormal occupancy. For example, when the density gradient continuously rises above the set threshold range, it can be considered that the device has not completed its release. The offset angle displacement direction describes the changing trend of the control dimension feature line in the spectral space. The positive or negative value of the angle and the direction of change indicate the temporal relationship of resource flow. If the current offset angle direction is inconsistent with the historical release trend, it indicates that the resource state change path has reversed, reflecting reverse occupancy behavior of unreleased resources. The historical release trend is based on a statistical model of the characteristic trajectory formed by the device during multiple rounds of normal task switching, recording the typical angle distribution during control resource release. By matching the current offset angle with this historical model, high-precision identification of abnormal release behavior can be achieved. This comprehensive judgment structure based on density gradient, preset threshold, and offset angle displacement direction enables the identification of resource occupancy status to not only rely on instantaneous signals, but also to achieve accurate dynamic judgment based on multi-dimensional features in time and space.

[0056] The occupancy feature parsing module, under the condition that the device is in a state where control resources are not released, inputs the dynamic occupancy spectrum of control resources into the convolutional attention residual network, extracts depth variables, and generates a set of real-time control resource occupancy parameters to express the true occupancy status of control resources inside the device.

[0057] In this embodiment, the implementation of the occupancy feature parsing module is as follows:

[0058] Under the condition that the device is in a state where control resources are not released, the dynamic occupancy spectrum of control resources is processed by temporal block and feature normalization and then input into the input end of the convolutional attention residual network. The spectrum data is passed sequentially in a fixed time window. Multiple sets of convolution kernels are used to perform cross-convolution operations on the time axis and the control dimension axis to extract the spatial feature correlation between the distribution of control resources and temporal changes in the spectrum.

[0059] When the device is in a state where control resources are not released, the dynamic occupancy spectrum of control resources needs to be divided into blocks along the time axis. Continuous subsequences are extracted with a fixed step size, each subsequence reflecting the trend of control state changes within a time window. Blocking avoids feature dilution caused by long data sequences. Next, numerical normalization is performed on the multi-dimensional feature channels within each subsequence, compressing them to a uniform scale range, thereby eliminating the interference of feature amplitude differences on the perception ability of the convolutional kernel. Subsequently, these preprocessed spectrum sub-blocks are sequentially input into the input of a convolutional attention residual network. The network uses multiple sets of convolutional kernels to perform cross-convolution operations along the time axis and the control dimension axis. The time axis convolution is used to capture the evolution trend of state changes, and the control dimension axis convolution is used to model the coupling patterns between multiple resources. Cross-convolution merges the two to form a spatiotemporal joint feature. This design enables the network to simultaneously perceive the resource release rate and the intensity of inter-dimensional interference, significantly improving the ability to identify implicit occupancy patterns in the spectrum.

[0060] Temporal segmentation refers to dividing continuously acquired spectral data into sub-windows of uniform length according to time sequence, so that each sub-window represents an independent task switching process interval, which is convenient for short-term pattern analysis. Feature normalization mapping of data in each dimension to the same numerical range according to the maximum and minimum values ​​avoids the dominance of high-amplitude channels in convolution calculations. The convolutional attention residual network consists of three parts: the convolutional layer is responsible for extracting local change features on both the time and control dimensions; the attention allocation structure dynamically allocates computational resources based on the response intensity and change density of each channel, so that the network focuses on important change regions; the residual connection structure superimposes the original input and the convolution output, preserving input information to avoid feature degradation. The fixed time window setting ensures that each round of the network processes a spectral sequence of uniform length, which is convenient for model stable learning. Cross-convolution operation simultaneously processes data changes in the time and control dimensions. For example, when the interface level and the cache refresh signal fluctuate at the same time, cross-convolution can identify them as joint features of potential conflict patterns, which helps in subsequent state recognition.

[0061] The multi-channel feature maps obtained by cross-convolution are input into the attention weight allocation layer. The weight distribution is dynamically adjusted based on the coupling degree between the channel response intensity and the control dimension. Then, the input signal and weighted features are superimposed through the residual structure to generate a high-dimensional feature matrix with multi-layer feature fusion capability.

[0062] The multi-channel feature maps output by cross-convolution represent the dynamic resource change patterns exhibited by different control dimensions in the time series. To further enhance feature discrimination capabilities, these multi-channel feature maps are input into an attention weight allocation layer. In this structure, the response intensity of each feature channel within the current time window is first calculated, such as the amplitude change of the cache refresh channel or the fluctuation frequency of the interface stability channel. Then, considering the coupling degree between different control dimensions, such as whether there is a synchronous trend between cache changes and thread blocking, a channel-level dynamic weighting operation is achieved by multiplying by the attention factor matrix. This strengthens key features with high coupling and high dynamic response while suppressing the influence of noise features. Subsequently, the initial input signal and the weighted feature maps are fused element-wise through a residual connection structure to obtain a high-dimensional feature matrix with joint perception capabilities between upper and lower layers. This matrix can simultaneously retain the original resource change contour and key feature enhancement information, thus providing rich semantic support for subsequent deep variable extraction. For example, if interface level fluctuations and thread blocking delays highly overlap, the attention mechanism will significantly increase their response weights, making the combination form a clear feature ridge in the high-dimensional feature matrix.

[0063] Channel response intensity refers to the degree of drastic change in a control dimension within a specific time window. It is typically quantified by the mean square value or maximum amplitude of the statistical convolutional feature map, used to determine the contribution of the current channel to the task switching state. Control dimension coupling degree characterizes whether there is a synchronous fluctuation relationship between two or more control resources during dynamic processes, and can be calculated based on the covariance matrix or phase consistency score between channels. The attention weight allocation layer assigns channel weights based on these two indicators, allowing the network to focus on control dimension combinations with actual occupancy risk. The residual structure superposition operation, by adding the initial spectral input element-wise to the weighted feature map, addresses the information loss problem that may occur during deep feature processing, while also making the model more robust to handling highly uncertain temporal features. The final output high-dimensional feature matrix not only possesses the collaborative features between control resources laterally but also has a multi-layered abstraction capability of longitudinal temporal change trends, providing a structured semantic foundation for deep identification of control resource occupancy states.

[0064] Deep variables representing the relationship between the coupling degree of control signals, resource retention strength and thread response latency are extracted from the high-dimensional feature matrix. Based on the extracted deep variables, a set of control resource occupancy parameters is constructed. The set of control resource occupancy parameters is used to express the actual occupancy status of control resources inside the device and is used for subsequent task scheduling judgment.

[0065] Extracting deep variables representing the relationship between control signal coupling, resource retention strength, and thread response latency from a high-dimensional feature matrix, and constructing a set of real-time control resource occupancy parameters based on these variables, transforms the originally scattered, complex, and multi-dimensional control resource state information into a structured set of parameters with clear physical meaning, quantifiability, and input capability. This process, through the extraction of deep variables, achieves abstract modeling of the dynamic behavior of internal control resources, making the unobservable internal occupancy state externalized in a processable data form, thus providing accurate basis for subsequent scheduling decisions. By constructing the set of real-time control resource occupancy parameters, not only can the current usage pressure and release process of control resources be comprehensively reflected, but the task scheduling model can also be assisted in identifying abnormal resource states or critical scheduling risks, improving the accuracy and stability of collaborative control. Therefore, this operation is the foundation for realizing intelligent perception and dynamic control of task scheduling, possessing key data bridging functions and algorithmic support value.

[0066] In this embodiment, deep variables characterizing the relationship between control signal coupling degree, resource retention strength, and thread response latency are extracted from the high-dimensional feature matrix, and a set of control resource occupancy parameters is constructed based on the extracted deep variables, specifically:

[0067] In the high-dimensional feature matrix, channels are divided based on the response amplitude and time change rate of the feature channels, into control signal coupling channel region, resource holding signal channel region and thread response delay channel region, respectively. Feature point masks are marked in each channel region to extract the target response distribution features within the region.

[0068] The process of channel segmentation in the high-dimensional feature matrix is ​​based on the joint analysis of two dimensions: the response amplitude and the rate of change over time for each feature channel within a continuous time window. First, the response curve for each feature channel over a continuous time period is extracted, and the average response intensity and rate of change for that channel are calculated. The response intensity measures the absolute magnitude of resource changes, while the rate of change identifies the characteristics of sudden or persistent changes. Then, all channels are divided into three regions based on their distribution along these two dimensions: the control signal coupling channel region, which refers to channels with high response amplitudes and strong synchronous change rates, representing the linkage between multiple signals in the control path; the resource holding signal channel region, which refers to channels with high response amplitudes but slow changes, characterizing the long-term occupation of resources such as caches and communication links; and the thread response delay channel region, which corresponds to channels with low response amplitudes but large change rates and intermittent sudden fluctuations, reflecting task thread interruption or blocking behavior. After channel segmentation, feature point masks are calibrated within each region based on local change trends. These masks are applied to the feature map of the corresponding channel to extract representative local response feature points, preserving the change boundaries and peak characteristics of the actual occupancy state. For example, in the resource holding signal channel region, the mask focuses on the slowly changing segments that remain consistently high, while in the thread response delay channel region, the mask calibrates for sudden rising edges. This method achieves accurate region feature extraction, which not only improves the discriminative ability of subsequent deep variable extraction but also avoids interference and dilution of channel response information during complex temporal evolution.

[0069] In the control signal coupling channel region, the channel response density is extracted, the signal attenuation ratio is calculated in the resource retention signal channel region, and the variation range of the blocking duration is statistically analyzed in the thread response delay channel region. In these ways, depth variables corresponding to the relationship between the control signal coupling degree, resource retention strength and thread response delay are generated.

[0070] In the control signal coupling channel region, channel response density is extracted by statistically analyzing the ratio of the sum of response intensities at all calibrated feature point mask coverage locations in this region to the time span. This reflects the tightness of coupling between control signals per unit time. Higher coupling indicates that multiple signal channels are highly synchronized in time and have strong control correlation. In the resource holding signal channel region, signal attenuation ratio is calculated by measuring the proportion of the time it takes for the signal to go from peak to stable low to the total transmission time. If this proportion is small and the tail plateau is persistent, it indicates slow resource release and that resources may be persistently occupied. In the thread response delay channel region, the variation range of blocking duration is statistically analyzed by identifying the start and end times of multiple consecutive thread blocking events and calculating the difference between their minimum and maximum durations. This reflects the latency fluctuations exhibited by threads during different task switching. The extraction methods of these three deep variables are all based on time series analysis of specific response behaviors, and are independently modeled within their respective regions to ensure semantic decoupling and clear expression between variables. For example, if the control signal coupling density increases significantly while the resource retention ratio decreases simultaneously, it may indicate that the control path has switched but the cache has not yet synchronized; conversely, if the thread blocking time fluctuates drastically, it suggests potential scheduling instability or abnormal lock resource release. By extracting these three sets of deep variables, a multi-dimensional model of the actual occupancy status of the device's internal control resources is achieved, providing an accurate basis for subsequent scheduling decisions.

[0071] The deep variables relating control signal coupling degree, resource retention strength and thread response latency are numerically normalized to generate a set of deep variables at the same data scale. A control resource occupancy real-time parameter set is then constructed in a matrix structure. This set of parameters expresses the actual occupancy status of control resources within the device and serves as the input required for constructing the task scheduling decision tensor.

[0072] Normalizing the deep variables related to the coupling degree of control signals, resource retention strength, and thread response latency is crucial to eliminating differences in physical dimensions, value ranges, and magnitudes of variation among the three types of variables, thus unifying their weighting scale in the computational model. The normalization process can employ min-max scaling or Z-score standardization. By calculating the maximum, minimum, mean, and standard deviation of each variable in historical samples, a linear transformation is applied to the current value, ensuring that all variable values ​​fall within the same distribution range. This prevents a single variable's excessively large or small value from dominating the model's judgment. After normalization, the three normalized deep variables are stacked column-wise to construct a fixed-dimensional matrix structure, making it input-friendly and resolvable, facilitating the construction of multi-dimensional input tensors for subsequent scheduling evaluation. The constructed control resource occupancy real-time parameter set not only preserves the dynamic characteristics of the original variables but also enhances the comparability between variables of different dimensions through a unified scale. For example, when a device is in a state of high coupling, high persistence, and long latency, the values ​​in the normalized parameter set will all approach the high value range. The corresponding scheduling model can then determine that the current resource pressure is high and the device is unschedulable, thereby guiding the system to perform scheduling suppression. Therefore, this parameter set constitutes the core data representation carrier for expressing the true occupancy status of the device's internal control resources, providing accurate input for the construction of the task scheduling decision tensor.

[0073] The scheduling decision tensor construction module constructs a task scheduling decision tensor based on the actual parameter set of control resource occupancy, combined with task type, historical execution time and scheduling success rate, and calculates the blocking intention coefficient.

[0074] In this embodiment, in the scheduling decision tensor construction module, the control resource occupancy real-time parameter group is expanded according to the control dimension and time series, a task association vector is established in combination with the task type, and time domain alignment is completed based on historical execution time to construct a data matrix that reflects the impact of control state on task scheduling efficiency. The data matrix is ​​fused with the scheduling success rate information obtained during task scheduling, response weights are set according to task type, and tensor construction is performed based on parameter coupling relationship to generate a task scheduling decision tensor with task type, control state parameters, and execution feedback information as dimensions. According to the weight change trend between the control state parameter dimension and the task type dimension in the task scheduling decision tensor, the gradient change region is extracted, and the blocking intention coefficient used for scheduling control is calculated in combination with the control disturbance amplitude and the task sensitivity change rate.

[0075] To model the correlation between control resource status and task scheduling, the control resource occupancy parameter set must first be expanded into a two-dimensional sequence array consisting of a control dimension and a time dimension. Each row represents the dynamic change process of a certain control resource, and each column represents the status value at different time points. This structure can be constructed by traversing the occupancy status values ​​of each resource dimension within a time window. Subsequently, based on the type of the current scheduled task, the set of task attribute tags corresponding to that type is extracted from the task type mapping library and encoded into a task association vector. This vector is used to embed the data matrix to express the task scheduling background. Then, combined with the time consumption data recorded during historical task execution, the expanded control status array is aligned with the task time consumption window along the time axis to ensure that the control resource status and task progress status at the same time point can be synchronously compared. Taking a functional test task as an example, if the cache dimension resource occupancy is significantly higher than the average and the duration is prolonged in the early stage of the task, while the scheduling response is delayed, a synchronous anomaly region will also be formed in the matrix, providing characteristic basis for subsequent scheduling judgments.

[0076] The control resource occupancy real-time parameter set is a dynamic set of states composed of multi-dimensional control variables. Expanding it into control dimensions and time series is to form a multi-channel temporal representation structure, allowing subsequent analysis to assess the scheduling impact from different resource dimensions. Control dimensions include, for example, signal path connectivity, thread response latency, and cache refresh efficiency; the time series expansion is used to quantify the state evolution trend of each control dimension during scheduling. The task association vector is a structured expression of the semantic information of the current test task, reflecting its scheduling sensitivity, resource dependence, and anomaly tolerance. After being fused with the state matrix, it provides semantic guidance information. The temporal alignment of historical execution time is used to map the intersection nodes of task progress and control state, a key means of identifying scheduling bottlenecks and resource contention locations. The final constructed data matrix is ​​an expression that integrates the temporal domain, control domain, and task semantics, specifically used to quantify the impact strength of different control states on scheduling efficiency, providing a data foundation for quantifying blocking intentions.

[0077] To achieve joint modeling of control state, task characteristics, and scheduling feedback, the constructed data matrix is ​​first numerically fused with task scheduling success rate data recorded during historical scheduling processes. This fusion process uses a task type matching mechanism to complete index binding, ensuring that each control state sample corresponds to the scheduling success rate index of the same type of task in the previous iteration. Subsequently, response weights are set based on the real-time task sensitivity level of the task type, assigning higher-priority tasks with higher influence factors, thereby achieving task-driven feature guidance in tensor construction. Next, a parameter coupling modeling mechanism is introduced into the fused dataset. By calculating the interaction strength between control parameters and the correlation strength between scheduling feedback and control state, a three-dimensional data representation is constructed, with the three axes of the tensor representing task type, control state parameters, and execution feedback information, respectively. For example, in multiple tests, it was found that the success rate of high-speed communication tasks continuously decreased under unstable interface levels. The coupling value of the corresponding channel in the tensor will be dynamically increased to enhance the accuracy of scheduling judgments.

[0078] The data matrix is ​​a two-dimensional set of states built upon control dimensions and time series. Its fused object scheduling success rate information originates from historical task execution records, representing the performance of specific tasks under different control states. Task types here are not only classification indexes but also influence the proportion of each task in parameter calculations within the tensor structure through response weights, ensuring enhanced risk control capabilities for critical tasks during scheduling decisions. Response weights are not fixed values ​​but dynamically adjusted based on the current task's sensitivity to resource consumption, scheduling fault tolerance, and equipment load status. Parameter coupling relationships characterize whether there are significant interaction features between control state parameters. For example, signal path discontinuity and cache refresh lag may exhibit positive coupling, appearing as high correlation values ​​in the scheduling decision tensor. The final generated task scheduling decision tensor possesses context-aware capabilities, providing a high-dimensional collaborative expression of task categories, control states, and historical feedback in scheduling decisions, forming the core foundation supporting subsequent blocking intention calculations.

[0079] In this embodiment, based on the weight change trend between the control state parameter dimension and the task type dimension in the task scheduling determination tensor, the gradient change region is extracted, and combined with the control disturbance amplitude and the task sensitivity change rate, the blocking intention coefficient for scheduling control is calculated, specifically:

[0080] In the task scheduling decision tensor, the weight distribution intervals of the control state parameter dimension and the task type dimension are selected. The weight change curve is expanded over time to calculate the gradient change rate within the continuous window. The sensitive interval of weight change is formed by filtering through the gradient threshold, which is used to identify the key nodes of sudden changes in resource state during task scheduling.

[0081] In the task scheduling decision tensor, the weight distribution intervals of the control state parameter dimension and the task type dimension are selected to locate the response characteristics of the task scheduling process to changes in control state. Specifically, the weight value sequence between task type and each control state parameter is first extracted through index mapping. Then, the weight change curve over time is unfolded into an ordered time series. A sliding time window is used to perform local analysis on this series, calculating the gradient rate of change of the weight within each window, i.e., the average rate of weight change between adjacent time points. This process can be implemented using first-order difference operations. The result is then compared with a preset gradient change threshold to filter out time intervals with sudden increases in the rate of change, identifying resource state change points sensitive to the scheduling process. For example, in a task switch, if the weight of the cache refresh delay parameter rises rapidly within a short period, the system marks this interval as a sensitive area to prompt adjustments to the scheduling strategy. These sensitive intervals represent the response boundaries of the scheduling system to resource changes and are key input variables for judging scheduling risk in subsequent blocking intention coefficient calculations. Technically, the weight change curve reflects the dynamic fluctuation of the task's dependence on control resources. Extracting its gradient change through time series analysis helps to accurately model the scheduling behavior's ability to perceive sudden changes in resource state.

[0082] Extract control disturbance amplitude data within the weight change sensitive interval, calculate the change difference of each control dimension in adjacent time periods, and generate a disturbance sensitivity matrix by combining the sensitivity change rate of task type. The disturbance sensitivity matrix is ​​used to characterize the relationship between the fluctuation intensity of control resource changes and time offset during task execution.

[0083] Extracting control disturbance amplitude data within the weight change sensitive interval is crucial for capturing the impact of sudden changes in control resource state on task execution stability during scheduling decisions. The implementation first involves perturbation detection of the time-series data of control state parameters within the previously identified weight change sensitive interval. This is achieved by calculating the numerical difference between two consecutive time points in the control dimension, forming a change difference sequence. Then, for each task type, the sensitivity change rate is quantified based on its historical response to control resource changes. This sensitivity change rate can be obtained through statistical modeling of the correlation between control parameter changes and task result offsets. The change difference and sensitivity change rate are then fused dimension-by-dimensionally to construct a disturbance sensitivity matrix. Each cell in the matrix represents the coupling strength between the fluctuation intensity of a specific control resource under a specific task and the task adaptability. For example, if interface level stability fluctuates significantly within a short period and a certain type of real-time task has historically shown extremely high execution sensitivity to this resource, the corresponding position in the disturbance sensitivity matrix is ​​assigned a higher numerical weight, characterizing it as a scheduling risk point due to resource mutation. In this way, the perturbation sensitivity matrix effectively correlates the intensity of resource state fluctuations with the tolerance boundary of task types to resource changes, forming one of the core bases for scheduling security assessment.

[0084] The perturbation sensitivity matrix is ​​input into the blocking calculation model, and the blocking intention coefficient is generated based on the coupling slope between the control perturbation amplitude and the rate of change of task sensitivity. The blocking intention coefficient is used to characterize whether the dynamic scheduling suppression condition needs to be triggered during the current task scheduling process.

[0085] The purpose of inputting the perturbation sensitivity matrix into the blocking computation model is to quantify the risk of control resource fluctuations interfering with task scheduling behavior and determine whether the scheduling process needs to be interrupted. Specifically, the numerical value corresponding to the task type for each control dimension in the perturbation sensitivity matrix is ​​first used as an input vector. A pre-trained blocking computation model is then used to perform a nonlinear fitting operation to establish the coupling relationship between the magnitude of control perturbation and the rate of change in task sensitivity. This model learns its risk distribution curve by analyzing the failure or delay behavior caused by the intensity of control resource fluctuations in historical scheduling processes, and uses the slope change as a representation of the perturbation intensity. When the coupling slope shows a steep upward trend within a specific interval, it means that a sudden change in control resources has a significant perturbation effect on the current task. In this case, the blocking computation model will output a high blocking intention coefficient. For example, when the cache refresh rate drops sharply within a unit of time, and the executing data synchronization task has historically shown extremely high sensitivity, the coupling slope calculated by the model will significantly deviate from the normal distribution range, thus increasing the blocking intention coefficient value. This blocking intention coefficient ultimately serves as a reference for scheduling interruption decisions, triggering dynamic scheduling suppression conditions during task switching to prevent control conflicts caused by scheduling advancement without resource release. This process ensures that the scheduling behavior has the ability to immediately detect sudden anomalies in resource status and has an adaptive suppression mechanism.

[0086] The multi-dimensional scheduling inhibition execution module constructs a scheduling inhibition structure based on the calculated blocking intention coefficient. It divides the blocking intention coefficient into prohibited scheduling segments, lazy scheduling segments, and allowed scheduling segments, and executes three types of scheduling inhibition behaviors: concurrent instruction cancellation, delay lock setting, and task advancement, respectively.

[0087] In this embodiment, the implementation of the multi-dimensional scheduling suppression execution module is as follows:

[0088] Based on the calculated blocking intention coefficient, a scheduling inhibition structure is constructed. By analyzing the amplitude fluctuation of the blocking intention coefficient in the process of controlling changes in resource occupation, a mapping model is established. The blocking intention coefficient is divided into multiple threshold intervals according to risk level in the numerical space, forming a scheduling inhibition structure that includes prohibited scheduling segments, lazy scheduling segments, and allowed scheduling segments.

[0089] In the design of scheduling suppression strategies, the blocking intention coefficient, as a core indicator for quantifying the coupling relationship between the current control resource occupancy status of equipment and scheduling risk, exhibits highly dynamic sensitivity in its numerical changes. To construct an implementable scheduling suppression structure, it is necessary to first continuously monitor and model the blocking intention coefficient during the execution of test tasks, calculate its amplitude fluctuation curve using the sliding window technique, and set multi-level mapping rules based on the risk evolution model. In specific implementation, cluster analysis can be performed based on historical fault data to extract the statistical characteristic distribution of the blocking intention coefficient under high-risk conditions, define risk level hierarchical boundaries, divide multiple risk level intervals in the numerical space, and map and bind each interval to a specific scheduling behavior, thereby forming a structured scheduling suppression structure to guide the selection of subsequent scheduling behaviors.

[0090] The amplitude fluctuation of the blocking intention coefficient reflects the changing trend of tension between the delay in releasing control resources and changes in task sensitivity. The mapping model is used to discretize the continuously changing blocking intention coefficient into executable state segments, which facilitates rapid judgment by the scheduling execution unit. The threshold interval is generated by logically mapping the risk level and combining it with a data-driven approach. It does not rely on fixed thresholds but is dynamically adjusted based on the current system operating status. The final constructed scheduling suppression structure includes three clearly defined scheduling decision regions: prohibited scheduling segment, lazy scheduling segment, and allowed scheduling segment. The control strategy corresponding to each region is implemented based on the specific quantification degree of the blocking intention, thereby ensuring the sensitivity and security of the scheduling response.

[0091] The blocking intention coefficient is mapped to the corresponding segment in the scheduling suppression structure. If the blocking intention coefficient is located in the prohibited scheduling segment, the concurrent instruction cancellation behavior is triggered. The instruction priority filtering and mutual exclusion conflict detection method is used to clear all pending concurrent control instructions in the scheduling instruction queue to prevent task conflict execution when the resource state has not been switched.

[0092] During task scheduling, when the blocking intention coefficient is mapped to the prohibited scheduling segment in the scheduling suppression structure, concurrent instruction cancellation behavior must be executed immediately to prevent tasks from running interleaved before the control resources have been switched over. Specifically, all instructions to be executed in the scheduling instruction queue are first marked for concurrency identification. Cross-validation is performed using the timing identifier, functional domain label, and call path field in the instruction metadata to extract the set of concurrent instructions. Then, a priority queue-based filtering mechanism is introduced, arranging the marked instructions in descending priority order and combining mutual exclusion conflict detection rules to determine their compatibility with the current resource occupancy state. Upon discovering instruction items with resource conflicts or logical overlap, an immediate removal operation is performed, and the retained instructions are set to a delayed suspension state, to be released only when subsequent scheduling conditions are met. This strategy achieves early removal of potential instruction interference items in areas with extremely high scheduling risk, ensuring that the device is not triggered by erroneous instructions and preventing state pollution between test tasks from the root cause.

[0093] The mapping segment of the blocking intention coefficient is used to trigger behavioral branches. The prohibition scheduling segment is set based on the judgment that the blocking intention value is in the high-risk range, which usually corresponds to the maximum latency during resource switching. The instruction priority screening mechanism relies on a multi-level weight model built from the statistical learning results of instruction historical execution feedback to quickly identify the urgency and risk level of instructions. The mutual exclusion conflict detection adopts a graph structure modeling method based on control variable mapping, and performs graph traversal analysis on the scope and dependency path of all control instructions to determine whether there is an intersection of non-parallel resource occupation. The instruction repeal behavior is essentially an active circuit breaker control of the scheduling link, which aims to avoid unpredictable resource states from causing device logic errors or data overwriting, thereby improving the determinism and safety boundary of the overall scheduling.

[0094] If the blocking intention coefficient is in the lazy scheduling segment, a delay lock setting behavior is executed, and a time offset is inserted into the task scheduling logic to postpone the scheduling action; if the blocking intention coefficient is in the allowed scheduling segment, a task advancement behavior is executed, and the scheduling instructions for subsequent tasks are executed immediately after the scheduling conditions are confirmed to be met through feedback.

[0095] When the blocking intention coefficient is mapped to the lazy scheduling segment, the scheduling process neither immediately stops nor directly proceeds. Instead, it enters a waiting strategy to avoid the risk of minor disturbances during resource switching. Specifically, a dynamic time offset is injected into the task scheduling logic. This time offset is calculated based on the precise value of the blocking intention coefficient within the lazy scheduling segment and the average resource release delay, forming an adaptive delay locking parameter. During execution, upon receiving a scheduling request, the scheduler does not immediately trigger resource calls but activates the locking logic, freezing the current task process for a set waiting period to reserve sufficient buffer time to ensure the stable completion of the resource release process. After the delay period ends, the system re-confirms the resource status and decides whether to proceed with task scheduling based on the feedback result. When the blocking intention coefficient is in the allowed scheduling segment, no locking or blocking behavior is triggered; the scheduling process directly proceeds. After confirming that the key conditions are met through the task execution feedback mechanism, the control command for the corresponding test task is immediately initiated, ensuring the efficient and continuous testing process.

[0096] The lazy scheduling segment serves as a buffer period to accommodate fluctuations during resource release. Its numerical range is typically set based on the standard deviation estimate of the resource release time distribution. The time offset, a core parameter of the lazy strategy, quantifies the dynamic characteristics of resource release and obtains the minimum safe scheduling delay by fitting it to historical resource occupancy and release curves. The delay lock setting behavior uses a logical freezing mechanism to dampen scheduling at a micro-timescale, preventing premature resource switching due to excessively rapid scheduling. Task advancement behavior is based on the safety judgment of the allowed scheduling segment, combined with feedback confirmation signals such as I / O interface idle status, thread blocking flag clearing, and cached data consistency verification results, forming a closed-loop scheduling process for task execution. This ensures efficient execution of scheduling behavior after the status is confirmed to be correct. This strategy achieves fine-grained division and dynamic response to scheduling risk intervals.

[0097] After completing the scheduling suppression behavior, the self-evolving closed-loop control module constructs a state convergence matrix based on the scheduling feedback signal and task results, generates a self-evolving tensor mapping by combining the actual control resource occupancy parameter set, and updates the actual control resource occupancy parameter set and the task scheduling decision tensor.

[0098] In this embodiment, the self-evolving closed-loop control module is implemented as follows:

[0099] After completing the scheduling suppression behavior, the system collects the control path on / off state recovery time, cache refresh completion flag, signal acknowledgment delay fall-off value, and interface level stabilization time based on the scheduling feedback signal. Combined with the task completion flag, output data consistency flag, and resource release delay value recorded in the task results, a state convergence matrix is ​​constructed to describe the dynamic regression characteristics of control resources after the task is completed.

[0100] After completing the scheduling suppression behavior, the recovery time of the control path on / off state is collected by acquiring the scheduling feedback signal and extracting the recovery time. This can be achieved by detecting the complete time period during which the control signal channel switches from the active state to the neutral state. The cache refresh completion flag can be obtained by reading the flag bit in the memory controller used to identify the cache write-clear state. The signal acknowledgment delay fallback value is obtained by comparing the signal acknowledgment response time in the task completion stage with the average value in the normal stable state. The interface level stabilization time is determined by the length of the time window required to maintain stability within the set error range based on continuous level sampling data. In the task results, the task completion flag is determined by the termination flag provided by the task execution engine, the output data consistency flag is determined by comparing the output data hash value with the preset result hash value to determine its data validity, and the resource release delay value can be calculated by comparing the time interval between the task end signal and each resource release confirmation signal. By organizing the above data in the control and time dimensions, a state convergence matrix can be constructed to describe the dynamic regression process of control resources after the task is completed, providing an accurate basis for subsequent state evolution analysis.

[0101] The recovery time of the control path connectivity status reflects the physical completion time of resource switching; the cache refresh completion flag is used to confirm the degree of data isolation; the signal acknowledgment delay fallback value is used to determine whether the communication layer has recovered to the initial low-load state; and the interface level stabilization time characterizes the stability of the hardware interface electrical characteristics after load switching. The task completion flag is used to confirm whether the current scheduling process has been successfully exited; the output data consistency flag detects whether the calculation results have deviated due to resource anomalies; and the resource release delay value reflects the timeliness of system resource reclamation. By modeling these key features in a unified matrix structure, it is possible not only to capture the state changes of each dimension after task execution but also to identify whether there is residual interference in the control resources under specific task conditions, providing complete feedback information for the subsequent construction of a dynamic control model.

[0102] The state convergence matrix and the actual parameter set of control resource occupancy are bidirectionally coupled in the control and time dimensions. By constructing a multi-head attention mapping structure, the response trend between the history of control resources and feedback is extracted, and a self-evolving tensor mapping containing the change trend of control resources and the evolution state is generated.

[0103] Bidirectional coupling of the state convergence matrix and the real-time parameter set of control resource occupancy in both the control and time dimensions can be achieved through matrix alignment and time window synchronization techniques. In the control dimension, indicators of the same dimension in both sets of data are mapped one-to-one to ensure consistency in the coupling relationship. In the time dimension, a sliding time window is used to align the timestamps of historical state data and feedback information, linking resource usage status and feedback behavior at the same time point. Based on this aligned data structure, a multi-head attention mapping structure is constructed, focusing each attention head on different control dimensions and time segments to learn the evolution trajectory of control resources and feedback fluctuation trends. For example, in a certain control dimension, the multi-head attention mechanism can simultaneously capture short-term release anomalies and long-term response delays, and generate response features that fuse different change patterns through weight distribution. The resulting multi-dimensional feature vectors are stacked and linearly mapped to form a tensor structure, thus constructing a self-evolving tensor mapping that expresses the evolutionary relationship of control resources.

[0104] The state convergence matrix reflects the regression performance of control resources after task completion, while the control resource occupancy real-time parameter set expresses the historical state of resource occupancy before task execution. The bidirectional coupling of these two in the control and time dimensions enables the model to simultaneously perceive resource changes before and after scheduling. The multi-head attention mapping structure, by concurrently focusing on local anomalies and global trends in multiple control signal dimensions, extracts key patterns of resource release lag, response fluctuations, and coupling strength evolution, enhancing dynamic prediction capabilities. The self-evolving tensor mapping, by encoding time-series characteristics and resource coupling features, forms a multi-dimensional data structure containing the logical relationships between control resource change trends, release delays, and occupancy feedback, providing intelligent support for subsequent parameter updates and scheduling strategy adjustments.

[0105] Based on the change magnitude and time offset gradient of the control dimension evolution path in the self-evolving tensor mapping, the corresponding parameters in the control resource occupancy real-time parameter set are updated. The updated control resource occupancy real-time parameter set is then used as input to jointly reconstruct the task scheduling decision tensor with the original scheduling strategy, so as to realize dynamic iterative control with scheduling state awareness.

[0106] During dynamic regulation, the self-evolving tensor mapping provides information on the evolutionary trend of control resources in time and the control dimension, which can be used to determine key features such as resource release delay, changes in coupling strength, and response offset. To update the actual parameter set of control resource occupancy, the magnitude of change in the evolution path of the control dimension is first extracted from the tensor. The direction and intensity of parameter changes are identified through difference calculation, and then combined with the offset gradient in the time dimension, the speed of resource change and lag characteristics on the time axis are determined. For parameters with drastic changes or significant response delays, exponential smoothing or weighted update mechanisms are used to dynamically correct the corresponding indicators in the actual parameter set, thereby improving its ability to fit the real resource state. Subsequently, the updated actual parameter set of control resource occupancy is fused with the static scheduling rules in the predetermined task scheduling strategy to construct a task scheduling decision tensor with feedback perception capabilities, enabling the scheduling model to perceive changes in resource state and dynamically adjust the strategy.

[0107] The magnitude of changes in the control dimension evolution path is used to characterize the structural transformation of control resources during the scheduling process, such as the transition from a highly coupled state to a resource release state. The time offset gradient reflects the delay in control resource response and has a direct impact on scheduling timing. Adjusting the actual control resource occupancy parameter set using these two indicators can improve the matching degree of the parameter set to the current system operating state. The updated parameter set not only includes the state characteristics of historical task execution but also incorporates feedback information after task completion. The task scheduling decision tensor generated after fusing the original scheduling strategy covers multiple layers of information in terms of dimensional composition, including control resources, task type, and feedback response. It has the ability to predict and adaptively adjust future scheduling results, realizing dynamic adaptive control driven by scheduling state awareness.

[0108] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0109] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0110] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0111] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0112] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0113] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0114] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An intelligent test equipment collaborative control system for vehicle domain control, characterized in that, The control resource dynamic spectrum judgment module, the occupation feature analysis module, the scheduling judgment tensor construction module, the multi-dimensional scheduling suppression execution module, and the self-evolution closed-loop regulation module are included. The control resource dynamic spectrum judgment module collects the control path on-off state, the cache refresh degree, the signal response delay, the interface level stability and the thread blocking time during the test task switching process, constructs a control resource dynamic occupation spectrum diagram, and determines whether the device is in a control resource unreleased state according to the feature intersection in the control resource dynamic occupation spectrum diagram. The occupation feature analysis module inputs the control resource dynamic occupation spectrum diagram into a convolution attention residual network to extract deep variables and generate a control resource occupation live parameter group for expressing the real occupation state of the internal control resource of the device when the device is in a control resource unreleased state. The scheduling judgment tensor construction module constructs a task scheduling judgment tensor based on the control resource occupation live parameter group, combines the task type, the historical execution time consumption and the scheduling success rate, calculates the blocking intention coefficient, and constructs a scheduling suppression structure. The multi-dimensional scheduling suppression execution module divides the blocking intention coefficient into a prohibited scheduling segment, an inert scheduling segment and an allowed scheduling segment based on the calculated blocking intention coefficient, and respectively executes three types of scheduling suppression behaviors, namely concurrent instruction abortion, delay lock setting and task promotion. The self-evolution closed-loop regulation module constructs a state convergence matrix according to the scheduling feedback signal and the task result after completing the scheduling suppression behavior, generates a self-evolution tensor mapping combined with the control resource occupation live parameter group, and updates the control resource occupation live parameter group and the task scheduling judgment tensor. 2.The intelligent test equipment cooperative control system of a vehicle-mounted domain controller according to claim 1, characterized in that, The implementation of the control resource dynamic spectrum judgment module is as follows: During the test task switching process, the on-off data of the control path is collected by monitoring the continuous conduction level change of the control instruction signal; the refresh ratio is collected by detecting the cache write rate and data emptying rate; the delay sequence is collected by recording the time difference between signal sending and response; the interface level waveform is formed by sampling the interface voltage fluctuation amplitude; and the blocking interval is calculated by recording the blocking start and end time of the task thread. The control path on-off data, the cache refresh ratio, the signal response delay sequence, the interface level waveform and the thread blocking interval are synchronized and aligned in time sequence to generate a multi-dimensional time sequence matrix, and a control resource dynamic occupation spectrum diagram is constructed through frequency domain decomposition and amplitude normalization, wherein the horizontal axis represents the task switching time length and the vertical axis represents the coupling strength of the control dimension. The feature line intersection points in the control resource dynamic occupation spectrum diagram are extracted, the density gradient and the offset angle of the intersection region are calculated, and when the density gradient continuously changes by more than a preset threshold and the offset angle displacement direction is inconsistent with the historical release trend, it is determined that the device is in a control resource unreleased state, and the determination result is output for subsequent occupation feature analysis processing.

3. The intelligent test equipment collaborative control system for vehicle domain control according to claim 1, characterized in that, The implementation of the occupation feature analysis module is as follows: Under the condition that the device is in the control resource unreleased state, the control resource dynamic occupation spectrum is time-sequentially blocked and normalized, and then input into the input end of the convolution attention residual network, so as to sequentially transfer the spectrum data in a fixed time window, cross-convolve on the time axis and the control dimension axis through multiple groups of convolution kernels, and extract the spatial feature correlation between the control resource distribution and the time variation in the spectrum; The multi-channel feature map obtained by cross-convolution is input into the attention weight distribution layer, the weight distribution is dynamically adjusted based on the channel response strength and the control dimension coupling degree, and then the input signal and the weighted feature are superimposed through the residual structure to generate a high-dimensional feature matrix with multi-layer feature fusion capability; Deep variables representing the coupling degree of control signals, resource retention strength and thread response delay relationship are extracted from the high-dimensional feature matrix, and a control resource occupation live parameter group is constructed according to the extracted deep variables. The control resource occupation live parameter group is used to express the real occupation state of the internal control resource of the device and is used for subsequent task scheduling decision.

4. The intelligent test equipment cooperative control system of a vehicle-mounted domain control according to claim 3, characterized in that, Deep variables representing the coupling degree of control signals, resource retention strength and thread response delay relationship are extracted from the high-dimensional feature matrix, and a control resource occupation live parameter group is constructed according to the extracted deep variables. Specifically: Based on the response amplitude and time variation rate of the feature channel, the channel is divided into a control signal coupling channel region, a resource retention signal channel region and a thread response delay channel region in the high-dimensional feature matrix, and a feature point mask is labeled in each channel region for extracting the target response distribution feature in the region; The channel response density is extracted in the control signal coupling channel region, the signal attenuation ratio is calculated in the resource retention signal channel region, and the change interval of the blocking duration range is counted in the thread response delay channel region, to generate deep variables corresponding to the coupling degree of control signals, the resource retention strength and the thread response delay relationship respectively; The deep variables of the coupling degree of control signals, the resource retention strength and the thread response delay relationship are respectively normalized to generate a set of deep variables in the same data scale, and a control resource occupation live parameter group is constructed in a matrix structure.

5. The intelligent test equipment collaborative control system for vehicle domain control according to claim 1, characterized in that, In the scheduling decision tensor construction module, the control resource occupation live parameter group is unfolded according to the control dimension and the time sequence, a task association vector is established in combination with the task type, and a data matrix reflecting the influence of the control state on the task scheduling efficiency is constructed according to the time domain alignment of the historical execution time; the data matrix and the scheduling success rate information obtained in the task scheduling process are fused, the response weight is set according to the task type, the tensor is constructed based on the parameter coupling relationship, and a task scheduling decision tensor with task type, control state parameter and execution feedback information as dimensions is generated; According to the weight change trend between the control state parameter dimension and the task type dimension in the task scheduling decision tensor, the gradient change region is extracted, and the blocking intention coefficient for scheduling control is calculated in combination with the control disturbance amplitude and the task sensitivity change rate.

6. The intelligent test equipment collaborative control system of a vehicle-mounted domain control according to claim 5, characterized in that, According to the task scheduling judgment, the weight change trend between the control state parameter dimension and the task type dimension in the tensor is determined, the gradient change area is extracted, and the blocking intention coefficient for scheduling control is calculated by combining the control disturbance amplitude and the task sensitivity change rate. Specifically, the weight distribution interval of the control state parameter dimension and the task type dimension is selected in the task scheduling judgment tensor, the time series of the weight change curve is expanded, the gradient change rate in the continuous window is calculated, and the sensitive interval of the weight change is formed by gradient threshold screening, which is used to identify the key nodes of the resource state mutation in the task scheduling process. In the weight change sensitive interval, the control disturbance amplitude data is extracted, the change difference of each control dimension in the adjacent period is calculated, and the disturbance sensitive matrix is generated by combining the sensitivity change rate of the task type. The disturbance sensitive matrix is used to depict the fluctuation intensity and time offset relationship of the control resource change in the task execution process. The disturbance sensitive matrix is input into the blocking calculation model, the blocking intention coefficient is generated according to the coupling slope between the control disturbance amplitude and the task sensitivity change rate, and the blocking intention coefficient is used to represent whether the dynamic scheduling inhibition condition needs to be triggered in the current task scheduling process. The implementation of the multi-dimensional scheduling inhibition execution module is specifically:

7. The intelligent test equipment collaborative control system for vehicle domain control according to claim 1, characterized in that, Based on the calculated blocking intention coefficient, the scheduling inhibition structure is constructed, the amplitude fluctuation of the blocking intention coefficient in the control resource occupation change process is analyzed, the mapping model is established, multiple division threshold intervals are set in the numerical space according to the risk level, and the scheduling inhibition structure including the prohibited scheduling segment, the inert scheduling segment and the allowed scheduling segment is formed. The blocking intention coefficient is mapped to the corresponding segment in the scheduling inhibition structure. If the blocking intention coefficient is located in the prohibited scheduling segment, the concurrent instruction abortion behavior is triggered, the instruction priority screening and mutual exclusion conflict detection method is used to clear all pending concurrent control instructions in the scheduling instruction queue, so as to prevent the task conflict execution under the condition of incomplete resource state switching. If the blocking intention coefficient is located in the inert scheduling segment, the delay lock setting behavior is executed, and the time offset is inserted in the task scheduling logic to delay the scheduling action. If the blocking intention coefficient is located in the allowed scheduling segment, the task pushing behavior is executed, and the scheduling instruction of the subsequent task is executed immediately after the feedback confirms that the scheduling condition is met. The implementation of the self-evolution closed-loop regulation module is specifically: 8.The intelligent test equipment cooperative control system of vehicle-mounted domain control according to claim 1, characterized in that, After completing the scheduling inhibition behavior, the state recovery time length of the control path on-off state, the cache refresh completion flag, the signal reply delay falling value and the interface level back to stable time are collected according to the scheduling feedback signal, and the state convergence matrix is constructed by combining the task completion identifier, the output data consistency identifier and the resource release delay value recorded in the task result, which is used to describe the dynamic regression characteristics of the control resource after the task is completed. The state convergence matrix and the control resource occupation parameter group are coupled in the control dimension and the time dimension, the response trend between the control resource history and the feedback is extracted by constructing a multi-head attention mapping structure, and a self-evolution tensor mapping including the control resource change trend and the evolution state is generated. ​ According to the control dimension evolution path change amplitude and the time offset gradient in the self-evolution tensor mapping, the corresponding parameter in the control resource occupation live parameter group is updated, and the updated control resource occupation live parameter group is used as input to jointly reconstruct the task scheduling decision tensor with the original scheduling strategy, so as to realize the dynamic iterative regulation and control of the scheduling state perception.

Images